2015-02-13 14:40:36 -05:00
|
|
|
// Copyright 2009 The Go Authors. All rights reserved.
|
|
|
|
|
// Use of this source code is governed by a BSD-style
|
|
|
|
|
// license that can be found in the LICENSE file.
|
|
|
|
|
|
|
|
|
|
package gc
|
|
|
|
|
|
|
|
|
|
import (
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
"cmd/compile/internal/types"
|
2017-04-18 12:53:25 -07:00
|
|
|
"cmd/internal/objabi"
|
2016-04-06 12:01:40 -07:00
|
|
|
"cmd/internal/sys"
|
2015-02-13 14:40:36 -05:00
|
|
|
"fmt"
|
|
|
|
|
"strings"
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
// The constant is known to runtime.
|
|
|
|
|
const (
|
|
|
|
|
tmpstringbufsize = 32
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
func walk(fn *Node) {
|
|
|
|
|
Curfn = fn
|
|
|
|
|
|
|
|
|
|
if Debug['W'] != 0 {
|
2015-05-27 10:42:55 -04:00
|
|
|
s := fmt.Sprintf("\nbefore %v", Curfn.Func.Nname.Sym)
|
2016-03-04 13:16:48 -08:00
|
|
|
dumplist(s, Curfn.Nbody)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-02 17:34:42 -08:00
|
|
|
lno := lineno
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Final typecheck for any unused variables.
|
2016-02-25 10:35:19 -08:00
|
|
|
for i, ln := range fn.Func.Dcl {
|
2017-04-25 18:14:12 -07:00
|
|
|
if ln.Op == ONAME && (ln.Class() == PAUTO || ln.Class() == PAUTOHEAP) {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
ln = typecheck(ln, Erv|Easgn)
|
2016-02-25 10:35:19 -08:00
|
|
|
fn.Func.Dcl[i] = ln
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Propagate the used flag for typeswitch variables up to the NONAME in it's definition.
|
2016-02-25 10:35:19 -08:00
|
|
|
for _, ln := range fn.Func.Dcl {
|
2017-04-27 15:17:57 -07:00
|
|
|
if ln.Op == ONAME && (ln.Class() == PAUTO || ln.Class() == PAUTOHEAP) && ln.Name.Defn != nil && ln.Name.Defn.Op == OTYPESW && ln.Name.Used() {
|
|
|
|
|
ln.Name.Defn.Left.Name.SetUsed(true)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-02-25 10:35:19 -08:00
|
|
|
for _, ln := range fn.Func.Dcl {
|
2017-04-27 15:17:57 -07:00
|
|
|
if ln.Op != ONAME || (ln.Class() != PAUTO && ln.Class() != PAUTOHEAP) || ln.Sym.Name[0] == '&' || ln.Name.Used() {
|
2015-02-13 14:40:36 -05:00
|
|
|
continue
|
|
|
|
|
}
|
2016-02-25 10:35:19 -08:00
|
|
|
if defn := ln.Name.Defn; defn != nil && defn.Op == OTYPESW {
|
2017-04-27 15:17:57 -07:00
|
|
|
if defn.Left.Name.Used() {
|
2015-02-13 14:40:36 -05:00
|
|
|
continue
|
|
|
|
|
}
|
2017-03-28 11:17:52 -07:00
|
|
|
yyerrorl(defn.Left.Pos, "%v declared and not used", ln.Sym)
|
2017-04-27 15:17:57 -07:00
|
|
|
defn.Left.Name.SetUsed(true) // suppress repeats
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2017-03-28 11:17:52 -07:00
|
|
|
yyerrorl(ln.Pos, "%v declared and not used", ln.Sym)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-02 17:34:42 -08:00
|
|
|
lineno = lno
|
2015-02-13 14:40:36 -05:00
|
|
|
if nerrors != 0 {
|
|
|
|
|
return
|
|
|
|
|
}
|
2016-03-07 22:54:46 -08:00
|
|
|
walkstmtlist(Curfn.Nbody.Slice())
|
2015-02-13 14:40:36 -05:00
|
|
|
if Debug['W'] != 0 {
|
2015-05-27 10:42:55 -04:00
|
|
|
s := fmt.Sprintf("after walk %v", Curfn.Func.Nname.Sym)
|
2016-03-04 13:16:48 -08:00
|
|
|
dumplist(s, Curfn.Nbody)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-09 10:38:45 -08:00
|
|
|
zeroResults()
|
2015-02-13 14:40:36 -05:00
|
|
|
heapmoves()
|
2016-04-24 13:50:26 -07:00
|
|
|
if Debug['W'] != 0 && Curfn.Func.Enter.Len() > 0 {
|
2015-05-27 10:42:55 -04:00
|
|
|
s := fmt.Sprintf("enter %v", Curfn.Func.Nname.Sym)
|
2016-03-04 13:16:48 -08:00
|
|
|
dumplist(s, Curfn.Func.Enter)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
func walkstmtlist(s []*Node) {
|
|
|
|
|
for i := range s {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
s[i] = walkstmt(s[i])
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
func samelist(a, b []*Node) bool {
|
|
|
|
|
if len(a) != len(b) {
|
|
|
|
|
return false
|
2016-02-26 14:28:48 -08:00
|
|
|
}
|
2016-03-07 22:54:46 -08:00
|
|
|
for i, n := range a {
|
|
|
|
|
if n != b[i] {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
2016-03-07 22:54:46 -08:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-03-06 18:42:58 +11:00
|
|
|
func paramoutheap(fn *Node) bool {
|
2016-02-25 10:35:19 -08:00
|
|
|
for _, ln := range fn.Func.Dcl {
|
2017-04-25 18:14:12 -07:00
|
|
|
switch ln.Class() {
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
case PPARAMOUT:
|
2017-02-27 19:56:38 +02:00
|
|
|
if ln.isParamStackCopy() || ln.Addrtaken() {
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
return true
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
case PAUTO:
|
2015-02-13 14:40:36 -05:00
|
|
|
// stop early - parameters are over
|
2015-03-06 18:42:58 +11:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2015-03-06 18:42:58 +11:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// adds "adjust" to all the argument locations for the call n.
|
|
|
|
|
// n must be a defer or go node that has already been walked.
|
|
|
|
|
func adjustargs(n *Node, adjust int) {
|
|
|
|
|
var arg *Node
|
|
|
|
|
var lhs *Node
|
|
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
callfunc := n.Left
|
2016-03-08 15:10:26 -08:00
|
|
|
for _, arg = range callfunc.List.Slice() {
|
2015-02-13 14:40:36 -05:00
|
|
|
if arg.Op != OAS {
|
2017-03-14 09:46:45 -07:00
|
|
|
Fatalf("call arg not assignment")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
lhs = arg.Left
|
|
|
|
|
if lhs.Op == ONAME {
|
|
|
|
|
// This is a temporary introduced by reorder1.
|
|
|
|
|
// The real store to the stack appears later in the arg list.
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
2016-10-24 14:33:22 -07:00
|
|
|
if lhs.Op != OINDREGSP {
|
2017-03-14 09:46:45 -07:00
|
|
|
Fatalf("call argument store does not use OINDREGSP")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// can't really check this in machine-indep code.
|
|
|
|
|
//if(lhs->val.u.reg != D_SP)
|
2017-03-14 09:46:45 -07:00
|
|
|
// Fatalf("call arg assign not indreg(SP)")
|
2015-02-13 14:40:36 -05:00
|
|
|
lhs.Xoffset += int64(adjust)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
// The result of walkstmt MUST be assigned back to n, e.g.
|
|
|
|
|
// n.Left = walkstmt(n.Left)
|
|
|
|
|
func walkstmt(n *Node) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n == nil {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
setlineno(n)
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
walkstmtlist(n.Ninit.Slice())
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
switch n.Op {
|
|
|
|
|
default:
|
|
|
|
|
if n.Op == ONAME {
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("%v is not a top level statement", n.Sym)
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("%v is not a top level statement", n.Op)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
Dump("nottop", n)
|
|
|
|
|
|
|
|
|
|
case OAS,
|
|
|
|
|
OASOP,
|
|
|
|
|
OAS2,
|
|
|
|
|
OAS2DOTTYPE,
|
|
|
|
|
OAS2RECV,
|
|
|
|
|
OAS2FUNC,
|
|
|
|
|
OAS2MAPR,
|
|
|
|
|
OCLOSE,
|
|
|
|
|
OCOPY,
|
|
|
|
|
OCALLMETH,
|
|
|
|
|
OCALLINTER,
|
|
|
|
|
OCALL,
|
|
|
|
|
OCALLFUNC,
|
|
|
|
|
ODELETE,
|
|
|
|
|
OSEND,
|
|
|
|
|
OPRINT,
|
|
|
|
|
OPRINTN,
|
|
|
|
|
OPANIC,
|
|
|
|
|
OEMPTY,
|
2015-04-03 12:23:28 -04:00
|
|
|
ORECOVER,
|
|
|
|
|
OGETG:
|
2017-04-25 18:02:43 -07:00
|
|
|
if n.Typecheck() == 0 {
|
2016-08-31 15:22:36 -07:00
|
|
|
Fatalf("missing typecheck: %+v", n)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
wascopy := n.Op == OCOPY
|
2015-02-23 16:07:24 -05:00
|
|
|
init := n.Ninit
|
2016-03-08 15:10:26 -08:00
|
|
|
n.Ninit.Set(nil)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = walkexpr(n, &init)
|
|
|
|
|
n = addinit(n, init.Slice())
|
|
|
|
|
if wascopy && n.Op == OCONVNOP {
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Op = OEMPTY // don't leave plain values as statements.
|
|
|
|
|
}
|
|
|
|
|
|
2016-05-07 21:34:04 +02:00
|
|
|
// special case for a receive where we throw away
|
2015-02-13 14:40:36 -05:00
|
|
|
// the value received.
|
|
|
|
|
case ORECV:
|
2017-04-25 18:02:43 -07:00
|
|
|
if n.Typecheck() == 0 {
|
2016-08-31 15:22:36 -07:00
|
|
|
Fatalf("missing typecheck: %+v", n)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-02-23 16:07:24 -05:00
|
|
|
init := n.Ninit
|
2016-03-08 15:10:26 -08:00
|
|
|
n.Ninit.Set(nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, &init)
|
2017-03-18 15:55:41 +00:00
|
|
|
n = mkcall1(chanfn("chanrecv1", 2, n.Left.Type), nil, &init, n.Left, nodnil())
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = walkexpr(n, &init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = addinit(n, init.Slice())
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OBREAK,
|
|
|
|
|
OCONTINUE,
|
|
|
|
|
OFALL,
|
|
|
|
|
OGOTO,
|
|
|
|
|
OLABEL,
|
|
|
|
|
ODCLCONST,
|
|
|
|
|
ODCLTYPE,
|
|
|
|
|
OCHECKNIL,
|
cmd/compile: recognize Syscall-like functions for liveness analysis
Consider this code:
func f(*int)
func g() {
p := new(int)
f(p)
}
where f is an assembly function.
In general liveness analysis assumes that during the call to f, p is dead
in this frame. If f has retained p, p will be found alive in f's frame and keep
the new(int) from being garbage collected. This is all correct and works.
We use the Go func declaration for f to give the assembly function
liveness information (the arguments are assumed live for the entire call).
Now consider this code:
func h1() {
p := new(int)
syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p)))
}
Here syscall.Syscall is taking the place of f, but because its arguments
are uintptr, the liveness analysis and the garbage collector ignore them.
Since p is no longer live in h once the call starts, if the garbage collector
scans the stack while the system call is blocked, it will find no reference
to the new(int) and reclaim it. If the kernel is going to write to *p once
the call finishes, reclaiming the memory is a mistake.
We can't change the arguments or the liveness information for
syscall.Syscall itself, both for compatibility and because sometimes the
arguments really are integers, and the garbage collector will get quite upset
if it finds an integer where it expects a pointer. The problem is that
these arguments are fundamentally untyped.
The solution we have taken in the syscall package's wrappers in past
releases is to insert a call to a dummy function named "use", to make
it look like the argument is live during the call to syscall.Syscall:
func h2() {
p := new(int)
syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p)))
use(unsafe.Pointer(p))
}
Keeping p alive during the call means that if the garbage collector
scans the stack during the system call now, it will find the reference to p.
Unfortunately, this approach is not available to users outside syscall,
because 'use' is unexported, and people also have to realize they need
to use it and do so. There is much existing code using syscall.Syscall
without a 'use'-like function. That code will fail very occasionally in
mysterious ways (see #13372).
This CL fixes all that existing code by making the compiler do the right
thing automatically, without any code modifications. That is, it takes h1
above, which is incorrect code today, and makes it correct code.
Specifically, if the compiler sees a foreign func definition (one
without a body) that has uintptr arguments, it marks those arguments
as "unsafe uintptrs". If it later sees the function being called
with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x
as having escaped, and it makes sure to hold x in a live temporary
variable until the call returns, so that the garbage collector cannot
reclaim whatever heap memory x points to.
For now I am leaving the explicit calls to use in package syscall,
but they can be removed early in a future cycle (likely Go 1.7).
The rule has no effect on escape analysis, only on liveness analysis.
Fixes #13372.
Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500
Reviewed-on: https://go-review.googlesource.com/18584
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-01-13 00:46:28 -05:00
|
|
|
OVARKILL,
|
|
|
|
|
OVARLIVE:
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
case ODCL:
|
|
|
|
|
v := n.Left
|
2017-04-25 18:14:12 -07:00
|
|
|
if v.Class() == PAUTOHEAP {
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
if compiling_runtime {
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("%v escapes to heap, not allowed in runtime.", v)
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
}
|
|
|
|
|
if prealloc[v] == nil {
|
|
|
|
|
prealloc[v] = callnew(v.Type)
|
|
|
|
|
}
|
2016-10-28 11:42:40 -07:00
|
|
|
nn := nod(OAS, v.Name.Param.Heapaddr, prealloc[v])
|
2017-02-27 19:56:38 +02:00
|
|
|
nn.SetColas(true)
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
nn = typecheck(nn, Etop)
|
|
|
|
|
return walkstmt(nn)
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-13 14:40:36 -05:00
|
|
|
case OBLOCK:
|
2016-03-07 22:54:46 -08:00
|
|
|
walkstmtlist(n.List.Slice())
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OXCASE:
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("case statement out of place")
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Op = OCASE
|
|
|
|
|
fallthrough
|
|
|
|
|
|
|
|
|
|
case OCASE:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right = walkstmt(n.Right)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ODEFER:
|
2017-03-15 22:55:21 -07:00
|
|
|
Curfn.Func.SetHasDefer(true)
|
2015-02-13 14:40:36 -05:00
|
|
|
switch n.Left.Op {
|
2015-04-01 09:38:44 -07:00
|
|
|
case OPRINT, OPRINTN:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkprintfunc(n.Left, &n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OCOPY:
|
2015-10-20 10:00:07 -07:00
|
|
|
n.Left = copyany(n.Left, &n.Ninit, true)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
default:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, &n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// make room for size & fn arguments.
|
|
|
|
|
adjustargs(n, 2*Widthptr)
|
|
|
|
|
|
2017-02-02 11:53:41 -05:00
|
|
|
case OFOR, OFORUNTIL:
|
2015-05-26 21:30:20 -04:00
|
|
|
if n.Left != nil {
|
2016-03-07 22:54:46 -08:00
|
|
|
walkstmtlist(n.Left.Ninit.Slice())
|
2015-05-26 21:30:20 -04:00
|
|
|
init := n.Left.Ninit
|
2016-03-08 15:10:26 -08:00
|
|
|
n.Left.Ninit.Set(nil)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, &init)
|
|
|
|
|
n.Left = addinit(n.Left, init.Slice())
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right = walkstmt(n.Right)
|
2016-03-07 22:54:46 -08:00
|
|
|
walkstmtlist(n.Nbody.Slice())
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OIF:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, &n.Ninit)
|
2016-03-07 22:54:46 -08:00
|
|
|
walkstmtlist(n.Nbody.Slice())
|
|
|
|
|
walkstmtlist(n.Rlist.Slice())
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OPROC:
|
|
|
|
|
switch n.Left.Op {
|
2015-04-01 09:38:44 -07:00
|
|
|
case OPRINT, OPRINTN:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkprintfunc(n.Left, &n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OCOPY:
|
2015-10-20 10:00:07 -07:00
|
|
|
n.Left = copyany(n.Left, &n.Ninit, true)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
default:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, &n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// make room for size & fn arguments.
|
|
|
|
|
adjustargs(n, 2*Widthptr)
|
|
|
|
|
|
|
|
|
|
case ORETURN:
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlist(n.List.Slice(), &n.Ninit)
|
2016-03-08 15:10:26 -08:00
|
|
|
if n.List.Len() == 0 {
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
}
|
2016-04-01 20:11:30 -07:00
|
|
|
if (Curfn.Type.FuncType().Outnamed && n.List.Len() > 1) || paramoutheap(Curfn) {
|
2015-02-13 14:40:36 -05:00
|
|
|
// assign to the function out parameters,
|
|
|
|
|
// so that reorder3 can fix up conflicts
|
2016-03-07 22:54:46 -08:00
|
|
|
var rl []*Node
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-10-26 14:57:36 -07:00
|
|
|
var cl Class
|
2016-02-25 10:35:19 -08:00
|
|
|
for _, ln := range Curfn.Func.Dcl {
|
2017-04-25 18:14:12 -07:00
|
|
|
cl = ln.Class()
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
if cl == PAUTO || cl == PAUTOHEAP {
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
if cl == PPARAMOUT {
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
if ln.isParamStackCopy() {
|
2016-10-28 11:42:40 -07:00
|
|
|
ln = walkexpr(typecheck(nod(OIND, ln.Name.Param.Heapaddr, nil), Erv), nil)
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
}
|
2016-03-07 22:54:46 -08:00
|
|
|
rl = append(rl, ln)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
if got, want := n.List.Len(), len(rl); got != want {
|
2015-12-14 12:37:26 -08:00
|
|
|
// order should have rewritten multi-value function calls
|
|
|
|
|
// with explicit OAS2FUNC nodes.
|
|
|
|
|
Fatalf("expected %v return arguments, have %v", want, got)
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
if samelist(rl, n.List.Slice()) {
|
2015-02-13 14:40:36 -05:00
|
|
|
// special return in disguise
|
2016-03-08 15:10:26 -08:00
|
|
|
n.List.Set(nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// move function calls out, to make reorder3's job easier.
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlistsafe(n.List.Slice(), &n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-08 10:26:20 -08:00
|
|
|
ll := ascompatee(n.Op, rl, n.List.Slice(), &n.Ninit)
|
2016-03-08 15:10:26 -08:00
|
|
|
n.List.Set(reorder3(ll))
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
2017-02-22 14:43:21 -08:00
|
|
|
ll := ascompatte(nil, false, Curfn.Type.Results(), n.List.Slice(), 1, &n.Ninit)
|
2016-03-08 15:10:26 -08:00
|
|
|
n.List.Set(ll)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ORETJMP:
|
|
|
|
|
break
|
|
|
|
|
|
|
|
|
|
case OSELECT:
|
|
|
|
|
walkselect(n)
|
|
|
|
|
|
|
|
|
|
case OSWITCH:
|
|
|
|
|
walkswitch(n)
|
|
|
|
|
|
|
|
|
|
case ORANGE:
|
2017-02-02 11:53:41 -05:00
|
|
|
n = walkrange(n)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OXFALL:
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("fallthrough statement out of place")
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Op = OFALL
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if n.Op == ONAME {
|
2016-08-31 15:22:36 -07:00
|
|
|
Fatalf("walkstmt ended up with name: %+v", n)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-05-20 15:16:34 -04:00
|
|
|
func isSmallMakeSlice(n *Node) bool {
|
|
|
|
|
if n.Op != OMAKESLICE {
|
|
|
|
|
return false
|
|
|
|
|
}
|
|
|
|
|
l := n.Left
|
|
|
|
|
r := n.Right
|
|
|
|
|
if r == nil {
|
|
|
|
|
r = l
|
|
|
|
|
}
|
|
|
|
|
t := n.Type
|
|
|
|
|
|
2016-09-15 14:34:20 +10:00
|
|
|
return smallintconst(l) && smallintconst(r) && (t.Elem().Width == 0 || r.Int64() < (1<<16)/t.Elem().Width)
|
2015-05-20 15:16:34 -04:00
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// walk the whole tree of the body of an
|
|
|
|
|
// expression or simple statement.
|
|
|
|
|
// the types expressions are calculated.
|
|
|
|
|
// compile-time constants are evaluated.
|
|
|
|
|
// complex side effects like statements are appended to init
|
2016-03-07 22:54:46 -08:00
|
|
|
func walkexprlist(s []*Node, init *Nodes) {
|
|
|
|
|
for i := range s {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
s[i] = walkexpr(s[i], init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
func walkexprlistsafe(s []*Node, init *Nodes) {
|
|
|
|
|
for i, n := range s {
|
|
|
|
|
s[i] = safeexpr(n, init)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
s[i] = walkexpr(s[i], init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
func walkexprlistcheap(s []*Node, init *Nodes) {
|
|
|
|
|
for i, n := range s {
|
|
|
|
|
s[i] = cheapexpr(n, init)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
s[i] = walkexpr(s[i], init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-10-25 23:34:35 +00:00
|
|
|
// Build name of function for interface conversion.
|
2016-03-04 22:02:27 +00:00
|
|
|
// Not all names are possible
|
2016-10-25 23:34:35 +00:00
|
|
|
// (e.g., we'll never generate convE2E or convE2I or convI2E).
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func convFuncName(from, to *types.Type) string {
|
|
|
|
|
tkind := to.Tie()
|
|
|
|
|
switch from.Tie() {
|
2016-03-04 22:02:27 +00:00
|
|
|
case 'I':
|
|
|
|
|
switch tkind {
|
|
|
|
|
case 'I':
|
|
|
|
|
return "convI2I"
|
|
|
|
|
}
|
|
|
|
|
case 'T':
|
|
|
|
|
switch tkind {
|
|
|
|
|
case 'E':
|
cmd/compile, runtime: specialize convT2x, don't alloc for zero vals
Prior to this CL, all runtime conversions
from a concrete value to an interface went
through one of two runtime calls: convT2E or convT2I.
However, in practice, basic types are very common.
Specializing convT2x for those basic types allows
for a more efficient implementation for those types.
For basic scalars and strings, allocation and copying
can use the same methods as normal code.
For pointer-free types, allocation can occur without
zeroing, and copying can take place without GC calls.
For slices, copying is cheaper and simpler.
This CL adds twelve runtime routines:
convT2E16, convT2I16
convT2E32, convT2I32
convT2E64, convT2I64
convT2Estring, convT2Istring
convT2Eslice, convT2Islice
convT2Enoptr, convT2Inoptr
While compiling make.bash, 93% of all convT2x calls
are now to one of these specialized convT2x call.
Within specialized convT2x routines, it is cheap to check
for a zero value, in a way that it is not in general.
When we detect a zero value there, we return a pointer
to zeroVal, rather than allocating.
name old time/op new time/op delta
ConvT2Ezero/zero/16-8 17.9ns ± 2% 3.0ns ± 3% -83.20% (p=0.000 n=56+56)
ConvT2Ezero/zero/32-8 17.8ns ± 2% 3.0ns ± 3% -83.15% (p=0.000 n=59+60)
ConvT2Ezero/zero/64-8 20.1ns ± 1% 3.0ns ± 2% -84.98% (p=0.000 n=57+57)
ConvT2Ezero/zero/str-8 32.6ns ± 1% 3.0ns ± 4% -90.70% (p=0.000 n=59+60)
ConvT2Ezero/zero/slice-8 36.7ns ± 2% 3.0ns ± 2% -91.78% (p=0.000 n=59+59)
ConvT2Ezero/zero/big-8 91.9ns ± 2% 85.9ns ± 2% -6.52% (p=0.000 n=57+57)
ConvT2Ezero/nonzero/16-8 17.7ns ± 2% 12.7ns ± 3% -28.38% (p=0.000 n=55+60)
ConvT2Ezero/nonzero/32-8 17.8ns ± 1% 12.7ns ± 1% -28.44% (p=0.000 n=54+57)
ConvT2Ezero/nonzero/64-8 20.0ns ± 1% 15.0ns ± 1% -24.90% (p=0.000 n=56+58)
ConvT2Ezero/nonzero/str-8 32.6ns ± 1% 25.7ns ± 1% -21.17% (p=0.000 n=58+55)
ConvT2Ezero/nonzero/slice-8 36.8ns ± 2% 30.4ns ± 1% -17.32% (p=0.000 n=60+52)
ConvT2Ezero/nonzero/big-8 92.1ns ± 2% 85.9ns ± 2% -6.70% (p=0.000 n=57+59)
Benchmarks on a real program (the compiler):
name old time/op new time/op delta
Template 227ms ± 5% 221ms ± 2% -2.48% (p=0.000 n=30+26)
Unicode 102ms ± 5% 100ms ± 3% -1.30% (p=0.009 n=30+26)
GoTypes 656ms ± 5% 659ms ± 4% ~ (p=0.208 n=30+30)
Compiler 2.82s ± 2% 2.82s ± 1% ~ (p=0.614 n=29+27)
Flate 128ms ± 2% 128ms ± 5% ~ (p=0.783 n=27+28)
GoParser 158ms ± 3% 158ms ± 3% ~ (p=0.261 n=28+30)
Reflect 408ms ± 7% 401ms ± 3% ~ (p=0.075 n=30+30)
Tar 123ms ± 6% 121ms ± 8% ~ (p=0.287 n=29+30)
XML 220ms ± 2% 220ms ± 4% ~ (p=0.805 n=29+29)
name old user-ns/op new user-ns/op delta
Template 281user-ms ± 4% 279user-ms ± 3% -0.87% (p=0.044 n=28+28)
Unicode 142user-ms ± 4% 141user-ms ± 3% -1.04% (p=0.015 n=30+27)
GoTypes 884user-ms ± 3% 886user-ms ± 2% ~ (p=0.532 n=30+30)
Compiler 3.94user-s ± 3% 3.92user-s ± 1% ~ (p=0.185 n=30+28)
Flate 165user-ms ± 2% 165user-ms ± 4% ~ (p=0.780 n=27+29)
GoParser 209user-ms ± 2% 208user-ms ± 3% ~ (p=0.453 n=28+30)
Reflect 533user-ms ± 6% 526user-ms ± 3% ~ (p=0.057 n=30+30)
Tar 156user-ms ± 6% 154user-ms ± 6% ~ (p=0.133 n=29+30)
XML 288user-ms ± 4% 288user-ms ± 4% ~ (p=0.633 n=30+30)
name old alloc/op new alloc/op delta
Template 41.0MB ± 0% 40.9MB ± 0% -0.11% (p=0.000 n=29+29)
Unicode 32.6MB ± 0% 32.6MB ± 0% ~ (p=0.572 n=29+30)
GoTypes 122MB ± 0% 122MB ± 0% -0.10% (p=0.000 n=30+30)
Compiler 482MB ± 0% 481MB ± 0% -0.07% (p=0.000 n=30+29)
Flate 26.6MB ± 0% 26.6MB ± 0% ~ (p=0.096 n=30+30)
GoParser 32.7MB ± 0% 32.6MB ± 0% -0.06% (p=0.011 n=28+28)
Reflect 84.2MB ± 0% 84.1MB ± 0% -0.17% (p=0.000 n=29+30)
Tar 27.7MB ± 0% 27.7MB ± 0% -0.05% (p=0.032 n=27+28)
XML 44.7MB ± 0% 44.7MB ± 0% ~ (p=0.131 n=28+30)
name old allocs/op new allocs/op delta
Template 373k ± 1% 370k ± 1% -0.76% (p=0.000 n=30+30)
Unicode 325k ± 1% 325k ± 1% ~ (p=0.383 n=29+30)
GoTypes 1.16M ± 0% 1.15M ± 0% -0.75% (p=0.000 n=29+30)
Compiler 4.15M ± 0% 4.13M ± 0% -0.59% (p=0.000 n=30+29)
Flate 238k ± 1% 237k ± 1% -0.62% (p=0.000 n=30+30)
GoParser 304k ± 1% 302k ± 1% -0.64% (p=0.000 n=30+28)
Reflect 1.00M ± 0% 0.99M ± 0% -1.10% (p=0.000 n=29+30)
Tar 245k ± 1% 244k ± 1% -0.59% (p=0.000 n=27+29)
XML 391k ± 1% 389k ± 1% -0.59% (p=0.000 n=29+30)
Change-Id: Id7f456d690567c2b0a96b0d6d64de8784b6e305f
Reviewed-on: https://go-review.googlesource.com/36476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-02-03 22:40:56 -08:00
|
|
|
switch {
|
|
|
|
|
case from.Size() == 2 && from.Align == 2:
|
|
|
|
|
return "convT2E16"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
case from.Size() == 4 && from.Align == 4 && !types.Haspointers(from):
|
cmd/compile, runtime: specialize convT2x, don't alloc for zero vals
Prior to this CL, all runtime conversions
from a concrete value to an interface went
through one of two runtime calls: convT2E or convT2I.
However, in practice, basic types are very common.
Specializing convT2x for those basic types allows
for a more efficient implementation for those types.
For basic scalars and strings, allocation and copying
can use the same methods as normal code.
For pointer-free types, allocation can occur without
zeroing, and copying can take place without GC calls.
For slices, copying is cheaper and simpler.
This CL adds twelve runtime routines:
convT2E16, convT2I16
convT2E32, convT2I32
convT2E64, convT2I64
convT2Estring, convT2Istring
convT2Eslice, convT2Islice
convT2Enoptr, convT2Inoptr
While compiling make.bash, 93% of all convT2x calls
are now to one of these specialized convT2x call.
Within specialized convT2x routines, it is cheap to check
for a zero value, in a way that it is not in general.
When we detect a zero value there, we return a pointer
to zeroVal, rather than allocating.
name old time/op new time/op delta
ConvT2Ezero/zero/16-8 17.9ns ± 2% 3.0ns ± 3% -83.20% (p=0.000 n=56+56)
ConvT2Ezero/zero/32-8 17.8ns ± 2% 3.0ns ± 3% -83.15% (p=0.000 n=59+60)
ConvT2Ezero/zero/64-8 20.1ns ± 1% 3.0ns ± 2% -84.98% (p=0.000 n=57+57)
ConvT2Ezero/zero/str-8 32.6ns ± 1% 3.0ns ± 4% -90.70% (p=0.000 n=59+60)
ConvT2Ezero/zero/slice-8 36.7ns ± 2% 3.0ns ± 2% -91.78% (p=0.000 n=59+59)
ConvT2Ezero/zero/big-8 91.9ns ± 2% 85.9ns ± 2% -6.52% (p=0.000 n=57+57)
ConvT2Ezero/nonzero/16-8 17.7ns ± 2% 12.7ns ± 3% -28.38% (p=0.000 n=55+60)
ConvT2Ezero/nonzero/32-8 17.8ns ± 1% 12.7ns ± 1% -28.44% (p=0.000 n=54+57)
ConvT2Ezero/nonzero/64-8 20.0ns ± 1% 15.0ns ± 1% -24.90% (p=0.000 n=56+58)
ConvT2Ezero/nonzero/str-8 32.6ns ± 1% 25.7ns ± 1% -21.17% (p=0.000 n=58+55)
ConvT2Ezero/nonzero/slice-8 36.8ns ± 2% 30.4ns ± 1% -17.32% (p=0.000 n=60+52)
ConvT2Ezero/nonzero/big-8 92.1ns ± 2% 85.9ns ± 2% -6.70% (p=0.000 n=57+59)
Benchmarks on a real program (the compiler):
name old time/op new time/op delta
Template 227ms ± 5% 221ms ± 2% -2.48% (p=0.000 n=30+26)
Unicode 102ms ± 5% 100ms ± 3% -1.30% (p=0.009 n=30+26)
GoTypes 656ms ± 5% 659ms ± 4% ~ (p=0.208 n=30+30)
Compiler 2.82s ± 2% 2.82s ± 1% ~ (p=0.614 n=29+27)
Flate 128ms ± 2% 128ms ± 5% ~ (p=0.783 n=27+28)
GoParser 158ms ± 3% 158ms ± 3% ~ (p=0.261 n=28+30)
Reflect 408ms ± 7% 401ms ± 3% ~ (p=0.075 n=30+30)
Tar 123ms ± 6% 121ms ± 8% ~ (p=0.287 n=29+30)
XML 220ms ± 2% 220ms ± 4% ~ (p=0.805 n=29+29)
name old user-ns/op new user-ns/op delta
Template 281user-ms ± 4% 279user-ms ± 3% -0.87% (p=0.044 n=28+28)
Unicode 142user-ms ± 4% 141user-ms ± 3% -1.04% (p=0.015 n=30+27)
GoTypes 884user-ms ± 3% 886user-ms ± 2% ~ (p=0.532 n=30+30)
Compiler 3.94user-s ± 3% 3.92user-s ± 1% ~ (p=0.185 n=30+28)
Flate 165user-ms ± 2% 165user-ms ± 4% ~ (p=0.780 n=27+29)
GoParser 209user-ms ± 2% 208user-ms ± 3% ~ (p=0.453 n=28+30)
Reflect 533user-ms ± 6% 526user-ms ± 3% ~ (p=0.057 n=30+30)
Tar 156user-ms ± 6% 154user-ms ± 6% ~ (p=0.133 n=29+30)
XML 288user-ms ± 4% 288user-ms ± 4% ~ (p=0.633 n=30+30)
name old alloc/op new alloc/op delta
Template 41.0MB ± 0% 40.9MB ± 0% -0.11% (p=0.000 n=29+29)
Unicode 32.6MB ± 0% 32.6MB ± 0% ~ (p=0.572 n=29+30)
GoTypes 122MB ± 0% 122MB ± 0% -0.10% (p=0.000 n=30+30)
Compiler 482MB ± 0% 481MB ± 0% -0.07% (p=0.000 n=30+29)
Flate 26.6MB ± 0% 26.6MB ± 0% ~ (p=0.096 n=30+30)
GoParser 32.7MB ± 0% 32.6MB ± 0% -0.06% (p=0.011 n=28+28)
Reflect 84.2MB ± 0% 84.1MB ± 0% -0.17% (p=0.000 n=29+30)
Tar 27.7MB ± 0% 27.7MB ± 0% -0.05% (p=0.032 n=27+28)
XML 44.7MB ± 0% 44.7MB ± 0% ~ (p=0.131 n=28+30)
name old allocs/op new allocs/op delta
Template 373k ± 1% 370k ± 1% -0.76% (p=0.000 n=30+30)
Unicode 325k ± 1% 325k ± 1% ~ (p=0.383 n=29+30)
GoTypes 1.16M ± 0% 1.15M ± 0% -0.75% (p=0.000 n=29+30)
Compiler 4.15M ± 0% 4.13M ± 0% -0.59% (p=0.000 n=30+29)
Flate 238k ± 1% 237k ± 1% -0.62% (p=0.000 n=30+30)
GoParser 304k ± 1% 302k ± 1% -0.64% (p=0.000 n=30+28)
Reflect 1.00M ± 0% 0.99M ± 0% -1.10% (p=0.000 n=29+30)
Tar 245k ± 1% 244k ± 1% -0.59% (p=0.000 n=27+29)
XML 391k ± 1% 389k ± 1% -0.59% (p=0.000 n=29+30)
Change-Id: Id7f456d690567c2b0a96b0d6d64de8784b6e305f
Reviewed-on: https://go-review.googlesource.com/36476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-02-03 22:40:56 -08:00
|
|
|
return "convT2E32"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
case from.Size() == 8 && from.Align == types.Types[TUINT64].Align && !types.Haspointers(from):
|
cmd/compile, runtime: specialize convT2x, don't alloc for zero vals
Prior to this CL, all runtime conversions
from a concrete value to an interface went
through one of two runtime calls: convT2E or convT2I.
However, in practice, basic types are very common.
Specializing convT2x for those basic types allows
for a more efficient implementation for those types.
For basic scalars and strings, allocation and copying
can use the same methods as normal code.
For pointer-free types, allocation can occur without
zeroing, and copying can take place without GC calls.
For slices, copying is cheaper and simpler.
This CL adds twelve runtime routines:
convT2E16, convT2I16
convT2E32, convT2I32
convT2E64, convT2I64
convT2Estring, convT2Istring
convT2Eslice, convT2Islice
convT2Enoptr, convT2Inoptr
While compiling make.bash, 93% of all convT2x calls
are now to one of these specialized convT2x call.
Within specialized convT2x routines, it is cheap to check
for a zero value, in a way that it is not in general.
When we detect a zero value there, we return a pointer
to zeroVal, rather than allocating.
name old time/op new time/op delta
ConvT2Ezero/zero/16-8 17.9ns ± 2% 3.0ns ± 3% -83.20% (p=0.000 n=56+56)
ConvT2Ezero/zero/32-8 17.8ns ± 2% 3.0ns ± 3% -83.15% (p=0.000 n=59+60)
ConvT2Ezero/zero/64-8 20.1ns ± 1% 3.0ns ± 2% -84.98% (p=0.000 n=57+57)
ConvT2Ezero/zero/str-8 32.6ns ± 1% 3.0ns ± 4% -90.70% (p=0.000 n=59+60)
ConvT2Ezero/zero/slice-8 36.7ns ± 2% 3.0ns ± 2% -91.78% (p=0.000 n=59+59)
ConvT2Ezero/zero/big-8 91.9ns ± 2% 85.9ns ± 2% -6.52% (p=0.000 n=57+57)
ConvT2Ezero/nonzero/16-8 17.7ns ± 2% 12.7ns ± 3% -28.38% (p=0.000 n=55+60)
ConvT2Ezero/nonzero/32-8 17.8ns ± 1% 12.7ns ± 1% -28.44% (p=0.000 n=54+57)
ConvT2Ezero/nonzero/64-8 20.0ns ± 1% 15.0ns ± 1% -24.90% (p=0.000 n=56+58)
ConvT2Ezero/nonzero/str-8 32.6ns ± 1% 25.7ns ± 1% -21.17% (p=0.000 n=58+55)
ConvT2Ezero/nonzero/slice-8 36.8ns ± 2% 30.4ns ± 1% -17.32% (p=0.000 n=60+52)
ConvT2Ezero/nonzero/big-8 92.1ns ± 2% 85.9ns ± 2% -6.70% (p=0.000 n=57+59)
Benchmarks on a real program (the compiler):
name old time/op new time/op delta
Template 227ms ± 5% 221ms ± 2% -2.48% (p=0.000 n=30+26)
Unicode 102ms ± 5% 100ms ± 3% -1.30% (p=0.009 n=30+26)
GoTypes 656ms ± 5% 659ms ± 4% ~ (p=0.208 n=30+30)
Compiler 2.82s ± 2% 2.82s ± 1% ~ (p=0.614 n=29+27)
Flate 128ms ± 2% 128ms ± 5% ~ (p=0.783 n=27+28)
GoParser 158ms ± 3% 158ms ± 3% ~ (p=0.261 n=28+30)
Reflect 408ms ± 7% 401ms ± 3% ~ (p=0.075 n=30+30)
Tar 123ms ± 6% 121ms ± 8% ~ (p=0.287 n=29+30)
XML 220ms ± 2% 220ms ± 4% ~ (p=0.805 n=29+29)
name old user-ns/op new user-ns/op delta
Template 281user-ms ± 4% 279user-ms ± 3% -0.87% (p=0.044 n=28+28)
Unicode 142user-ms ± 4% 141user-ms ± 3% -1.04% (p=0.015 n=30+27)
GoTypes 884user-ms ± 3% 886user-ms ± 2% ~ (p=0.532 n=30+30)
Compiler 3.94user-s ± 3% 3.92user-s ± 1% ~ (p=0.185 n=30+28)
Flate 165user-ms ± 2% 165user-ms ± 4% ~ (p=0.780 n=27+29)
GoParser 209user-ms ± 2% 208user-ms ± 3% ~ (p=0.453 n=28+30)
Reflect 533user-ms ± 6% 526user-ms ± 3% ~ (p=0.057 n=30+30)
Tar 156user-ms ± 6% 154user-ms ± 6% ~ (p=0.133 n=29+30)
XML 288user-ms ± 4% 288user-ms ± 4% ~ (p=0.633 n=30+30)
name old alloc/op new alloc/op delta
Template 41.0MB ± 0% 40.9MB ± 0% -0.11% (p=0.000 n=29+29)
Unicode 32.6MB ± 0% 32.6MB ± 0% ~ (p=0.572 n=29+30)
GoTypes 122MB ± 0% 122MB ± 0% -0.10% (p=0.000 n=30+30)
Compiler 482MB ± 0% 481MB ± 0% -0.07% (p=0.000 n=30+29)
Flate 26.6MB ± 0% 26.6MB ± 0% ~ (p=0.096 n=30+30)
GoParser 32.7MB ± 0% 32.6MB ± 0% -0.06% (p=0.011 n=28+28)
Reflect 84.2MB ± 0% 84.1MB ± 0% -0.17% (p=0.000 n=29+30)
Tar 27.7MB ± 0% 27.7MB ± 0% -0.05% (p=0.032 n=27+28)
XML 44.7MB ± 0% 44.7MB ± 0% ~ (p=0.131 n=28+30)
name old allocs/op new allocs/op delta
Template 373k ± 1% 370k ± 1% -0.76% (p=0.000 n=30+30)
Unicode 325k ± 1% 325k ± 1% ~ (p=0.383 n=29+30)
GoTypes 1.16M ± 0% 1.15M ± 0% -0.75% (p=0.000 n=29+30)
Compiler 4.15M ± 0% 4.13M ± 0% -0.59% (p=0.000 n=30+29)
Flate 238k ± 1% 237k ± 1% -0.62% (p=0.000 n=30+30)
GoParser 304k ± 1% 302k ± 1% -0.64% (p=0.000 n=30+28)
Reflect 1.00M ± 0% 0.99M ± 0% -1.10% (p=0.000 n=29+30)
Tar 245k ± 1% 244k ± 1% -0.59% (p=0.000 n=27+29)
XML 391k ± 1% 389k ± 1% -0.59% (p=0.000 n=29+30)
Change-Id: Id7f456d690567c2b0a96b0d6d64de8784b6e305f
Reviewed-on: https://go-review.googlesource.com/36476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-02-03 22:40:56 -08:00
|
|
|
return "convT2E64"
|
|
|
|
|
case from.IsString():
|
|
|
|
|
return "convT2Estring"
|
|
|
|
|
case from.IsSlice():
|
|
|
|
|
return "convT2Eslice"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
case !types.Haspointers(from):
|
cmd/compile, runtime: specialize convT2x, don't alloc for zero vals
Prior to this CL, all runtime conversions
from a concrete value to an interface went
through one of two runtime calls: convT2E or convT2I.
However, in practice, basic types are very common.
Specializing convT2x for those basic types allows
for a more efficient implementation for those types.
For basic scalars and strings, allocation and copying
can use the same methods as normal code.
For pointer-free types, allocation can occur without
zeroing, and copying can take place without GC calls.
For slices, copying is cheaper and simpler.
This CL adds twelve runtime routines:
convT2E16, convT2I16
convT2E32, convT2I32
convT2E64, convT2I64
convT2Estring, convT2Istring
convT2Eslice, convT2Islice
convT2Enoptr, convT2Inoptr
While compiling make.bash, 93% of all convT2x calls
are now to one of these specialized convT2x call.
Within specialized convT2x routines, it is cheap to check
for a zero value, in a way that it is not in general.
When we detect a zero value there, we return a pointer
to zeroVal, rather than allocating.
name old time/op new time/op delta
ConvT2Ezero/zero/16-8 17.9ns ± 2% 3.0ns ± 3% -83.20% (p=0.000 n=56+56)
ConvT2Ezero/zero/32-8 17.8ns ± 2% 3.0ns ± 3% -83.15% (p=0.000 n=59+60)
ConvT2Ezero/zero/64-8 20.1ns ± 1% 3.0ns ± 2% -84.98% (p=0.000 n=57+57)
ConvT2Ezero/zero/str-8 32.6ns ± 1% 3.0ns ± 4% -90.70% (p=0.000 n=59+60)
ConvT2Ezero/zero/slice-8 36.7ns ± 2% 3.0ns ± 2% -91.78% (p=0.000 n=59+59)
ConvT2Ezero/zero/big-8 91.9ns ± 2% 85.9ns ± 2% -6.52% (p=0.000 n=57+57)
ConvT2Ezero/nonzero/16-8 17.7ns ± 2% 12.7ns ± 3% -28.38% (p=0.000 n=55+60)
ConvT2Ezero/nonzero/32-8 17.8ns ± 1% 12.7ns ± 1% -28.44% (p=0.000 n=54+57)
ConvT2Ezero/nonzero/64-8 20.0ns ± 1% 15.0ns ± 1% -24.90% (p=0.000 n=56+58)
ConvT2Ezero/nonzero/str-8 32.6ns ± 1% 25.7ns ± 1% -21.17% (p=0.000 n=58+55)
ConvT2Ezero/nonzero/slice-8 36.8ns ± 2% 30.4ns ± 1% -17.32% (p=0.000 n=60+52)
ConvT2Ezero/nonzero/big-8 92.1ns ± 2% 85.9ns ± 2% -6.70% (p=0.000 n=57+59)
Benchmarks on a real program (the compiler):
name old time/op new time/op delta
Template 227ms ± 5% 221ms ± 2% -2.48% (p=0.000 n=30+26)
Unicode 102ms ± 5% 100ms ± 3% -1.30% (p=0.009 n=30+26)
GoTypes 656ms ± 5% 659ms ± 4% ~ (p=0.208 n=30+30)
Compiler 2.82s ± 2% 2.82s ± 1% ~ (p=0.614 n=29+27)
Flate 128ms ± 2% 128ms ± 5% ~ (p=0.783 n=27+28)
GoParser 158ms ± 3% 158ms ± 3% ~ (p=0.261 n=28+30)
Reflect 408ms ± 7% 401ms ± 3% ~ (p=0.075 n=30+30)
Tar 123ms ± 6% 121ms ± 8% ~ (p=0.287 n=29+30)
XML 220ms ± 2% 220ms ± 4% ~ (p=0.805 n=29+29)
name old user-ns/op new user-ns/op delta
Template 281user-ms ± 4% 279user-ms ± 3% -0.87% (p=0.044 n=28+28)
Unicode 142user-ms ± 4% 141user-ms ± 3% -1.04% (p=0.015 n=30+27)
GoTypes 884user-ms ± 3% 886user-ms ± 2% ~ (p=0.532 n=30+30)
Compiler 3.94user-s ± 3% 3.92user-s ± 1% ~ (p=0.185 n=30+28)
Flate 165user-ms ± 2% 165user-ms ± 4% ~ (p=0.780 n=27+29)
GoParser 209user-ms ± 2% 208user-ms ± 3% ~ (p=0.453 n=28+30)
Reflect 533user-ms ± 6% 526user-ms ± 3% ~ (p=0.057 n=30+30)
Tar 156user-ms ± 6% 154user-ms ± 6% ~ (p=0.133 n=29+30)
XML 288user-ms ± 4% 288user-ms ± 4% ~ (p=0.633 n=30+30)
name old alloc/op new alloc/op delta
Template 41.0MB ± 0% 40.9MB ± 0% -0.11% (p=0.000 n=29+29)
Unicode 32.6MB ± 0% 32.6MB ± 0% ~ (p=0.572 n=29+30)
GoTypes 122MB ± 0% 122MB ± 0% -0.10% (p=0.000 n=30+30)
Compiler 482MB ± 0% 481MB ± 0% -0.07% (p=0.000 n=30+29)
Flate 26.6MB ± 0% 26.6MB ± 0% ~ (p=0.096 n=30+30)
GoParser 32.7MB ± 0% 32.6MB ± 0% -0.06% (p=0.011 n=28+28)
Reflect 84.2MB ± 0% 84.1MB ± 0% -0.17% (p=0.000 n=29+30)
Tar 27.7MB ± 0% 27.7MB ± 0% -0.05% (p=0.032 n=27+28)
XML 44.7MB ± 0% 44.7MB ± 0% ~ (p=0.131 n=28+30)
name old allocs/op new allocs/op delta
Template 373k ± 1% 370k ± 1% -0.76% (p=0.000 n=30+30)
Unicode 325k ± 1% 325k ± 1% ~ (p=0.383 n=29+30)
GoTypes 1.16M ± 0% 1.15M ± 0% -0.75% (p=0.000 n=29+30)
Compiler 4.15M ± 0% 4.13M ± 0% -0.59% (p=0.000 n=30+29)
Flate 238k ± 1% 237k ± 1% -0.62% (p=0.000 n=30+30)
GoParser 304k ± 1% 302k ± 1% -0.64% (p=0.000 n=30+28)
Reflect 1.00M ± 0% 0.99M ± 0% -1.10% (p=0.000 n=29+30)
Tar 245k ± 1% 244k ± 1% -0.59% (p=0.000 n=27+29)
XML 391k ± 1% 389k ± 1% -0.59% (p=0.000 n=29+30)
Change-Id: Id7f456d690567c2b0a96b0d6d64de8784b6e305f
Reviewed-on: https://go-review.googlesource.com/36476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-02-03 22:40:56 -08:00
|
|
|
return "convT2Enoptr"
|
|
|
|
|
}
|
2016-03-04 22:02:27 +00:00
|
|
|
return "convT2E"
|
|
|
|
|
case 'I':
|
cmd/compile, runtime: specialize convT2x, don't alloc for zero vals
Prior to this CL, all runtime conversions
from a concrete value to an interface went
through one of two runtime calls: convT2E or convT2I.
However, in practice, basic types are very common.
Specializing convT2x for those basic types allows
for a more efficient implementation for those types.
For basic scalars and strings, allocation and copying
can use the same methods as normal code.
For pointer-free types, allocation can occur without
zeroing, and copying can take place without GC calls.
For slices, copying is cheaper and simpler.
This CL adds twelve runtime routines:
convT2E16, convT2I16
convT2E32, convT2I32
convT2E64, convT2I64
convT2Estring, convT2Istring
convT2Eslice, convT2Islice
convT2Enoptr, convT2Inoptr
While compiling make.bash, 93% of all convT2x calls
are now to one of these specialized convT2x call.
Within specialized convT2x routines, it is cheap to check
for a zero value, in a way that it is not in general.
When we detect a zero value there, we return a pointer
to zeroVal, rather than allocating.
name old time/op new time/op delta
ConvT2Ezero/zero/16-8 17.9ns ± 2% 3.0ns ± 3% -83.20% (p=0.000 n=56+56)
ConvT2Ezero/zero/32-8 17.8ns ± 2% 3.0ns ± 3% -83.15% (p=0.000 n=59+60)
ConvT2Ezero/zero/64-8 20.1ns ± 1% 3.0ns ± 2% -84.98% (p=0.000 n=57+57)
ConvT2Ezero/zero/str-8 32.6ns ± 1% 3.0ns ± 4% -90.70% (p=0.000 n=59+60)
ConvT2Ezero/zero/slice-8 36.7ns ± 2% 3.0ns ± 2% -91.78% (p=0.000 n=59+59)
ConvT2Ezero/zero/big-8 91.9ns ± 2% 85.9ns ± 2% -6.52% (p=0.000 n=57+57)
ConvT2Ezero/nonzero/16-8 17.7ns ± 2% 12.7ns ± 3% -28.38% (p=0.000 n=55+60)
ConvT2Ezero/nonzero/32-8 17.8ns ± 1% 12.7ns ± 1% -28.44% (p=0.000 n=54+57)
ConvT2Ezero/nonzero/64-8 20.0ns ± 1% 15.0ns ± 1% -24.90% (p=0.000 n=56+58)
ConvT2Ezero/nonzero/str-8 32.6ns ± 1% 25.7ns ± 1% -21.17% (p=0.000 n=58+55)
ConvT2Ezero/nonzero/slice-8 36.8ns ± 2% 30.4ns ± 1% -17.32% (p=0.000 n=60+52)
ConvT2Ezero/nonzero/big-8 92.1ns ± 2% 85.9ns ± 2% -6.70% (p=0.000 n=57+59)
Benchmarks on a real program (the compiler):
name old time/op new time/op delta
Template 227ms ± 5% 221ms ± 2% -2.48% (p=0.000 n=30+26)
Unicode 102ms ± 5% 100ms ± 3% -1.30% (p=0.009 n=30+26)
GoTypes 656ms ± 5% 659ms ± 4% ~ (p=0.208 n=30+30)
Compiler 2.82s ± 2% 2.82s ± 1% ~ (p=0.614 n=29+27)
Flate 128ms ± 2% 128ms ± 5% ~ (p=0.783 n=27+28)
GoParser 158ms ± 3% 158ms ± 3% ~ (p=0.261 n=28+30)
Reflect 408ms ± 7% 401ms ± 3% ~ (p=0.075 n=30+30)
Tar 123ms ± 6% 121ms ± 8% ~ (p=0.287 n=29+30)
XML 220ms ± 2% 220ms ± 4% ~ (p=0.805 n=29+29)
name old user-ns/op new user-ns/op delta
Template 281user-ms ± 4% 279user-ms ± 3% -0.87% (p=0.044 n=28+28)
Unicode 142user-ms ± 4% 141user-ms ± 3% -1.04% (p=0.015 n=30+27)
GoTypes 884user-ms ± 3% 886user-ms ± 2% ~ (p=0.532 n=30+30)
Compiler 3.94user-s ± 3% 3.92user-s ± 1% ~ (p=0.185 n=30+28)
Flate 165user-ms ± 2% 165user-ms ± 4% ~ (p=0.780 n=27+29)
GoParser 209user-ms ± 2% 208user-ms ± 3% ~ (p=0.453 n=28+30)
Reflect 533user-ms ± 6% 526user-ms ± 3% ~ (p=0.057 n=30+30)
Tar 156user-ms ± 6% 154user-ms ± 6% ~ (p=0.133 n=29+30)
XML 288user-ms ± 4% 288user-ms ± 4% ~ (p=0.633 n=30+30)
name old alloc/op new alloc/op delta
Template 41.0MB ± 0% 40.9MB ± 0% -0.11% (p=0.000 n=29+29)
Unicode 32.6MB ± 0% 32.6MB ± 0% ~ (p=0.572 n=29+30)
GoTypes 122MB ± 0% 122MB ± 0% -0.10% (p=0.000 n=30+30)
Compiler 482MB ± 0% 481MB ± 0% -0.07% (p=0.000 n=30+29)
Flate 26.6MB ± 0% 26.6MB ± 0% ~ (p=0.096 n=30+30)
GoParser 32.7MB ± 0% 32.6MB ± 0% -0.06% (p=0.011 n=28+28)
Reflect 84.2MB ± 0% 84.1MB ± 0% -0.17% (p=0.000 n=29+30)
Tar 27.7MB ± 0% 27.7MB ± 0% -0.05% (p=0.032 n=27+28)
XML 44.7MB ± 0% 44.7MB ± 0% ~ (p=0.131 n=28+30)
name old allocs/op new allocs/op delta
Template 373k ± 1% 370k ± 1% -0.76% (p=0.000 n=30+30)
Unicode 325k ± 1% 325k ± 1% ~ (p=0.383 n=29+30)
GoTypes 1.16M ± 0% 1.15M ± 0% -0.75% (p=0.000 n=29+30)
Compiler 4.15M ± 0% 4.13M ± 0% -0.59% (p=0.000 n=30+29)
Flate 238k ± 1% 237k ± 1% -0.62% (p=0.000 n=30+30)
GoParser 304k ± 1% 302k ± 1% -0.64% (p=0.000 n=30+28)
Reflect 1.00M ± 0% 0.99M ± 0% -1.10% (p=0.000 n=29+30)
Tar 245k ± 1% 244k ± 1% -0.59% (p=0.000 n=27+29)
XML 391k ± 1% 389k ± 1% -0.59% (p=0.000 n=29+30)
Change-Id: Id7f456d690567c2b0a96b0d6d64de8784b6e305f
Reviewed-on: https://go-review.googlesource.com/36476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-02-03 22:40:56 -08:00
|
|
|
switch {
|
|
|
|
|
case from.Size() == 2 && from.Align == 2:
|
|
|
|
|
return "convT2I16"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
case from.Size() == 4 && from.Align == 4 && !types.Haspointers(from):
|
cmd/compile, runtime: specialize convT2x, don't alloc for zero vals
Prior to this CL, all runtime conversions
from a concrete value to an interface went
through one of two runtime calls: convT2E or convT2I.
However, in practice, basic types are very common.
Specializing convT2x for those basic types allows
for a more efficient implementation for those types.
For basic scalars and strings, allocation and copying
can use the same methods as normal code.
For pointer-free types, allocation can occur without
zeroing, and copying can take place without GC calls.
For slices, copying is cheaper and simpler.
This CL adds twelve runtime routines:
convT2E16, convT2I16
convT2E32, convT2I32
convT2E64, convT2I64
convT2Estring, convT2Istring
convT2Eslice, convT2Islice
convT2Enoptr, convT2Inoptr
While compiling make.bash, 93% of all convT2x calls
are now to one of these specialized convT2x call.
Within specialized convT2x routines, it is cheap to check
for a zero value, in a way that it is not in general.
When we detect a zero value there, we return a pointer
to zeroVal, rather than allocating.
name old time/op new time/op delta
ConvT2Ezero/zero/16-8 17.9ns ± 2% 3.0ns ± 3% -83.20% (p=0.000 n=56+56)
ConvT2Ezero/zero/32-8 17.8ns ± 2% 3.0ns ± 3% -83.15% (p=0.000 n=59+60)
ConvT2Ezero/zero/64-8 20.1ns ± 1% 3.0ns ± 2% -84.98% (p=0.000 n=57+57)
ConvT2Ezero/zero/str-8 32.6ns ± 1% 3.0ns ± 4% -90.70% (p=0.000 n=59+60)
ConvT2Ezero/zero/slice-8 36.7ns ± 2% 3.0ns ± 2% -91.78% (p=0.000 n=59+59)
ConvT2Ezero/zero/big-8 91.9ns ± 2% 85.9ns ± 2% -6.52% (p=0.000 n=57+57)
ConvT2Ezero/nonzero/16-8 17.7ns ± 2% 12.7ns ± 3% -28.38% (p=0.000 n=55+60)
ConvT2Ezero/nonzero/32-8 17.8ns ± 1% 12.7ns ± 1% -28.44% (p=0.000 n=54+57)
ConvT2Ezero/nonzero/64-8 20.0ns ± 1% 15.0ns ± 1% -24.90% (p=0.000 n=56+58)
ConvT2Ezero/nonzero/str-8 32.6ns ± 1% 25.7ns ± 1% -21.17% (p=0.000 n=58+55)
ConvT2Ezero/nonzero/slice-8 36.8ns ± 2% 30.4ns ± 1% -17.32% (p=0.000 n=60+52)
ConvT2Ezero/nonzero/big-8 92.1ns ± 2% 85.9ns ± 2% -6.70% (p=0.000 n=57+59)
Benchmarks on a real program (the compiler):
name old time/op new time/op delta
Template 227ms ± 5% 221ms ± 2% -2.48% (p=0.000 n=30+26)
Unicode 102ms ± 5% 100ms ± 3% -1.30% (p=0.009 n=30+26)
GoTypes 656ms ± 5% 659ms ± 4% ~ (p=0.208 n=30+30)
Compiler 2.82s ± 2% 2.82s ± 1% ~ (p=0.614 n=29+27)
Flate 128ms ± 2% 128ms ± 5% ~ (p=0.783 n=27+28)
GoParser 158ms ± 3% 158ms ± 3% ~ (p=0.261 n=28+30)
Reflect 408ms ± 7% 401ms ± 3% ~ (p=0.075 n=30+30)
Tar 123ms ± 6% 121ms ± 8% ~ (p=0.287 n=29+30)
XML 220ms ± 2% 220ms ± 4% ~ (p=0.805 n=29+29)
name old user-ns/op new user-ns/op delta
Template 281user-ms ± 4% 279user-ms ± 3% -0.87% (p=0.044 n=28+28)
Unicode 142user-ms ± 4% 141user-ms ± 3% -1.04% (p=0.015 n=30+27)
GoTypes 884user-ms ± 3% 886user-ms ± 2% ~ (p=0.532 n=30+30)
Compiler 3.94user-s ± 3% 3.92user-s ± 1% ~ (p=0.185 n=30+28)
Flate 165user-ms ± 2% 165user-ms ± 4% ~ (p=0.780 n=27+29)
GoParser 209user-ms ± 2% 208user-ms ± 3% ~ (p=0.453 n=28+30)
Reflect 533user-ms ± 6% 526user-ms ± 3% ~ (p=0.057 n=30+30)
Tar 156user-ms ± 6% 154user-ms ± 6% ~ (p=0.133 n=29+30)
XML 288user-ms ± 4% 288user-ms ± 4% ~ (p=0.633 n=30+30)
name old alloc/op new alloc/op delta
Template 41.0MB ± 0% 40.9MB ± 0% -0.11% (p=0.000 n=29+29)
Unicode 32.6MB ± 0% 32.6MB ± 0% ~ (p=0.572 n=29+30)
GoTypes 122MB ± 0% 122MB ± 0% -0.10% (p=0.000 n=30+30)
Compiler 482MB ± 0% 481MB ± 0% -0.07% (p=0.000 n=30+29)
Flate 26.6MB ± 0% 26.6MB ± 0% ~ (p=0.096 n=30+30)
GoParser 32.7MB ± 0% 32.6MB ± 0% -0.06% (p=0.011 n=28+28)
Reflect 84.2MB ± 0% 84.1MB ± 0% -0.17% (p=0.000 n=29+30)
Tar 27.7MB ± 0% 27.7MB ± 0% -0.05% (p=0.032 n=27+28)
XML 44.7MB ± 0% 44.7MB ± 0% ~ (p=0.131 n=28+30)
name old allocs/op new allocs/op delta
Template 373k ± 1% 370k ± 1% -0.76% (p=0.000 n=30+30)
Unicode 325k ± 1% 325k ± 1% ~ (p=0.383 n=29+30)
GoTypes 1.16M ± 0% 1.15M ± 0% -0.75% (p=0.000 n=29+30)
Compiler 4.15M ± 0% 4.13M ± 0% -0.59% (p=0.000 n=30+29)
Flate 238k ± 1% 237k ± 1% -0.62% (p=0.000 n=30+30)
GoParser 304k ± 1% 302k ± 1% -0.64% (p=0.000 n=30+28)
Reflect 1.00M ± 0% 0.99M ± 0% -1.10% (p=0.000 n=29+30)
Tar 245k ± 1% 244k ± 1% -0.59% (p=0.000 n=27+29)
XML 391k ± 1% 389k ± 1% -0.59% (p=0.000 n=29+30)
Change-Id: Id7f456d690567c2b0a96b0d6d64de8784b6e305f
Reviewed-on: https://go-review.googlesource.com/36476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-02-03 22:40:56 -08:00
|
|
|
return "convT2I32"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
case from.Size() == 8 && from.Align == types.Types[TUINT64].Align && !types.Haspointers(from):
|
cmd/compile, runtime: specialize convT2x, don't alloc for zero vals
Prior to this CL, all runtime conversions
from a concrete value to an interface went
through one of two runtime calls: convT2E or convT2I.
However, in practice, basic types are very common.
Specializing convT2x for those basic types allows
for a more efficient implementation for those types.
For basic scalars and strings, allocation and copying
can use the same methods as normal code.
For pointer-free types, allocation can occur without
zeroing, and copying can take place without GC calls.
For slices, copying is cheaper and simpler.
This CL adds twelve runtime routines:
convT2E16, convT2I16
convT2E32, convT2I32
convT2E64, convT2I64
convT2Estring, convT2Istring
convT2Eslice, convT2Islice
convT2Enoptr, convT2Inoptr
While compiling make.bash, 93% of all convT2x calls
are now to one of these specialized convT2x call.
Within specialized convT2x routines, it is cheap to check
for a zero value, in a way that it is not in general.
When we detect a zero value there, we return a pointer
to zeroVal, rather than allocating.
name old time/op new time/op delta
ConvT2Ezero/zero/16-8 17.9ns ± 2% 3.0ns ± 3% -83.20% (p=0.000 n=56+56)
ConvT2Ezero/zero/32-8 17.8ns ± 2% 3.0ns ± 3% -83.15% (p=0.000 n=59+60)
ConvT2Ezero/zero/64-8 20.1ns ± 1% 3.0ns ± 2% -84.98% (p=0.000 n=57+57)
ConvT2Ezero/zero/str-8 32.6ns ± 1% 3.0ns ± 4% -90.70% (p=0.000 n=59+60)
ConvT2Ezero/zero/slice-8 36.7ns ± 2% 3.0ns ± 2% -91.78% (p=0.000 n=59+59)
ConvT2Ezero/zero/big-8 91.9ns ± 2% 85.9ns ± 2% -6.52% (p=0.000 n=57+57)
ConvT2Ezero/nonzero/16-8 17.7ns ± 2% 12.7ns ± 3% -28.38% (p=0.000 n=55+60)
ConvT2Ezero/nonzero/32-8 17.8ns ± 1% 12.7ns ± 1% -28.44% (p=0.000 n=54+57)
ConvT2Ezero/nonzero/64-8 20.0ns ± 1% 15.0ns ± 1% -24.90% (p=0.000 n=56+58)
ConvT2Ezero/nonzero/str-8 32.6ns ± 1% 25.7ns ± 1% -21.17% (p=0.000 n=58+55)
ConvT2Ezero/nonzero/slice-8 36.8ns ± 2% 30.4ns ± 1% -17.32% (p=0.000 n=60+52)
ConvT2Ezero/nonzero/big-8 92.1ns ± 2% 85.9ns ± 2% -6.70% (p=0.000 n=57+59)
Benchmarks on a real program (the compiler):
name old time/op new time/op delta
Template 227ms ± 5% 221ms ± 2% -2.48% (p=0.000 n=30+26)
Unicode 102ms ± 5% 100ms ± 3% -1.30% (p=0.009 n=30+26)
GoTypes 656ms ± 5% 659ms ± 4% ~ (p=0.208 n=30+30)
Compiler 2.82s ± 2% 2.82s ± 1% ~ (p=0.614 n=29+27)
Flate 128ms ± 2% 128ms ± 5% ~ (p=0.783 n=27+28)
GoParser 158ms ± 3% 158ms ± 3% ~ (p=0.261 n=28+30)
Reflect 408ms ± 7% 401ms ± 3% ~ (p=0.075 n=30+30)
Tar 123ms ± 6% 121ms ± 8% ~ (p=0.287 n=29+30)
XML 220ms ± 2% 220ms ± 4% ~ (p=0.805 n=29+29)
name old user-ns/op new user-ns/op delta
Template 281user-ms ± 4% 279user-ms ± 3% -0.87% (p=0.044 n=28+28)
Unicode 142user-ms ± 4% 141user-ms ± 3% -1.04% (p=0.015 n=30+27)
GoTypes 884user-ms ± 3% 886user-ms ± 2% ~ (p=0.532 n=30+30)
Compiler 3.94user-s ± 3% 3.92user-s ± 1% ~ (p=0.185 n=30+28)
Flate 165user-ms ± 2% 165user-ms ± 4% ~ (p=0.780 n=27+29)
GoParser 209user-ms ± 2% 208user-ms ± 3% ~ (p=0.453 n=28+30)
Reflect 533user-ms ± 6% 526user-ms ± 3% ~ (p=0.057 n=30+30)
Tar 156user-ms ± 6% 154user-ms ± 6% ~ (p=0.133 n=29+30)
XML 288user-ms ± 4% 288user-ms ± 4% ~ (p=0.633 n=30+30)
name old alloc/op new alloc/op delta
Template 41.0MB ± 0% 40.9MB ± 0% -0.11% (p=0.000 n=29+29)
Unicode 32.6MB ± 0% 32.6MB ± 0% ~ (p=0.572 n=29+30)
GoTypes 122MB ± 0% 122MB ± 0% -0.10% (p=0.000 n=30+30)
Compiler 482MB ± 0% 481MB ± 0% -0.07% (p=0.000 n=30+29)
Flate 26.6MB ± 0% 26.6MB ± 0% ~ (p=0.096 n=30+30)
GoParser 32.7MB ± 0% 32.6MB ± 0% -0.06% (p=0.011 n=28+28)
Reflect 84.2MB ± 0% 84.1MB ± 0% -0.17% (p=0.000 n=29+30)
Tar 27.7MB ± 0% 27.7MB ± 0% -0.05% (p=0.032 n=27+28)
XML 44.7MB ± 0% 44.7MB ± 0% ~ (p=0.131 n=28+30)
name old allocs/op new allocs/op delta
Template 373k ± 1% 370k ± 1% -0.76% (p=0.000 n=30+30)
Unicode 325k ± 1% 325k ± 1% ~ (p=0.383 n=29+30)
GoTypes 1.16M ± 0% 1.15M ± 0% -0.75% (p=0.000 n=29+30)
Compiler 4.15M ± 0% 4.13M ± 0% -0.59% (p=0.000 n=30+29)
Flate 238k ± 1% 237k ± 1% -0.62% (p=0.000 n=30+30)
GoParser 304k ± 1% 302k ± 1% -0.64% (p=0.000 n=30+28)
Reflect 1.00M ± 0% 0.99M ± 0% -1.10% (p=0.000 n=29+30)
Tar 245k ± 1% 244k ± 1% -0.59% (p=0.000 n=27+29)
XML 391k ± 1% 389k ± 1% -0.59% (p=0.000 n=29+30)
Change-Id: Id7f456d690567c2b0a96b0d6d64de8784b6e305f
Reviewed-on: https://go-review.googlesource.com/36476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-02-03 22:40:56 -08:00
|
|
|
return "convT2I64"
|
|
|
|
|
case from.IsString():
|
|
|
|
|
return "convT2Istring"
|
|
|
|
|
case from.IsSlice():
|
|
|
|
|
return "convT2Islice"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
case !types.Haspointers(from):
|
cmd/compile, runtime: specialize convT2x, don't alloc for zero vals
Prior to this CL, all runtime conversions
from a concrete value to an interface went
through one of two runtime calls: convT2E or convT2I.
However, in practice, basic types are very common.
Specializing convT2x for those basic types allows
for a more efficient implementation for those types.
For basic scalars and strings, allocation and copying
can use the same methods as normal code.
For pointer-free types, allocation can occur without
zeroing, and copying can take place without GC calls.
For slices, copying is cheaper and simpler.
This CL adds twelve runtime routines:
convT2E16, convT2I16
convT2E32, convT2I32
convT2E64, convT2I64
convT2Estring, convT2Istring
convT2Eslice, convT2Islice
convT2Enoptr, convT2Inoptr
While compiling make.bash, 93% of all convT2x calls
are now to one of these specialized convT2x call.
Within specialized convT2x routines, it is cheap to check
for a zero value, in a way that it is not in general.
When we detect a zero value there, we return a pointer
to zeroVal, rather than allocating.
name old time/op new time/op delta
ConvT2Ezero/zero/16-8 17.9ns ± 2% 3.0ns ± 3% -83.20% (p=0.000 n=56+56)
ConvT2Ezero/zero/32-8 17.8ns ± 2% 3.0ns ± 3% -83.15% (p=0.000 n=59+60)
ConvT2Ezero/zero/64-8 20.1ns ± 1% 3.0ns ± 2% -84.98% (p=0.000 n=57+57)
ConvT2Ezero/zero/str-8 32.6ns ± 1% 3.0ns ± 4% -90.70% (p=0.000 n=59+60)
ConvT2Ezero/zero/slice-8 36.7ns ± 2% 3.0ns ± 2% -91.78% (p=0.000 n=59+59)
ConvT2Ezero/zero/big-8 91.9ns ± 2% 85.9ns ± 2% -6.52% (p=0.000 n=57+57)
ConvT2Ezero/nonzero/16-8 17.7ns ± 2% 12.7ns ± 3% -28.38% (p=0.000 n=55+60)
ConvT2Ezero/nonzero/32-8 17.8ns ± 1% 12.7ns ± 1% -28.44% (p=0.000 n=54+57)
ConvT2Ezero/nonzero/64-8 20.0ns ± 1% 15.0ns ± 1% -24.90% (p=0.000 n=56+58)
ConvT2Ezero/nonzero/str-8 32.6ns ± 1% 25.7ns ± 1% -21.17% (p=0.000 n=58+55)
ConvT2Ezero/nonzero/slice-8 36.8ns ± 2% 30.4ns ± 1% -17.32% (p=0.000 n=60+52)
ConvT2Ezero/nonzero/big-8 92.1ns ± 2% 85.9ns ± 2% -6.70% (p=0.000 n=57+59)
Benchmarks on a real program (the compiler):
name old time/op new time/op delta
Template 227ms ± 5% 221ms ± 2% -2.48% (p=0.000 n=30+26)
Unicode 102ms ± 5% 100ms ± 3% -1.30% (p=0.009 n=30+26)
GoTypes 656ms ± 5% 659ms ± 4% ~ (p=0.208 n=30+30)
Compiler 2.82s ± 2% 2.82s ± 1% ~ (p=0.614 n=29+27)
Flate 128ms ± 2% 128ms ± 5% ~ (p=0.783 n=27+28)
GoParser 158ms ± 3% 158ms ± 3% ~ (p=0.261 n=28+30)
Reflect 408ms ± 7% 401ms ± 3% ~ (p=0.075 n=30+30)
Tar 123ms ± 6% 121ms ± 8% ~ (p=0.287 n=29+30)
XML 220ms ± 2% 220ms ± 4% ~ (p=0.805 n=29+29)
name old user-ns/op new user-ns/op delta
Template 281user-ms ± 4% 279user-ms ± 3% -0.87% (p=0.044 n=28+28)
Unicode 142user-ms ± 4% 141user-ms ± 3% -1.04% (p=0.015 n=30+27)
GoTypes 884user-ms ± 3% 886user-ms ± 2% ~ (p=0.532 n=30+30)
Compiler 3.94user-s ± 3% 3.92user-s ± 1% ~ (p=0.185 n=30+28)
Flate 165user-ms ± 2% 165user-ms ± 4% ~ (p=0.780 n=27+29)
GoParser 209user-ms ± 2% 208user-ms ± 3% ~ (p=0.453 n=28+30)
Reflect 533user-ms ± 6% 526user-ms ± 3% ~ (p=0.057 n=30+30)
Tar 156user-ms ± 6% 154user-ms ± 6% ~ (p=0.133 n=29+30)
XML 288user-ms ± 4% 288user-ms ± 4% ~ (p=0.633 n=30+30)
name old alloc/op new alloc/op delta
Template 41.0MB ± 0% 40.9MB ± 0% -0.11% (p=0.000 n=29+29)
Unicode 32.6MB ± 0% 32.6MB ± 0% ~ (p=0.572 n=29+30)
GoTypes 122MB ± 0% 122MB ± 0% -0.10% (p=0.000 n=30+30)
Compiler 482MB ± 0% 481MB ± 0% -0.07% (p=0.000 n=30+29)
Flate 26.6MB ± 0% 26.6MB ± 0% ~ (p=0.096 n=30+30)
GoParser 32.7MB ± 0% 32.6MB ± 0% -0.06% (p=0.011 n=28+28)
Reflect 84.2MB ± 0% 84.1MB ± 0% -0.17% (p=0.000 n=29+30)
Tar 27.7MB ± 0% 27.7MB ± 0% -0.05% (p=0.032 n=27+28)
XML 44.7MB ± 0% 44.7MB ± 0% ~ (p=0.131 n=28+30)
name old allocs/op new allocs/op delta
Template 373k ± 1% 370k ± 1% -0.76% (p=0.000 n=30+30)
Unicode 325k ± 1% 325k ± 1% ~ (p=0.383 n=29+30)
GoTypes 1.16M ± 0% 1.15M ± 0% -0.75% (p=0.000 n=29+30)
Compiler 4.15M ± 0% 4.13M ± 0% -0.59% (p=0.000 n=30+29)
Flate 238k ± 1% 237k ± 1% -0.62% (p=0.000 n=30+30)
GoParser 304k ± 1% 302k ± 1% -0.64% (p=0.000 n=30+28)
Reflect 1.00M ± 0% 0.99M ± 0% -1.10% (p=0.000 n=29+30)
Tar 245k ± 1% 244k ± 1% -0.59% (p=0.000 n=27+29)
XML 391k ± 1% 389k ± 1% -0.59% (p=0.000 n=29+30)
Change-Id: Id7f456d690567c2b0a96b0d6d64de8784b6e305f
Reviewed-on: https://go-review.googlesource.com/36476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-02-03 22:40:56 -08:00
|
|
|
return "convT2Inoptr"
|
|
|
|
|
}
|
2016-03-04 22:02:27 +00:00
|
|
|
return "convT2I"
|
|
|
|
|
}
|
|
|
|
|
}
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
Fatalf("unknown conv func %c2%c", from.Tie(), to.Tie())
|
2016-03-04 22:02:27 +00:00
|
|
|
panic("unreachable")
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
// The result of walkexpr MUST be assigned back to n, e.g.
|
|
|
|
|
// n.Left = walkexpr(n.Left, init)
|
|
|
|
|
func walkexpr(n *Node, init *Nodes) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n == nil {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if init == &n.Ninit {
|
|
|
|
|
// not okay to use n->ninit when walking n,
|
|
|
|
|
// because we might replace n with some other node
|
|
|
|
|
// and would lose the init list.
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("walkexpr init == &n->ninit")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
if n.Ninit.Len() != 0 {
|
2016-03-07 22:54:46 -08:00
|
|
|
walkstmtlist(n.Ninit.Slice())
|
|
|
|
|
init.AppendNodes(&n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
lno := setlineno(n)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
if Debug['w'] > 1 {
|
|
|
|
|
Dump("walk-before", n)
|
|
|
|
|
}
|
|
|
|
|
|
2017-04-25 18:02:43 -07:00
|
|
|
if n.Typecheck() != 1 {
|
2016-08-31 15:22:36 -07:00
|
|
|
Fatalf("missed typecheck: %+v", n)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-04-25 18:14:12 -07:00
|
|
|
if n.Op == ONAME && n.Class() == PAUTOHEAP {
|
2016-10-28 11:42:40 -07:00
|
|
|
nn := nod(OIND, n.Name.Param.Heapaddr, nil)
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
nn = typecheck(nn, Erv)
|
2016-08-11 09:54:04 -04:00
|
|
|
nn = walkexpr(nn, init)
|
2017-02-27 19:56:38 +02:00
|
|
|
nn.Left.SetNonNil(true)
|
2016-08-11 09:54:04 -04:00
|
|
|
return nn
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
}
|
|
|
|
|
|
2015-09-20 23:28:34 +02:00
|
|
|
opswitch:
|
2015-02-13 14:40:36 -05:00
|
|
|
switch n.Op {
|
|
|
|
|
default:
|
|
|
|
|
Dump("walk", n)
|
2016-09-09 21:08:46 -07:00
|
|
|
Fatalf("walkexpr: switch 1 unknown op %+S", n)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-03-24 16:15:10 -07:00
|
|
|
case ONONAME, OINDREGSP, OEMPTY, OGETG:
|
|
|
|
|
|
|
|
|
|
case OTYPE, ONAME, OLITERAL:
|
|
|
|
|
// TODO(mdempsky): Just return n; see discussion on CL 38655.
|
2017-03-28 07:12:57 -07:00
|
|
|
// Perhaps refactor to use Node.mayBeShared for these instead.
|
2017-04-13 06:11:36 -07:00
|
|
|
// If these return early, make sure to still call
|
|
|
|
|
// stringsym for constant strings.
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-03-24 16:22:08 -07:00
|
|
|
case ONOT, OMINUS, OPLUS, OCOM, OREAL, OIMAG, ODOTMETH, ODOTINTER,
|
2017-03-27 14:48:24 -07:00
|
|
|
OIND, OSPTR, OITAB, OIDATA, OADDR:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-03-24 16:22:08 -07:00
|
|
|
case OEFACE, OAND, OSUB, OMUL, OLT, OLE, OGE, OGT, OADD, OOR, OXOR:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2017-03-24 16:22:08 -07:00
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ODOT:
|
|
|
|
|
usefield(n)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-03-27 14:48:24 -07:00
|
|
|
case ODOTTYPE, ODOTTYPE2:
|
|
|
|
|
n.Left = walkexpr(n.Left, init)
|
|
|
|
|
// Set up interface type addresses for back end.
|
|
|
|
|
n.Right = typename(n.Type)
|
|
|
|
|
if n.Op == ODOTTYPE {
|
|
|
|
|
n.Right.Right = typename(n.Left.Type)
|
|
|
|
|
}
|
2017-03-28 13:52:33 -07:00
|
|
|
if !n.Type.IsInterface() && !n.Left.Type.IsEmptyInterface() {
|
|
|
|
|
n.List.Set1(itabname(n.Type, n.Left.Type))
|
|
|
|
|
}
|
2017-03-27 14:48:24 -07:00
|
|
|
|
2015-02-13 14:40:36 -05:00
|
|
|
case ODOTPTR:
|
|
|
|
|
usefield(n)
|
2016-03-30 10:57:47 -07:00
|
|
|
if n.Op == ODOTPTR && n.Left.Type.Elem().Width == 0 {
|
2015-02-13 14:40:36 -05:00
|
|
|
// No actual copy will be generated, so emit an explicit nil check.
|
|
|
|
|
n.Left = cheapexpr(n.Left, init)
|
|
|
|
|
|
|
|
|
|
checknil(n.Left, init)
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case OLEN, OCAP:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// replace len(*[10]int) with 10.
|
|
|
|
|
// delayed until now to preserve side effects.
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Left.Type
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-30 15:09:25 -07:00
|
|
|
if t.IsPtr() {
|
2016-03-30 10:57:47 -07:00
|
|
|
t = t.Elem()
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-30 14:45:47 -07:00
|
|
|
if t.IsArray() {
|
2015-02-13 14:40:36 -05:00
|
|
|
safeexpr(n.Left, init)
|
2017-02-11 02:09:57 +09:00
|
|
|
nodconst(n, n.Type, t.NumElem())
|
2017-04-25 18:02:43 -07:00
|
|
|
n.SetTypecheck(1)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case OLSH, ORSH:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
|
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Left.Type
|
2017-02-27 19:56:38 +02:00
|
|
|
n.SetBounded(bounded(n.Right, 8*t.Width))
|
2015-02-17 22:13:49 -05:00
|
|
|
if Debug['m'] != 0 && n.Etype != 0 && !Isconst(n.Right, CTINT) {
|
2015-02-13 14:40:36 -05:00
|
|
|
Warn("shift bounds check elided")
|
|
|
|
|
}
|
|
|
|
|
|
2017-02-04 21:20:23 -08:00
|
|
|
case OCOMPLEX:
|
|
|
|
|
// Use results from call expression as arguments for complex.
|
|
|
|
|
if n.Left == nil && n.Right == nil {
|
2016-03-08 15:10:26 -08:00
|
|
|
n.Left = n.List.First()
|
|
|
|
|
n.Right = n.List.Second()
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
|
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case OEQ, ONE:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
|
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Disable safemode while compiling this code: the code we
|
|
|
|
|
// generate internally can refer to unsafe.Pointer.
|
|
|
|
|
// In this case it can happen if we need to generate an ==
|
|
|
|
|
// for a struct containing a reflect.Value, which itself has
|
|
|
|
|
// an unexported field of type unsafe.Pointer.
|
2015-02-23 16:07:24 -05:00
|
|
|
old_safemode := safemode
|
2016-04-13 18:37:18 -07:00
|
|
|
safemode = false
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = walkcompare(n, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
safemode = old_safemode
|
|
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case OANDAND, OOROR:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-03-27 12:34:45 -04:00
|
|
|
// cannot put side effects from n.Right on init,
|
|
|
|
|
// because they cannot run before n.Left is checked.
|
|
|
|
|
// save elsewhere and store on the eventual n.Right.
|
2016-03-07 22:54:46 -08:00
|
|
|
var ll Nodes
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right = walkexpr(n.Right, &ll)
|
|
|
|
|
n.Right = addinit(n.Right, ll.Slice())
|
2016-05-26 18:08:24 -07:00
|
|
|
n = walkinrange(n, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case OPRINT, OPRINTN:
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlist(n.List.Slice(), init)
|
2015-02-13 14:40:36 -05:00
|
|
|
n = walkprint(n, init)
|
|
|
|
|
|
|
|
|
|
case OPANIC:
|
|
|
|
|
n = mkcall("gopanic", nil, init, n.Left)
|
|
|
|
|
|
|
|
|
|
case ORECOVER:
|
2016-09-16 11:00:54 +10:00
|
|
|
n = mkcall("gorecover", n.Type, init, nod(OADDR, nodfp, nil))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-03-24 16:15:10 -07:00
|
|
|
case OCLOSUREVAR, OCFUNC:
|
2017-02-27 19:56:38 +02:00
|
|
|
n.SetAddable(true)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OCALLINTER:
|
2016-03-10 16:15:26 -05:00
|
|
|
usemethod(n)
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Left.Type
|
2016-03-08 15:10:26 -08:00
|
|
|
if n.List.Len() != 0 && n.List.First().Op == OAS {
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlist(n.List.Slice(), init)
|
2017-02-27 19:56:38 +02:00
|
|
|
ll := ascompatte(n, n.Isddd(), t.Params(), n.List.Slice(), 0, init)
|
2016-03-08 15:10:26 -08:00
|
|
|
n.List.Set(reorder1(ll))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OCALLFUNC:
|
|
|
|
|
if n.Left.Op == OCLOSURE {
|
|
|
|
|
// Transform direct call of a closure to call of a normal function.
|
|
|
|
|
// transformclosure already did all preparation work.
|
|
|
|
|
|
2015-07-23 14:17:07 -04:00
|
|
|
// Prepend captured variables to argument list.
|
2016-09-14 13:19:20 -07:00
|
|
|
n.List.Prepend(n.Left.Func.Enter.Slice()...)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-02-26 14:28:48 -08:00
|
|
|
n.Left.Func.Enter.Set(nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Replace OCLOSURE with ONAME/PFUNC.
|
2015-05-27 10:42:55 -04:00
|
|
|
n.Left = n.Left.Func.Closure.Func.Nname
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Update type of OCALLFUNC node.
|
|
|
|
|
// Output arguments had not changed, but their offsets could.
|
2016-03-17 01:47:16 -07:00
|
|
|
if n.Left.Type.Results().NumFields() == 1 {
|
2016-03-13 23:02:38 -07:00
|
|
|
n.Type = n.Left.Type.Results().Field(0).Type
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-03-08 16:31:28 -08:00
|
|
|
n.Type = n.Left.Type.Results()
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Left.Type
|
2016-03-08 15:10:26 -08:00
|
|
|
if n.List.Len() != 0 && n.List.First().Op == OAS {
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlist(n.List.Slice(), init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-02-27 19:56:38 +02:00
|
|
|
ll := ascompatte(n, n.Isddd(), t.Params(), n.List.Slice(), 0, init)
|
2016-03-08 15:10:26 -08:00
|
|
|
n.List.Set(reorder1(ll))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OCALLMETH:
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Left.Type
|
2016-03-08 15:10:26 -08:00
|
|
|
if n.List.Len() != 0 && n.List.First().Op == OAS {
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlist(n.List.Slice(), init)
|
2017-02-22 14:43:21 -08:00
|
|
|
ll := ascompatte(n, false, t.Recvs(), []*Node{n.Left.Left}, 0, init)
|
2017-02-27 19:56:38 +02:00
|
|
|
lr := ascompatte(n, n.Isddd(), t.Params(), n.List.Slice(), 0, init)
|
2016-03-08 10:26:20 -08:00
|
|
|
ll = append(ll, lr...)
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Left.Left = nil
|
2017-03-03 13:38:49 -08:00
|
|
|
updateHasCall(n.Left)
|
2016-03-08 15:10:26 -08:00
|
|
|
n.List.Set(reorder1(ll))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OAS:
|
2016-03-07 22:54:46 -08:00
|
|
|
init.AppendNodes(&n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Left = safeexpr(n.Left, init)
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
if oaslit(n, init) {
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-10-18 10:26:28 -04:00
|
|
|
if n.Right == nil {
|
|
|
|
|
// TODO(austin): Check all "implicit zeroing"
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-12-19 10:30:44 -08:00
|
|
|
if !instrumenting && iszero(n.Right) {
|
2016-12-19 08:19:50 -08:00
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-13 14:40:36 -05:00
|
|
|
switch n.Right.Op {
|
|
|
|
|
default:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ORECV:
|
2015-03-27 12:34:45 -04:00
|
|
|
// x = <-c; n.Left is x, n.Right.Left is c.
|
2015-03-20 00:06:10 -04:00
|
|
|
// orderstmt made sure x is addressable.
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right.Left = walkexpr(n.Right.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
n1 := nod(OADDR, n.Left, nil)
|
2015-02-23 16:07:24 -05:00
|
|
|
r := n.Right.Left // the channel
|
2017-03-18 15:55:41 +00:00
|
|
|
n = mkcall1(chanfn("chanrecv1", 2, r.Type), nil, init, r, n1)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = walkexpr(n, init)
|
2015-09-20 23:28:34 +02:00
|
|
|
break opswitch
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
|
|
|
|
|
case OAPPEND:
|
|
|
|
|
// x = append(...)
|
|
|
|
|
r := n.Right
|
2017-02-27 19:56:38 +02:00
|
|
|
if r.Type.Elem().NotInHeap() {
|
cmd/compile: add go:notinheap type pragma
This adds a //go:notinheap pragma for declarations of types that must
not be heap allocated. We ensure these rules by disallowing new(T),
make([]T), append([]T), or implicit allocation of T, by disallowing
conversions to notinheap types, and by propagating notinheap to any
struct or array that contains notinheap elements.
The utility of this pragma is that we can eliminate write barriers for
writes to pointers to go:notinheap types, since the write barrier is
guaranteed to be a no-op. This will let us mark several scheduler and
memory allocator structures as go:notinheap, which will let us
disallow write barriers in the scheduler and memory allocator much
more thoroughly and also eliminate some problematic hybrid write
barriers.
This also makes go:nowritebarrierrec and go:yeswritebarrierrec much
more powerful. Currently we use go:nowritebarrier all over the place,
but it's almost never what you actually want: when write barriers are
illegal, they're typically illegal for a whole dynamic scope. Partly
this is because go:nowritebarrier has been around longer, but it's
also because go:nowritebarrierrec couldn't be used in situations that
had no-op write barriers or where some nested scope did allow write
barriers. go:notinheap eliminates many no-op write barriers and
go:yeswritebarrierrec makes it possible to opt back in to write
barriers, so these two changes will let us use go:nowritebarrierrec
far more liberally.
This updates #13386, which is about controlling pointers from non-GC'd
memory to GC'd memory. That would require some additional pragma (or
pragmas), but could build on this pragma.
Change-Id: I6314f8f4181535dd166887c9ec239977b54940bd
Reviewed-on: https://go-review.googlesource.com/30939
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-10-11 22:53:27 -04:00
|
|
|
yyerror("%v is go:notinheap; heap allocation disallowed", r.Type.Elem())
|
|
|
|
|
}
|
2017-02-27 19:56:38 +02:00
|
|
|
if r.Isddd() {
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
r = appendslice(r, init) // also works for append(slice, string).
|
|
|
|
|
} else {
|
|
|
|
|
r = walkappend(r, init, n)
|
|
|
|
|
}
|
|
|
|
|
n.Right = r
|
|
|
|
|
if r.Op == OAPPEND {
|
|
|
|
|
// Left in place for back end.
|
|
|
|
|
// Do not add a new write barrier.
|
2017-03-27 14:48:24 -07:00
|
|
|
// Set up address of type for back end.
|
|
|
|
|
r.Left = typename(r.Type.Elem())
|
2015-09-20 23:28:34 +02:00
|
|
|
break opswitch
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
}
|
|
|
|
|
// Otherwise, lowered for race detector.
|
|
|
|
|
// Treat as ordinary assignment.
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if n.Left != nil && n.Right != nil {
|
2016-03-26 08:17:43 -07:00
|
|
|
n = convas(n, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
case OAS2:
|
2016-03-07 22:54:46 -08:00
|
|
|
init.AppendNodes(&n.Ninit)
|
|
|
|
|
walkexprlistsafe(n.List.Slice(), init)
|
|
|
|
|
walkexprlistsafe(n.Rlist.Slice(), init)
|
2016-03-08 10:26:20 -08:00
|
|
|
ll := ascompatee(OAS, n.List.Slice(), n.Rlist.Slice(), init)
|
2015-02-13 14:40:36 -05:00
|
|
|
ll = reorder3(ll)
|
|
|
|
|
n = liststmt(ll)
|
|
|
|
|
|
2016-10-06 15:43:47 -04:00
|
|
|
// a,b,... = fn()
|
2015-02-13 14:40:36 -05:00
|
|
|
case OAS2FUNC:
|
2016-03-07 22:54:46 -08:00
|
|
|
init.AppendNodes(&n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
r := n.Rlist.First()
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlistsafe(n.List.Slice(), init)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = walkexpr(r, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-10-06 15:43:47 -04:00
|
|
|
if isIntrinsicCall(r) {
|
|
|
|
|
n.Rlist.Set1(r)
|
|
|
|
|
break
|
|
|
|
|
}
|
2016-10-27 09:23:48 -07:00
|
|
|
init.Append(r)
|
2016-10-06 15:43:47 -04:00
|
|
|
|
2016-10-27 09:23:48 -07:00
|
|
|
ll := ascompatet(n.Op, n.List, r.Type)
|
|
|
|
|
n = liststmt(ll)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-24 21:38:58 +02:00
|
|
|
// x, y = <-c
|
2015-02-13 14:40:36 -05:00
|
|
|
// orderstmt made sure x is addressable.
|
|
|
|
|
case OAS2RECV:
|
2016-03-07 22:54:46 -08:00
|
|
|
init.AppendNodes(&n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
r := n.Rlist.First()
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlistsafe(n.List.Slice(), init)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r.Left = walkexpr(r.Left, init)
|
2015-02-23 16:07:24 -05:00
|
|
|
var n1 *Node
|
2016-03-08 15:10:26 -08:00
|
|
|
if isblank(n.List.First()) {
|
2015-02-13 14:40:36 -05:00
|
|
|
n1 = nodnil()
|
|
|
|
|
} else {
|
2016-09-16 11:00:54 +10:00
|
|
|
n1 = nod(OADDR, n.List.First(), nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
n1.Etype = 1 // addr does not escape
|
2015-02-23 16:07:24 -05:00
|
|
|
fn := chanfn("chanrecv2", 2, r.Left.Type)
|
2016-08-26 10:50:12 -07:00
|
|
|
ok := n.List.Second()
|
2017-03-18 15:55:41 +00:00
|
|
|
call := mkcall1(fn, ok.Type, init, r.Left, n1)
|
2016-09-16 11:00:54 +10:00
|
|
|
n = nod(OAS, ok, call)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = typecheck(n, Etop)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-03-24 16:22:08 -07:00
|
|
|
// a,b = m[i]
|
2015-02-13 14:40:36 -05:00
|
|
|
case OAS2MAPR:
|
2016-03-07 22:54:46 -08:00
|
|
|
init.AppendNodes(&n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
r := n.Rlist.First()
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlistsafe(n.List.Slice(), init)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r.Left = walkexpr(r.Left, init)
|
|
|
|
|
r.Right = walkexpr(r.Right, init)
|
2015-02-23 16:07:24 -05:00
|
|
|
t := r.Left.Type
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-03-14 11:11:28 -07:00
|
|
|
fast := mapfast(t)
|
2015-02-23 16:07:24 -05:00
|
|
|
var key *Node
|
2017-03-14 11:11:28 -07:00
|
|
|
if fast != mapslow {
|
2015-02-13 14:40:36 -05:00
|
|
|
// fast versions take key by value
|
|
|
|
|
key = r.Right
|
|
|
|
|
} else {
|
|
|
|
|
// standard version takes key by reference
|
|
|
|
|
// orderexpr made sure key is addressable.
|
2016-09-16 11:00:54 +10:00
|
|
|
key = nod(OADDR, r.Right, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// from:
|
|
|
|
|
// a,b = m[i]
|
|
|
|
|
// to:
|
|
|
|
|
// var,b = mapaccess2*(t, m, i)
|
|
|
|
|
// a = *var
|
2016-03-08 15:10:26 -08:00
|
|
|
a := n.List.First()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-04-19 08:31:04 -07:00
|
|
|
if w := t.Val().Width; w <= 1024 { // 1024 must match ../../../../runtime/hashmap.go:maxZero
|
2017-03-14 11:11:28 -07:00
|
|
|
fn := mapfn(mapaccess2[fast], t)
|
2016-04-19 08:31:04 -07:00
|
|
|
r = mkcall1(fn, fn.Type.Results(), init, typename(t), r.Left, key)
|
|
|
|
|
} else {
|
|
|
|
|
fn := mapfn("mapaccess2_fat", t)
|
|
|
|
|
z := zeroaddr(w)
|
|
|
|
|
r = mkcall1(fn, fn.Type.Results(), init, typename(t), r.Left, key, z)
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// mapaccess2* returns a typed bool, but due to spec changes,
|
|
|
|
|
// the boolean result of i.(T) is now untyped so we make it the
|
|
|
|
|
// same type as the variable on the lhs.
|
2016-08-26 10:50:12 -07:00
|
|
|
if ok := n.List.Second(); !isblank(ok) && ok.Type.IsBoolean() {
|
|
|
|
|
r.Type.Field(1).Type = ok.Type
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-10 10:13:42 -08:00
|
|
|
n.Rlist.Set1(r)
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Op = OAS2FUNC
|
|
|
|
|
|
|
|
|
|
// don't generate a = *var if a is _
|
|
|
|
|
if !isblank(a) {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
var_ := temp(types.NewPtr(t.Val()))
|
2017-04-25 18:02:43 -07:00
|
|
|
var_.SetTypecheck(1)
|
2017-02-27 19:56:38 +02:00
|
|
|
var_.SetNonNil(true) // mapaccess always returns a non-nil pointer
|
2017-02-19 15:57:58 +01:00
|
|
|
n.List.SetFirst(var_)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = walkexpr(n, init)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(n)
|
2016-09-16 11:00:54 +10:00
|
|
|
n = nod(OAS, a, nod(OIND, var_, nil))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = typecheck(n, Etop)
|
|
|
|
|
n = walkexpr(n, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ODELETE:
|
2016-03-07 22:54:46 -08:00
|
|
|
init.AppendNodes(&n.Ninit)
|
2016-03-08 15:10:26 -08:00
|
|
|
map_ := n.List.First()
|
|
|
|
|
key := n.List.Second()
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
map_ = walkexpr(map_, init)
|
|
|
|
|
key = walkexpr(key, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
t := map_.Type
|
2017-03-14 11:11:28 -07:00
|
|
|
fast := mapfast(t)
|
|
|
|
|
if fast == mapslow {
|
|
|
|
|
// orderstmt made sure key is addressable.
|
|
|
|
|
key = nod(OADDR, key, nil)
|
|
|
|
|
}
|
|
|
|
|
n = mkcall1(mapfndel(mapdelete[fast], t), nil, init, typename(t), map_, key)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OAS2DOTTYPE:
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlistsafe(n.List.Slice(), init)
|
2017-03-27 14:48:24 -07:00
|
|
|
n.Rlist.SetFirst(walkexpr(n.Rlist.First(), init))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OCONVIFACE:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-18 16:20:20 -07:00
|
|
|
// Optimize convT2E or convT2I as a two-word copy when T is pointer-shaped.
|
|
|
|
|
if isdirectiface(n.Left.Type) {
|
|
|
|
|
var t *Node
|
2016-04-01 13:36:24 -07:00
|
|
|
if n.Type.IsEmptyInterface() {
|
2016-03-18 16:20:20 -07:00
|
|
|
t = typename(n.Left.Type)
|
|
|
|
|
} else {
|
|
|
|
|
t = itabname(n.Left.Type, n.Type)
|
|
|
|
|
}
|
2016-09-16 11:00:54 +10:00
|
|
|
l := nod(OEFACE, t, n.Left)
|
2015-02-13 14:40:36 -05:00
|
|
|
l.Type = n.Type
|
2017-04-25 18:02:43 -07:00
|
|
|
l.SetTypecheck(n.Typecheck())
|
2015-02-13 14:40:36 -05:00
|
|
|
n = l
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: convert constants to interfaces without allocating
The order pass is responsible for ensuring that
values passed to runtime functions, including
convT2E/convT2I, are addressable.
Prior to this CL, this was always accomplished
by creating a temp, which frequently escaped to
the heap, causing allocations, perhaps most
notably in code like:
fmt.Println(1, 2, 3) // allocates three times
None of the runtime routines modify the contents
of the pointers they receive, so in the case of
constants, instead of creating a temp value,
we can create a static value.
(Marking the static value as read-only provides
protection against accidental attempts by the runtime
to modify the constant data.)
This improves code generation for code like:
panic("abc")
c <- 2 // c is a chan int
which can now simply refer to "abc" and 2,
rather than going by way of a temporary.
It also allows us to optimize convT2E/convT2I,
by recognizing static readonly values
and directly constructing the interface.
This CL adds ~0.5% to binary size, despite
decreasing the size of many functions,
because it also adds many static symbols.
This binary size regression could be recovered in
future (but currently unplanned) work.
There is a lot of content-duplication in these
symbols; this statement generates six new symbols,
three containing an int 1 and three containing
a pointer to the string "a":
fmt.Println(1, 1, 1, "a", "a", "a")
These symbols could be made content-addressable.
Furthermore, these symbols are small, so the
alignment and naming overhead is large.
As with the go.strings section, these symbols
could be hidden and have their alignment reduced.
The changes to test/live.go make it impossible
(at least with current optimization techniques)
to place the values being passed to the runtime
in static symbols, preserving autotmp creation.
Fixes #18704
Benchmarks from fmt and go-kit's logging package:
github.com/go-kit/kit/log
name old time/op new time/op delta
JSONLoggerSimple-8 1.91µs ± 2% 2.11µs ±22% ~ (p=1.000 n=9+10)
JSONLoggerContextual-8 2.60µs ± 6% 2.43µs ± 2% -6.29% (p=0.000 n=9+10)
Discard-8 101ns ± 2% 34ns ±14% -66.33% (p=0.000 n=10+9)
OneWith-8 161ns ± 1% 102ns ±16% -36.78% (p=0.000 n=10+10)
TwoWith-8 175ns ± 3% 106ns ± 7% -39.36% (p=0.000 n=10+9)
TenWith-8 293ns ± 3% 227ns ±15% -22.44% (p=0.000 n=9+10)
LogfmtLoggerSimple-8 704ns ± 2% 608ns ± 2% -13.65% (p=0.000 n=10+9)
LogfmtLoggerContextual-8 962ns ± 1% 860ns ±17% -10.57% (p=0.003 n=9+10)
NopLoggerSimple-8 188ns ± 1% 120ns ± 1% -36.39% (p=0.000 n=9+10)
NopLoggerContextual-8 379ns ± 1% 243ns ± 0% -35.77% (p=0.000 n=9+10)
ValueBindingTimestamp-8 577ns ± 1% 499ns ± 1% -13.51% (p=0.000 n=10+10)
ValueBindingCaller-8 898ns ± 2% 844ns ± 2% -6.00% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
JSONLoggerSimple-8 904B ± 0% 872B ± 0% -3.54% (p=0.000 n=10+10)
JSONLoggerContextual-8 1.20kB ± 0% 1.14kB ± 0% -5.33% (p=0.000 n=10+10)
Discard-8 64.0B ± 0% 32.0B ± 0% -50.00% (p=0.000 n=10+10)
OneWith-8 96.0B ± 0% 64.0B ± 0% -33.33% (p=0.000 n=10+10)
TwoWith-8 160B ± 0% 128B ± 0% -20.00% (p=0.000 n=10+10)
TenWith-8 672B ± 0% 640B ± 0% -4.76% (p=0.000 n=10+10)
LogfmtLoggerSimple-8 128B ± 0% 96B ± 0% -25.00% (p=0.000 n=10+10)
LogfmtLoggerContextual-8 304B ± 0% 240B ± 0% -21.05% (p=0.000 n=10+10)
NopLoggerSimple-8 128B ± 0% 96B ± 0% -25.00% (p=0.000 n=10+10)
NopLoggerContextual-8 304B ± 0% 240B ± 0% -21.05% (p=0.000 n=10+10)
ValueBindingTimestamp-8 159B ± 0% 127B ± 0% -20.13% (p=0.000 n=10+10)
ValueBindingCaller-8 112B ± 0% 80B ± 0% -28.57% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
JSONLoggerSimple-8 19.0 ± 0% 17.0 ± 0% -10.53% (p=0.000 n=10+10)
JSONLoggerContextual-8 25.0 ± 0% 21.0 ± 0% -16.00% (p=0.000 n=10+10)
Discard-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
OneWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
TwoWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
TenWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
LogfmtLoggerSimple-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
LogfmtLoggerContextual-8 7.00 ± 0% 3.00 ± 0% -57.14% (p=0.000 n=10+10)
NopLoggerSimple-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
NopLoggerContextual-8 7.00 ± 0% 3.00 ± 0% -57.14% (p=0.000 n=10+10)
ValueBindingTimestamp-8 5.00 ± 0% 3.00 ± 0% -40.00% (p=0.000 n=10+10)
ValueBindingCaller-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
fmt
name old time/op new time/op delta
SprintfPadding-8 88.9ns ± 3% 79.1ns ± 1% -11.09% (p=0.000 n=10+7)
SprintfEmpty-8 12.6ns ± 3% 12.8ns ± 3% ~ (p=0.136 n=10+10)
SprintfString-8 38.7ns ± 5% 26.9ns ± 6% -30.65% (p=0.000 n=10+10)
SprintfTruncateString-8 56.7ns ± 2% 47.0ns ± 3% -17.05% (p=0.000 n=10+10)
SprintfQuoteString-8 164ns ± 2% 153ns ± 2% -7.01% (p=0.000 n=10+10)
SprintfInt-8 38.9ns ±15% 26.5ns ± 2% -31.93% (p=0.000 n=10+9)
SprintfIntInt-8 60.3ns ± 9% 38.2ns ± 1% -36.67% (p=0.000 n=10+8)
SprintfPrefixedInt-8 58.6ns ±13% 51.2ns ±11% -12.66% (p=0.001 n=10+10)
SprintfFloat-8 71.4ns ± 3% 64.2ns ± 3% -10.08% (p=0.000 n=8+10)
SprintfComplex-8 175ns ± 3% 159ns ± 2% -9.03% (p=0.000 n=10+10)
SprintfBoolean-8 33.5ns ± 4% 25.7ns ± 5% -23.28% (p=0.000 n=10+10)
SprintfHexString-8 65.3ns ± 3% 51.7ns ± 5% -20.86% (p=0.000 n=10+9)
SprintfHexBytes-8 67.2ns ± 5% 67.9ns ± 4% ~ (p=0.383 n=10+10)
SprintfBytes-8 129ns ± 7% 124ns ± 7% ~ (p=0.074 n=9+10)
SprintfStringer-8 127ns ± 4% 126ns ± 8% ~ (p=0.506 n=9+10)
SprintfStructure-8 357ns ± 3% 359ns ± 3% ~ (p=0.469 n=10+10)
ManyArgs-8 203ns ± 6% 126ns ± 3% -37.94% (p=0.000 n=10+10)
FprintInt-8 119ns ±10% 74ns ± 3% -37.54% (p=0.000 n=10+10)
FprintfBytes-8 122ns ± 4% 120ns ± 3% ~ (p=0.124 n=10+10)
FprintIntNoAlloc-8 78.2ns ± 5% 74.1ns ± 3% -5.28% (p=0.000 n=10+10)
ScanInts-8 349µs ± 1% 349µs ± 0% ~ (p=0.606 n=9+8)
ScanRecursiveInt-8 43.8ms ± 7% 40.1ms ± 2% -8.42% (p=0.000 n=10+10)
ScanRecursiveIntReaderWrapper-8 43.5ms ± 4% 40.4ms ± 2% -7.16% (p=0.000 n=10+9)
name old alloc/op new alloc/op delta
SprintfPadding-8 24.0B ± 0% 16.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfEmpty-8 0.00B 0.00B ~ (all equal)
SprintfString-8 21.0B ± 0% 5.0B ± 0% -76.19% (p=0.000 n=10+10)
SprintfTruncateString-8 32.0B ± 0% 16.0B ± 0% -50.00% (p=0.000 n=10+10)
SprintfQuoteString-8 48.0B ± 0% 32.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfInt-8 16.0B ± 0% 1.0B ± 0% -93.75% (p=0.000 n=10+10)
SprintfIntInt-8 24.0B ± 0% 3.0B ± 0% -87.50% (p=0.000 n=10+10)
SprintfPrefixedInt-8 72.0B ± 0% 64.0B ± 0% -11.11% (p=0.000 n=10+10)
SprintfFloat-8 16.0B ± 0% 8.0B ± 0% -50.00% (p=0.000 n=10+10)
SprintfComplex-8 48.0B ± 0% 32.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfBoolean-8 8.00B ± 0% 4.00B ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexString-8 96.0B ± 0% 80.0B ± 0% -16.67% (p=0.000 n=10+10)
SprintfHexBytes-8 112B ± 0% 112B ± 0% ~ (all equal)
SprintfBytes-8 96.0B ± 0% 96.0B ± 0% ~ (all equal)
SprintfStringer-8 32.0B ± 0% 32.0B ± 0% ~ (all equal)
SprintfStructure-8 256B ± 0% 256B ± 0% ~ (all equal)
ManyArgs-8 80.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10)
FprintInt-8 8.00B ± 0% 0.00B -100.00% (p=0.000 n=10+10)
FprintfBytes-8 32.0B ± 0% 32.0B ± 0% ~ (all equal)
FprintIntNoAlloc-8 0.00B 0.00B ~ (all equal)
ScanInts-8 15.2kB ± 0% 15.2kB ± 0% ~ (p=0.248 n=9+10)
ScanRecursiveInt-8 21.6kB ± 0% 21.6kB ± 0% ~ (all equal)
ScanRecursiveIntReaderWrapper-8 21.7kB ± 0% 21.7kB ± 0% ~ (all equal)
name old allocs/op new allocs/op delta
SprintfPadding-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfEmpty-8 0.00 0.00 ~ (all equal)
SprintfString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfTruncateString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfQuoteString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfInt-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfIntInt-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
SprintfPrefixedInt-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfFloat-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfComplex-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfBoolean-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexBytes-8 2.00 ± 0% 2.00 ± 0% ~ (all equal)
SprintfBytes-8 2.00 ± 0% 2.00 ± 0% ~ (all equal)
SprintfStringer-8 4.00 ± 0% 4.00 ± 0% ~ (all equal)
SprintfStructure-8 7.00 ± 0% 7.00 ± 0% ~ (all equal)
ManyArgs-8 8.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
FprintInt-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
FprintfBytes-8 1.00 ± 0% 1.00 ± 0% ~ (all equal)
FprintIntNoAlloc-8 0.00 0.00 ~ (all equal)
ScanInts-8 1.60k ± 0% 1.60k ± 0% ~ (all equal)
ScanRecursiveInt-8 1.71k ± 0% 1.71k ± 0% ~ (all equal)
ScanRecursiveIntReaderWrapper-8 1.71k ± 0% 1.71k ± 0% ~ (all equal)
Change-Id: I7ba72a25fea4140a0ba40a9f443103ed87cc69b5
Reviewed-on: https://go-review.googlesource.com/35554
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-01-21 13:41:06 -08:00
|
|
|
|
2017-01-21 19:52:09 -08:00
|
|
|
if staticbytes == nil {
|
2017-03-30 13:19:18 -07:00
|
|
|
staticbytes = newname(Runtimepkg.Lookup("staticbytes"))
|
2017-04-25 18:14:12 -07:00
|
|
|
staticbytes.SetClass(PEXTERN)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
staticbytes.Type = types.NewArray(types.Types[TUINT8], 256)
|
2017-03-30 13:19:18 -07:00
|
|
|
zerobase = newname(Runtimepkg.Lookup("zerobase"))
|
2017-04-25 18:14:12 -07:00
|
|
|
zerobase.SetClass(PEXTERN)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
zerobase.Type = types.Types[TUINTPTR]
|
2017-01-21 19:52:09 -08:00
|
|
|
}
|
|
|
|
|
|
2017-01-22 10:17:12 -08:00
|
|
|
// Optimize convT2{E,I} for many cases in which T is not pointer-shaped,
|
|
|
|
|
// by using an existing addressable value identical to n.Left
|
|
|
|
|
// or creating one on the stack.
|
cmd/compile: convert constants to interfaces without allocating
The order pass is responsible for ensuring that
values passed to runtime functions, including
convT2E/convT2I, are addressable.
Prior to this CL, this was always accomplished
by creating a temp, which frequently escaped to
the heap, causing allocations, perhaps most
notably in code like:
fmt.Println(1, 2, 3) // allocates three times
None of the runtime routines modify the contents
of the pointers they receive, so in the case of
constants, instead of creating a temp value,
we can create a static value.
(Marking the static value as read-only provides
protection against accidental attempts by the runtime
to modify the constant data.)
This improves code generation for code like:
panic("abc")
c <- 2 // c is a chan int
which can now simply refer to "abc" and 2,
rather than going by way of a temporary.
It also allows us to optimize convT2E/convT2I,
by recognizing static readonly values
and directly constructing the interface.
This CL adds ~0.5% to binary size, despite
decreasing the size of many functions,
because it also adds many static symbols.
This binary size regression could be recovered in
future (but currently unplanned) work.
There is a lot of content-duplication in these
symbols; this statement generates six new symbols,
three containing an int 1 and three containing
a pointer to the string "a":
fmt.Println(1, 1, 1, "a", "a", "a")
These symbols could be made content-addressable.
Furthermore, these symbols are small, so the
alignment and naming overhead is large.
As with the go.strings section, these symbols
could be hidden and have their alignment reduced.
The changes to test/live.go make it impossible
(at least with current optimization techniques)
to place the values being passed to the runtime
in static symbols, preserving autotmp creation.
Fixes #18704
Benchmarks from fmt and go-kit's logging package:
github.com/go-kit/kit/log
name old time/op new time/op delta
JSONLoggerSimple-8 1.91µs ± 2% 2.11µs ±22% ~ (p=1.000 n=9+10)
JSONLoggerContextual-8 2.60µs ± 6% 2.43µs ± 2% -6.29% (p=0.000 n=9+10)
Discard-8 101ns ± 2% 34ns ±14% -66.33% (p=0.000 n=10+9)
OneWith-8 161ns ± 1% 102ns ±16% -36.78% (p=0.000 n=10+10)
TwoWith-8 175ns ± 3% 106ns ± 7% -39.36% (p=0.000 n=10+9)
TenWith-8 293ns ± 3% 227ns ±15% -22.44% (p=0.000 n=9+10)
LogfmtLoggerSimple-8 704ns ± 2% 608ns ± 2% -13.65% (p=0.000 n=10+9)
LogfmtLoggerContextual-8 962ns ± 1% 860ns ±17% -10.57% (p=0.003 n=9+10)
NopLoggerSimple-8 188ns ± 1% 120ns ± 1% -36.39% (p=0.000 n=9+10)
NopLoggerContextual-8 379ns ± 1% 243ns ± 0% -35.77% (p=0.000 n=9+10)
ValueBindingTimestamp-8 577ns ± 1% 499ns ± 1% -13.51% (p=0.000 n=10+10)
ValueBindingCaller-8 898ns ± 2% 844ns ± 2% -6.00% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
JSONLoggerSimple-8 904B ± 0% 872B ± 0% -3.54% (p=0.000 n=10+10)
JSONLoggerContextual-8 1.20kB ± 0% 1.14kB ± 0% -5.33% (p=0.000 n=10+10)
Discard-8 64.0B ± 0% 32.0B ± 0% -50.00% (p=0.000 n=10+10)
OneWith-8 96.0B ± 0% 64.0B ± 0% -33.33% (p=0.000 n=10+10)
TwoWith-8 160B ± 0% 128B ± 0% -20.00% (p=0.000 n=10+10)
TenWith-8 672B ± 0% 640B ± 0% -4.76% (p=0.000 n=10+10)
LogfmtLoggerSimple-8 128B ± 0% 96B ± 0% -25.00% (p=0.000 n=10+10)
LogfmtLoggerContextual-8 304B ± 0% 240B ± 0% -21.05% (p=0.000 n=10+10)
NopLoggerSimple-8 128B ± 0% 96B ± 0% -25.00% (p=0.000 n=10+10)
NopLoggerContextual-8 304B ± 0% 240B ± 0% -21.05% (p=0.000 n=10+10)
ValueBindingTimestamp-8 159B ± 0% 127B ± 0% -20.13% (p=0.000 n=10+10)
ValueBindingCaller-8 112B ± 0% 80B ± 0% -28.57% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
JSONLoggerSimple-8 19.0 ± 0% 17.0 ± 0% -10.53% (p=0.000 n=10+10)
JSONLoggerContextual-8 25.0 ± 0% 21.0 ± 0% -16.00% (p=0.000 n=10+10)
Discard-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
OneWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
TwoWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
TenWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
LogfmtLoggerSimple-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
LogfmtLoggerContextual-8 7.00 ± 0% 3.00 ± 0% -57.14% (p=0.000 n=10+10)
NopLoggerSimple-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
NopLoggerContextual-8 7.00 ± 0% 3.00 ± 0% -57.14% (p=0.000 n=10+10)
ValueBindingTimestamp-8 5.00 ± 0% 3.00 ± 0% -40.00% (p=0.000 n=10+10)
ValueBindingCaller-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
fmt
name old time/op new time/op delta
SprintfPadding-8 88.9ns ± 3% 79.1ns ± 1% -11.09% (p=0.000 n=10+7)
SprintfEmpty-8 12.6ns ± 3% 12.8ns ± 3% ~ (p=0.136 n=10+10)
SprintfString-8 38.7ns ± 5% 26.9ns ± 6% -30.65% (p=0.000 n=10+10)
SprintfTruncateString-8 56.7ns ± 2% 47.0ns ± 3% -17.05% (p=0.000 n=10+10)
SprintfQuoteString-8 164ns ± 2% 153ns ± 2% -7.01% (p=0.000 n=10+10)
SprintfInt-8 38.9ns ±15% 26.5ns ± 2% -31.93% (p=0.000 n=10+9)
SprintfIntInt-8 60.3ns ± 9% 38.2ns ± 1% -36.67% (p=0.000 n=10+8)
SprintfPrefixedInt-8 58.6ns ±13% 51.2ns ±11% -12.66% (p=0.001 n=10+10)
SprintfFloat-8 71.4ns ± 3% 64.2ns ± 3% -10.08% (p=0.000 n=8+10)
SprintfComplex-8 175ns ± 3% 159ns ± 2% -9.03% (p=0.000 n=10+10)
SprintfBoolean-8 33.5ns ± 4% 25.7ns ± 5% -23.28% (p=0.000 n=10+10)
SprintfHexString-8 65.3ns ± 3% 51.7ns ± 5% -20.86% (p=0.000 n=10+9)
SprintfHexBytes-8 67.2ns ± 5% 67.9ns ± 4% ~ (p=0.383 n=10+10)
SprintfBytes-8 129ns ± 7% 124ns ± 7% ~ (p=0.074 n=9+10)
SprintfStringer-8 127ns ± 4% 126ns ± 8% ~ (p=0.506 n=9+10)
SprintfStructure-8 357ns ± 3% 359ns ± 3% ~ (p=0.469 n=10+10)
ManyArgs-8 203ns ± 6% 126ns ± 3% -37.94% (p=0.000 n=10+10)
FprintInt-8 119ns ±10% 74ns ± 3% -37.54% (p=0.000 n=10+10)
FprintfBytes-8 122ns ± 4% 120ns ± 3% ~ (p=0.124 n=10+10)
FprintIntNoAlloc-8 78.2ns ± 5% 74.1ns ± 3% -5.28% (p=0.000 n=10+10)
ScanInts-8 349µs ± 1% 349µs ± 0% ~ (p=0.606 n=9+8)
ScanRecursiveInt-8 43.8ms ± 7% 40.1ms ± 2% -8.42% (p=0.000 n=10+10)
ScanRecursiveIntReaderWrapper-8 43.5ms ± 4% 40.4ms ± 2% -7.16% (p=0.000 n=10+9)
name old alloc/op new alloc/op delta
SprintfPadding-8 24.0B ± 0% 16.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfEmpty-8 0.00B 0.00B ~ (all equal)
SprintfString-8 21.0B ± 0% 5.0B ± 0% -76.19% (p=0.000 n=10+10)
SprintfTruncateString-8 32.0B ± 0% 16.0B ± 0% -50.00% (p=0.000 n=10+10)
SprintfQuoteString-8 48.0B ± 0% 32.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfInt-8 16.0B ± 0% 1.0B ± 0% -93.75% (p=0.000 n=10+10)
SprintfIntInt-8 24.0B ± 0% 3.0B ± 0% -87.50% (p=0.000 n=10+10)
SprintfPrefixedInt-8 72.0B ± 0% 64.0B ± 0% -11.11% (p=0.000 n=10+10)
SprintfFloat-8 16.0B ± 0% 8.0B ± 0% -50.00% (p=0.000 n=10+10)
SprintfComplex-8 48.0B ± 0% 32.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfBoolean-8 8.00B ± 0% 4.00B ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexString-8 96.0B ± 0% 80.0B ± 0% -16.67% (p=0.000 n=10+10)
SprintfHexBytes-8 112B ± 0% 112B ± 0% ~ (all equal)
SprintfBytes-8 96.0B ± 0% 96.0B ± 0% ~ (all equal)
SprintfStringer-8 32.0B ± 0% 32.0B ± 0% ~ (all equal)
SprintfStructure-8 256B ± 0% 256B ± 0% ~ (all equal)
ManyArgs-8 80.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10)
FprintInt-8 8.00B ± 0% 0.00B -100.00% (p=0.000 n=10+10)
FprintfBytes-8 32.0B ± 0% 32.0B ± 0% ~ (all equal)
FprintIntNoAlloc-8 0.00B 0.00B ~ (all equal)
ScanInts-8 15.2kB ± 0% 15.2kB ± 0% ~ (p=0.248 n=9+10)
ScanRecursiveInt-8 21.6kB ± 0% 21.6kB ± 0% ~ (all equal)
ScanRecursiveIntReaderWrapper-8 21.7kB ± 0% 21.7kB ± 0% ~ (all equal)
name old allocs/op new allocs/op delta
SprintfPadding-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfEmpty-8 0.00 0.00 ~ (all equal)
SprintfString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfTruncateString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfQuoteString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfInt-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfIntInt-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
SprintfPrefixedInt-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfFloat-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfComplex-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfBoolean-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexBytes-8 2.00 ± 0% 2.00 ± 0% ~ (all equal)
SprintfBytes-8 2.00 ± 0% 2.00 ± 0% ~ (all equal)
SprintfStringer-8 4.00 ± 0% 4.00 ± 0% ~ (all equal)
SprintfStructure-8 7.00 ± 0% 7.00 ± 0% ~ (all equal)
ManyArgs-8 8.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
FprintInt-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
FprintfBytes-8 1.00 ± 0% 1.00 ± 0% ~ (all equal)
FprintIntNoAlloc-8 0.00 0.00 ~ (all equal)
ScanInts-8 1.60k ± 0% 1.60k ± 0% ~ (all equal)
ScanRecursiveInt-8 1.71k ± 0% 1.71k ± 0% ~ (all equal)
ScanRecursiveIntReaderWrapper-8 1.71k ± 0% 1.71k ± 0% ~ (all equal)
Change-Id: I7ba72a25fea4140a0ba40a9f443103ed87cc69b5
Reviewed-on: https://go-review.googlesource.com/35554
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-01-21 13:41:06 -08:00
|
|
|
var value *Node
|
|
|
|
|
switch {
|
2017-01-22 10:17:12 -08:00
|
|
|
case n.Left.Type.Size() == 0:
|
|
|
|
|
// n.Left is zero-sized. Use zerobase.
|
2017-02-22 21:19:29 -08:00
|
|
|
cheapexpr(n.Left, init) // Evaluate n.Left for side-effects. See issue 19246.
|
2017-01-22 10:17:12 -08:00
|
|
|
value = zerobase
|
2017-01-21 19:52:09 -08:00
|
|
|
case n.Left.Type.IsBoolean() || (n.Left.Type.Size() == 1 && n.Left.Type.IsInteger()):
|
2017-01-22 10:17:12 -08:00
|
|
|
// n.Left is a bool/byte. Use staticbytes[n.Left].
|
2017-02-24 17:21:54 -05:00
|
|
|
n.Left = cheapexpr(n.Left, init)
|
2017-01-21 19:52:09 -08:00
|
|
|
value = nod(OINDEX, staticbytes, byteindex(n.Left))
|
2017-02-27 19:56:38 +02:00
|
|
|
value.SetBounded(true)
|
2017-04-25 18:14:12 -07:00
|
|
|
case n.Left.Class() == PEXTERN && n.Left.Name != nil && n.Left.Name.Readonly():
|
2017-01-22 10:17:12 -08:00
|
|
|
// n.Left is a readonly global; use it directly.
|
|
|
|
|
value = n.Left
|
|
|
|
|
case !n.Left.Type.IsInterface() && n.Esc == EscNone && n.Left.Type.Width <= 1024:
|
|
|
|
|
// n.Left does not escape. Use a stack temporary initialized to n.Left.
|
|
|
|
|
value = temp(n.Left.Type)
|
|
|
|
|
init.Append(typecheck(nod(OAS, value, n.Left), Etop))
|
cmd/compile: convert constants to interfaces without allocating
The order pass is responsible for ensuring that
values passed to runtime functions, including
convT2E/convT2I, are addressable.
Prior to this CL, this was always accomplished
by creating a temp, which frequently escaped to
the heap, causing allocations, perhaps most
notably in code like:
fmt.Println(1, 2, 3) // allocates three times
None of the runtime routines modify the contents
of the pointers they receive, so in the case of
constants, instead of creating a temp value,
we can create a static value.
(Marking the static value as read-only provides
protection against accidental attempts by the runtime
to modify the constant data.)
This improves code generation for code like:
panic("abc")
c <- 2 // c is a chan int
which can now simply refer to "abc" and 2,
rather than going by way of a temporary.
It also allows us to optimize convT2E/convT2I,
by recognizing static readonly values
and directly constructing the interface.
This CL adds ~0.5% to binary size, despite
decreasing the size of many functions,
because it also adds many static symbols.
This binary size regression could be recovered in
future (but currently unplanned) work.
There is a lot of content-duplication in these
symbols; this statement generates six new symbols,
three containing an int 1 and three containing
a pointer to the string "a":
fmt.Println(1, 1, 1, "a", "a", "a")
These symbols could be made content-addressable.
Furthermore, these symbols are small, so the
alignment and naming overhead is large.
As with the go.strings section, these symbols
could be hidden and have their alignment reduced.
The changes to test/live.go make it impossible
(at least with current optimization techniques)
to place the values being passed to the runtime
in static symbols, preserving autotmp creation.
Fixes #18704
Benchmarks from fmt and go-kit's logging package:
github.com/go-kit/kit/log
name old time/op new time/op delta
JSONLoggerSimple-8 1.91µs ± 2% 2.11µs ±22% ~ (p=1.000 n=9+10)
JSONLoggerContextual-8 2.60µs ± 6% 2.43µs ± 2% -6.29% (p=0.000 n=9+10)
Discard-8 101ns ± 2% 34ns ±14% -66.33% (p=0.000 n=10+9)
OneWith-8 161ns ± 1% 102ns ±16% -36.78% (p=0.000 n=10+10)
TwoWith-8 175ns ± 3% 106ns ± 7% -39.36% (p=0.000 n=10+9)
TenWith-8 293ns ± 3% 227ns ±15% -22.44% (p=0.000 n=9+10)
LogfmtLoggerSimple-8 704ns ± 2% 608ns ± 2% -13.65% (p=0.000 n=10+9)
LogfmtLoggerContextual-8 962ns ± 1% 860ns ±17% -10.57% (p=0.003 n=9+10)
NopLoggerSimple-8 188ns ± 1% 120ns ± 1% -36.39% (p=0.000 n=9+10)
NopLoggerContextual-8 379ns ± 1% 243ns ± 0% -35.77% (p=0.000 n=9+10)
ValueBindingTimestamp-8 577ns ± 1% 499ns ± 1% -13.51% (p=0.000 n=10+10)
ValueBindingCaller-8 898ns ± 2% 844ns ± 2% -6.00% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
JSONLoggerSimple-8 904B ± 0% 872B ± 0% -3.54% (p=0.000 n=10+10)
JSONLoggerContextual-8 1.20kB ± 0% 1.14kB ± 0% -5.33% (p=0.000 n=10+10)
Discard-8 64.0B ± 0% 32.0B ± 0% -50.00% (p=0.000 n=10+10)
OneWith-8 96.0B ± 0% 64.0B ± 0% -33.33% (p=0.000 n=10+10)
TwoWith-8 160B ± 0% 128B ± 0% -20.00% (p=0.000 n=10+10)
TenWith-8 672B ± 0% 640B ± 0% -4.76% (p=0.000 n=10+10)
LogfmtLoggerSimple-8 128B ± 0% 96B ± 0% -25.00% (p=0.000 n=10+10)
LogfmtLoggerContextual-8 304B ± 0% 240B ± 0% -21.05% (p=0.000 n=10+10)
NopLoggerSimple-8 128B ± 0% 96B ± 0% -25.00% (p=0.000 n=10+10)
NopLoggerContextual-8 304B ± 0% 240B ± 0% -21.05% (p=0.000 n=10+10)
ValueBindingTimestamp-8 159B ± 0% 127B ± 0% -20.13% (p=0.000 n=10+10)
ValueBindingCaller-8 112B ± 0% 80B ± 0% -28.57% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
JSONLoggerSimple-8 19.0 ± 0% 17.0 ± 0% -10.53% (p=0.000 n=10+10)
JSONLoggerContextual-8 25.0 ± 0% 21.0 ± 0% -16.00% (p=0.000 n=10+10)
Discard-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
OneWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
TwoWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
TenWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
LogfmtLoggerSimple-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
LogfmtLoggerContextual-8 7.00 ± 0% 3.00 ± 0% -57.14% (p=0.000 n=10+10)
NopLoggerSimple-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
NopLoggerContextual-8 7.00 ± 0% 3.00 ± 0% -57.14% (p=0.000 n=10+10)
ValueBindingTimestamp-8 5.00 ± 0% 3.00 ± 0% -40.00% (p=0.000 n=10+10)
ValueBindingCaller-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
fmt
name old time/op new time/op delta
SprintfPadding-8 88.9ns ± 3% 79.1ns ± 1% -11.09% (p=0.000 n=10+7)
SprintfEmpty-8 12.6ns ± 3% 12.8ns ± 3% ~ (p=0.136 n=10+10)
SprintfString-8 38.7ns ± 5% 26.9ns ± 6% -30.65% (p=0.000 n=10+10)
SprintfTruncateString-8 56.7ns ± 2% 47.0ns ± 3% -17.05% (p=0.000 n=10+10)
SprintfQuoteString-8 164ns ± 2% 153ns ± 2% -7.01% (p=0.000 n=10+10)
SprintfInt-8 38.9ns ±15% 26.5ns ± 2% -31.93% (p=0.000 n=10+9)
SprintfIntInt-8 60.3ns ± 9% 38.2ns ± 1% -36.67% (p=0.000 n=10+8)
SprintfPrefixedInt-8 58.6ns ±13% 51.2ns ±11% -12.66% (p=0.001 n=10+10)
SprintfFloat-8 71.4ns ± 3% 64.2ns ± 3% -10.08% (p=0.000 n=8+10)
SprintfComplex-8 175ns ± 3% 159ns ± 2% -9.03% (p=0.000 n=10+10)
SprintfBoolean-8 33.5ns ± 4% 25.7ns ± 5% -23.28% (p=0.000 n=10+10)
SprintfHexString-8 65.3ns ± 3% 51.7ns ± 5% -20.86% (p=0.000 n=10+9)
SprintfHexBytes-8 67.2ns ± 5% 67.9ns ± 4% ~ (p=0.383 n=10+10)
SprintfBytes-8 129ns ± 7% 124ns ± 7% ~ (p=0.074 n=9+10)
SprintfStringer-8 127ns ± 4% 126ns ± 8% ~ (p=0.506 n=9+10)
SprintfStructure-8 357ns ± 3% 359ns ± 3% ~ (p=0.469 n=10+10)
ManyArgs-8 203ns ± 6% 126ns ± 3% -37.94% (p=0.000 n=10+10)
FprintInt-8 119ns ±10% 74ns ± 3% -37.54% (p=0.000 n=10+10)
FprintfBytes-8 122ns ± 4% 120ns ± 3% ~ (p=0.124 n=10+10)
FprintIntNoAlloc-8 78.2ns ± 5% 74.1ns ± 3% -5.28% (p=0.000 n=10+10)
ScanInts-8 349µs ± 1% 349µs ± 0% ~ (p=0.606 n=9+8)
ScanRecursiveInt-8 43.8ms ± 7% 40.1ms ± 2% -8.42% (p=0.000 n=10+10)
ScanRecursiveIntReaderWrapper-8 43.5ms ± 4% 40.4ms ± 2% -7.16% (p=0.000 n=10+9)
name old alloc/op new alloc/op delta
SprintfPadding-8 24.0B ± 0% 16.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfEmpty-8 0.00B 0.00B ~ (all equal)
SprintfString-8 21.0B ± 0% 5.0B ± 0% -76.19% (p=0.000 n=10+10)
SprintfTruncateString-8 32.0B ± 0% 16.0B ± 0% -50.00% (p=0.000 n=10+10)
SprintfQuoteString-8 48.0B ± 0% 32.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfInt-8 16.0B ± 0% 1.0B ± 0% -93.75% (p=0.000 n=10+10)
SprintfIntInt-8 24.0B ± 0% 3.0B ± 0% -87.50% (p=0.000 n=10+10)
SprintfPrefixedInt-8 72.0B ± 0% 64.0B ± 0% -11.11% (p=0.000 n=10+10)
SprintfFloat-8 16.0B ± 0% 8.0B ± 0% -50.00% (p=0.000 n=10+10)
SprintfComplex-8 48.0B ± 0% 32.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfBoolean-8 8.00B ± 0% 4.00B ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexString-8 96.0B ± 0% 80.0B ± 0% -16.67% (p=0.000 n=10+10)
SprintfHexBytes-8 112B ± 0% 112B ± 0% ~ (all equal)
SprintfBytes-8 96.0B ± 0% 96.0B ± 0% ~ (all equal)
SprintfStringer-8 32.0B ± 0% 32.0B ± 0% ~ (all equal)
SprintfStructure-8 256B ± 0% 256B ± 0% ~ (all equal)
ManyArgs-8 80.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10)
FprintInt-8 8.00B ± 0% 0.00B -100.00% (p=0.000 n=10+10)
FprintfBytes-8 32.0B ± 0% 32.0B ± 0% ~ (all equal)
FprintIntNoAlloc-8 0.00B 0.00B ~ (all equal)
ScanInts-8 15.2kB ± 0% 15.2kB ± 0% ~ (p=0.248 n=9+10)
ScanRecursiveInt-8 21.6kB ± 0% 21.6kB ± 0% ~ (all equal)
ScanRecursiveIntReaderWrapper-8 21.7kB ± 0% 21.7kB ± 0% ~ (all equal)
name old allocs/op new allocs/op delta
SprintfPadding-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfEmpty-8 0.00 0.00 ~ (all equal)
SprintfString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfTruncateString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfQuoteString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfInt-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfIntInt-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
SprintfPrefixedInt-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfFloat-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfComplex-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfBoolean-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexBytes-8 2.00 ± 0% 2.00 ± 0% ~ (all equal)
SprintfBytes-8 2.00 ± 0% 2.00 ± 0% ~ (all equal)
SprintfStringer-8 4.00 ± 0% 4.00 ± 0% ~ (all equal)
SprintfStructure-8 7.00 ± 0% 7.00 ± 0% ~ (all equal)
ManyArgs-8 8.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
FprintInt-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
FprintfBytes-8 1.00 ± 0% 1.00 ± 0% ~ (all equal)
FprintIntNoAlloc-8 0.00 0.00 ~ (all equal)
ScanInts-8 1.60k ± 0% 1.60k ± 0% ~ (all equal)
ScanRecursiveInt-8 1.71k ± 0% 1.71k ± 0% ~ (all equal)
ScanRecursiveIntReaderWrapper-8 1.71k ± 0% 1.71k ± 0% ~ (all equal)
Change-Id: I7ba72a25fea4140a0ba40a9f443103ed87cc69b5
Reviewed-on: https://go-review.googlesource.com/35554
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-01-21 13:41:06 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if value != nil {
|
2017-01-22 10:17:12 -08:00
|
|
|
// Value is identical to n.Left.
|
|
|
|
|
// Construct the interface directly: {type/itab, &value}.
|
2016-09-17 15:04:36 -07:00
|
|
|
var t *Node
|
|
|
|
|
if n.Type.IsEmptyInterface() {
|
|
|
|
|
t = typename(n.Left.Type)
|
|
|
|
|
} else {
|
|
|
|
|
t = itabname(n.Left.Type, n.Type)
|
|
|
|
|
}
|
cmd/compile: convert constants to interfaces without allocating
The order pass is responsible for ensuring that
values passed to runtime functions, including
convT2E/convT2I, are addressable.
Prior to this CL, this was always accomplished
by creating a temp, which frequently escaped to
the heap, causing allocations, perhaps most
notably in code like:
fmt.Println(1, 2, 3) // allocates three times
None of the runtime routines modify the contents
of the pointers they receive, so in the case of
constants, instead of creating a temp value,
we can create a static value.
(Marking the static value as read-only provides
protection against accidental attempts by the runtime
to modify the constant data.)
This improves code generation for code like:
panic("abc")
c <- 2 // c is a chan int
which can now simply refer to "abc" and 2,
rather than going by way of a temporary.
It also allows us to optimize convT2E/convT2I,
by recognizing static readonly values
and directly constructing the interface.
This CL adds ~0.5% to binary size, despite
decreasing the size of many functions,
because it also adds many static symbols.
This binary size regression could be recovered in
future (but currently unplanned) work.
There is a lot of content-duplication in these
symbols; this statement generates six new symbols,
three containing an int 1 and three containing
a pointer to the string "a":
fmt.Println(1, 1, 1, "a", "a", "a")
These symbols could be made content-addressable.
Furthermore, these symbols are small, so the
alignment and naming overhead is large.
As with the go.strings section, these symbols
could be hidden and have their alignment reduced.
The changes to test/live.go make it impossible
(at least with current optimization techniques)
to place the values being passed to the runtime
in static symbols, preserving autotmp creation.
Fixes #18704
Benchmarks from fmt and go-kit's logging package:
github.com/go-kit/kit/log
name old time/op new time/op delta
JSONLoggerSimple-8 1.91µs ± 2% 2.11µs ±22% ~ (p=1.000 n=9+10)
JSONLoggerContextual-8 2.60µs ± 6% 2.43µs ± 2% -6.29% (p=0.000 n=9+10)
Discard-8 101ns ± 2% 34ns ±14% -66.33% (p=0.000 n=10+9)
OneWith-8 161ns ± 1% 102ns ±16% -36.78% (p=0.000 n=10+10)
TwoWith-8 175ns ± 3% 106ns ± 7% -39.36% (p=0.000 n=10+9)
TenWith-8 293ns ± 3% 227ns ±15% -22.44% (p=0.000 n=9+10)
LogfmtLoggerSimple-8 704ns ± 2% 608ns ± 2% -13.65% (p=0.000 n=10+9)
LogfmtLoggerContextual-8 962ns ± 1% 860ns ±17% -10.57% (p=0.003 n=9+10)
NopLoggerSimple-8 188ns ± 1% 120ns ± 1% -36.39% (p=0.000 n=9+10)
NopLoggerContextual-8 379ns ± 1% 243ns ± 0% -35.77% (p=0.000 n=9+10)
ValueBindingTimestamp-8 577ns ± 1% 499ns ± 1% -13.51% (p=0.000 n=10+10)
ValueBindingCaller-8 898ns ± 2% 844ns ± 2% -6.00% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
JSONLoggerSimple-8 904B ± 0% 872B ± 0% -3.54% (p=0.000 n=10+10)
JSONLoggerContextual-8 1.20kB ± 0% 1.14kB ± 0% -5.33% (p=0.000 n=10+10)
Discard-8 64.0B ± 0% 32.0B ± 0% -50.00% (p=0.000 n=10+10)
OneWith-8 96.0B ± 0% 64.0B ± 0% -33.33% (p=0.000 n=10+10)
TwoWith-8 160B ± 0% 128B ± 0% -20.00% (p=0.000 n=10+10)
TenWith-8 672B ± 0% 640B ± 0% -4.76% (p=0.000 n=10+10)
LogfmtLoggerSimple-8 128B ± 0% 96B ± 0% -25.00% (p=0.000 n=10+10)
LogfmtLoggerContextual-8 304B ± 0% 240B ± 0% -21.05% (p=0.000 n=10+10)
NopLoggerSimple-8 128B ± 0% 96B ± 0% -25.00% (p=0.000 n=10+10)
NopLoggerContextual-8 304B ± 0% 240B ± 0% -21.05% (p=0.000 n=10+10)
ValueBindingTimestamp-8 159B ± 0% 127B ± 0% -20.13% (p=0.000 n=10+10)
ValueBindingCaller-8 112B ± 0% 80B ± 0% -28.57% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
JSONLoggerSimple-8 19.0 ± 0% 17.0 ± 0% -10.53% (p=0.000 n=10+10)
JSONLoggerContextual-8 25.0 ± 0% 21.0 ± 0% -16.00% (p=0.000 n=10+10)
Discard-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
OneWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
TwoWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
TenWith-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
LogfmtLoggerSimple-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
LogfmtLoggerContextual-8 7.00 ± 0% 3.00 ± 0% -57.14% (p=0.000 n=10+10)
NopLoggerSimple-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
NopLoggerContextual-8 7.00 ± 0% 3.00 ± 0% -57.14% (p=0.000 n=10+10)
ValueBindingTimestamp-8 5.00 ± 0% 3.00 ± 0% -40.00% (p=0.000 n=10+10)
ValueBindingCaller-8 4.00 ± 0% 2.00 ± 0% -50.00% (p=0.000 n=10+10)
fmt
name old time/op new time/op delta
SprintfPadding-8 88.9ns ± 3% 79.1ns ± 1% -11.09% (p=0.000 n=10+7)
SprintfEmpty-8 12.6ns ± 3% 12.8ns ± 3% ~ (p=0.136 n=10+10)
SprintfString-8 38.7ns ± 5% 26.9ns ± 6% -30.65% (p=0.000 n=10+10)
SprintfTruncateString-8 56.7ns ± 2% 47.0ns ± 3% -17.05% (p=0.000 n=10+10)
SprintfQuoteString-8 164ns ± 2% 153ns ± 2% -7.01% (p=0.000 n=10+10)
SprintfInt-8 38.9ns ±15% 26.5ns ± 2% -31.93% (p=0.000 n=10+9)
SprintfIntInt-8 60.3ns ± 9% 38.2ns ± 1% -36.67% (p=0.000 n=10+8)
SprintfPrefixedInt-8 58.6ns ±13% 51.2ns ±11% -12.66% (p=0.001 n=10+10)
SprintfFloat-8 71.4ns ± 3% 64.2ns ± 3% -10.08% (p=0.000 n=8+10)
SprintfComplex-8 175ns ± 3% 159ns ± 2% -9.03% (p=0.000 n=10+10)
SprintfBoolean-8 33.5ns ± 4% 25.7ns ± 5% -23.28% (p=0.000 n=10+10)
SprintfHexString-8 65.3ns ± 3% 51.7ns ± 5% -20.86% (p=0.000 n=10+9)
SprintfHexBytes-8 67.2ns ± 5% 67.9ns ± 4% ~ (p=0.383 n=10+10)
SprintfBytes-8 129ns ± 7% 124ns ± 7% ~ (p=0.074 n=9+10)
SprintfStringer-8 127ns ± 4% 126ns ± 8% ~ (p=0.506 n=9+10)
SprintfStructure-8 357ns ± 3% 359ns ± 3% ~ (p=0.469 n=10+10)
ManyArgs-8 203ns ± 6% 126ns ± 3% -37.94% (p=0.000 n=10+10)
FprintInt-8 119ns ±10% 74ns ± 3% -37.54% (p=0.000 n=10+10)
FprintfBytes-8 122ns ± 4% 120ns ± 3% ~ (p=0.124 n=10+10)
FprintIntNoAlloc-8 78.2ns ± 5% 74.1ns ± 3% -5.28% (p=0.000 n=10+10)
ScanInts-8 349µs ± 1% 349µs ± 0% ~ (p=0.606 n=9+8)
ScanRecursiveInt-8 43.8ms ± 7% 40.1ms ± 2% -8.42% (p=0.000 n=10+10)
ScanRecursiveIntReaderWrapper-8 43.5ms ± 4% 40.4ms ± 2% -7.16% (p=0.000 n=10+9)
name old alloc/op new alloc/op delta
SprintfPadding-8 24.0B ± 0% 16.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfEmpty-8 0.00B 0.00B ~ (all equal)
SprintfString-8 21.0B ± 0% 5.0B ± 0% -76.19% (p=0.000 n=10+10)
SprintfTruncateString-8 32.0B ± 0% 16.0B ± 0% -50.00% (p=0.000 n=10+10)
SprintfQuoteString-8 48.0B ± 0% 32.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfInt-8 16.0B ± 0% 1.0B ± 0% -93.75% (p=0.000 n=10+10)
SprintfIntInt-8 24.0B ± 0% 3.0B ± 0% -87.50% (p=0.000 n=10+10)
SprintfPrefixedInt-8 72.0B ± 0% 64.0B ± 0% -11.11% (p=0.000 n=10+10)
SprintfFloat-8 16.0B ± 0% 8.0B ± 0% -50.00% (p=0.000 n=10+10)
SprintfComplex-8 48.0B ± 0% 32.0B ± 0% -33.33% (p=0.000 n=10+10)
SprintfBoolean-8 8.00B ± 0% 4.00B ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexString-8 96.0B ± 0% 80.0B ± 0% -16.67% (p=0.000 n=10+10)
SprintfHexBytes-8 112B ± 0% 112B ± 0% ~ (all equal)
SprintfBytes-8 96.0B ± 0% 96.0B ± 0% ~ (all equal)
SprintfStringer-8 32.0B ± 0% 32.0B ± 0% ~ (all equal)
SprintfStructure-8 256B ± 0% 256B ± 0% ~ (all equal)
ManyArgs-8 80.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10)
FprintInt-8 8.00B ± 0% 0.00B -100.00% (p=0.000 n=10+10)
FprintfBytes-8 32.0B ± 0% 32.0B ± 0% ~ (all equal)
FprintIntNoAlloc-8 0.00B 0.00B ~ (all equal)
ScanInts-8 15.2kB ± 0% 15.2kB ± 0% ~ (p=0.248 n=9+10)
ScanRecursiveInt-8 21.6kB ± 0% 21.6kB ± 0% ~ (all equal)
ScanRecursiveIntReaderWrapper-8 21.7kB ± 0% 21.7kB ± 0% ~ (all equal)
name old allocs/op new allocs/op delta
SprintfPadding-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfEmpty-8 0.00 0.00 ~ (all equal)
SprintfString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfTruncateString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfQuoteString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfInt-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfIntInt-8 3.00 ± 0% 1.00 ± 0% -66.67% (p=0.000 n=10+10)
SprintfPrefixedInt-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfFloat-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfComplex-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfBoolean-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexString-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.000 n=10+10)
SprintfHexBytes-8 2.00 ± 0% 2.00 ± 0% ~ (all equal)
SprintfBytes-8 2.00 ± 0% 2.00 ± 0% ~ (all equal)
SprintfStringer-8 4.00 ± 0% 4.00 ± 0% ~ (all equal)
SprintfStructure-8 7.00 ± 0% 7.00 ± 0% ~ (all equal)
ManyArgs-8 8.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
FprintInt-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
FprintfBytes-8 1.00 ± 0% 1.00 ± 0% ~ (all equal)
FprintIntNoAlloc-8 0.00 0.00 ~ (all equal)
ScanInts-8 1.60k ± 0% 1.60k ± 0% ~ (all equal)
ScanRecursiveInt-8 1.71k ± 0% 1.71k ± 0% ~ (all equal)
ScanRecursiveIntReaderWrapper-8 1.71k ± 0% 1.71k ± 0% ~ (all equal)
Change-Id: I7ba72a25fea4140a0ba40a9f443103ed87cc69b5
Reviewed-on: https://go-review.googlesource.com/35554
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-01-21 13:41:06 -08:00
|
|
|
l := nod(OEFACE, t, typecheck(nod(OADDR, value, nil), Erv))
|
2016-09-17 15:04:36 -07:00
|
|
|
l.Type = n.Type
|
2017-04-25 18:02:43 -07:00
|
|
|
l.SetTypecheck(n.Typecheck())
|
2016-09-17 15:04:36 -07:00
|
|
|
n = l
|
|
|
|
|
break
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-10-25 23:34:35 +00:00
|
|
|
// Implement interface to empty interface conversion.
|
|
|
|
|
// tmp = i.itab
|
|
|
|
|
// if tmp != nil {
|
|
|
|
|
// tmp = tmp.type
|
|
|
|
|
// }
|
|
|
|
|
// e = iface{tmp, i.data}
|
|
|
|
|
if n.Type.IsEmptyInterface() && n.Left.Type.IsInterface() && !n.Left.Type.IsEmptyInterface() {
|
|
|
|
|
// Evaluate the input interface.
|
|
|
|
|
c := temp(n.Left.Type)
|
|
|
|
|
init.Append(nod(OAS, c, n.Left))
|
|
|
|
|
|
|
|
|
|
// Get the itab out of the interface.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
tmp := temp(types.NewPtr(types.Types[TUINT8]))
|
2016-10-25 23:34:35 +00:00
|
|
|
init.Append(nod(OAS, tmp, typecheck(nod(OITAB, c, nil), Erv)))
|
|
|
|
|
|
|
|
|
|
// Get the type out of the itab.
|
|
|
|
|
nif := nod(OIF, typecheck(nod(ONE, tmp, nodnil()), Erv), nil)
|
|
|
|
|
nif.Nbody.Set1(nod(OAS, tmp, itabType(tmp)))
|
|
|
|
|
init.Append(nif)
|
|
|
|
|
|
|
|
|
|
// Build the result.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
e := nod(OEFACE, tmp, ifaceData(c, types.NewPtr(types.Types[TUINT8])))
|
2016-10-25 23:34:35 +00:00
|
|
|
e.Type = n.Type // assign type manually, typecheck doesn't understand OEFACE.
|
2017-04-25 18:02:43 -07:00
|
|
|
e.SetTypecheck(1)
|
2016-10-25 23:34:35 +00:00
|
|
|
n = e
|
|
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-09 20:29:21 -08:00
|
|
|
var ll []*Node
|
2016-04-01 13:36:24 -07:00
|
|
|
if n.Type.IsEmptyInterface() {
|
2016-03-30 14:45:47 -07:00
|
|
|
if !n.Left.Type.IsInterface() {
|
2016-03-18 17:21:33 -07:00
|
|
|
ll = append(ll, typename(n.Left.Type))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-18 17:21:33 -07:00
|
|
|
} else {
|
2016-03-30 14:45:47 -07:00
|
|
|
if n.Left.Type.IsInterface() {
|
2016-03-18 17:21:33 -07:00
|
|
|
ll = append(ll, typename(n.Type))
|
|
|
|
|
} else {
|
|
|
|
|
ll = append(ll, itabname(n.Left.Type, n.Type))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-30 14:45:47 -07:00
|
|
|
if n.Left.Type.IsInterface() {
|
2016-03-09 20:29:21 -08:00
|
|
|
ll = append(ll, n.Left)
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
|
|
|
|
// regular types are passed by reference to avoid C vararg calls
|
2015-03-27 12:34:45 -04:00
|
|
|
// orderexpr arranged for n.Left to be a temporary for all
|
2015-02-13 14:40:36 -05:00
|
|
|
// the conversions it could see. comparison of an interface
|
|
|
|
|
// with a non-interface, especially in a switch on interface value
|
|
|
|
|
// with non-interface cases, is not visible to orderstmt, so we
|
|
|
|
|
// have to fall back on allocating a temp here.
|
2015-02-17 22:13:49 -05:00
|
|
|
if islvalue(n.Left) {
|
2016-09-16 11:00:54 +10:00
|
|
|
ll = append(ll, nod(OADDR, n.Left, nil))
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-09-16 11:00:54 +10:00
|
|
|
ll = append(ll, nod(OADDR, copyexpr(n.Left, n.Left.Type, init), nil))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-03-27 11:21:14 -04:00
|
|
|
dowidth(n.Left.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook(convFuncName(n.Left.Type, n.Type))
|
2016-09-17 15:04:36 -07:00
|
|
|
fn = substArgTypes(fn, n.Left.Type, n.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
dowidth(fn.Type)
|
2016-09-16 11:00:54 +10:00
|
|
|
n = nod(OCALL, fn, nil)
|
2016-03-09 20:29:21 -08:00
|
|
|
n.List.Set(ll)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = typecheck(n, Erv)
|
|
|
|
|
n = walkexpr(n, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case OCONV, OCONVNOP:
|
2017-03-17 13:35:36 -07:00
|
|
|
if thearch.LinkArch.Family == sys.ARM || thearch.LinkArch.Family == sys.MIPS {
|
2016-03-30 15:09:25 -07:00
|
|
|
if n.Left.Type.IsFloat() {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Type.Etype == TINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall("float64toint64", n.Type, init, conv(n.Left, types.Types[TFLOAT64]))
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if n.Type.Etype == TUINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall("float64touint64", n.Type, init, conv(n.Left, types.Types[TFLOAT64]))
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-30 15:09:25 -07:00
|
|
|
if n.Type.IsFloat() {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Left.Type.Etype == TINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = conv(mkcall("int64tofloat64", types.Types[TFLOAT64], init, conv(n.Left, types.Types[TINT64])), n.Type)
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if n.Left.Type.Etype == TUINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = conv(mkcall("uint64tofloat64", types.Types[TFLOAT64], init, conv(n.Left, types.Types[TUINT64])), n.Type)
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-17 13:35:36 -07:00
|
|
|
if thearch.LinkArch.Family == sys.I386 {
|
2016-07-21 10:37:59 -07:00
|
|
|
if n.Left.Type.IsFloat() {
|
|
|
|
|
if n.Type.Etype == TINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall("float64toint64", n.Type, init, conv(n.Left, types.Types[TFLOAT64]))
|
2016-07-21 10:37:59 -07:00
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if n.Type.Etype == TUINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall("float64touint64", n.Type, init, conv(n.Left, types.Types[TFLOAT64]))
|
2016-07-21 10:37:59 -07:00
|
|
|
break
|
|
|
|
|
}
|
2016-08-16 12:14:54 -04:00
|
|
|
if n.Type.Etype == TUINT32 || n.Type.Etype == TUINT || n.Type.Etype == TUINTPTR {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall("float64touint32", n.Type, init, conv(n.Left, types.Types[TFLOAT64]))
|
2016-07-21 10:37:59 -07:00
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
if n.Type.IsFloat() {
|
|
|
|
|
if n.Left.Type.Etype == TINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = conv(mkcall("int64tofloat64", types.Types[TFLOAT64], init, conv(n.Left, types.Types[TINT64])), n.Type)
|
2016-07-21 10:37:59 -07:00
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if n.Left.Type.Etype == TUINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = conv(mkcall("uint64tofloat64", types.Types[TFLOAT64], init, conv(n.Left, types.Types[TUINT64])), n.Type)
|
2016-07-21 10:37:59 -07:00
|
|
|
break
|
|
|
|
|
}
|
2016-08-16 12:14:54 -04:00
|
|
|
if n.Left.Type.Etype == TUINT32 || n.Left.Type.Etype == TUINT || n.Left.Type.Etype == TUINTPTR {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = conv(mkcall("uint32tofloat64", types.Types[TFLOAT64], init, conv(n.Left, types.Types[TUINT32])), n.Type)
|
2016-07-21 10:37:59 -07:00
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OANDNOT:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Op = OAND
|
2016-09-16 11:00:54 +10:00
|
|
|
n.Right = nod(OCOM, n.Right, nil)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right = typecheck(n.Right, Erv)
|
|
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case ODIV, OMOD:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
|
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// rewrite complex div into function call.
|
2015-09-24 23:21:18 +02:00
|
|
|
et := n.Left.Type.Etype
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 00:33:29 +10:00
|
|
|
if isComplex[et] && n.Op == ODIV {
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Type
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall("complex128div", types.Types[TCOMPLEX128], init, conv(n.Left, types.Types[TCOMPLEX128]), conv(n.Right, types.Types[TCOMPLEX128]))
|
2015-02-13 14:40:36 -05:00
|
|
|
n = conv(n, t)
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Nothing to do for float divisions.
|
2016-09-16 00:33:29 +10:00
|
|
|
if isFloat[et] {
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-02-13 16:00:09 -08:00
|
|
|
// rewrite 64-bit div and mod on 32-bit architectures.
|
|
|
|
|
// TODO: Remove this code once we can introduce
|
|
|
|
|
// runtime calls late in SSA processing.
|
|
|
|
|
if Widthreg < 8 && (et == TINT64 || et == TUINT64) {
|
|
|
|
|
if n.Right.Op == OLITERAL {
|
|
|
|
|
// Leave div/mod by constant powers of 2.
|
|
|
|
|
// The SSA backend will handle those.
|
|
|
|
|
switch et {
|
|
|
|
|
case TINT64:
|
|
|
|
|
c := n.Right.Int64()
|
|
|
|
|
if c < 0 {
|
|
|
|
|
c = -c
|
|
|
|
|
}
|
|
|
|
|
if c != 0 && c&(c-1) == 0 {
|
|
|
|
|
break opswitch
|
|
|
|
|
}
|
|
|
|
|
case TUINT64:
|
|
|
|
|
c := uint64(n.Right.Int64())
|
|
|
|
|
if c != 0 && c&(c-1) == 0 {
|
|
|
|
|
break opswitch
|
|
|
|
|
}
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-03-06 12:02:24 -08:00
|
|
|
var fn string
|
2015-02-13 14:40:36 -05:00
|
|
|
if et == TINT64 {
|
2015-03-06 12:02:24 -08:00
|
|
|
fn = "int64"
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2015-03-06 12:02:24 -08:00
|
|
|
fn = "uint64"
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
if n.Op == ODIV {
|
2015-03-06 12:02:24 -08:00
|
|
|
fn += "div"
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2015-03-06 12:02:24 -08:00
|
|
|
fn += "mod"
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall(fn, n.Type, init, conv(n.Left, types.Types[et]), conv(n.Right, types.Types[et]))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
case OINDEX:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// save the original node for bounds checking elision.
|
|
|
|
|
// If it was a ODIV/OMOD walk might rewrite it.
|
2015-02-23 16:07:24 -05:00
|
|
|
r := n.Right
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// if range of type cannot exceed static array bound,
|
|
|
|
|
// disable bounds check.
|
2017-02-27 19:56:38 +02:00
|
|
|
if n.Bounded() {
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Left.Type
|
2016-03-30 15:09:25 -07:00
|
|
|
if t != nil && t.IsPtr() {
|
2016-03-30 10:57:47 -07:00
|
|
|
t = t.Elem()
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-30 14:45:47 -07:00
|
|
|
if t.IsArray() {
|
2017-02-27 19:56:38 +02:00
|
|
|
n.SetBounded(bounded(r, t.NumElem()))
|
|
|
|
|
if Debug['m'] != 0 && n.Bounded() && !Isconst(n.Right, CTINT) {
|
2015-02-13 14:40:36 -05:00
|
|
|
Warn("index bounds check elided")
|
|
|
|
|
}
|
2017-02-27 19:56:38 +02:00
|
|
|
if smallintconst(n.Right) && !n.Bounded() {
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("index out of bounds")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-02-17 22:13:49 -05:00
|
|
|
} else if Isconst(n.Left, CTSTR) {
|
2017-02-27 19:56:38 +02:00
|
|
|
n.SetBounded(bounded(r, int64(len(n.Left.Val().U.(string)))))
|
|
|
|
|
if Debug['m'] != 0 && n.Bounded() && !Isconst(n.Right, CTINT) {
|
2015-02-13 14:40:36 -05:00
|
|
|
Warn("index bounds check elided")
|
|
|
|
|
}
|
2017-02-27 19:56:38 +02:00
|
|
|
if smallintconst(n.Right) && !n.Bounded() {
|
2015-10-10 15:24:34 -04:00
|
|
|
yyerror("index out of bounds")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
if Isconst(n.Right, CTINT) {
|
2016-09-16 00:33:29 +10:00
|
|
|
if n.Right.Val().U.(*Mpint).CmpInt64(0) < 0 || n.Right.Val().U.(*Mpint).Cmp(maxintval[TINT]) > 0 {
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("index out of bounds")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
case OINDEXMAP:
|
2016-10-11 08:36:38 -07:00
|
|
|
// Replace m[k] with *map{access1,assign}(maptype, m, &k)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
|
|
|
|
n.Right = walkexpr(n.Right, init)
|
2016-10-11 08:36:38 -07:00
|
|
|
map_ := n.Left
|
|
|
|
|
key := n.Right
|
|
|
|
|
t := map_.Type
|
|
|
|
|
if n.Etype == 1 {
|
|
|
|
|
// This m[k] expression is on the left-hand side of an assignment.
|
2017-03-14 11:11:28 -07:00
|
|
|
fast := mapfast(t)
|
|
|
|
|
if fast == mapslow {
|
2017-03-12 14:47:59 -07:00
|
|
|
// standard version takes key by reference.
|
|
|
|
|
// orderexpr made sure key is addressable.
|
|
|
|
|
key = nod(OADDR, key, nil)
|
|
|
|
|
}
|
2017-03-14 11:11:28 -07:00
|
|
|
n = mkcall1(mapfn(mapassign[fast], t), nil, init, typename(t), map_, key)
|
2016-10-11 08:36:38 -07:00
|
|
|
} else {
|
|
|
|
|
// m[k] is not the target of an assignment.
|
2017-03-14 11:11:28 -07:00
|
|
|
fast := mapfast(t)
|
|
|
|
|
if fast == mapslow {
|
2016-10-11 08:36:38 -07:00
|
|
|
// standard version takes key by reference.
|
|
|
|
|
// orderexpr made sure key is addressable.
|
|
|
|
|
key = nod(OADDR, key, nil)
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-10-11 08:36:38 -07:00
|
|
|
if w := t.Val().Width; w <= 1024 { // 1024 must match ../../../../runtime/hashmap.go:maxZero
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall1(mapfn(mapaccess1[fast], t), types.NewPtr(t.Val()), init, typename(t), map_, key)
|
2016-10-11 08:36:38 -07:00
|
|
|
} else {
|
|
|
|
|
z := zeroaddr(w)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall1(mapfn("mapaccess1_fat", t), types.NewPtr(t.Val()), init, typename(t), map_, key, z)
|
2016-10-11 08:36:38 -07:00
|
|
|
}
|
2016-04-19 08:31:04 -07:00
|
|
|
}
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n.Type = types.NewPtr(t.Val())
|
2017-02-27 19:56:38 +02:00
|
|
|
n.SetNonNil(true) // mapaccess1* and mapassign always return non-nil pointers.
|
2016-09-16 11:00:54 +10:00
|
|
|
n = nod(OIND, n, nil)
|
2016-03-28 21:48:47 -07:00
|
|
|
n.Type = t.Val()
|
2017-04-25 18:02:43 -07:00
|
|
|
n.SetTypecheck(1)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ORECV:
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("walkexpr ORECV") // should see inside OAS only
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-04-21 11:55:33 -07:00
|
|
|
case OSLICE, OSLICEARR, OSLICESTR, OSLICE3, OSLICE3ARR:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
2016-04-21 11:55:33 -07:00
|
|
|
low, high, max := n.SliceBounds()
|
|
|
|
|
low = walkexpr(low, init)
|
|
|
|
|
if low != nil && iszero(low) {
|
|
|
|
|
// Reduce x[0:j] to x[:j] and x[0:j:k] to x[:j:k].
|
|
|
|
|
low = nil
|
|
|
|
|
}
|
|
|
|
|
high = walkexpr(high, init)
|
|
|
|
|
max = walkexpr(max, init)
|
|
|
|
|
n.SetSliceBounds(low, high, max)
|
|
|
|
|
if n.Op.IsSlice3() {
|
|
|
|
|
if max != nil && max.Op == OCAP && samesafeexpr(n.Left, max.Left) {
|
|
|
|
|
// Reduce x[i:j:cap(x)] to x[i:j].
|
|
|
|
|
if n.Op == OSLICE3 {
|
|
|
|
|
n.Op = OSLICE
|
|
|
|
|
} else {
|
|
|
|
|
n.Op = OSLICEARR
|
|
|
|
|
}
|
|
|
|
|
n = reduceSlice(n)
|
cmd/internal/gc: optimize slice + write barrier
The code generated for a slice x[i:j] or x[i:j:k] computes the entire
new slice (base, len, cap) and then uses it as the evaluation of the
slice expression.
If the slice is part of an update x = x[i:j] or x = x[i:j:k], there are
opportunities to avoid computing some of these fields.
For x = x[0:i], we know that only the len is changing;
base can be ignored completely, and cap can be left unmodified.
For x = x[0:i:j], we know that only len and cap are changing;
base can be ignored completely.
For x = x[i:i], we know that the resulting cap is zero, and we don't
adjust the base during a slice producing a zero-cap result,
so again base can be ignored completely.
No write to base, no write barrier.
The old slice code was trying to work at a Go syntax level, mainly
because that was how you wrote code just once instead of once
per architecture. Now the compiler is factored a bit better and we
can implement slice during code generation but still have one copy
of the code. So the new code is working at that lower level.
(It must, to update only parts of the result.)
This CL by itself:
name old mean new mean delta
BinaryTree17 5.81s × (0.98,1.03) 5.71s × (0.96,1.05) ~ (p=0.101)
Fannkuch11 4.35s × (1.00,1.00) 4.39s × (1.00,1.00) +0.79% (p=0.000)
FmtFprintfEmpty 86.0ns × (0.94,1.11) 82.6ns × (0.98,1.04) -3.86% (p=0.048)
FmtFprintfString 276ns × (0.98,1.04) 273ns × (0.98,1.02) ~ (p=0.235)
FmtFprintfInt 274ns × (0.98,1.06) 270ns × (0.99,1.01) ~ (p=0.119)
FmtFprintfIntInt 506ns × (0.99,1.01) 475ns × (0.99,1.01) -6.02% (p=0.000)
FmtFprintfPrefixedInt 391ns × (0.99,1.01) 393ns × (1.00,1.01) ~ (p=0.139)
FmtFprintfFloat 566ns × (0.99,1.01) 574ns × (1.00,1.01) +1.33% (p=0.001)
FmtManyArgs 1.91µs × (0.99,1.01) 1.87µs × (0.99,1.02) -1.83% (p=0.000)
GobDecode 15.3ms × (0.99,1.02) 15.0ms × (0.98,1.05) -1.84% (p=0.042)
GobEncode 11.5ms × (0.97,1.03) 11.4ms × (0.99,1.03) ~ (p=0.152)
Gzip 645ms × (0.99,1.01) 647ms × (0.99,1.01) ~ (p=0.265)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.90% (p=0.000)
HTTPClientServer 90.5µs × (0.97,1.04) 88.5µs × (0.99,1.03) -2.27% (p=0.014)
JSONEncode 32.0ms × (0.98,1.03) 29.6ms × (0.98,1.01) -7.51% (p=0.000)
JSONDecode 114ms × (0.99,1.01) 104ms × (1.00,1.01) -8.60% (p=0.000)
Mandelbrot200 6.04ms × (1.00,1.01) 6.02ms × (1.00,1.00) ~ (p=0.057)
GoParse 6.47ms × (0.97,1.05) 6.37ms × (0.97,1.04) ~ (p=0.105)
RegexpMatchEasy0_32 171ns × (0.93,1.07) 152ns × (0.99,1.01) -11.09% (p=0.000)
RegexpMatchEasy0_1K 550ns × (0.98,1.01) 530ns × (1.00,1.00) -3.78% (p=0.000)
RegexpMatchEasy1_32 135ns × (0.99,1.02) 134ns × (0.99,1.01) -1.33% (p=0.002)
RegexpMatchEasy1_1K 879ns × (1.00,1.01) 865ns × (1.00,1.00) -1.58% (p=0.000)
RegexpMatchMedium_32 243ns × (1.00,1.00) 233ns × (1.00,1.00) -4.30% (p=0.000)
RegexpMatchMedium_1K 70.3µs × (1.00,1.00) 69.5µs × (1.00,1.00) -1.13% (p=0.000)
RegexpMatchHard_32 3.82µs × (1.00,1.01) 3.74µs × (1.00,1.00) -1.95% (p=0.000)
RegexpMatchHard_1K 117µs × (1.00,1.00) 115µs × (1.00,1.00) -1.69% (p=0.000)
Revcomp 917ms × (0.97,1.04) 920ms × (0.97,1.04) ~ (p=0.786)
Template 114ms × (0.99,1.01) 117ms × (0.99,1.01) +2.58% (p=0.000)
TimeParse 622ns × (0.99,1.01) 615ns × (0.99,1.00) -1.06% (p=0.000)
TimeFormat 665ns × (0.99,1.01) 654ns × (0.99,1.00) -1.70% (p=0.000)
This CL and previous CL (append) combined:
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.71s × (0.96,1.05) ~ (p=0.638)
Fannkuch11 4.41s × (0.98,1.03) 4.39s × (1.00,1.00) ~ (p=0.474)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 82.6ns × (0.98,1.04) -10.89% (p=0.004)
FmtFprintfString 281ns × (0.96,1.08) 273ns × (0.98,1.02) ~ (p=0.078)
FmtFprintfInt 288ns × (0.97,1.06) 270ns × (0.99,1.01) -6.37% (p=0.000)
FmtFprintfIntInt 493ns × (0.97,1.04) 475ns × (0.99,1.01) -3.53% (p=0.002)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 393ns × (1.00,1.01) -7.07% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 574ns × (1.00,1.01) -4.02% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.87µs × (0.99,1.02) ~ (p=0.305)
GobDecode 14.8ms × (0.98,1.03) 15.0ms × (0.98,1.05) ~ (p=0.237)
GobEncode 12.3ms × (0.98,1.01) 11.4ms × (0.99,1.03) -6.95% (p=0.000)
Gzip 656ms × (0.99,1.05) 647ms × (0.99,1.01) ~ (p=0.101)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.58% (p=0.001)
HTTPClientServer 91.2µs × (0.97,1.04) 88.5µs × (0.99,1.03) -3.02% (p=0.003)
JSONEncode 32.6ms × (0.97,1.08) 29.6ms × (0.98,1.01) -9.10% (p=0.000)
JSONDecode 114ms × (0.97,1.05) 104ms × (1.00,1.01) -8.74% (p=0.000)
Mandelbrot200 6.11ms × (0.98,1.04) 6.02ms × (1.00,1.00) ~ (p=0.090)
GoParse 6.66ms × (0.97,1.04) 6.37ms × (0.97,1.04) -4.41% (p=0.000)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 152ns × (0.99,1.01) -4.69% (p=0.000)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 530ns × (1.00,1.00) -1.57% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 134ns × (0.99,1.01) -2.91% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 865ns × (1.00,1.00) -0.51% (p=0.012)
RegexpMatchMedium_32 252ns × (0.99,1.01) 233ns × (1.00,1.00) -7.85% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 69.5µs × (1.00,1.00) -4.43% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.74µs × (1.00,1.00) -2.74% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 115µs × (1.00,1.00) -2.24% (p=0.000)
Revcomp 920ms × (0.97,1.07) 920ms × (0.97,1.04) ~ (p=0.998)
Template 129ms × (0.98,1.03) 117ms × (0.99,1.01) -9.79% (p=0.000)
TimeParse 619ns × (0.99,1.01) 615ns × (0.99,1.00) -0.57% (p=0.011)
TimeFormat 661ns × (0.98,1.04) 654ns × (0.99,1.00) ~ (p=0.223)
Change-Id: If054d81ab2c71d8d62cf54b5b1fac2af66b387fc
Reviewed-on: https://go-review.googlesource.com/9813
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-05-06 12:35:53 -04:00
|
|
|
}
|
2016-04-21 11:55:33 -07:00
|
|
|
} else {
|
cmd/internal/gc: optimize slice + write barrier
The code generated for a slice x[i:j] or x[i:j:k] computes the entire
new slice (base, len, cap) and then uses it as the evaluation of the
slice expression.
If the slice is part of an update x = x[i:j] or x = x[i:j:k], there are
opportunities to avoid computing some of these fields.
For x = x[0:i], we know that only the len is changing;
base can be ignored completely, and cap can be left unmodified.
For x = x[0:i:j], we know that only len and cap are changing;
base can be ignored completely.
For x = x[i:i], we know that the resulting cap is zero, and we don't
adjust the base during a slice producing a zero-cap result,
so again base can be ignored completely.
No write to base, no write barrier.
The old slice code was trying to work at a Go syntax level, mainly
because that was how you wrote code just once instead of once
per architecture. Now the compiler is factored a bit better and we
can implement slice during code generation but still have one copy
of the code. So the new code is working at that lower level.
(It must, to update only parts of the result.)
This CL by itself:
name old mean new mean delta
BinaryTree17 5.81s × (0.98,1.03) 5.71s × (0.96,1.05) ~ (p=0.101)
Fannkuch11 4.35s × (1.00,1.00) 4.39s × (1.00,1.00) +0.79% (p=0.000)
FmtFprintfEmpty 86.0ns × (0.94,1.11) 82.6ns × (0.98,1.04) -3.86% (p=0.048)
FmtFprintfString 276ns × (0.98,1.04) 273ns × (0.98,1.02) ~ (p=0.235)
FmtFprintfInt 274ns × (0.98,1.06) 270ns × (0.99,1.01) ~ (p=0.119)
FmtFprintfIntInt 506ns × (0.99,1.01) 475ns × (0.99,1.01) -6.02% (p=0.000)
FmtFprintfPrefixedInt 391ns × (0.99,1.01) 393ns × (1.00,1.01) ~ (p=0.139)
FmtFprintfFloat 566ns × (0.99,1.01) 574ns × (1.00,1.01) +1.33% (p=0.001)
FmtManyArgs 1.91µs × (0.99,1.01) 1.87µs × (0.99,1.02) -1.83% (p=0.000)
GobDecode 15.3ms × (0.99,1.02) 15.0ms × (0.98,1.05) -1.84% (p=0.042)
GobEncode 11.5ms × (0.97,1.03) 11.4ms × (0.99,1.03) ~ (p=0.152)
Gzip 645ms × (0.99,1.01) 647ms × (0.99,1.01) ~ (p=0.265)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.90% (p=0.000)
HTTPClientServer 90.5µs × (0.97,1.04) 88.5µs × (0.99,1.03) -2.27% (p=0.014)
JSONEncode 32.0ms × (0.98,1.03) 29.6ms × (0.98,1.01) -7.51% (p=0.000)
JSONDecode 114ms × (0.99,1.01) 104ms × (1.00,1.01) -8.60% (p=0.000)
Mandelbrot200 6.04ms × (1.00,1.01) 6.02ms × (1.00,1.00) ~ (p=0.057)
GoParse 6.47ms × (0.97,1.05) 6.37ms × (0.97,1.04) ~ (p=0.105)
RegexpMatchEasy0_32 171ns × (0.93,1.07) 152ns × (0.99,1.01) -11.09% (p=0.000)
RegexpMatchEasy0_1K 550ns × (0.98,1.01) 530ns × (1.00,1.00) -3.78% (p=0.000)
RegexpMatchEasy1_32 135ns × (0.99,1.02) 134ns × (0.99,1.01) -1.33% (p=0.002)
RegexpMatchEasy1_1K 879ns × (1.00,1.01) 865ns × (1.00,1.00) -1.58% (p=0.000)
RegexpMatchMedium_32 243ns × (1.00,1.00) 233ns × (1.00,1.00) -4.30% (p=0.000)
RegexpMatchMedium_1K 70.3µs × (1.00,1.00) 69.5µs × (1.00,1.00) -1.13% (p=0.000)
RegexpMatchHard_32 3.82µs × (1.00,1.01) 3.74µs × (1.00,1.00) -1.95% (p=0.000)
RegexpMatchHard_1K 117µs × (1.00,1.00) 115µs × (1.00,1.00) -1.69% (p=0.000)
Revcomp 917ms × (0.97,1.04) 920ms × (0.97,1.04) ~ (p=0.786)
Template 114ms × (0.99,1.01) 117ms × (0.99,1.01) +2.58% (p=0.000)
TimeParse 622ns × (0.99,1.01) 615ns × (0.99,1.00) -1.06% (p=0.000)
TimeFormat 665ns × (0.99,1.01) 654ns × (0.99,1.00) -1.70% (p=0.000)
This CL and previous CL (append) combined:
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.71s × (0.96,1.05) ~ (p=0.638)
Fannkuch11 4.41s × (0.98,1.03) 4.39s × (1.00,1.00) ~ (p=0.474)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 82.6ns × (0.98,1.04) -10.89% (p=0.004)
FmtFprintfString 281ns × (0.96,1.08) 273ns × (0.98,1.02) ~ (p=0.078)
FmtFprintfInt 288ns × (0.97,1.06) 270ns × (0.99,1.01) -6.37% (p=0.000)
FmtFprintfIntInt 493ns × (0.97,1.04) 475ns × (0.99,1.01) -3.53% (p=0.002)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 393ns × (1.00,1.01) -7.07% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 574ns × (1.00,1.01) -4.02% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.87µs × (0.99,1.02) ~ (p=0.305)
GobDecode 14.8ms × (0.98,1.03) 15.0ms × (0.98,1.05) ~ (p=0.237)
GobEncode 12.3ms × (0.98,1.01) 11.4ms × (0.99,1.03) -6.95% (p=0.000)
Gzip 656ms × (0.99,1.05) 647ms × (0.99,1.01) ~ (p=0.101)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.58% (p=0.001)
HTTPClientServer 91.2µs × (0.97,1.04) 88.5µs × (0.99,1.03) -3.02% (p=0.003)
JSONEncode 32.6ms × (0.97,1.08) 29.6ms × (0.98,1.01) -9.10% (p=0.000)
JSONDecode 114ms × (0.97,1.05) 104ms × (1.00,1.01) -8.74% (p=0.000)
Mandelbrot200 6.11ms × (0.98,1.04) 6.02ms × (1.00,1.00) ~ (p=0.090)
GoParse 6.66ms × (0.97,1.04) 6.37ms × (0.97,1.04) -4.41% (p=0.000)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 152ns × (0.99,1.01) -4.69% (p=0.000)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 530ns × (1.00,1.00) -1.57% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 134ns × (0.99,1.01) -2.91% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 865ns × (1.00,1.00) -0.51% (p=0.012)
RegexpMatchMedium_32 252ns × (0.99,1.01) 233ns × (1.00,1.00) -7.85% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 69.5µs × (1.00,1.00) -4.43% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.74µs × (1.00,1.00) -2.74% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 115µs × (1.00,1.00) -2.24% (p=0.000)
Revcomp 920ms × (0.97,1.07) 920ms × (0.97,1.04) ~ (p=0.998)
Template 129ms × (0.98,1.03) 117ms × (0.99,1.01) -9.79% (p=0.000)
TimeParse 619ns × (0.99,1.01) 615ns × (0.99,1.00) -0.57% (p=0.011)
TimeFormat 661ns × (0.98,1.04) 654ns × (0.99,1.00) ~ (p=0.223)
Change-Id: If054d81ab2c71d8d62cf54b5b1fac2af66b387fc
Reviewed-on: https://go-review.googlesource.com/9813
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-05-06 12:35:53 -04:00
|
|
|
n = reduceSlice(n)
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ONEW:
|
2015-05-20 15:16:34 -04:00
|
|
|
if n.Esc == EscNone {
|
2016-03-30 10:57:47 -07:00
|
|
|
if n.Type.Elem().Width >= 1<<16 {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("large ONEW with EscNone: %v", n)
|
2015-05-20 15:16:34 -04:00
|
|
|
}
|
2016-03-30 10:57:47 -07:00
|
|
|
r := temp(n.Type.Elem())
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(OAS, r, nil) // zero temp
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Etop)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(r)
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(OADDR, r.Left, nil)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Erv)
|
2015-02-13 14:40:36 -05:00
|
|
|
n = r
|
|
|
|
|
} else {
|
2016-03-30 10:57:47 -07:00
|
|
|
n = callnew(n.Type.Elem())
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
case OCMPSTR:
|
|
|
|
|
// s + "badgerbadgerbadger" == "badgerbadgerbadger"
|
2016-03-08 15:10:26 -08:00
|
|
|
if (Op(n.Etype) == OEQ || Op(n.Etype) == ONE) && Isconst(n.Right, CTSTR) && n.Left.Op == OADDSTR && n.Left.List.Len() == 2 && Isconst(n.Left.List.Second(), CTSTR) && strlit(n.Right) == strlit(n.Left.List.Second()) {
|
2015-09-24 23:21:18 +02:00
|
|
|
// TODO(marvin): Fix Node.EType type union.
|
2016-09-16 11:00:54 +10:00
|
|
|
r := nod(Op(n.Etype), nod(OLEN, n.Left.List.First(), nil), nodintconst(0))
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Erv)
|
|
|
|
|
r = walkexpr(r, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
r.Type = n.Type
|
|
|
|
|
n = r
|
2015-09-20 23:28:34 +02:00
|
|
|
break
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-06-25 13:40:43 -07:00
|
|
|
// Rewrite comparisons to short constant strings as length+byte-wise comparisons.
|
|
|
|
|
var cs, ncs *Node // const string, non-const string
|
|
|
|
|
switch {
|
|
|
|
|
case Isconst(n.Left, CTSTR) && Isconst(n.Right, CTSTR):
|
|
|
|
|
// ignore; will be constant evaluated
|
|
|
|
|
case Isconst(n.Left, CTSTR):
|
|
|
|
|
cs = n.Left
|
|
|
|
|
ncs = n.Right
|
|
|
|
|
case Isconst(n.Right, CTSTR):
|
|
|
|
|
cs = n.Right
|
|
|
|
|
ncs = n.Left
|
|
|
|
|
}
|
|
|
|
|
if cs != nil {
|
|
|
|
|
cmp := Op(n.Etype)
|
|
|
|
|
// maxRewriteLen was chosen empirically.
|
|
|
|
|
// It is the value that minimizes cmd/go file size
|
|
|
|
|
// across most architectures.
|
|
|
|
|
// See the commit description for CL 26758 for details.
|
|
|
|
|
maxRewriteLen := 6
|
2017-03-28 15:30:31 -05:00
|
|
|
// Some architectures can load unaligned byte sequence as 1 word.
|
|
|
|
|
// So we can cover longer strings with the same amount of code.
|
|
|
|
|
canCombineLoads := false
|
|
|
|
|
combine64bit := false
|
|
|
|
|
// TODO: does this improve performance on any other architectures?
|
|
|
|
|
switch thearch.LinkArch.Family {
|
|
|
|
|
case sys.AMD64:
|
|
|
|
|
// Larger compare require longer instructions, so keep this reasonably low.
|
|
|
|
|
// Data from CL 26758 shows that longer strings are rare.
|
|
|
|
|
// If we really want we can do 16 byte SSE comparisons in the future.
|
|
|
|
|
maxRewriteLen = 16
|
|
|
|
|
canCombineLoads = true
|
|
|
|
|
combine64bit = true
|
|
|
|
|
case sys.I386:
|
|
|
|
|
maxRewriteLen = 8
|
|
|
|
|
canCombineLoads = true
|
|
|
|
|
}
|
2016-06-25 13:40:43 -07:00
|
|
|
var and Op
|
|
|
|
|
switch cmp {
|
|
|
|
|
case OEQ:
|
|
|
|
|
and = OANDAND
|
|
|
|
|
case ONE:
|
|
|
|
|
and = OOROR
|
|
|
|
|
default:
|
|
|
|
|
// Don't do byte-wise comparisons for <, <=, etc.
|
|
|
|
|
// They're fairly complicated.
|
|
|
|
|
// Length-only checks are ok, though.
|
|
|
|
|
maxRewriteLen = 0
|
|
|
|
|
}
|
|
|
|
|
if s := cs.Val().U.(string); len(s) <= maxRewriteLen {
|
|
|
|
|
if len(s) > 0 {
|
|
|
|
|
ncs = safeexpr(ncs, init)
|
|
|
|
|
}
|
|
|
|
|
// TODO(marvin): Fix Node.EType type union.
|
2016-09-16 11:00:54 +10:00
|
|
|
r := nod(cmp, nod(OLEN, ncs, nil), nodintconst(int64(len(s))))
|
2017-03-28 15:30:31 -05:00
|
|
|
remains := len(s)
|
|
|
|
|
for i := 0; remains > 0; {
|
|
|
|
|
if remains == 1 || !canCombineLoads {
|
|
|
|
|
cb := nodintconst(int64(s[i]))
|
|
|
|
|
ncb := nod(OINDEX, ncs, nodintconst(int64(i)))
|
|
|
|
|
r = nod(and, r, nod(cmp, ncb, cb))
|
|
|
|
|
remains--
|
|
|
|
|
i++
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
var step int
|
2017-04-07 08:53:38 -07:00
|
|
|
var convType *types.Type
|
2017-03-28 15:30:31 -05:00
|
|
|
switch {
|
|
|
|
|
case remains >= 8 && combine64bit:
|
2017-04-07 08:53:38 -07:00
|
|
|
convType = types.Types[TINT64]
|
2017-03-28 15:30:31 -05:00
|
|
|
step = 8
|
|
|
|
|
case remains >= 4:
|
2017-04-07 08:53:38 -07:00
|
|
|
convType = types.Types[TUINT32]
|
2017-03-28 15:30:31 -05:00
|
|
|
step = 4
|
|
|
|
|
case remains >= 2:
|
2017-04-07 08:53:38 -07:00
|
|
|
convType = types.Types[TUINT16]
|
2017-03-28 15:30:31 -05:00
|
|
|
step = 2
|
|
|
|
|
}
|
|
|
|
|
ncsubstr := nod(OINDEX, ncs, nodintconst(int64(i)))
|
|
|
|
|
ncsubstr = conv(ncsubstr, convType)
|
|
|
|
|
csubstr := int64(s[i])
|
|
|
|
|
// Calculate large constant from bytes as sequence of shifts and ors.
|
|
|
|
|
// Like this: uint32(s[0]) | uint32(s[1])<<8 | uint32(s[2])<<16 ...
|
|
|
|
|
// ssa will combine this into a single large load.
|
|
|
|
|
for offset := 1; offset < step; offset++ {
|
|
|
|
|
b := nod(OINDEX, ncs, nodintconst(int64(i+offset)))
|
|
|
|
|
b = conv(b, convType)
|
|
|
|
|
b = nod(OLSH, b, nodintconst(int64(8*offset)))
|
|
|
|
|
ncsubstr = nod(OOR, ncsubstr, b)
|
|
|
|
|
csubstr = csubstr | int64(s[i+offset])<<uint8(8*offset)
|
|
|
|
|
}
|
|
|
|
|
csubstrPart := nodintconst(csubstr)
|
|
|
|
|
// Compare "step" bytes as once
|
|
|
|
|
r = nod(and, r, nod(cmp, csubstrPart, ncsubstr))
|
|
|
|
|
remains -= step
|
|
|
|
|
i += step
|
2016-06-25 13:40:43 -07:00
|
|
|
}
|
|
|
|
|
r = typecheck(r, Erv)
|
|
|
|
|
r = walkexpr(r, init)
|
|
|
|
|
r.Type = n.Type
|
|
|
|
|
n = r
|
|
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
var r *Node
|
2015-09-24 23:21:18 +02:00
|
|
|
// TODO(marvin): Fix Node.EType type union.
|
|
|
|
|
if Op(n.Etype) == OEQ || Op(n.Etype) == ONE {
|
2015-02-13 14:40:36 -05:00
|
|
|
// prepare for rewrite below
|
|
|
|
|
n.Left = cheapexpr(n.Left, init)
|
|
|
|
|
n.Right = cheapexpr(n.Right, init)
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
r = mkcall("eqstring", types.Types[TBOOL], init, conv(n.Left, types.Types[TSTRING]), conv(n.Right, types.Types[TSTRING]))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// quick check of len before full compare for == or !=
|
|
|
|
|
// eqstring assumes that the lengths are equal
|
2015-09-24 23:21:18 +02:00
|
|
|
// TODO(marvin): Fix Node.EType type union.
|
|
|
|
|
if Op(n.Etype) == OEQ {
|
2015-02-13 14:40:36 -05:00
|
|
|
// len(left) == len(right) && eqstring(left, right)
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(OANDAND, nod(OEQ, nod(OLEN, n.Left, nil), nod(OLEN, n.Right, nil)), r)
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
|
|
|
|
// len(left) != len(right) || !eqstring(left, right)
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(ONOT, r, nil)
|
|
|
|
|
r = nod(OOROR, nod(ONE, nod(OLEN, n.Left, nil), nod(OLEN, n.Right, nil)), r)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Erv)
|
|
|
|
|
r = walkexpr(r, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
|
|
|
|
// sys_cmpstring(s1, s2) :: 0
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
r = mkcall("cmpstring", types.Types[TINT], init, conv(n.Left, types.Types[TSTRING]), conv(n.Right, types.Types[TSTRING]))
|
2015-09-24 23:21:18 +02:00
|
|
|
// TODO(marvin): Fix Node.EType type union.
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(Op(n.Etype), r, nodintconst(0))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Erv)
|
2016-03-30 14:56:08 -07:00
|
|
|
if !n.Type.IsBoolean() {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("cmp %v", n.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
r.Type = n.Type
|
|
|
|
|
n = r
|
|
|
|
|
|
|
|
|
|
case OADDSTR:
|
|
|
|
|
n = addstr(n, init)
|
|
|
|
|
|
|
|
|
|
case OAPPEND:
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
// order should make sure we only see OAS(node, OAPPEND), which we handle above.
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("append outside assignment")
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OCOPY:
|
2016-09-21 09:44:40 -07:00
|
|
|
n = copyany(n, init, instrumenting && !compiling_runtime)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// cannot use chanfn - closechan takes any, not chan any
|
|
|
|
|
case OCLOSE:
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook("closechan")
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
fn = substArgTypes(fn, n.Left.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
n = mkcall1(fn, nil, init, n.Left)
|
|
|
|
|
|
|
|
|
|
case OMAKECHAN:
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall1(chanfn("makechan", 1, n.Type), n.Type, init, typename(n.Type), conv(n.Left, types.Types[TINT64]))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OMAKEMAP:
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Type
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
a := nodnil() // hmap buffer
|
|
|
|
|
r := nodnil() // bucket buffer
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Esc == EscNone {
|
|
|
|
|
// Allocate hmap buffer on stack.
|
2016-02-23 07:46:01 +00:00
|
|
|
var_ := temp(hmap(t))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OAS, var_, nil) // zero temp
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
a = typecheck(a, Etop)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(a)
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OADDR, var_, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Allocate one bucket on stack.
|
|
|
|
|
// Maximum key/value size is 128 bytes, larger objects
|
|
|
|
|
// are stored with an indirection. So max bucket size is 2048+eps.
|
|
|
|
|
var_ = temp(mapbucket(t))
|
|
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(OAS, var_, nil) // zero temp
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Etop)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(r)
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(OADDR, var_, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook("makemap")
|
2016-03-28 21:48:47 -07:00
|
|
|
fn = substArgTypes(fn, hmap(t), mapbucket(t), t.Key(), t.Val())
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall1(fn, n.Type, init, typename(n.Type), conv(n.Left, types.Types[TINT64]), a, r)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OMAKESLICE:
|
2015-02-23 16:07:24 -05:00
|
|
|
l := n.Left
|
|
|
|
|
r := n.Right
|
2015-02-13 14:40:36 -05:00
|
|
|
if r == nil {
|
|
|
|
|
r = safeexpr(l, init)
|
|
|
|
|
l = r
|
|
|
|
|
}
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Type
|
2015-05-20 15:16:34 -04:00
|
|
|
if n.Esc == EscNone {
|
|
|
|
|
if !isSmallMakeSlice(n) {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("non-small OMAKESLICE with EscNone: %v", n)
|
2015-05-20 15:16:34 -04:00
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
// var arr [r]T
|
|
|
|
|
// n = arr[:l]
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
t = types.NewArray(t.Elem(), nonnegintconst(r)) // [r]T
|
2015-02-23 16:07:24 -05:00
|
|
|
var_ := temp(t)
|
2016-09-16 11:00:54 +10:00
|
|
|
a := nod(OAS, var_, nil) // zero temp
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
a = typecheck(a, Etop)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(a)
|
2016-09-16 11:00:54 +10:00
|
|
|
r := nod(OSLICE, var_, nil) // arr[:l]
|
2016-04-21 11:55:33 -07:00
|
|
|
r.SetSliceBounds(nil, l, nil)
|
|
|
|
|
r = conv(r, n.Type) // in case n.Type is named.
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Erv)
|
|
|
|
|
r = walkexpr(r, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
n = r
|
|
|
|
|
} else {
|
2016-08-25 14:17:52 +02:00
|
|
|
// n escapes; set up a call to makeslice.
|
|
|
|
|
// When len and cap can fit into int, use makeslice instead of
|
|
|
|
|
// makeslice64, which is faster and shorter on 32 bit platforms.
|
|
|
|
|
|
2017-02-27 19:56:38 +02:00
|
|
|
if t.Elem().NotInHeap() {
|
cmd/compile: add go:notinheap type pragma
This adds a //go:notinheap pragma for declarations of types that must
not be heap allocated. We ensure these rules by disallowing new(T),
make([]T), append([]T), or implicit allocation of T, by disallowing
conversions to notinheap types, and by propagating notinheap to any
struct or array that contains notinheap elements.
The utility of this pragma is that we can eliminate write barriers for
writes to pointers to go:notinheap types, since the write barrier is
guaranteed to be a no-op. This will let us mark several scheduler and
memory allocator structures as go:notinheap, which will let us
disallow write barriers in the scheduler and memory allocator much
more thoroughly and also eliminate some problematic hybrid write
barriers.
This also makes go:nowritebarrierrec and go:yeswritebarrierrec much
more powerful. Currently we use go:nowritebarrier all over the place,
but it's almost never what you actually want: when write barriers are
illegal, they're typically illegal for a whole dynamic scope. Partly
this is because go:nowritebarrier has been around longer, but it's
also because go:nowritebarrierrec couldn't be used in situations that
had no-op write barriers or where some nested scope did allow write
barriers. go:notinheap eliminates many no-op write barriers and
go:yeswritebarrierrec makes it possible to opt back in to write
barriers, so these two changes will let us use go:nowritebarrierrec
far more liberally.
This updates #13386, which is about controlling pointers from non-GC'd
memory to GC'd memory. That would require some additional pragma (or
pragmas), but could build on this pragma.
Change-Id: I6314f8f4181535dd166887c9ec239977b54940bd
Reviewed-on: https://go-review.googlesource.com/30939
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-10-11 22:53:27 -04:00
|
|
|
yyerror("%v is go:notinheap; heap allocation disallowed", t.Elem())
|
|
|
|
|
}
|
|
|
|
|
|
2016-08-25 14:17:52 +02:00
|
|
|
len, cap := l, r
|
|
|
|
|
|
|
|
|
|
fnname := "makeslice64"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
argtype := types.Types[TINT64]
|
2016-08-25 14:17:52 +02:00
|
|
|
|
|
|
|
|
// typechecking guarantees that TIDEAL len/cap are positive and fit in an int.
|
|
|
|
|
// The case of len or cap overflow when converting TUINT or TUINTPTR to TINT
|
|
|
|
|
// will be handled by the negative range checks in makeslice during runtime.
|
2016-09-16 00:33:29 +10:00
|
|
|
if (len.Type.IsKind(TIDEAL) || maxintval[len.Type.Etype].Cmp(maxintval[TUINT]) <= 0) &&
|
|
|
|
|
(cap.Type.IsKind(TIDEAL) || maxintval[cap.Type.Etype].Cmp(maxintval[TUINT]) <= 0) {
|
2016-08-25 14:17:52 +02:00
|
|
|
fnname = "makeslice"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
argtype = types.Types[TINT]
|
2016-08-25 14:17:52 +02:00
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-08-25 14:17:52 +02:00
|
|
|
fn := syslook(fnname)
|
2016-03-30 10:57:47 -07:00
|
|
|
fn = substArgTypes(fn, t.Elem()) // any-1
|
2016-08-25 14:17:52 +02:00
|
|
|
n = mkcall1(fn, t, init, typename(t.Elem()), conv(len, argtype), conv(cap, argtype))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
case ORUNESTR:
|
2015-02-23 16:07:24 -05:00
|
|
|
a := nodnil()
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Esc == EscNone {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
t := types.NewArray(types.Types[TUINT8], 4)
|
2015-02-23 16:07:24 -05:00
|
|
|
var_ := temp(t)
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OADDR, var_, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// intstring(*[4]byte, rune)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall("intstring", n.Type, init, a, conv(n.Left, types.Types[TINT64]))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OARRAYBYTESTR:
|
2015-02-23 16:07:24 -05:00
|
|
|
a := nodnil()
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Esc == EscNone {
|
|
|
|
|
// Create temporary buffer for string on stack.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
t := types.NewArray(types.Types[TUINT8], tmpstringbufsize)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OADDR, temp(t), nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// slicebytetostring(*[32]byte, []byte) string;
|
|
|
|
|
n = mkcall("slicebytetostring", n.Type, init, a, n.Left)
|
|
|
|
|
|
|
|
|
|
// slicebytetostringtmp([]byte) string;
|
|
|
|
|
case OARRAYBYTESTRTMP:
|
2016-09-10 22:44:00 +02:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
|
|
|
|
|
|
|
|
|
if !instrumenting {
|
|
|
|
|
// Let the backend handle OARRAYBYTESTRTMP directly
|
|
|
|
|
// to avoid a function call to slicebytetostringtmp.
|
|
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-13 14:40:36 -05:00
|
|
|
n = mkcall("slicebytetostringtmp", n.Type, init, n.Left)
|
|
|
|
|
|
|
|
|
|
// slicerunetostring(*[32]byte, []rune) string;
|
|
|
|
|
case OARRAYRUNESTR:
|
2015-02-23 16:07:24 -05:00
|
|
|
a := nodnil()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
if n.Esc == EscNone {
|
|
|
|
|
// Create temporary buffer for string on stack.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
t := types.NewArray(types.Types[TUINT8], tmpstringbufsize)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OADDR, temp(t), nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
n = mkcall("slicerunetostring", n.Type, init, a, n.Left)
|
|
|
|
|
|
|
|
|
|
// stringtoslicebyte(*32[byte], string) []byte;
|
|
|
|
|
case OSTRARRAYBYTE:
|
2015-02-23 16:07:24 -05:00
|
|
|
a := nodnil()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
if n.Esc == EscNone {
|
|
|
|
|
// Create temporary buffer for slice on stack.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
t := types.NewArray(types.Types[TUINT8], tmpstringbufsize)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OADDR, temp(t), nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = mkcall("stringtoslicebyte", n.Type, init, a, conv(n.Left, types.Types[TSTRING]))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OSTRARRAYBYTETMP:
|
2016-10-27 23:31:38 +02:00
|
|
|
// []byte(string) conversion that creates a slice
|
|
|
|
|
// referring to the actual string bytes.
|
|
|
|
|
// This conversion is handled later by the backend and
|
|
|
|
|
// is only for use by internal compiler optimizations
|
|
|
|
|
// that know that the slice won't be mutated.
|
|
|
|
|
// The only such case today is:
|
|
|
|
|
// for i, c := range []byte(string)
|
|
|
|
|
n.Left = walkexpr(n.Left, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// stringtoslicerune(*[32]rune, string) []rune
|
|
|
|
|
case OSTRARRAYRUNE:
|
2015-02-23 16:07:24 -05:00
|
|
|
a := nodnil()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
if n.Esc == EscNone {
|
|
|
|
|
// Create temporary buffer for slice on stack.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
t := types.NewArray(types.Types[TINT32], tmpstringbufsize)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OADDR, temp(t), nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
n = mkcall("stringtoslicerune", n.Type, init, a, n.Left)
|
|
|
|
|
|
|
|
|
|
// ifaceeq(i1 any-1, i2 any-2) (ret bool);
|
|
|
|
|
case OCMPIFACE:
|
2016-09-15 14:34:20 +10:00
|
|
|
if !eqtype(n.Left.Type, n.Right.Type) {
|
2016-04-27 15:10:10 +10:00
|
|
|
Fatalf("ifaceeq %v %v %v", n.Op, n.Left.Type, n.Right.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-02-23 16:07:24 -05:00
|
|
|
var fn *Node
|
2016-04-01 13:36:24 -07:00
|
|
|
if n.Left.Type.IsEmptyInterface() {
|
2016-03-04 15:19:06 -08:00
|
|
|
fn = syslook("efaceeq")
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-03-04 15:19:06 -08:00
|
|
|
fn = syslook("ifaceeq")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
n.Right = cheapexpr(n.Right, init)
|
|
|
|
|
n.Left = cheapexpr(n.Left, init)
|
2017-03-24 14:03:15 -07:00
|
|
|
lt := nod(OITAB, n.Left, nil)
|
|
|
|
|
rt := nod(OITAB, n.Right, nil)
|
|
|
|
|
ld := nod(OIDATA, n.Left, nil)
|
|
|
|
|
rd := nod(OIDATA, n.Right, nil)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
ld.Type = types.Types[TUNSAFEPTR]
|
|
|
|
|
rd.Type = types.Types[TUNSAFEPTR]
|
2017-04-25 18:02:43 -07:00
|
|
|
ld.SetTypecheck(1)
|
|
|
|
|
rd.SetTypecheck(1)
|
2017-03-24 14:03:15 -07:00
|
|
|
call := mkcall1(fn, n.Type, init, lt, ld, rd)
|
|
|
|
|
|
|
|
|
|
// Check itable/type before full compare.
|
|
|
|
|
// Note: short-circuited because order matters.
|
2015-09-24 23:21:18 +02:00
|
|
|
// TODO(marvin): Fix Node.EType type union.
|
2017-03-24 14:03:15 -07:00
|
|
|
var cmp *Node
|
2015-09-24 23:21:18 +02:00
|
|
|
if Op(n.Etype) == OEQ {
|
2017-03-24 14:03:15 -07:00
|
|
|
cmp = nod(OANDAND, nod(OEQ, lt, rt), call)
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2017-03-24 14:03:15 -07:00
|
|
|
cmp = nod(OOROR, nod(ONE, lt, rt), nod(ONOT, call, nil))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2017-03-24 14:03:15 -07:00
|
|
|
cmp = typecheck(cmp, Erv)
|
|
|
|
|
cmp = walkexpr(cmp, init)
|
|
|
|
|
cmp.Type = n.Type
|
|
|
|
|
n = cmp
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-06-19 07:20:28 -07:00
|
|
|
case OARRAYLIT, OSLICELIT, OMAPLIT, OSTRUCTLIT, OPTRLIT:
|
2016-04-19 12:08:33 -07:00
|
|
|
if isStaticCompositeLiteral(n) {
|
2016-04-18 11:17:55 -07:00
|
|
|
// n can be directly represented in the read-only data section.
|
|
|
|
|
// Make direct reference to the static data. See issue 12841.
|
2016-06-20 08:18:22 -07:00
|
|
|
vstat := staticname(n.Type)
|
2017-02-27 19:56:38 +02:00
|
|
|
vstat.Name.SetReadonly(true)
|
2016-06-19 14:12:59 -07:00
|
|
|
fixedlit(inInitFunction, initKindStatic, n, vstat, init)
|
2016-04-18 11:17:55 -07:00
|
|
|
n = vstat
|
|
|
|
|
n = typecheck(n, Erv)
|
|
|
|
|
break
|
|
|
|
|
}
|
2015-02-23 16:07:24 -05:00
|
|
|
var_ := temp(n.Type)
|
2016-09-04 09:34:03 -07:00
|
|
|
anylit(n, var_, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
n = var_
|
|
|
|
|
|
|
|
|
|
case OSEND:
|
2015-02-23 16:07:24 -05:00
|
|
|
n1 := n.Right
|
2016-03-30 10:57:47 -07:00
|
|
|
n1 = assignconv(n1, n.Left.Type.Elem(), "chan send")
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n1 = walkexpr(n1, init)
|
2016-09-16 11:00:54 +10:00
|
|
|
n1 = nod(OADDR, n1, nil)
|
2017-03-18 15:55:41 +00:00
|
|
|
n = mkcall1(chanfn("chansend1", 2, n.Left.Type), nil, init, n.Left, n1)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OCLOSURE:
|
|
|
|
|
n = walkclosure(n, init)
|
|
|
|
|
|
|
|
|
|
case OCALLPART:
|
|
|
|
|
n = walkpartialcall(n, init)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Expressions that are constant at run time but not
|
|
|
|
|
// considered const by the language spec are not turned into
|
|
|
|
|
// constants until walk. For example, if n is y%1 == 0, the
|
|
|
|
|
// walk of y%1 may have replaced it by 0.
|
|
|
|
|
// Check whether n with its updated args is itself now a constant.
|
2015-02-23 16:07:24 -05:00
|
|
|
t := n.Type
|
2015-02-13 14:40:36 -05:00
|
|
|
evconst(n)
|
2017-03-24 13:30:19 -07:00
|
|
|
if n.Type != t {
|
|
|
|
|
Fatalf("evconst changed Type: %v had type %v, now %v", n, t, n.Type)
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Op == OLITERAL {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = typecheck(n, Erv)
|
2017-04-13 06:11:36 -07:00
|
|
|
// Emit string symbol now to avoid emitting
|
|
|
|
|
// any concurrently during the backend.
|
|
|
|
|
if s, ok := n.Val().U.(string); ok {
|
|
|
|
|
_ = stringsym(s)
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-03 13:38:49 -08:00
|
|
|
updateHasCall(n)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
if Debug['w'] != 0 && n != nil {
|
|
|
|
|
Dump("walk", n)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
lineno = lno
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-04-21 11:55:33 -07:00
|
|
|
// TODO(josharian): combine this with its caller and simplify
|
cmd/internal/gc: optimize slice + write barrier
The code generated for a slice x[i:j] or x[i:j:k] computes the entire
new slice (base, len, cap) and then uses it as the evaluation of the
slice expression.
If the slice is part of an update x = x[i:j] or x = x[i:j:k], there are
opportunities to avoid computing some of these fields.
For x = x[0:i], we know that only the len is changing;
base can be ignored completely, and cap can be left unmodified.
For x = x[0:i:j], we know that only len and cap are changing;
base can be ignored completely.
For x = x[i:i], we know that the resulting cap is zero, and we don't
adjust the base during a slice producing a zero-cap result,
so again base can be ignored completely.
No write to base, no write barrier.
The old slice code was trying to work at a Go syntax level, mainly
because that was how you wrote code just once instead of once
per architecture. Now the compiler is factored a bit better and we
can implement slice during code generation but still have one copy
of the code. So the new code is working at that lower level.
(It must, to update only parts of the result.)
This CL by itself:
name old mean new mean delta
BinaryTree17 5.81s × (0.98,1.03) 5.71s × (0.96,1.05) ~ (p=0.101)
Fannkuch11 4.35s × (1.00,1.00) 4.39s × (1.00,1.00) +0.79% (p=0.000)
FmtFprintfEmpty 86.0ns × (0.94,1.11) 82.6ns × (0.98,1.04) -3.86% (p=0.048)
FmtFprintfString 276ns × (0.98,1.04) 273ns × (0.98,1.02) ~ (p=0.235)
FmtFprintfInt 274ns × (0.98,1.06) 270ns × (0.99,1.01) ~ (p=0.119)
FmtFprintfIntInt 506ns × (0.99,1.01) 475ns × (0.99,1.01) -6.02% (p=0.000)
FmtFprintfPrefixedInt 391ns × (0.99,1.01) 393ns × (1.00,1.01) ~ (p=0.139)
FmtFprintfFloat 566ns × (0.99,1.01) 574ns × (1.00,1.01) +1.33% (p=0.001)
FmtManyArgs 1.91µs × (0.99,1.01) 1.87µs × (0.99,1.02) -1.83% (p=0.000)
GobDecode 15.3ms × (0.99,1.02) 15.0ms × (0.98,1.05) -1.84% (p=0.042)
GobEncode 11.5ms × (0.97,1.03) 11.4ms × (0.99,1.03) ~ (p=0.152)
Gzip 645ms × (0.99,1.01) 647ms × (0.99,1.01) ~ (p=0.265)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.90% (p=0.000)
HTTPClientServer 90.5µs × (0.97,1.04) 88.5µs × (0.99,1.03) -2.27% (p=0.014)
JSONEncode 32.0ms × (0.98,1.03) 29.6ms × (0.98,1.01) -7.51% (p=0.000)
JSONDecode 114ms × (0.99,1.01) 104ms × (1.00,1.01) -8.60% (p=0.000)
Mandelbrot200 6.04ms × (1.00,1.01) 6.02ms × (1.00,1.00) ~ (p=0.057)
GoParse 6.47ms × (0.97,1.05) 6.37ms × (0.97,1.04) ~ (p=0.105)
RegexpMatchEasy0_32 171ns × (0.93,1.07) 152ns × (0.99,1.01) -11.09% (p=0.000)
RegexpMatchEasy0_1K 550ns × (0.98,1.01) 530ns × (1.00,1.00) -3.78% (p=0.000)
RegexpMatchEasy1_32 135ns × (0.99,1.02) 134ns × (0.99,1.01) -1.33% (p=0.002)
RegexpMatchEasy1_1K 879ns × (1.00,1.01) 865ns × (1.00,1.00) -1.58% (p=0.000)
RegexpMatchMedium_32 243ns × (1.00,1.00) 233ns × (1.00,1.00) -4.30% (p=0.000)
RegexpMatchMedium_1K 70.3µs × (1.00,1.00) 69.5µs × (1.00,1.00) -1.13% (p=0.000)
RegexpMatchHard_32 3.82µs × (1.00,1.01) 3.74µs × (1.00,1.00) -1.95% (p=0.000)
RegexpMatchHard_1K 117µs × (1.00,1.00) 115µs × (1.00,1.00) -1.69% (p=0.000)
Revcomp 917ms × (0.97,1.04) 920ms × (0.97,1.04) ~ (p=0.786)
Template 114ms × (0.99,1.01) 117ms × (0.99,1.01) +2.58% (p=0.000)
TimeParse 622ns × (0.99,1.01) 615ns × (0.99,1.00) -1.06% (p=0.000)
TimeFormat 665ns × (0.99,1.01) 654ns × (0.99,1.00) -1.70% (p=0.000)
This CL and previous CL (append) combined:
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.71s × (0.96,1.05) ~ (p=0.638)
Fannkuch11 4.41s × (0.98,1.03) 4.39s × (1.00,1.00) ~ (p=0.474)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 82.6ns × (0.98,1.04) -10.89% (p=0.004)
FmtFprintfString 281ns × (0.96,1.08) 273ns × (0.98,1.02) ~ (p=0.078)
FmtFprintfInt 288ns × (0.97,1.06) 270ns × (0.99,1.01) -6.37% (p=0.000)
FmtFprintfIntInt 493ns × (0.97,1.04) 475ns × (0.99,1.01) -3.53% (p=0.002)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 393ns × (1.00,1.01) -7.07% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 574ns × (1.00,1.01) -4.02% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.87µs × (0.99,1.02) ~ (p=0.305)
GobDecode 14.8ms × (0.98,1.03) 15.0ms × (0.98,1.05) ~ (p=0.237)
GobEncode 12.3ms × (0.98,1.01) 11.4ms × (0.99,1.03) -6.95% (p=0.000)
Gzip 656ms × (0.99,1.05) 647ms × (0.99,1.01) ~ (p=0.101)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.58% (p=0.001)
HTTPClientServer 91.2µs × (0.97,1.04) 88.5µs × (0.99,1.03) -3.02% (p=0.003)
JSONEncode 32.6ms × (0.97,1.08) 29.6ms × (0.98,1.01) -9.10% (p=0.000)
JSONDecode 114ms × (0.97,1.05) 104ms × (1.00,1.01) -8.74% (p=0.000)
Mandelbrot200 6.11ms × (0.98,1.04) 6.02ms × (1.00,1.00) ~ (p=0.090)
GoParse 6.66ms × (0.97,1.04) 6.37ms × (0.97,1.04) -4.41% (p=0.000)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 152ns × (0.99,1.01) -4.69% (p=0.000)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 530ns × (1.00,1.00) -1.57% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 134ns × (0.99,1.01) -2.91% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 865ns × (1.00,1.00) -0.51% (p=0.012)
RegexpMatchMedium_32 252ns × (0.99,1.01) 233ns × (1.00,1.00) -7.85% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 69.5µs × (1.00,1.00) -4.43% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.74µs × (1.00,1.00) -2.74% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 115µs × (1.00,1.00) -2.24% (p=0.000)
Revcomp 920ms × (0.97,1.07) 920ms × (0.97,1.04) ~ (p=0.998)
Template 129ms × (0.98,1.03) 117ms × (0.99,1.01) -9.79% (p=0.000)
TimeParse 619ns × (0.99,1.01) 615ns × (0.99,1.00) -0.57% (p=0.011)
TimeFormat 661ns × (0.98,1.04) 654ns × (0.99,1.00) ~ (p=0.223)
Change-Id: If054d81ab2c71d8d62cf54b5b1fac2af66b387fc
Reviewed-on: https://go-review.googlesource.com/9813
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-05-06 12:35:53 -04:00
|
|
|
func reduceSlice(n *Node) *Node {
|
2016-04-21 11:55:33 -07:00
|
|
|
low, high, max := n.SliceBounds()
|
|
|
|
|
if high != nil && high.Op == OLEN && samesafeexpr(n.Left, high.Left) {
|
cmd/internal/gc: optimize slice + write barrier
The code generated for a slice x[i:j] or x[i:j:k] computes the entire
new slice (base, len, cap) and then uses it as the evaluation of the
slice expression.
If the slice is part of an update x = x[i:j] or x = x[i:j:k], there are
opportunities to avoid computing some of these fields.
For x = x[0:i], we know that only the len is changing;
base can be ignored completely, and cap can be left unmodified.
For x = x[0:i:j], we know that only len and cap are changing;
base can be ignored completely.
For x = x[i:i], we know that the resulting cap is zero, and we don't
adjust the base during a slice producing a zero-cap result,
so again base can be ignored completely.
No write to base, no write barrier.
The old slice code was trying to work at a Go syntax level, mainly
because that was how you wrote code just once instead of once
per architecture. Now the compiler is factored a bit better and we
can implement slice during code generation but still have one copy
of the code. So the new code is working at that lower level.
(It must, to update only parts of the result.)
This CL by itself:
name old mean new mean delta
BinaryTree17 5.81s × (0.98,1.03) 5.71s × (0.96,1.05) ~ (p=0.101)
Fannkuch11 4.35s × (1.00,1.00) 4.39s × (1.00,1.00) +0.79% (p=0.000)
FmtFprintfEmpty 86.0ns × (0.94,1.11) 82.6ns × (0.98,1.04) -3.86% (p=0.048)
FmtFprintfString 276ns × (0.98,1.04) 273ns × (0.98,1.02) ~ (p=0.235)
FmtFprintfInt 274ns × (0.98,1.06) 270ns × (0.99,1.01) ~ (p=0.119)
FmtFprintfIntInt 506ns × (0.99,1.01) 475ns × (0.99,1.01) -6.02% (p=0.000)
FmtFprintfPrefixedInt 391ns × (0.99,1.01) 393ns × (1.00,1.01) ~ (p=0.139)
FmtFprintfFloat 566ns × (0.99,1.01) 574ns × (1.00,1.01) +1.33% (p=0.001)
FmtManyArgs 1.91µs × (0.99,1.01) 1.87µs × (0.99,1.02) -1.83% (p=0.000)
GobDecode 15.3ms × (0.99,1.02) 15.0ms × (0.98,1.05) -1.84% (p=0.042)
GobEncode 11.5ms × (0.97,1.03) 11.4ms × (0.99,1.03) ~ (p=0.152)
Gzip 645ms × (0.99,1.01) 647ms × (0.99,1.01) ~ (p=0.265)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.90% (p=0.000)
HTTPClientServer 90.5µs × (0.97,1.04) 88.5µs × (0.99,1.03) -2.27% (p=0.014)
JSONEncode 32.0ms × (0.98,1.03) 29.6ms × (0.98,1.01) -7.51% (p=0.000)
JSONDecode 114ms × (0.99,1.01) 104ms × (1.00,1.01) -8.60% (p=0.000)
Mandelbrot200 6.04ms × (1.00,1.01) 6.02ms × (1.00,1.00) ~ (p=0.057)
GoParse 6.47ms × (0.97,1.05) 6.37ms × (0.97,1.04) ~ (p=0.105)
RegexpMatchEasy0_32 171ns × (0.93,1.07) 152ns × (0.99,1.01) -11.09% (p=0.000)
RegexpMatchEasy0_1K 550ns × (0.98,1.01) 530ns × (1.00,1.00) -3.78% (p=0.000)
RegexpMatchEasy1_32 135ns × (0.99,1.02) 134ns × (0.99,1.01) -1.33% (p=0.002)
RegexpMatchEasy1_1K 879ns × (1.00,1.01) 865ns × (1.00,1.00) -1.58% (p=0.000)
RegexpMatchMedium_32 243ns × (1.00,1.00) 233ns × (1.00,1.00) -4.30% (p=0.000)
RegexpMatchMedium_1K 70.3µs × (1.00,1.00) 69.5µs × (1.00,1.00) -1.13% (p=0.000)
RegexpMatchHard_32 3.82µs × (1.00,1.01) 3.74µs × (1.00,1.00) -1.95% (p=0.000)
RegexpMatchHard_1K 117µs × (1.00,1.00) 115µs × (1.00,1.00) -1.69% (p=0.000)
Revcomp 917ms × (0.97,1.04) 920ms × (0.97,1.04) ~ (p=0.786)
Template 114ms × (0.99,1.01) 117ms × (0.99,1.01) +2.58% (p=0.000)
TimeParse 622ns × (0.99,1.01) 615ns × (0.99,1.00) -1.06% (p=0.000)
TimeFormat 665ns × (0.99,1.01) 654ns × (0.99,1.00) -1.70% (p=0.000)
This CL and previous CL (append) combined:
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.71s × (0.96,1.05) ~ (p=0.638)
Fannkuch11 4.41s × (0.98,1.03) 4.39s × (1.00,1.00) ~ (p=0.474)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 82.6ns × (0.98,1.04) -10.89% (p=0.004)
FmtFprintfString 281ns × (0.96,1.08) 273ns × (0.98,1.02) ~ (p=0.078)
FmtFprintfInt 288ns × (0.97,1.06) 270ns × (0.99,1.01) -6.37% (p=0.000)
FmtFprintfIntInt 493ns × (0.97,1.04) 475ns × (0.99,1.01) -3.53% (p=0.002)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 393ns × (1.00,1.01) -7.07% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 574ns × (1.00,1.01) -4.02% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.87µs × (0.99,1.02) ~ (p=0.305)
GobDecode 14.8ms × (0.98,1.03) 15.0ms × (0.98,1.05) ~ (p=0.237)
GobEncode 12.3ms × (0.98,1.01) 11.4ms × (0.99,1.03) -6.95% (p=0.000)
Gzip 656ms × (0.99,1.05) 647ms × (0.99,1.01) ~ (p=0.101)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.58% (p=0.001)
HTTPClientServer 91.2µs × (0.97,1.04) 88.5µs × (0.99,1.03) -3.02% (p=0.003)
JSONEncode 32.6ms × (0.97,1.08) 29.6ms × (0.98,1.01) -9.10% (p=0.000)
JSONDecode 114ms × (0.97,1.05) 104ms × (1.00,1.01) -8.74% (p=0.000)
Mandelbrot200 6.11ms × (0.98,1.04) 6.02ms × (1.00,1.00) ~ (p=0.090)
GoParse 6.66ms × (0.97,1.04) 6.37ms × (0.97,1.04) -4.41% (p=0.000)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 152ns × (0.99,1.01) -4.69% (p=0.000)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 530ns × (1.00,1.00) -1.57% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 134ns × (0.99,1.01) -2.91% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 865ns × (1.00,1.00) -0.51% (p=0.012)
RegexpMatchMedium_32 252ns × (0.99,1.01) 233ns × (1.00,1.00) -7.85% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 69.5µs × (1.00,1.00) -4.43% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.74µs × (1.00,1.00) -2.74% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 115µs × (1.00,1.00) -2.24% (p=0.000)
Revcomp 920ms × (0.97,1.07) 920ms × (0.97,1.04) ~ (p=0.998)
Template 129ms × (0.98,1.03) 117ms × (0.99,1.01) -9.79% (p=0.000)
TimeParse 619ns × (0.99,1.01) 615ns × (0.99,1.00) -0.57% (p=0.011)
TimeFormat 661ns × (0.98,1.04) 654ns × (0.99,1.00) ~ (p=0.223)
Change-Id: If054d81ab2c71d8d62cf54b5b1fac2af66b387fc
Reviewed-on: https://go-review.googlesource.com/9813
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-05-06 12:35:53 -04:00
|
|
|
// Reduce x[i:len(x)] to x[i:].
|
2016-04-21 11:55:33 -07:00
|
|
|
high = nil
|
cmd/internal/gc: optimize slice + write barrier
The code generated for a slice x[i:j] or x[i:j:k] computes the entire
new slice (base, len, cap) and then uses it as the evaluation of the
slice expression.
If the slice is part of an update x = x[i:j] or x = x[i:j:k], there are
opportunities to avoid computing some of these fields.
For x = x[0:i], we know that only the len is changing;
base can be ignored completely, and cap can be left unmodified.
For x = x[0:i:j], we know that only len and cap are changing;
base can be ignored completely.
For x = x[i:i], we know that the resulting cap is zero, and we don't
adjust the base during a slice producing a zero-cap result,
so again base can be ignored completely.
No write to base, no write barrier.
The old slice code was trying to work at a Go syntax level, mainly
because that was how you wrote code just once instead of once
per architecture. Now the compiler is factored a bit better and we
can implement slice during code generation but still have one copy
of the code. So the new code is working at that lower level.
(It must, to update only parts of the result.)
This CL by itself:
name old mean new mean delta
BinaryTree17 5.81s × (0.98,1.03) 5.71s × (0.96,1.05) ~ (p=0.101)
Fannkuch11 4.35s × (1.00,1.00) 4.39s × (1.00,1.00) +0.79% (p=0.000)
FmtFprintfEmpty 86.0ns × (0.94,1.11) 82.6ns × (0.98,1.04) -3.86% (p=0.048)
FmtFprintfString 276ns × (0.98,1.04) 273ns × (0.98,1.02) ~ (p=0.235)
FmtFprintfInt 274ns × (0.98,1.06) 270ns × (0.99,1.01) ~ (p=0.119)
FmtFprintfIntInt 506ns × (0.99,1.01) 475ns × (0.99,1.01) -6.02% (p=0.000)
FmtFprintfPrefixedInt 391ns × (0.99,1.01) 393ns × (1.00,1.01) ~ (p=0.139)
FmtFprintfFloat 566ns × (0.99,1.01) 574ns × (1.00,1.01) +1.33% (p=0.001)
FmtManyArgs 1.91µs × (0.99,1.01) 1.87µs × (0.99,1.02) -1.83% (p=0.000)
GobDecode 15.3ms × (0.99,1.02) 15.0ms × (0.98,1.05) -1.84% (p=0.042)
GobEncode 11.5ms × (0.97,1.03) 11.4ms × (0.99,1.03) ~ (p=0.152)
Gzip 645ms × (0.99,1.01) 647ms × (0.99,1.01) ~ (p=0.265)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.90% (p=0.000)
HTTPClientServer 90.5µs × (0.97,1.04) 88.5µs × (0.99,1.03) -2.27% (p=0.014)
JSONEncode 32.0ms × (0.98,1.03) 29.6ms × (0.98,1.01) -7.51% (p=0.000)
JSONDecode 114ms × (0.99,1.01) 104ms × (1.00,1.01) -8.60% (p=0.000)
Mandelbrot200 6.04ms × (1.00,1.01) 6.02ms × (1.00,1.00) ~ (p=0.057)
GoParse 6.47ms × (0.97,1.05) 6.37ms × (0.97,1.04) ~ (p=0.105)
RegexpMatchEasy0_32 171ns × (0.93,1.07) 152ns × (0.99,1.01) -11.09% (p=0.000)
RegexpMatchEasy0_1K 550ns × (0.98,1.01) 530ns × (1.00,1.00) -3.78% (p=0.000)
RegexpMatchEasy1_32 135ns × (0.99,1.02) 134ns × (0.99,1.01) -1.33% (p=0.002)
RegexpMatchEasy1_1K 879ns × (1.00,1.01) 865ns × (1.00,1.00) -1.58% (p=0.000)
RegexpMatchMedium_32 243ns × (1.00,1.00) 233ns × (1.00,1.00) -4.30% (p=0.000)
RegexpMatchMedium_1K 70.3µs × (1.00,1.00) 69.5µs × (1.00,1.00) -1.13% (p=0.000)
RegexpMatchHard_32 3.82µs × (1.00,1.01) 3.74µs × (1.00,1.00) -1.95% (p=0.000)
RegexpMatchHard_1K 117µs × (1.00,1.00) 115µs × (1.00,1.00) -1.69% (p=0.000)
Revcomp 917ms × (0.97,1.04) 920ms × (0.97,1.04) ~ (p=0.786)
Template 114ms × (0.99,1.01) 117ms × (0.99,1.01) +2.58% (p=0.000)
TimeParse 622ns × (0.99,1.01) 615ns × (0.99,1.00) -1.06% (p=0.000)
TimeFormat 665ns × (0.99,1.01) 654ns × (0.99,1.00) -1.70% (p=0.000)
This CL and previous CL (append) combined:
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.71s × (0.96,1.05) ~ (p=0.638)
Fannkuch11 4.41s × (0.98,1.03) 4.39s × (1.00,1.00) ~ (p=0.474)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 82.6ns × (0.98,1.04) -10.89% (p=0.004)
FmtFprintfString 281ns × (0.96,1.08) 273ns × (0.98,1.02) ~ (p=0.078)
FmtFprintfInt 288ns × (0.97,1.06) 270ns × (0.99,1.01) -6.37% (p=0.000)
FmtFprintfIntInt 493ns × (0.97,1.04) 475ns × (0.99,1.01) -3.53% (p=0.002)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 393ns × (1.00,1.01) -7.07% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 574ns × (1.00,1.01) -4.02% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.87µs × (0.99,1.02) ~ (p=0.305)
GobDecode 14.8ms × (0.98,1.03) 15.0ms × (0.98,1.05) ~ (p=0.237)
GobEncode 12.3ms × (0.98,1.01) 11.4ms × (0.99,1.03) -6.95% (p=0.000)
Gzip 656ms × (0.99,1.05) 647ms × (0.99,1.01) ~ (p=0.101)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.58% (p=0.001)
HTTPClientServer 91.2µs × (0.97,1.04) 88.5µs × (0.99,1.03) -3.02% (p=0.003)
JSONEncode 32.6ms × (0.97,1.08) 29.6ms × (0.98,1.01) -9.10% (p=0.000)
JSONDecode 114ms × (0.97,1.05) 104ms × (1.00,1.01) -8.74% (p=0.000)
Mandelbrot200 6.11ms × (0.98,1.04) 6.02ms × (1.00,1.00) ~ (p=0.090)
GoParse 6.66ms × (0.97,1.04) 6.37ms × (0.97,1.04) -4.41% (p=0.000)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 152ns × (0.99,1.01) -4.69% (p=0.000)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 530ns × (1.00,1.00) -1.57% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 134ns × (0.99,1.01) -2.91% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 865ns × (1.00,1.00) -0.51% (p=0.012)
RegexpMatchMedium_32 252ns × (0.99,1.01) 233ns × (1.00,1.00) -7.85% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 69.5µs × (1.00,1.00) -4.43% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.74µs × (1.00,1.00) -2.74% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 115µs × (1.00,1.00) -2.24% (p=0.000)
Revcomp 920ms × (0.97,1.07) 920ms × (0.97,1.04) ~ (p=0.998)
Template 129ms × (0.98,1.03) 117ms × (0.99,1.01) -9.79% (p=0.000)
TimeParse 619ns × (0.99,1.01) 615ns × (0.99,1.00) -0.57% (p=0.011)
TimeFormat 661ns × (0.98,1.04) 654ns × (0.99,1.00) ~ (p=0.223)
Change-Id: If054d81ab2c71d8d62cf54b5b1fac2af66b387fc
Reviewed-on: https://go-review.googlesource.com/9813
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-05-06 12:35:53 -04:00
|
|
|
}
|
2016-04-21 11:55:33 -07:00
|
|
|
n.SetSliceBounds(low, high, max)
|
|
|
|
|
if (n.Op == OSLICE || n.Op == OSLICESTR) && low == nil && high == nil {
|
cmd/internal/gc: optimize slice + write barrier
The code generated for a slice x[i:j] or x[i:j:k] computes the entire
new slice (base, len, cap) and then uses it as the evaluation of the
slice expression.
If the slice is part of an update x = x[i:j] or x = x[i:j:k], there are
opportunities to avoid computing some of these fields.
For x = x[0:i], we know that only the len is changing;
base can be ignored completely, and cap can be left unmodified.
For x = x[0:i:j], we know that only len and cap are changing;
base can be ignored completely.
For x = x[i:i], we know that the resulting cap is zero, and we don't
adjust the base during a slice producing a zero-cap result,
so again base can be ignored completely.
No write to base, no write barrier.
The old slice code was trying to work at a Go syntax level, mainly
because that was how you wrote code just once instead of once
per architecture. Now the compiler is factored a bit better and we
can implement slice during code generation but still have one copy
of the code. So the new code is working at that lower level.
(It must, to update only parts of the result.)
This CL by itself:
name old mean new mean delta
BinaryTree17 5.81s × (0.98,1.03) 5.71s × (0.96,1.05) ~ (p=0.101)
Fannkuch11 4.35s × (1.00,1.00) 4.39s × (1.00,1.00) +0.79% (p=0.000)
FmtFprintfEmpty 86.0ns × (0.94,1.11) 82.6ns × (0.98,1.04) -3.86% (p=0.048)
FmtFprintfString 276ns × (0.98,1.04) 273ns × (0.98,1.02) ~ (p=0.235)
FmtFprintfInt 274ns × (0.98,1.06) 270ns × (0.99,1.01) ~ (p=0.119)
FmtFprintfIntInt 506ns × (0.99,1.01) 475ns × (0.99,1.01) -6.02% (p=0.000)
FmtFprintfPrefixedInt 391ns × (0.99,1.01) 393ns × (1.00,1.01) ~ (p=0.139)
FmtFprintfFloat 566ns × (0.99,1.01) 574ns × (1.00,1.01) +1.33% (p=0.001)
FmtManyArgs 1.91µs × (0.99,1.01) 1.87µs × (0.99,1.02) -1.83% (p=0.000)
GobDecode 15.3ms × (0.99,1.02) 15.0ms × (0.98,1.05) -1.84% (p=0.042)
GobEncode 11.5ms × (0.97,1.03) 11.4ms × (0.99,1.03) ~ (p=0.152)
Gzip 645ms × (0.99,1.01) 647ms × (0.99,1.01) ~ (p=0.265)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.90% (p=0.000)
HTTPClientServer 90.5µs × (0.97,1.04) 88.5µs × (0.99,1.03) -2.27% (p=0.014)
JSONEncode 32.0ms × (0.98,1.03) 29.6ms × (0.98,1.01) -7.51% (p=0.000)
JSONDecode 114ms × (0.99,1.01) 104ms × (1.00,1.01) -8.60% (p=0.000)
Mandelbrot200 6.04ms × (1.00,1.01) 6.02ms × (1.00,1.00) ~ (p=0.057)
GoParse 6.47ms × (0.97,1.05) 6.37ms × (0.97,1.04) ~ (p=0.105)
RegexpMatchEasy0_32 171ns × (0.93,1.07) 152ns × (0.99,1.01) -11.09% (p=0.000)
RegexpMatchEasy0_1K 550ns × (0.98,1.01) 530ns × (1.00,1.00) -3.78% (p=0.000)
RegexpMatchEasy1_32 135ns × (0.99,1.02) 134ns × (0.99,1.01) -1.33% (p=0.002)
RegexpMatchEasy1_1K 879ns × (1.00,1.01) 865ns × (1.00,1.00) -1.58% (p=0.000)
RegexpMatchMedium_32 243ns × (1.00,1.00) 233ns × (1.00,1.00) -4.30% (p=0.000)
RegexpMatchMedium_1K 70.3µs × (1.00,1.00) 69.5µs × (1.00,1.00) -1.13% (p=0.000)
RegexpMatchHard_32 3.82µs × (1.00,1.01) 3.74µs × (1.00,1.00) -1.95% (p=0.000)
RegexpMatchHard_1K 117µs × (1.00,1.00) 115µs × (1.00,1.00) -1.69% (p=0.000)
Revcomp 917ms × (0.97,1.04) 920ms × (0.97,1.04) ~ (p=0.786)
Template 114ms × (0.99,1.01) 117ms × (0.99,1.01) +2.58% (p=0.000)
TimeParse 622ns × (0.99,1.01) 615ns × (0.99,1.00) -1.06% (p=0.000)
TimeFormat 665ns × (0.99,1.01) 654ns × (0.99,1.00) -1.70% (p=0.000)
This CL and previous CL (append) combined:
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.71s × (0.96,1.05) ~ (p=0.638)
Fannkuch11 4.41s × (0.98,1.03) 4.39s × (1.00,1.00) ~ (p=0.474)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 82.6ns × (0.98,1.04) -10.89% (p=0.004)
FmtFprintfString 281ns × (0.96,1.08) 273ns × (0.98,1.02) ~ (p=0.078)
FmtFprintfInt 288ns × (0.97,1.06) 270ns × (0.99,1.01) -6.37% (p=0.000)
FmtFprintfIntInt 493ns × (0.97,1.04) 475ns × (0.99,1.01) -3.53% (p=0.002)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 393ns × (1.00,1.01) -7.07% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 574ns × (1.00,1.01) -4.02% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.87µs × (0.99,1.02) ~ (p=0.305)
GobDecode 14.8ms × (0.98,1.03) 15.0ms × (0.98,1.05) ~ (p=0.237)
GobEncode 12.3ms × (0.98,1.01) 11.4ms × (0.99,1.03) -6.95% (p=0.000)
Gzip 656ms × (0.99,1.05) 647ms × (0.99,1.01) ~ (p=0.101)
Gunzip 142ms × (1.00,1.00) 143ms × (1.00,1.01) +0.58% (p=0.001)
HTTPClientServer 91.2µs × (0.97,1.04) 88.5µs × (0.99,1.03) -3.02% (p=0.003)
JSONEncode 32.6ms × (0.97,1.08) 29.6ms × (0.98,1.01) -9.10% (p=0.000)
JSONDecode 114ms × (0.97,1.05) 104ms × (1.00,1.01) -8.74% (p=0.000)
Mandelbrot200 6.11ms × (0.98,1.04) 6.02ms × (1.00,1.00) ~ (p=0.090)
GoParse 6.66ms × (0.97,1.04) 6.37ms × (0.97,1.04) -4.41% (p=0.000)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 152ns × (0.99,1.01) -4.69% (p=0.000)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 530ns × (1.00,1.00) -1.57% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 134ns × (0.99,1.01) -2.91% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 865ns × (1.00,1.00) -0.51% (p=0.012)
RegexpMatchMedium_32 252ns × (0.99,1.01) 233ns × (1.00,1.00) -7.85% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 69.5µs × (1.00,1.00) -4.43% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.74µs × (1.00,1.00) -2.74% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 115µs × (1.00,1.00) -2.24% (p=0.000)
Revcomp 920ms × (0.97,1.07) 920ms × (0.97,1.04) ~ (p=0.998)
Template 129ms × (0.98,1.03) 117ms × (0.99,1.01) -9.79% (p=0.000)
TimeParse 619ns × (0.99,1.01) 615ns × (0.99,1.00) -0.57% (p=0.011)
TimeFormat 661ns × (0.98,1.04) 654ns × (0.99,1.00) ~ (p=0.223)
Change-Id: If054d81ab2c71d8d62cf54b5b1fac2af66b387fc
Reviewed-on: https://go-review.googlesource.com/9813
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-05-06 12:35:53 -04:00
|
|
|
// Reduce x[:] to x.
|
|
|
|
|
if Debug_slice > 0 {
|
|
|
|
|
Warn("slice: omit slice operation")
|
|
|
|
|
}
|
|
|
|
|
return n.Left
|
|
|
|
|
}
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
func ascompatee1(op Op, l *Node, r *Node, init *Nodes) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
// convas will turn map assigns into function calls,
|
|
|
|
|
// making it impossible for reorder3 to work.
|
2016-09-16 11:00:54 +10:00
|
|
|
n := nod(OAS, l, r)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
if l.Op == OINDEXMAP {
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return convas(n, init)
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-08 10:26:20 -08:00
|
|
|
func ascompatee(op Op, nl, nr []*Node, init *Nodes) []*Node {
|
2015-10-22 09:51:12 +09:00
|
|
|
// check assign expression list to
|
|
|
|
|
// a expression list. called in
|
|
|
|
|
// expr-list = expr-list
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// ensure order of evaluation for function calls
|
2016-03-08 15:10:26 -08:00
|
|
|
for i := range nl {
|
|
|
|
|
nl[i] = safeexpr(nl[i], init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-08 15:10:26 -08:00
|
|
|
for i1 := range nr {
|
|
|
|
|
nr[i1] = safeexpr(nr[i1], init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
var nn []*Node
|
2016-03-09 12:39:36 -08:00
|
|
|
i := 0
|
|
|
|
|
for ; i < len(nl); i++ {
|
|
|
|
|
if i >= len(nr) {
|
|
|
|
|
break
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
// Do not generate 'x = x' during return. See issue 4014.
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
if op == ORETURN && samesafeexpr(nl[i], nr[i]) {
|
2015-02-13 14:40:36 -05:00
|
|
|
continue
|
|
|
|
|
}
|
2016-03-09 12:39:36 -08:00
|
|
|
nn = append(nn, ascompatee1(op, nl[i], nr[i], init))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// cannot happen: caller checked that lists had same length
|
2016-03-09 12:39:36 -08:00
|
|
|
if i < len(nl) || i < len(nr) {
|
2016-03-08 10:26:20 -08:00
|
|
|
var nln, nrn Nodes
|
|
|
|
|
nln.Set(nl)
|
|
|
|
|
nrn.Set(nr)
|
2017-04-23 05:10:21 -07:00
|
|
|
Fatalf("error in shape across %+v %v %+v / %d %d [%s]", nln, op, nrn, len(nl), len(nr), Curfn.funcname())
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
return nn
|
|
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// l is an lv and rt is the type of an rv
|
|
|
|
|
// return 1 if this implies a function call
|
|
|
|
|
// evaluating the lv or a function call
|
|
|
|
|
// in the conversion of the types
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func fncall(l *Node, rt *types.Type) bool {
|
2017-03-03 13:38:49 -08:00
|
|
|
if l.HasCall() || l.Op == OINDEXMAP {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2017-02-15 18:43:34 -08:00
|
|
|
if needwritebarrier(l) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-09-15 14:34:20 +10:00
|
|
|
if eqtype(l.Type, rt) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-09 19:32:10 -08:00
|
|
|
// check assign type list to
|
|
|
|
|
// a expression list. called in
|
|
|
|
|
// expr-list = func()
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func ascompatet(op Op, nl Nodes, nr *types.Type) []*Node {
|
2017-04-05 19:38:21 -07:00
|
|
|
if nl.Len() != nr.NumFields() {
|
|
|
|
|
Fatalf("ascompatet: assignment count mismatch: %d = %d", nl.Len(), nr.NumFields())
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-10-27 09:23:48 -07:00
|
|
|
var nn, mm Nodes
|
2017-04-05 19:38:21 -07:00
|
|
|
for i, l := range nl.Slice() {
|
2015-02-13 14:40:36 -05:00
|
|
|
if isblank(l) {
|
|
|
|
|
continue
|
|
|
|
|
}
|
2017-04-05 19:38:21 -07:00
|
|
|
r := nr.Field(i)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// any lv that causes a fn call must be
|
|
|
|
|
// deferred until all the return arguments
|
|
|
|
|
// have been pulled from the output arguments
|
2015-02-17 22:13:49 -05:00
|
|
|
if fncall(l, r.Type) {
|
2016-03-09 19:32:10 -08:00
|
|
|
tmp := temp(r.Type)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
tmp = typecheck(tmp, Erv)
|
2016-09-16 11:00:54 +10:00
|
|
|
a := nod(OAS, l, tmp)
|
2016-10-27 09:23:48 -07:00
|
|
|
a = convas(a, &mm)
|
|
|
|
|
mm.Append(a)
|
2015-02-13 14:40:36 -05:00
|
|
|
l = tmp
|
|
|
|
|
}
|
|
|
|
|
|
2016-10-27 09:23:48 -07:00
|
|
|
a := nod(OAS, l, nodarg(r, 0))
|
|
|
|
|
a = convas(a, &nn)
|
2017-03-03 13:38:49 -08:00
|
|
|
updateHasCall(a)
|
|
|
|
|
if a.HasCall() {
|
2015-02-13 14:40:36 -05:00
|
|
|
Dump("ascompatet ucount", a)
|
2017-04-05 19:38:21 -07:00
|
|
|
Fatalf("ascompatet: too many function calls evaluating parameters")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-10-27 09:23:48 -07:00
|
|
|
nn.Append(a)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-10-27 09:23:48 -07:00
|
|
|
return append(nn.Slice(), mm.Slice()...)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-04-27 06:41:50 -07:00
|
|
|
// nodarg returns a Node for the function argument denoted by t,
|
|
|
|
|
// which is either the entire function argument or result struct (t is a struct *types.Type)
|
|
|
|
|
// or a specific argument (t is a *types.Field within a struct *types.Type).
|
|
|
|
|
//
|
|
|
|
|
// If fp is 0, the node is for use by a caller invoking the given
|
|
|
|
|
// function, preparing the arguments before the call
|
|
|
|
|
// or retrieving the results after the call.
|
|
|
|
|
// In this case, the node will correspond to an outgoing argument
|
|
|
|
|
// slot like 8(SP).
|
|
|
|
|
//
|
|
|
|
|
// If fp is 1, the node is for use by the function itself
|
|
|
|
|
// (the callee), to retrieve its arguments or write its results.
|
|
|
|
|
// In this case the node will be an ONAME with an appropriate
|
|
|
|
|
// type and offset.
|
|
|
|
|
func nodarg(t interface{}, fp int) *Node {
|
|
|
|
|
var n *Node
|
|
|
|
|
|
|
|
|
|
var funarg types.Funarg
|
|
|
|
|
switch t := t.(type) {
|
|
|
|
|
default:
|
|
|
|
|
Fatalf("bad nodarg %T(%v)", t, t)
|
|
|
|
|
|
|
|
|
|
case *types.Type:
|
|
|
|
|
// Entire argument struct, not just one arg
|
|
|
|
|
if !t.IsFuncArgStruct() {
|
|
|
|
|
Fatalf("nodarg: bad type %v", t)
|
|
|
|
|
}
|
|
|
|
|
funarg = t.StructType().Funarg
|
|
|
|
|
|
|
|
|
|
// Build fake variable name for whole arg struct.
|
|
|
|
|
n = newname(lookup(".args"))
|
|
|
|
|
n.Type = t
|
|
|
|
|
first := t.Field(0)
|
|
|
|
|
if first == nil {
|
|
|
|
|
Fatalf("nodarg: bad struct")
|
|
|
|
|
}
|
|
|
|
|
if first.Offset == BADWIDTH {
|
|
|
|
|
Fatalf("nodarg: offset not computed for %v", t)
|
|
|
|
|
}
|
|
|
|
|
n.Xoffset = first.Offset
|
|
|
|
|
|
|
|
|
|
case *types.Field:
|
|
|
|
|
funarg = t.Funarg
|
|
|
|
|
if fp == 1 {
|
|
|
|
|
// NOTE(rsc): This should be using t.Nname directly,
|
|
|
|
|
// except in the case where t.Nname.Sym is the blank symbol and
|
|
|
|
|
// so the assignment would be discarded during code generation.
|
|
|
|
|
// In that case we need to make a new node, and there is no harm
|
|
|
|
|
// in optimization passes to doing so. But otherwise we should
|
|
|
|
|
// definitely be using the actual declaration and not a newly built node.
|
|
|
|
|
// The extra Fatalf checks here are verifying that this is the case,
|
|
|
|
|
// without changing the actual logic (at time of writing, it's getting
|
|
|
|
|
// toward time for the Go 1.7 beta).
|
|
|
|
|
// At some quieter time (assuming we've never seen these Fatalfs happen)
|
|
|
|
|
// we could change this code to use "expect" directly.
|
|
|
|
|
expect := asNode(t.Nname)
|
|
|
|
|
if expect.isParamHeapCopy() {
|
|
|
|
|
expect = expect.Name.Param.Stackcopy
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
for _, n := range Curfn.Func.Dcl {
|
|
|
|
|
if (n.Class() == PPARAM || n.Class() == PPARAMOUT) && !t.Sym.IsBlank() && n.Sym == t.Sym {
|
|
|
|
|
if n != expect {
|
|
|
|
|
Fatalf("nodarg: unexpected node: %v (%p %v) vs %v (%p %v)", n, n, n.Op, asNode(t.Nname), asNode(t.Nname), asNode(t.Nname).Op)
|
|
|
|
|
}
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if !expect.Sym.IsBlank() {
|
|
|
|
|
Fatalf("nodarg: did not find node in dcl list: %v", expect)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Build fake name for individual variable.
|
|
|
|
|
// This is safe because if there was a real declared name
|
|
|
|
|
// we'd have used it above.
|
|
|
|
|
n = newname(lookup("__"))
|
|
|
|
|
n.Type = t.Type
|
|
|
|
|
if t.Offset == BADWIDTH {
|
|
|
|
|
Fatalf("nodarg: offset not computed for %v", t)
|
|
|
|
|
}
|
|
|
|
|
n.Xoffset = t.Offset
|
|
|
|
|
n.Orig = asNode(t.Nname)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Rewrite argument named _ to __,
|
|
|
|
|
// or else the assignment to _ will be
|
|
|
|
|
// discarded during code generation.
|
|
|
|
|
if isblank(n) {
|
|
|
|
|
n.Sym = lookup("__")
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch fp {
|
|
|
|
|
default:
|
|
|
|
|
Fatalf("bad fp")
|
|
|
|
|
|
|
|
|
|
case 0: // preparing arguments for call
|
|
|
|
|
n.Op = OINDREGSP
|
|
|
|
|
n.Xoffset += Ctxt.FixedFrameSize()
|
|
|
|
|
|
|
|
|
|
case 1: // reading arguments inside call
|
|
|
|
|
n.SetClass(PPARAM)
|
|
|
|
|
if funarg == types.FunargResults {
|
|
|
|
|
n.SetClass(PPARAMOUT)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
n.SetTypecheck(1)
|
|
|
|
|
n.SetAddrtaken(true) // keep optimizers at bay
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// package all the arguments that match a ... T parameter into a []T.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func mkdotargslice(typ *types.Type, args []*Node, init *Nodes, ddd *Node) *Node {
|
cmd/internal/gc: improve flow of input params to output params
This includes the following information in the per-function summary:
outK = paramJ encoded in outK bits for paramJ
outK = *paramJ encoded in outK bits for paramJ
heap = paramJ EscHeap
heap = *paramJ EscContentEscapes
Note that (currently) if the address of a parameter is taken and
returned, necessarily a heap allocation occurred to contain that
reference, and the heap can never refer to stack, therefore the
parameter and everything downstream from it escapes to the heap.
The per-function summary information now has a tuneable number of bits
(2 is probably noticeably better than 1, 3 is likely overkill, but it
is now easy to check and the -m debugging output includes information
that allows you to figure out if more would be better.)
A new test was added to check pointer flow through struct-typed and
*struct-typed parameters and returns; some of these are sensitive to
the number of summary bits, and ought to yield better results with a
more competent escape analysis algorithm. Another new test checks
(some) correctness with array parameters, results, and operations.
The old analysis inferred a piece of plan9 runtime was non-escaping by
counteracting overconservative analysis with buggy analysis; with the
bug fixed, the result was too conservative (and it's not easy to fix
in this framework) so the source code was tweaked to get the desired
result. A test was added against the discovered bug.
The escape analysis was further improved splitting the "level" into
3 parts, one tracking the conventional "level" and the other two
computing the highest-level-suffix-from-copy, which is used to
generally model the cancelling effect of indirection applied to
address-of.
With the improved escape analysis enabled, it was necessary to
modify one of the runtime tests because it now attempts to allocate
too much on the (small, fixed-size) G0 (system) stack and this
failed the test.
Compiling src/std after touching src/runtime/*.go with -m logging
turned on shows 420 fewer heap allocation sites (10538 vs 10968).
Profiling allocations in src/html/template with
for i in {1..5} ;
do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go;
go tool pprof -alloc_objects -text mastx.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
|
|
|
esc := uint16(EscUnknown)
|
2015-02-13 14:40:36 -05:00
|
|
|
if ddd != nil {
|
2015-03-10 09:58:01 +11:00
|
|
|
esc = ddd.Esc
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-02-22 14:43:21 -08:00
|
|
|
if len(args) == 0 {
|
|
|
|
|
n := nodnil()
|
|
|
|
|
n.Type = typ
|
|
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-02-22 14:43:21 -08:00
|
|
|
n := nod(OCOMPLIT, nil, typenod(typ))
|
|
|
|
|
if ddd != nil && prealloc[ddd] != nil {
|
|
|
|
|
prealloc[n] = prealloc[ddd] // temporary to use
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2017-02-22 14:43:21 -08:00
|
|
|
n.List.Set(args)
|
|
|
|
|
n.Esc = esc
|
|
|
|
|
n = typecheck(n, Erv)
|
|
|
|
|
if n.Type == nil {
|
|
|
|
|
Fatalf("mkdotargslice: typecheck failed")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2017-02-22 14:43:21 -08:00
|
|
|
n = walkexpr(n, init)
|
|
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// check assign expression list to
|
|
|
|
|
// a type list. called in
|
|
|
|
|
// return expr-list
|
|
|
|
|
// func(expr-list)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func ascompatte(call *Node, isddd bool, lhs *types.Type, rhs []*Node, fp int, init *Nodes) []*Node {
|
2016-03-08 10:26:20 -08:00
|
|
|
var nn []*Node
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// f(g()) where g has multiple return values
|
2017-02-22 14:43:21 -08:00
|
|
|
if len(rhs) == 1 && rhs[0].Type.IsFuncArgStruct() {
|
2015-02-13 14:40:36 -05:00
|
|
|
// optimization - can do block copy
|
2017-02-22 14:43:21 -08:00
|
|
|
if eqtypenoname(rhs[0].Type, lhs) {
|
|
|
|
|
nl := nodarg(lhs, fp)
|
|
|
|
|
nr := nod(OCONVNOP, rhs[0], nil)
|
|
|
|
|
nr.Type = nl.Type
|
|
|
|
|
nn = []*Node{convas(nod(OAS, nl, nr), init)}
|
2015-02-13 14:40:36 -05:00
|
|
|
goto ret
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// conversions involved.
|
|
|
|
|
// copy into temporaries.
|
2017-02-22 14:43:21 -08:00
|
|
|
var tmps []*Node
|
|
|
|
|
for _, nr := range rhs[0].Type.FieldSlice() {
|
|
|
|
|
tmps = append(tmps, temp(nr.Type))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a := nod(OAS2, nil, nil)
|
2017-02-22 14:43:21 -08:00
|
|
|
a.List.Set(tmps)
|
|
|
|
|
a.Rlist.Set(rhs)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
a = typecheck(a, Etop)
|
|
|
|
|
a = walkstmt(a)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(a)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-02-22 14:43:21 -08:00
|
|
|
rhs = tmps
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-02-22 14:43:21 -08:00
|
|
|
// For each parameter (LHS), assign its corresponding argument (RHS).
|
|
|
|
|
// If there's a ... parameter (which is only valid as the final
|
|
|
|
|
// parameter) and this is not a ... call expression,
|
|
|
|
|
// then assign the remaining arguments as a slice.
|
|
|
|
|
for i, nl := range lhs.FieldSlice() {
|
|
|
|
|
var nr *Node
|
2017-02-27 19:56:38 +02:00
|
|
|
if nl.Isddd() && !isddd {
|
2017-02-22 14:43:21 -08:00
|
|
|
nr = mkdotargslice(nl.Type, rhs[i:], init, call.Right)
|
|
|
|
|
} else {
|
|
|
|
|
nr = rhs[i]
|
2016-03-09 19:32:10 -08:00
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-02-22 14:43:21 -08:00
|
|
|
a := nod(OAS, nodarg(nl, fp), nr)
|
2016-03-09 19:32:10 -08:00
|
|
|
a = convas(a, init)
|
|
|
|
|
nn = append(nn, a)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
ret:
|
2016-03-08 10:26:20 -08:00
|
|
|
for _, n := range nn {
|
2017-04-25 18:02:43 -07:00
|
|
|
n.SetTypecheck(1)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
return nn
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// generate code for print
|
2016-03-07 22:54:46 -08:00
|
|
|
func walkprint(nn *Node, init *Nodes) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
var r *Node
|
|
|
|
|
var n *Node
|
|
|
|
|
var on *Node
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
var t *types.Type
|
|
|
|
|
var et types.EType
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-09-24 23:21:18 +02:00
|
|
|
op := nn.Op
|
2015-02-23 16:07:24 -05:00
|
|
|
all := nn.List
|
2016-03-07 22:54:46 -08:00
|
|
|
var calls []*Node
|
2015-02-23 16:07:24 -05:00
|
|
|
notfirst := false
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Hoist all the argument evaluation up before the lock.
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlistcheap(all.Slice(), init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
calls = append(calls, mkcall("printlock", nil, init))
|
2016-03-08 15:10:26 -08:00
|
|
|
for i1, n1 := range all.Slice() {
|
2015-02-17 22:13:49 -05:00
|
|
|
if notfirst {
|
2016-03-07 22:54:46 -08:00
|
|
|
calls = append(calls, mkcall("printsp", nil, init))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
notfirst = op == OPRINTN
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
n = n1
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Op == OLITERAL {
|
2015-05-27 00:47:05 -04:00
|
|
|
switch n.Val().Ctype() {
|
2015-02-13 14:40:36 -05:00
|
|
|
case CTRUNE:
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = defaultlit(n, types.Runetype)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case CTINT:
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = defaultlit(n, types.Types[TINT64])
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case CTFLT:
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = defaultlit(n, types.Types[TFLOAT64])
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if n.Op != OLITERAL && n.Type != nil && n.Type.Etype == TIDEAL {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n = defaultlit(n, types.Types[TINT64])
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = defaultlit(n, nil)
|
2016-03-09 12:39:36 -08:00
|
|
|
all.SetIndex(i1, n)
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Type == nil || n.Type.Etype == TFORW {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
t = n.Type
|
2015-09-24 23:21:18 +02:00
|
|
|
et = n.Type.Etype
|
2016-03-30 14:45:47 -07:00
|
|
|
if n.Type.IsInterface() {
|
2016-04-01 13:36:24 -07:00
|
|
|
if n.Type.IsEmptyInterface() {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printeface")
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printiface")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
on = substArgTypes(on, n.Type) // any-1
|
2016-03-30 15:09:25 -07:00
|
|
|
} else if n.Type.IsPtr() || et == TCHAN || et == TMAP || et == TFUNC || et == TUNSAFEPTR {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printpointer")
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
on = substArgTypes(on, n.Type) // any-1
|
2016-03-30 14:45:47 -07:00
|
|
|
} else if n.Type.IsSlice() {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printslice")
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
on = substArgTypes(on, n.Type) // any-1
|
2016-09-16 00:33:29 +10:00
|
|
|
} else if isInt[et] {
|
2015-02-13 14:40:36 -05:00
|
|
|
if et == TUINT64 {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
if isRuntimePkg(t.Sym.Pkg) && t.Sym.Name == "hex" {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printhex")
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printuint")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
} else {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printint")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-09-16 00:33:29 +10:00
|
|
|
} else if isFloat[et] {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printfloat")
|
2016-09-16 00:33:29 +10:00
|
|
|
} else if isComplex[et] {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printcomplex")
|
2015-02-13 14:40:36 -05:00
|
|
|
} else if et == TBOOL {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printbool")
|
2015-02-13 14:40:36 -05:00
|
|
|
} else if et == TSTRING {
|
2016-03-04 15:19:06 -08:00
|
|
|
on = syslook("printstring")
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
|
|
|
|
badtype(OPRINT, n.Type, nil)
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-13 23:02:38 -07:00
|
|
|
t = on.Type.Params().Field(0).Type
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-15 14:34:20 +10:00
|
|
|
if !eqtype(t, n.Type) {
|
2016-09-16 11:00:54 +10:00
|
|
|
n = nod(OCONV, n, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Type = t
|
|
|
|
|
}
|
|
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(OCALL, on, nil)
|
2016-03-08 15:10:26 -08:00
|
|
|
r.List.Append(n)
|
2016-03-07 22:54:46 -08:00
|
|
|
calls = append(calls, r)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if op == OPRINTN {
|
2016-03-07 22:54:46 -08:00
|
|
|
calls = append(calls, mkcall("printnl", nil, nil))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
calls = append(calls, mkcall("printunlock", nil, init))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-19 17:02:01 -07:00
|
|
|
typecheckslice(calls, Etop)
|
2015-02-13 14:40:36 -05:00
|
|
|
walkexprlist(calls, init)
|
|
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(OEMPTY, nil, nil)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Etop)
|
|
|
|
|
r = walkexpr(r, init)
|
2016-03-08 15:10:26 -08:00
|
|
|
r.Ninit.Set(calls)
|
2015-02-13 14:40:36 -05:00
|
|
|
return r
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func callnew(t *types.Type) *Node {
|
2017-02-27 19:56:38 +02:00
|
|
|
if t.NotInHeap() {
|
cmd/compile: add go:notinheap type pragma
This adds a //go:notinheap pragma for declarations of types that must
not be heap allocated. We ensure these rules by disallowing new(T),
make([]T), append([]T), or implicit allocation of T, by disallowing
conversions to notinheap types, and by propagating notinheap to any
struct or array that contains notinheap elements.
The utility of this pragma is that we can eliminate write barriers for
writes to pointers to go:notinheap types, since the write barrier is
guaranteed to be a no-op. This will let us mark several scheduler and
memory allocator structures as go:notinheap, which will let us
disallow write barriers in the scheduler and memory allocator much
more thoroughly and also eliminate some problematic hybrid write
barriers.
This also makes go:nowritebarrierrec and go:yeswritebarrierrec much
more powerful. Currently we use go:nowritebarrier all over the place,
but it's almost never what you actually want: when write barriers are
illegal, they're typically illegal for a whole dynamic scope. Partly
this is because go:nowritebarrier has been around longer, but it's
also because go:nowritebarrierrec couldn't be used in situations that
had no-op write barriers or where some nested scope did allow write
barriers. go:notinheap eliminates many no-op write barriers and
go:yeswritebarrierrec makes it possible to opt back in to write
barriers, so these two changes will let us use go:nowritebarrierrec
far more liberally.
This updates #13386, which is about controlling pointers from non-GC'd
memory to GC'd memory. That would require some additional pragma (or
pragmas), but could build on this pragma.
Change-Id: I6314f8f4181535dd166887c9ec239977b54940bd
Reviewed-on: https://go-review.googlesource.com/30939
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-10-11 22:53:27 -04:00
|
|
|
yyerror("%v is go:notinheap; heap allocation disallowed", t)
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
dowidth(t)
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook("newobject")
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
fn = substArgTypes(fn, t)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
v := mkcall1(fn, types.NewPtr(t), nil, typename(t))
|
2017-02-27 19:56:38 +02:00
|
|
|
v.SetNonNil(true)
|
2016-04-19 21:06:53 -07:00
|
|
|
return v
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-06-29 15:17:14 -04:00
|
|
|
func iscallret(n *Node) bool {
|
|
|
|
|
n = outervalue(n)
|
2016-10-24 14:33:22 -07:00
|
|
|
return n.Op == OINDREGSP
|
2015-06-29 15:17:14 -04:00
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
func isstack(n *Node) bool {
|
2015-02-13 14:40:36 -05:00
|
|
|
n = outervalue(n)
|
|
|
|
|
|
|
|
|
|
// If n is *autotmp and autotmp = &foo, replace n with foo.
|
|
|
|
|
// We introduce such temps when initializing struct literals.
|
2016-10-28 13:33:57 -04:00
|
|
|
if n.Op == OIND && n.Left.Op == ONAME && n.Left.IsAutoTmp() {
|
2015-05-26 22:19:27 -04:00
|
|
|
defn := n.Left.Name.Defn
|
2015-02-13 14:40:36 -05:00
|
|
|
if defn != nil && defn.Op == OAS && defn.Right.Op == OADDR {
|
|
|
|
|
n = defn.Right.Left
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch n.Op {
|
2016-10-24 14:33:22 -07:00
|
|
|
case OINDREGSP:
|
|
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ONAME:
|
2017-04-25 18:14:12 -07:00
|
|
|
switch n.Class() {
|
2015-04-01 09:38:44 -07:00
|
|
|
case PAUTO, PPARAM, PPARAMOUT:
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-01 21:46:08 -08:00
|
|
|
// isReflectHeaderDataField reports whether l is an expression p.Data
|
|
|
|
|
// where p has type reflect.SliceHeader or reflect.StringHeader.
|
|
|
|
|
func isReflectHeaderDataField(l *Node) bool {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
if l.Type != types.Types[TUINTPTR] {
|
2017-03-01 21:46:08 -08:00
|
|
|
return false
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
var tsym *types.Sym
|
2017-03-01 21:46:08 -08:00
|
|
|
switch l.Op {
|
|
|
|
|
case ODOT:
|
|
|
|
|
tsym = l.Left.Type.Sym
|
|
|
|
|
case ODOTPTR:
|
|
|
|
|
tsym = l.Left.Type.Elem().Sym
|
|
|
|
|
default:
|
|
|
|
|
return false
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if tsym == nil || l.Sym.Name != "Data" || tsym.Pkg.Path != "reflect" {
|
|
|
|
|
return false
|
|
|
|
|
}
|
|
|
|
|
return tsym.Name == "SliceHeader" || tsym.Name == "StringHeader"
|
|
|
|
|
}
|
|
|
|
|
|
2017-02-15 18:43:34 -08:00
|
|
|
// Do we need a write barrier for assigning to l?
|
|
|
|
|
func needwritebarrier(l *Node) bool {
|
2016-04-13 18:37:18 -07:00
|
|
|
if !use_writebarrier {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if l == nil || isblank(l) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-01 21:46:08 -08:00
|
|
|
// No write barrier for write to stack.
|
|
|
|
|
if isstack(l) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-01 21:46:08 -08:00
|
|
|
// Package unsafe's documentation says storing pointers into
|
|
|
|
|
// reflect.SliceHeader and reflect.StringHeader's Data fields
|
|
|
|
|
// is valid, even though they have type uintptr (#19168).
|
|
|
|
|
if isReflectHeaderDataField(l) {
|
|
|
|
|
return true
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// No write barrier for write of non-pointers.
|
|
|
|
|
dowidth(l.Type)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
if !types.Haspointers(l.Type) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: add go:notinheap type pragma
This adds a //go:notinheap pragma for declarations of types that must
not be heap allocated. We ensure these rules by disallowing new(T),
make([]T), append([]T), or implicit allocation of T, by disallowing
conversions to notinheap types, and by propagating notinheap to any
struct or array that contains notinheap elements.
The utility of this pragma is that we can eliminate write barriers for
writes to pointers to go:notinheap types, since the write barrier is
guaranteed to be a no-op. This will let us mark several scheduler and
memory allocator structures as go:notinheap, which will let us
disallow write barriers in the scheduler and memory allocator much
more thoroughly and also eliminate some problematic hybrid write
barriers.
This also makes go:nowritebarrierrec and go:yeswritebarrierrec much
more powerful. Currently we use go:nowritebarrier all over the place,
but it's almost never what you actually want: when write barriers are
illegal, they're typically illegal for a whole dynamic scope. Partly
this is because go:nowritebarrier has been around longer, but it's
also because go:nowritebarrierrec couldn't be used in situations that
had no-op write barriers or where some nested scope did allow write
barriers. go:notinheap eliminates many no-op write barriers and
go:yeswritebarrierrec makes it possible to opt back in to write
barriers, so these two changes will let us use go:nowritebarrierrec
far more liberally.
This updates #13386, which is about controlling pointers from non-GC'd
memory to GC'd memory. That would require some additional pragma (or
pragmas), but could build on this pragma.
Change-Id: I6314f8f4181535dd166887c9ec239977b54940bd
Reviewed-on: https://go-review.googlesource.com/30939
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-10-11 22:53:27 -04:00
|
|
|
// No write barrier if this is a pointer to a go:notinheap
|
|
|
|
|
// type, since the write barrier's inheap(ptr) check will fail.
|
2017-02-27 19:56:38 +02:00
|
|
|
if l.Type.IsPtr() && l.Type.Elem().NotInHeap() {
|
cmd/compile: add go:notinheap type pragma
This adds a //go:notinheap pragma for declarations of types that must
not be heap allocated. We ensure these rules by disallowing new(T),
make([]T), append([]T), or implicit allocation of T, by disallowing
conversions to notinheap types, and by propagating notinheap to any
struct or array that contains notinheap elements.
The utility of this pragma is that we can eliminate write barriers for
writes to pointers to go:notinheap types, since the write barrier is
guaranteed to be a no-op. This will let us mark several scheduler and
memory allocator structures as go:notinheap, which will let us
disallow write barriers in the scheduler and memory allocator much
more thoroughly and also eliminate some problematic hybrid write
barriers.
This also makes go:nowritebarrierrec and go:yeswritebarrierrec much
more powerful. Currently we use go:nowritebarrier all over the place,
but it's almost never what you actually want: when write barriers are
illegal, they're typically illegal for a whole dynamic scope. Partly
this is because go:nowritebarrier has been around longer, but it's
also because go:nowritebarrierrec couldn't be used in situations that
had no-op write barriers or where some nested scope did allow write
barriers. go:notinheap eliminates many no-op write barriers and
go:yeswritebarrierrec makes it possible to opt back in to write
barriers, so these two changes will let us use go:nowritebarrierrec
far more liberally.
This updates #13386, which is about controlling pointers from non-GC'd
memory to GC'd memory. That would require some additional pragma (or
pragmas), but could build on this pragma.
Change-Id: I6314f8f4181535dd166887c9ec239977b54940bd
Reviewed-on: https://go-review.googlesource.com/30939
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-10-11 22:53:27 -04:00
|
|
|
return false
|
|
|
|
|
}
|
|
|
|
|
|
2016-10-18 10:26:28 -04:00
|
|
|
// TODO: We can eliminate write barriers if we know *both* the
|
|
|
|
|
// current and new content of the slot must already be shaded.
|
|
|
|
|
// We know a pointer is shaded if it's nil, or points to
|
|
|
|
|
// static data, a global (variable or function), or the stack.
|
|
|
|
|
// The nil optimization could be particularly useful for
|
|
|
|
|
// writes to just-allocated objects. Unfortunately, knowing
|
|
|
|
|
// the "current" value of the slot requires flow analysis.
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Otherwise, be conservative and use write barrier.
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
func convas(n *Node, init *Nodes) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Op != OAS {
|
2016-04-27 15:10:10 +10:00
|
|
|
Fatalf("convas: not OAS %v", n.Op)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-04-25 18:02:43 -07:00
|
|
|
n.SetTypecheck(1)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
var lt *types.Type
|
|
|
|
|
var rt *types.Type
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Left == nil || n.Right == nil {
|
|
|
|
|
goto out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
lt = n.Left.Type
|
|
|
|
|
rt = n.Right.Type
|
|
|
|
|
if lt == nil || rt == nil {
|
|
|
|
|
goto out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if isblank(n.Left) {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right = defaultlit(n.Right, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
goto out
|
|
|
|
|
}
|
|
|
|
|
|
2016-09-15 14:34:20 +10:00
|
|
|
if !eqtype(lt, rt) {
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Right = assignconv(n.Right, lt, "assignment")
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2017-03-28 12:10:38 -07:00
|
|
|
dowidth(n.Right.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
out:
|
2017-03-03 13:38:49 -08:00
|
|
|
updateHasCall(n)
|
2015-02-13 14:40:36 -05:00
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// from ascompat[te]
|
|
|
|
|
// evaluating actual function arguments.
|
|
|
|
|
// f(a,b)
|
|
|
|
|
// if there is exactly one function expr,
|
|
|
|
|
// then it is done first. otherwise must
|
|
|
|
|
// make temp variables
|
2016-03-08 10:26:20 -08:00
|
|
|
func reorder1(all []*Node) []*Node {
|
2015-02-23 16:07:24 -05:00
|
|
|
c := 0 // function calls
|
|
|
|
|
t := 0 // total parameters
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-08 10:26:20 -08:00
|
|
|
for _, n := range all {
|
2015-02-13 14:40:36 -05:00
|
|
|
t++
|
2017-03-03 13:38:49 -08:00
|
|
|
updateHasCall(n)
|
|
|
|
|
if n.HasCall() {
|
2015-02-13 14:40:36 -05:00
|
|
|
c++
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if c == 0 || t == 1 {
|
|
|
|
|
return all
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-08 10:26:20 -08:00
|
|
|
var g []*Node // fncalls assigned to tempnames
|
|
|
|
|
var f *Node // last fncall assigned to stack
|
|
|
|
|
var r []*Node // non fncalls and tempnames assigned to stack
|
2015-02-23 16:07:24 -05:00
|
|
|
d := 0
|
|
|
|
|
var a *Node
|
2016-03-08 10:26:20 -08:00
|
|
|
for _, n := range all {
|
2017-03-03 13:38:49 -08:00
|
|
|
if !n.HasCall() {
|
2016-03-08 10:26:20 -08:00
|
|
|
r = append(r, n)
|
2015-02-13 14:40:36 -05:00
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
d++
|
|
|
|
|
if d == c {
|
|
|
|
|
f = n
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// make assignment of fncall to tempname
|
|
|
|
|
a = temp(n.Right.Type)
|
|
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OAS, a, n.Right)
|
2016-03-08 10:26:20 -08:00
|
|
|
g = append(g, a)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// put normal arg assignment on list
|
|
|
|
|
// with fncall replaced by tempname
|
|
|
|
|
n.Right = a.Left
|
|
|
|
|
|
2016-03-08 10:26:20 -08:00
|
|
|
r = append(r, n)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if f != nil {
|
2016-03-08 10:26:20 -08:00
|
|
|
g = append(g, f)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-08 10:26:20 -08:00
|
|
|
return append(g, r...)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// from ascompat[ee]
|
|
|
|
|
// a,b = c,d
|
|
|
|
|
// simultaneous assignment. there cannot
|
|
|
|
|
// be later use of an earlier lvalue.
|
|
|
|
|
//
|
|
|
|
|
// function calls have been removed.
|
2016-03-07 22:54:46 -08:00
|
|
|
func reorder3(all []*Node) []*Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
var l *Node
|
|
|
|
|
|
|
|
|
|
// If a needed expression may be affected by an
|
|
|
|
|
// earlier assignment, make an early copy of that
|
|
|
|
|
// expression and use the copy instead.
|
2016-03-07 22:54:46 -08:00
|
|
|
var early []*Node
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
var mapinit Nodes
|
|
|
|
|
for i, n := range all {
|
|
|
|
|
l = n.Left
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Save subexpressions needed on left side.
|
|
|
|
|
// Drill through non-dereferences.
|
|
|
|
|
for {
|
|
|
|
|
if l.Op == ODOT || l.Op == OPAREN {
|
|
|
|
|
l = l.Left
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-30 14:45:47 -07:00
|
|
|
if l.Op == OINDEX && l.Left.Type.IsArray() {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
l.Right = reorder3save(l.Right, all, i, &early)
|
2015-02-13 14:40:36 -05:00
|
|
|
l = l.Left
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch l.Op {
|
|
|
|
|
default:
|
2016-04-27 19:34:17 +10:00
|
|
|
Fatalf("reorder3 unexpected lvalue %#v", l.Op)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ONAME:
|
|
|
|
|
break
|
|
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case OINDEX, OINDEXMAP:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
l.Left = reorder3save(l.Left, all, i, &early)
|
|
|
|
|
l.Right = reorder3save(l.Right, all, i, &early)
|
2015-02-13 14:40:36 -05:00
|
|
|
if l.Op == OINDEXMAP {
|
2016-03-07 22:54:46 -08:00
|
|
|
all[i] = convas(all[i], &mapinit)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case OIND, ODOTPTR:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
l.Left = reorder3save(l.Left, all, i, &early)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Save expression on right side.
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
all[i].Right = reorder3save(all[i].Right, all, i, &early)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
early = append(mapinit.Slice(), early...)
|
|
|
|
|
return append(early, all...)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// if the evaluation of *np would be affected by the
|
2015-06-10 15:31:53 -07:00
|
|
|
// assignments in all up to but not including the ith assignment,
|
2015-10-22 09:51:12 +09:00
|
|
|
// copy into a temporary during *early and
|
|
|
|
|
// replace *np with that temp.
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
// The result of reorder3save MUST be assigned back to n, e.g.
|
|
|
|
|
// n.Left = reorder3save(n.Left, all, i, early)
|
|
|
|
|
func reorder3save(n *Node, all []*Node, i int, early *[]*Node) *Node {
|
2015-06-10 15:31:53 -07:00
|
|
|
if !aliased(n, all, i) {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
q := temp(n.Type)
|
2016-09-16 11:00:54 +10:00
|
|
|
q = nod(OAS, q, n)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
q = typecheck(q, Etop)
|
2016-03-07 22:54:46 -08:00
|
|
|
*early = append(*early, q)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return q.Left
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// what's the outer value that a write to n affects?
|
|
|
|
|
// outer value means containing struct or array.
|
2015-02-13 14:40:36 -05:00
|
|
|
func outervalue(n *Node) *Node {
|
|
|
|
|
for {
|
|
|
|
|
if n.Op == OXDOT {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("OXDOT in walk")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
if n.Op == ODOT || n.Op == OPAREN || n.Op == OCONVNOP {
|
|
|
|
|
n = n.Left
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-30 14:45:47 -07:00
|
|
|
if n.Op == OINDEX && n.Left.Type != nil && n.Left.Type.IsArray() {
|
2015-02-13 14:40:36 -05:00
|
|
|
n = n.Left
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// Is it possible that the computation of n might be
|
2015-06-10 15:31:53 -07:00
|
|
|
// affected by writes in as up to but not including the ith element?
|
2016-03-07 22:54:46 -08:00
|
|
|
func aliased(n *Node, all []*Node, i int) bool {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n == nil {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-31 13:58:33 -07:00
|
|
|
// Treat all fields of a struct as referring to the whole struct.
|
|
|
|
|
// We could do better but we would have to keep track of the fields.
|
|
|
|
|
for n.Op == ODOT {
|
|
|
|
|
n = n.Left
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-13 14:40:36 -05:00
|
|
|
// Look for obvious aliasing: a variable being assigned
|
|
|
|
|
// during the all list and appearing in n.
|
|
|
|
|
// Also record whether there are any writes to main memory.
|
|
|
|
|
// Also record whether there are any writes to variables
|
|
|
|
|
// whose addresses have been taken.
|
2015-02-23 16:07:24 -05:00
|
|
|
memwrite := 0
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
varwrite := 0
|
|
|
|
|
var a *Node
|
2016-03-07 22:54:46 -08:00
|
|
|
for _, an := range all[:i] {
|
|
|
|
|
a = outervalue(an.Left)
|
2016-03-31 13:58:33 -07:00
|
|
|
|
|
|
|
|
for a.Op == ODOT {
|
|
|
|
|
a = a.Left
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-13 14:40:36 -05:00
|
|
|
if a.Op != ONAME {
|
|
|
|
|
memwrite = 1
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
2017-04-25 18:14:12 -07:00
|
|
|
switch n.Class() {
|
2015-02-13 14:40:36 -05:00
|
|
|
default:
|
|
|
|
|
varwrite = 1
|
|
|
|
|
continue
|
|
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case PAUTO, PPARAM, PPARAMOUT:
|
2017-02-27 19:56:38 +02:00
|
|
|
if n.Addrtaken() {
|
2015-02-13 14:40:36 -05:00
|
|
|
varwrite = 1
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
if vmatch2(a, n) {
|
2015-02-13 14:40:36 -05:00
|
|
|
// Direct hit.
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// The variables being written do not appear in n.
|
|
|
|
|
// However, n might refer to computed addresses
|
|
|
|
|
// that are being written.
|
|
|
|
|
|
|
|
|
|
// If no computed addresses are affected by the writes, no aliasing.
|
2015-02-17 22:13:49 -05:00
|
|
|
if memwrite == 0 && varwrite == 0 {
|
|
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// If n does not refer to computed addresses
|
|
|
|
|
// (that is, if n only refers to variables whose addresses
|
|
|
|
|
// have not been taken), no aliasing.
|
2015-02-17 22:13:49 -05:00
|
|
|
if varexpr(n) {
|
|
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Otherwise, both the writes and n refer to computed memory addresses.
|
|
|
|
|
// Assume that they might conflict.
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// does the evaluation of n only refer to variables
|
|
|
|
|
// whose addresses have not been taken?
|
|
|
|
|
// (and no other memory)
|
2015-02-17 22:13:49 -05:00
|
|
|
func varexpr(n *Node) bool {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n == nil {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch n.Op {
|
|
|
|
|
case OLITERAL:
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case ONAME:
|
2017-04-25 18:14:12 -07:00
|
|
|
switch n.Class() {
|
2015-04-01 09:38:44 -07:00
|
|
|
case PAUTO, PPARAM, PPARAMOUT:
|
2017-02-27 19:56:38 +02:00
|
|
|
if !n.Addrtaken() {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OADD,
|
|
|
|
|
OSUB,
|
|
|
|
|
OOR,
|
|
|
|
|
OXOR,
|
|
|
|
|
OMUL,
|
|
|
|
|
ODIV,
|
|
|
|
|
OMOD,
|
|
|
|
|
OLSH,
|
|
|
|
|
ORSH,
|
|
|
|
|
OAND,
|
|
|
|
|
OANDNOT,
|
|
|
|
|
OPLUS,
|
|
|
|
|
OMINUS,
|
|
|
|
|
OCOM,
|
|
|
|
|
OPAREN,
|
|
|
|
|
OANDAND,
|
|
|
|
|
OOROR,
|
|
|
|
|
OCONV,
|
|
|
|
|
OCONVNOP,
|
|
|
|
|
OCONVIFACE,
|
|
|
|
|
ODOTTYPE:
|
2015-02-17 22:13:49 -05:00
|
|
|
return varexpr(n.Left) && varexpr(n.Right)
|
cmd/compile: change ODOT and friends to use Sym, not Right
The Node type ODOT and its variants all represent a selector, with a
simple name to the right of the dot. Before this change this was
represented by using an ONAME Node in the Right field. This ONAME node
served no useful purpose. This CL changes these Node types to store the
symbol in the Sym field instead, thus not requiring allocating a Node
for each selector.
When compiling x/tools/go/types this CL eliminates nearly 5000 calls to
newname and reduces the total number of Nodes allocated by about 6.6%.
It seems to cut compilation time by 1 to 2 percent.
Getting this right was somewhat subtle, and I added two dubious changes
to produce the exact same output as before. One is to ishairy in
inl.go: the ONAME node increased the cost of ODOT and friends by 1, and
I retained that, although really ODOT is not more expensive than any
other node. The other is to varexpr in walk.go: because the ONAME in
the Right field of an ODOT has no class, varexpr would always return
false for an ODOT, although in fact for some ODOT's it seemingly ought
to return true; I added an && false for now. I will send separate CLs,
that will break toolstash -cmp, to clean these up.
This CL passes toolstash -cmp.
Change-Id: I4af8a10cc59078c436130ce472f25abc3a9b2f80
Reviewed-on: https://go-review.googlesource.com/20890
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-03-18 16:52:30 -07:00
|
|
|
|
|
|
|
|
case ODOT: // but not ODOTPTR
|
2016-03-31 13:58:33 -07:00
|
|
|
// Should have been handled in aliased.
|
|
|
|
|
Fatalf("varexpr unexpected ODOT")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Be conservative.
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// is the name l mentioned in r?
|
2015-02-17 22:13:49 -05:00
|
|
|
func vmatch2(l *Node, r *Node) bool {
|
2015-02-13 14:40:36 -05:00
|
|
|
if r == nil {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
switch r.Op {
|
|
|
|
|
// match each right given left
|
|
|
|
|
case ONAME:
|
2015-02-17 22:13:49 -05:00
|
|
|
return l == r
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
case OLITERAL:
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
if vmatch2(l, r.Left) {
|
|
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-02-17 22:13:49 -05:00
|
|
|
if vmatch2(l, r.Right) {
|
|
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-08 15:10:26 -08:00
|
|
|
for _, n := range r.List.Slice() {
|
|
|
|
|
if vmatch2(l, n) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-10-22 09:51:12 +09:00
|
|
|
// is any name mentioned in l also mentioned in r?
|
|
|
|
|
// called by sinit.go
|
2015-02-17 22:13:49 -05:00
|
|
|
func vmatch1(l *Node, r *Node) bool {
|
2015-10-22 09:51:12 +09:00
|
|
|
// isolate all left sides
|
2015-02-13 14:40:36 -05:00
|
|
|
if l == nil || r == nil {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
switch l.Op {
|
|
|
|
|
case ONAME:
|
2017-04-25 18:14:12 -07:00
|
|
|
switch l.Class() {
|
2016-05-25 10:29:50 -04:00
|
|
|
case PPARAM, PAUTO:
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
|
|
|
|
|
default:
|
2017-03-03 13:38:49 -08:00
|
|
|
// assignment to non-stack variable must be
|
|
|
|
|
// delayed if right has function calls.
|
|
|
|
|
if r.HasCall() {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return vmatch2(l, r)
|
|
|
|
|
|
|
|
|
|
case OLITERAL:
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
if vmatch1(l.Left, r) {
|
|
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-02-17 22:13:49 -05:00
|
|
|
if vmatch1(l.Right, r) {
|
|
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-08 15:10:26 -08:00
|
|
|
for _, n := range l.List.Slice() {
|
|
|
|
|
if vmatch1(n, r) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-09 18:54:26 -08:00
|
|
|
// paramstoheap returns code to allocate memory for heap-escaped parameters
|
2017-04-27 09:27:52 -07:00
|
|
|
// and to copy non-result parameters' values from the stack.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func paramstoheap(params *types.Type) []*Node {
|
2016-02-26 14:28:48 -08:00
|
|
|
var nn []*Node
|
2016-03-17 01:32:18 -07:00
|
|
|
for _, t := range params.Fields().Slice() {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
v := asNode(t.Nname)
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
if v != nil && v.Sym != nil && strings.HasPrefix(v.Sym.Name, "~r") { // unnamed result
|
|
|
|
|
v = nil
|
|
|
|
|
}
|
|
|
|
|
if v == nil {
|
2015-02-13 14:40:36 -05:00
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
if stackcopy := v.Name.Param.Stackcopy; stackcopy != nil {
|
2016-09-16 11:00:54 +10:00
|
|
|
nn = append(nn, walkstmt(nod(ODCL, v, nil)))
|
2017-04-25 18:14:12 -07:00
|
|
|
if stackcopy.Class() == PPARAM {
|
2016-09-16 11:00:54 +10:00
|
|
|
nn = append(nn, walkstmt(typecheck(nod(OAS, v, stackcopy), Etop)))
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return nn
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-09 10:38:45 -08:00
|
|
|
// zeroResults zeros the return values at the start of the function.
|
|
|
|
|
// We need to do this very early in the function. Defer might stop a
|
|
|
|
|
// panic and show the return values as they exist at the time of
|
|
|
|
|
// panic. For precise stacks, the garbage collector assumes results
|
|
|
|
|
// are always live, so we need to zero them before any allocations,
|
|
|
|
|
// even allocations to move params/results to the heap.
|
|
|
|
|
// The generated code is added to Curfn's Enter list.
|
|
|
|
|
func zeroResults() {
|
|
|
|
|
lno := lineno
|
|
|
|
|
lineno = Curfn.Pos
|
|
|
|
|
for _, f := range Curfn.Type.Results().Fields().Slice() {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
if v := asNode(f.Nname); v != nil && v.Name.Param.Heapaddr != nil {
|
2017-03-09 10:38:45 -08:00
|
|
|
// The local which points to the return value is the
|
|
|
|
|
// thing that needs zeroing. This is already handled
|
|
|
|
|
// by a Needzero annotation in plive.go:livenessepilogue.
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
// Zero the stack location containing f.
|
|
|
|
|
Curfn.Func.Enter.Append(nod(OAS, nodarg(f, 1), nil))
|
|
|
|
|
}
|
|
|
|
|
lineno = lno
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-09 18:54:26 -08:00
|
|
|
// returnsfromheap returns code to copy values for heap-escaped parameters
|
|
|
|
|
// back to the stack.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func returnsfromheap(params *types.Type) []*Node {
|
2016-02-26 14:28:48 -08:00
|
|
|
var nn []*Node
|
2016-03-17 01:32:18 -07:00
|
|
|
for _, t := range params.Fields().Slice() {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
v := asNode(t.Nname)
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
if v == nil {
|
2015-02-13 14:40:36 -05:00
|
|
|
continue
|
|
|
|
|
}
|
2017-04-25 18:14:12 -07:00
|
|
|
if stackcopy := v.Name.Param.Stackcopy; stackcopy != nil && stackcopy.Class() == PPARAMOUT {
|
2016-09-16 11:00:54 +10:00
|
|
|
nn = append(nn, walkstmt(typecheck(nod(OAS, stackcopy, v), Etop)))
|
cmd/compile: fix liveness computation for heap-escaped parameters
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes #15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return nn
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-09 18:54:26 -08:00
|
|
|
// heapmoves generates code to handle migrating heap-escaped parameters
|
|
|
|
|
// between the stack and the heap. The generated code is added to Curfn's
|
|
|
|
|
// Enter and Exit lists.
|
2015-02-13 14:40:36 -05:00
|
|
|
func heapmoves() {
|
2015-02-23 16:07:24 -05:00
|
|
|
lno := lineno
|
2016-12-07 17:40:46 -08:00
|
|
|
lineno = Curfn.Pos
|
2016-05-25 10:01:58 -04:00
|
|
|
nn := paramstoheap(Curfn.Type.Recvs())
|
|
|
|
|
nn = append(nn, paramstoheap(Curfn.Type.Params())...)
|
|
|
|
|
nn = append(nn, paramstoheap(Curfn.Type.Results())...)
|
2016-02-26 14:28:48 -08:00
|
|
|
Curfn.Func.Enter.Append(nn...)
|
2015-03-25 19:33:01 -07:00
|
|
|
lineno = Curfn.Func.Endlineno
|
2016-03-09 18:54:26 -08:00
|
|
|
Curfn.Func.Exit.Append(returnsfromheap(Curfn.Type.Results())...)
|
2015-02-13 14:40:36 -05:00
|
|
|
lineno = lno
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func vmkcall(fn *Node, t *types.Type, init *Nodes, va []*Node) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
if fn.Type == nil || fn.Type.Etype != TFUNC {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("mkcall %v %v", fn, fn.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-17 01:47:16 -07:00
|
|
|
n := fn.Type.Params().NumFields()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
r := nod(OCALL, fn, nil)
|
2016-03-08 15:10:26 -08:00
|
|
|
r.List.Set(va[:n])
|
2016-03-17 01:47:16 -07:00
|
|
|
if fn.Type.Results().NumFields() > 0 {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Erv|Efnstruct)
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Etop)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = walkexpr(r, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
r.Type = t
|
|
|
|
|
return r
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func mkcall(name string, t *types.Type, init *Nodes, args ...*Node) *Node {
|
2016-03-04 15:19:06 -08:00
|
|
|
return vmkcall(syslook(name), t, init, args)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func mkcall1(fn *Node, t *types.Type, init *Nodes, args ...*Node) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
return vmkcall(fn, t, init, args)
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func conv(n *Node, t *types.Type) *Node {
|
2016-09-15 14:34:20 +10:00
|
|
|
if eqtype(n.Type, t) {
|
2015-02-13 14:40:36 -05:00
|
|
|
return n
|
|
|
|
|
}
|
2016-09-16 11:00:54 +10:00
|
|
|
n = nod(OCONV, n, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
n.Type = t
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = typecheck(n, Erv)
|
2015-02-13 14:40:36 -05:00
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
2017-01-21 19:52:09 -08:00
|
|
|
// byteindex converts n, which is byte-sized, to a uint8.
|
|
|
|
|
// We cannot use conv, because we allow converting bool to uint8 here,
|
|
|
|
|
// which is forbidden in user code.
|
|
|
|
|
func byteindex(n *Node) *Node {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
if eqtype(n.Type, types.Types[TUINT8]) {
|
2017-01-21 19:52:09 -08:00
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
n = nod(OCONV, n, nil)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
n.Type = types.Types[TUINT8]
|
2017-04-25 18:02:43 -07:00
|
|
|
n.SetTypecheck(1)
|
2017-01-21 19:52:09 -08:00
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func chanfn(name string, n int, t *types.Type) *Node {
|
2016-03-30 14:56:08 -07:00
|
|
|
if !t.IsChan() {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("chanfn %v", t)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook(name)
|
2015-03-08 13:33:49 -04:00
|
|
|
switch n {
|
|
|
|
|
default:
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("chanfn %d", n)
|
2015-03-08 13:33:49 -04:00
|
|
|
case 1:
|
2016-03-30 10:57:47 -07:00
|
|
|
fn = substArgTypes(fn, t.Elem())
|
2015-03-08 13:33:49 -04:00
|
|
|
case 2:
|
2016-03-30 10:57:47 -07:00
|
|
|
fn = substArgTypes(fn, t.Elem(), t.Elem())
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
return fn
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func mapfn(name string, t *types.Type) *Node {
|
2016-03-30 14:56:08 -07:00
|
|
|
if !t.IsMap() {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("mapfn %v", t)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook(name)
|
2016-03-28 21:48:47 -07:00
|
|
|
fn = substArgTypes(fn, t.Key(), t.Val(), t.Key(), t.Val())
|
2015-02-13 14:40:36 -05:00
|
|
|
return fn
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func mapfndel(name string, t *types.Type) *Node {
|
2016-03-30 14:56:08 -07:00
|
|
|
if !t.IsMap() {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("mapfn %v", t)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook(name)
|
2016-03-28 21:48:47 -07:00
|
|
|
fn = substArgTypes(fn, t.Key(), t.Val(), t.Key())
|
2015-02-13 14:40:36 -05:00
|
|
|
return fn
|
|
|
|
|
}
|
|
|
|
|
|
2017-03-14 11:11:28 -07:00
|
|
|
const (
|
|
|
|
|
mapslow = iota
|
|
|
|
|
mapfast32
|
|
|
|
|
mapfast64
|
|
|
|
|
mapfaststr
|
|
|
|
|
nmapfast
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
type mapnames [nmapfast]string
|
|
|
|
|
|
|
|
|
|
func mkmapnames(base string) mapnames {
|
|
|
|
|
return mapnames{base, base + "_fast32", base + "_fast64", base + "_faststr"}
|
2017-02-09 14:00:23 -08:00
|
|
|
}
|
|
|
|
|
|
2017-03-14 11:11:28 -07:00
|
|
|
var mapaccess1 mapnames = mkmapnames("mapaccess1")
|
|
|
|
|
var mapaccess2 mapnames = mkmapnames("mapaccess2")
|
|
|
|
|
var mapassign mapnames = mkmapnames("mapassign")
|
|
|
|
|
var mapdelete mapnames = mkmapnames("mapdelete")
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func mapfast(t *types.Type) int {
|
2017-03-12 14:47:59 -07:00
|
|
|
// Check ../../runtime/hashmap.go:maxValueSize before changing.
|
|
|
|
|
if t.Val().Width > 128 {
|
2017-03-14 11:11:28 -07:00
|
|
|
return mapslow
|
2017-03-12 14:47:59 -07:00
|
|
|
}
|
|
|
|
|
switch algtype(t.Key()) {
|
|
|
|
|
case AMEM32:
|
2017-03-14 11:11:28 -07:00
|
|
|
return mapfast32
|
2017-03-12 14:47:59 -07:00
|
|
|
case AMEM64:
|
2017-03-14 11:11:28 -07:00
|
|
|
return mapfast64
|
2017-03-12 14:47:59 -07:00
|
|
|
case ASTRING:
|
2017-03-14 11:11:28 -07:00
|
|
|
return mapfaststr
|
2017-03-12 14:47:59 -07:00
|
|
|
}
|
2017-03-14 11:11:28 -07:00
|
|
|
return mapslow
|
2017-03-12 14:47:59 -07:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func writebarrierfn(name string, l *types.Type, r *types.Type) *Node {
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook(name)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
fn = substArgTypes(fn, l, r)
|
2015-02-13 14:40:36 -05:00
|
|
|
return fn
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
func addstr(n *Node, init *Nodes) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
// orderexpr rewrote OADDSTR to have a list of strings.
|
2016-03-08 15:10:26 -08:00
|
|
|
c := n.List.Len()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
if c < 2 {
|
2017-03-14 09:46:45 -07:00
|
|
|
Fatalf("addstr count %d too small", c)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
buf := nodnil()
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Esc == EscNone {
|
2015-02-23 16:07:24 -05:00
|
|
|
sz := int64(0)
|
2016-03-08 15:10:26 -08:00
|
|
|
for _, n1 := range n.List.Slice() {
|
|
|
|
|
if n1.Op == OLITERAL {
|
|
|
|
|
sz += int64(len(n1.Val().U.(string)))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Don't allocate the buffer if the result won't fit.
|
|
|
|
|
if sz < tmpstringbufsize {
|
|
|
|
|
// Create temporary buffer for result string on stack.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
t := types.NewArray(types.Types[TUINT8], tmpstringbufsize)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
buf = nod(OADDR, temp(t), nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// build list of string arguments
|
2016-03-03 20:48:00 -08:00
|
|
|
args := []*Node{buf}
|
2016-03-08 15:10:26 -08:00
|
|
|
for _, n2 := range n.List.Slice() {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
args = append(args, conv(n2, types.Types[TSTRING]))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-03-06 12:02:24 -08:00
|
|
|
var fn string
|
2015-02-13 14:40:36 -05:00
|
|
|
if c <= 5 {
|
|
|
|
|
// small numbers of strings use direct runtime helpers.
|
|
|
|
|
// note: orderexpr knows this cutoff too.
|
2015-03-06 12:02:24 -08:00
|
|
|
fn = fmt.Sprintf("concatstring%d", c)
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
|
|
|
|
// large numbers of strings are passed to the runtime as a slice.
|
2015-03-06 12:02:24 -08:00
|
|
|
fn = "concatstrings"
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
t := types.NewSlice(types.Types[TSTRING])
|
2016-09-16 11:00:54 +10:00
|
|
|
slice := nod(OCOMPLIT, nil, typenod(t))
|
2015-05-26 23:05:35 -04:00
|
|
|
if prealloc[n] != nil {
|
|
|
|
|
prealloc[slice] = prealloc[n]
|
|
|
|
|
}
|
2016-03-08 15:10:26 -08:00
|
|
|
slice.List.Set(args[1:]) // skip buf arg
|
2016-04-23 22:59:01 -07:00
|
|
|
args = []*Node{buf, slice}
|
2015-02-13 14:40:36 -05:00
|
|
|
slice.Esc = EscNone
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-04 15:19:06 -08:00
|
|
|
cat := syslook(fn)
|
2016-09-16 11:00:54 +10:00
|
|
|
r := nod(OCALL, cat, nil)
|
2016-03-08 15:10:26 -08:00
|
|
|
r.List.Set(args)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
r = typecheck(r, Erv)
|
|
|
|
|
r = walkexpr(r, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
r.Type = n.Type
|
|
|
|
|
|
|
|
|
|
return r
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// expand append(l1, l2...) to
|
|
|
|
|
// init {
|
|
|
|
|
// s := l1
|
2016-03-13 15:22:45 -07:00
|
|
|
// n := len(s) + len(l2)
|
|
|
|
|
// // Compare as uint so growslice can panic on overflow.
|
|
|
|
|
// if uint(n) > uint(cap(s)) {
|
|
|
|
|
// s = growslice(s, n)
|
2015-02-13 14:40:36 -05:00
|
|
|
// }
|
2016-03-13 15:22:45 -07:00
|
|
|
// s = s[:n]
|
2015-02-13 14:40:36 -05:00
|
|
|
// memmove(&s[len(l1)], &l2[0], len(l2)*sizeof(T))
|
|
|
|
|
// }
|
|
|
|
|
// s
|
|
|
|
|
//
|
|
|
|
|
// l2 is allowed to be a string.
|
2016-03-07 22:54:46 -08:00
|
|
|
func appendslice(n *Node, init *Nodes) *Node {
|
|
|
|
|
walkexprlistsafe(n.List.Slice(), init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// walkexprlistsafe will leave OINDEX (s[n]) alone if both s
|
|
|
|
|
// and n are name or literal, but those may index the slice we're
|
2016-03-01 23:21:55 +00:00
|
|
|
// modifying here. Fix explicitly.
|
2016-03-09 12:39:36 -08:00
|
|
|
ls := n.List.Slice()
|
|
|
|
|
for i1, n1 := range ls {
|
|
|
|
|
ls[i1] = cheapexpr(n1, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
l1 := n.List.First()
|
|
|
|
|
l2 := n.List.Second()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-03 20:48:00 -08:00
|
|
|
var l []*Node
|
2016-03-13 15:22:45 -07:00
|
|
|
|
|
|
|
|
// var s []T
|
|
|
|
|
s := temp(l1.Type)
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, s, l1)) // s = l1
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-13 15:22:45 -07:00
|
|
|
// n := len(s) + len(l2)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
nn := temp(types.Types[TINT])
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, nn, nod(OADD, nod(OLEN, s, nil), nod(OLEN, l2, nil))))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-13 15:22:45 -07:00
|
|
|
// if uint(n) > uint(cap(s))
|
2016-09-16 11:00:54 +10:00
|
|
|
nif := nod(OIF, nil, nil)
|
|
|
|
|
nif.Left = nod(OGT, nod(OCONV, nn, nil), nod(OCONV, nod(OCAP, s, nil), nil))
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
nif.Left.Left.Type = types.Types[TUINT]
|
|
|
|
|
nif.Left.Right.Type = types.Types[TUINT]
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-13 15:22:45 -07:00
|
|
|
// instantiate growslice(Type*, []any, int) []any
|
|
|
|
|
fn := syslook("growslice")
|
2016-03-30 10:57:47 -07:00
|
|
|
fn = substArgTypes(fn, s.Type.Elem(), s.Type.Elem())
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-13 15:22:45 -07:00
|
|
|
// s = growslice(T, s, n)
|
2016-09-16 11:00:54 +10:00
|
|
|
nif.Nbody.Set1(nod(OAS, s, mkcall1(fn, s.Type, &nif.Ninit, typename(s.Type.Elem()), s, nn)))
|
2016-03-03 20:48:00 -08:00
|
|
|
l = append(l, nif)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-13 15:22:45 -07:00
|
|
|
// s = s[:n]
|
2016-09-16 11:00:54 +10:00
|
|
|
nt := nod(OSLICE, s, nil)
|
2016-04-21 11:55:33 -07:00
|
|
|
nt.SetSliceBounds(nil, nn, nil)
|
2016-03-13 15:22:45 -07:00
|
|
|
nt.Etype = 1
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, s, nt))
|
2016-03-13 15:22:45 -07:00
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
if types.Haspointers(l1.Type.Elem()) {
|
2016-03-13 15:22:45 -07:00
|
|
|
// copy(s[len(l1):], l2)
|
2016-09-16 11:00:54 +10:00
|
|
|
nptr1 := nod(OSLICE, s, nil)
|
|
|
|
|
nptr1.SetSliceBounds(nod(OLEN, l1, nil), nil, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
nptr1.Etype = 1
|
2015-02-23 16:07:24 -05:00
|
|
|
nptr2 := l2
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook("typedslicecopy")
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
fn = substArgTypes(fn, l1.Type, l2.Type)
|
2016-03-07 22:54:46 -08:00
|
|
|
var ln Nodes
|
|
|
|
|
ln.Set(l)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
nt := mkcall1(fn, types.Types[TINT], &ln, typename(l1.Type.Elem()), nptr1, nptr2)
|
2016-03-07 22:54:46 -08:00
|
|
|
l = append(ln.Slice(), nt)
|
2016-09-21 09:44:40 -07:00
|
|
|
} else if instrumenting && !compiling_runtime {
|
2015-02-13 14:40:36 -05:00
|
|
|
// rely on runtime to instrument copy.
|
2016-03-13 15:22:45 -07:00
|
|
|
// copy(s[len(l1):], l2)
|
2016-09-16 11:00:54 +10:00
|
|
|
nptr1 := nod(OSLICE, s, nil)
|
|
|
|
|
nptr1.SetSliceBounds(nod(OLEN, l1, nil), nil, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
nptr1.Etype = 1
|
2015-02-23 16:07:24 -05:00
|
|
|
nptr2 := l2
|
|
|
|
|
var fn *Node
|
2016-03-30 14:56:08 -07:00
|
|
|
if l2.Type.IsString() {
|
2016-03-04 15:19:06 -08:00
|
|
|
fn = syslook("slicestringcopy")
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-03-04 15:19:06 -08:00
|
|
|
fn = syslook("slicecopy")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
fn = substArgTypes(fn, l1.Type, l2.Type)
|
2016-03-07 22:54:46 -08:00
|
|
|
var ln Nodes
|
|
|
|
|
ln.Set(l)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
nt := mkcall1(fn, types.Types[TINT], &ln, nptr1, nptr2, nodintconst(s.Type.Elem().Width))
|
2016-03-07 22:54:46 -08:00
|
|
|
l = append(ln.Slice(), nt)
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
|
|
|
|
// memmove(&s[len(l1)], &l2[0], len(l2)*sizeof(T))
|
2016-09-16 11:00:54 +10:00
|
|
|
nptr1 := nod(OINDEX, s, nod(OLEN, l1, nil))
|
2017-02-27 19:56:38 +02:00
|
|
|
nptr1.SetBounded(true)
|
2016-03-13 15:22:45 -07:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
nptr1 = nod(OADDR, nptr1, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
nptr2 := nod(OSPTR, l2, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook("memmove")
|
2016-03-30 10:57:47 -07:00
|
|
|
fn = substArgTypes(fn, s.Type.Elem(), s.Type.Elem())
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-07 22:54:46 -08:00
|
|
|
var ln Nodes
|
|
|
|
|
ln.Set(l)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
nwid := cheapexpr(conv(nod(OLEN, l2, nil), types.Types[TUINTPTR]), &ln)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
nwid = nod(OMUL, nwid, nodintconst(s.Type.Elem().Width))
|
2016-03-07 22:54:46 -08:00
|
|
|
nt := mkcall1(fn, nil, &ln, nptr1, nptr2, nwid)
|
|
|
|
|
l = append(ln.Slice(), nt)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-19 17:02:01 -07:00
|
|
|
typecheckslice(l, Etop)
|
2015-02-13 14:40:36 -05:00
|
|
|
walkstmtlist(l)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(l...)
|
2015-02-13 14:40:36 -05:00
|
|
|
return s
|
|
|
|
|
}
|
|
|
|
|
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
// Rewrite append(src, x, y, z) so that any side effects in
|
|
|
|
|
// x, y, z (including runtime panics) are evaluated in
|
|
|
|
|
// initialization statements before the append.
|
|
|
|
|
// For normal code generation, stop there and leave the
|
|
|
|
|
// rest to cgen_append.
|
|
|
|
|
//
|
|
|
|
|
// For race detector, expand append(src, a [, b]* ) to
|
2015-02-13 14:40:36 -05:00
|
|
|
//
|
|
|
|
|
// init {
|
|
|
|
|
// s := src
|
|
|
|
|
// const argc = len(args) - 1
|
|
|
|
|
// if cap(s) - len(s) < argc {
|
2015-06-25 19:27:20 -04:00
|
|
|
// s = growslice(s, len(s)+argc)
|
2015-02-13 14:40:36 -05:00
|
|
|
// }
|
|
|
|
|
// n := len(s)
|
|
|
|
|
// s = s[:n+argc]
|
|
|
|
|
// s[n] = a
|
|
|
|
|
// s[n+1] = b
|
|
|
|
|
// ...
|
|
|
|
|
// }
|
|
|
|
|
// s
|
2016-03-07 22:54:46 -08:00
|
|
|
func walkappend(n *Node, init *Nodes, dst *Node) *Node {
|
2016-03-08 15:10:26 -08:00
|
|
|
if !samesafeexpr(dst, n.List.First()) {
|
2017-02-19 15:57:58 +01:00
|
|
|
n.List.SetFirst(safeexpr(n.List.First(), init))
|
|
|
|
|
n.List.SetFirst(walkexpr(n.List.First(), init))
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
}
|
2016-03-07 22:54:46 -08:00
|
|
|
walkexprlistsafe(n.List.Slice()[1:], init)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// walkexprlistsafe will leave OINDEX (s[n]) alone if both s
|
|
|
|
|
// and n are name or literal, but those may index the slice we're
|
2016-03-01 23:21:55 +00:00
|
|
|
// modifying here. Fix explicitly.
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
// Using cheapexpr also makes sure that the evaluation
|
|
|
|
|
// of all arguments (and especially any panics) happen
|
|
|
|
|
// before we begin to modify the slice in a visible way.
|
2016-03-09 12:39:36 -08:00
|
|
|
ls := n.List.Slice()[1:]
|
|
|
|
|
for i, n := range ls {
|
|
|
|
|
ls[i] = cheapexpr(n, init)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
nsrc := n.List.First()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-08 15:10:26 -08:00
|
|
|
argc := n.List.Len() - 1
|
2015-02-13 14:40:36 -05:00
|
|
|
if argc < 1 {
|
|
|
|
|
return nsrc
|
|
|
|
|
}
|
|
|
|
|
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
// General case, with no function calls left as arguments.
|
2015-10-21 07:04:10 -07:00
|
|
|
// Leave for gen, except that instrumentation requires old form.
|
2016-09-21 09:44:40 -07:00
|
|
|
if !instrumenting || compiling_runtime {
|
cmd/internal/gc: optimize append + write barrier
The code generated for x = append(x, v) is roughly:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
}
t[len(t)] = v
len(t)++
x = t
We used to generate this code as Go pseudocode during walk.
Generate it instead as actual instructions during gen.
Doing so lets us apply a few optimizations. The most important
is that when, as in the above example, the source slice and the
destination slice are the same, the code can instead do:
t := x
if len(t)+1 > cap(t) {
t = grow(t)
x = {base(t), len(t)+1, cap(t)}
} else {
len(x)++
}
t[len(t)] = v
That is, in the fast path that does not reallocate the array,
only the updated length needs to be written back to x,
not the array pointer and not the capacity. This is more like
what you'd write by hand in C. It's faster in general, since
the fast path elides two of the three stores, but it's especially
faster when the form of x is such that the base pointer write
would turn into a write barrier. No write, no barrier.
name old mean new mean delta
BinaryTree17 5.68s × (0.97,1.04) 5.81s × (0.98,1.03) +2.35% (p=0.023)
Fannkuch11 4.41s × (0.98,1.03) 4.35s × (1.00,1.00) ~ (p=0.090)
FmtFprintfEmpty 92.7ns × (0.91,1.16) 86.0ns × (0.94,1.11) -7.31% (p=0.038)
FmtFprintfString 281ns × (0.96,1.08) 276ns × (0.98,1.04) ~ (p=0.219)
FmtFprintfInt 288ns × (0.97,1.06) 274ns × (0.98,1.06) -4.94% (p=0.002)
FmtFprintfIntInt 493ns × (0.97,1.04) 506ns × (0.99,1.01) +2.65% (p=0.009)
FmtFprintfPrefixedInt 423ns × (0.97,1.04) 391ns × (0.99,1.01) -7.52% (p=0.000)
FmtFprintfFloat 598ns × (0.99,1.01) 566ns × (0.99,1.01) -5.27% (p=0.000)
FmtManyArgs 1.89µs × (0.98,1.05) 1.91µs × (0.99,1.01) ~ (p=0.231)
GobDecode 14.8ms × (0.98,1.03) 15.3ms × (0.99,1.02) +3.01% (p=0.000)
GobEncode 12.3ms × (0.98,1.01) 11.5ms × (0.97,1.03) -5.93% (p=0.000)
Gzip 656ms × (0.99,1.05) 645ms × (0.99,1.01) ~ (p=0.055)
Gunzip 142ms × (1.00,1.00) 142ms × (1.00,1.00) -0.32% (p=0.034)
HTTPClientServer 91.2µs × (0.97,1.04) 90.5µs × (0.97,1.04) ~ (p=0.468)
JSONEncode 32.6ms × (0.97,1.08) 32.0ms × (0.98,1.03) ~ (p=0.190)
JSONDecode 114ms × (0.97,1.05) 114ms × (0.99,1.01) ~ (p=0.887)
Mandelbrot200 6.11ms × (0.98,1.04) 6.04ms × (1.00,1.01) ~ (p=0.167)
GoParse 6.66ms × (0.97,1.04) 6.47ms × (0.97,1.05) -2.81% (p=0.014)
RegexpMatchEasy0_32 159ns × (0.99,1.00) 171ns × (0.93,1.07) +7.19% (p=0.002)
RegexpMatchEasy0_1K 538ns × (1.00,1.01) 550ns × (0.98,1.01) +2.30% (p=0.000)
RegexpMatchEasy1_32 138ns × (1.00,1.00) 135ns × (0.99,1.02) -1.60% (p=0.000)
RegexpMatchEasy1_1K 869ns × (0.99,1.01) 879ns × (1.00,1.01) +1.08% (p=0.000)
RegexpMatchMedium_32 252ns × (0.99,1.01) 243ns × (1.00,1.00) -3.71% (p=0.000)
RegexpMatchMedium_1K 72.7µs × (1.00,1.00) 70.3µs × (1.00,1.00) -3.34% (p=0.000)
RegexpMatchHard_32 3.85µs × (1.00,1.00) 3.82µs × (1.00,1.01) -0.81% (p=0.000)
RegexpMatchHard_1K 118µs × (1.00,1.00) 117µs × (1.00,1.00) -0.56% (p=0.000)
Revcomp 920ms × (0.97,1.07) 917ms × (0.97,1.04) ~ (p=0.808)
Template 129ms × (0.98,1.03) 114ms × (0.99,1.01) -12.06% (p=0.000)
TimeParse 619ns × (0.99,1.01) 622ns × (0.99,1.01) ~ (p=0.062)
TimeFormat 661ns × (0.98,1.04) 665ns × (0.99,1.01) ~ (p=0.524)
See next CL for combination with a similar optimization for slice.
The benchmarks that are slower in this CL are still faster overall
with the combination of the two.
Change-Id: I2a7421658091b2488c64741b4db15ab6c3b4cb7e
Reviewed-on: https://go-review.googlesource.com/9812
Reviewed-by: David Chase <drchase@google.com>
2015-05-06 12:34:30 -04:00
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-03 20:48:00 -08:00
|
|
|
var l []*Node
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
ns := temp(nsrc.Type)
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, ns, nsrc)) // s = src
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-15 14:34:20 +10:00
|
|
|
na := nodintconst(int64(argc)) // const argc
|
2016-09-16 11:00:54 +10:00
|
|
|
nx := nod(OIF, nil, nil) // if cap(s) - len(s) < argc
|
|
|
|
|
nx.Left = nod(OLT, nod(OSUB, nod(OCAP, ns, nil), nod(OLEN, ns, nil)), na)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook("growslice") // growslice(<type>, old []T, mincap int) (ret []T)
|
2016-03-30 10:57:47 -07:00
|
|
|
fn = substArgTypes(fn, ns.Type.Elem(), ns.Type.Elem())
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
nx.Nbody.Set1(nod(OAS, ns,
|
2016-04-19 15:38:59 -07:00
|
|
|
mkcall1(fn, ns.Type, &nx.Ninit, typename(ns.Type.Elem()), ns,
|
2016-09-16 11:00:54 +10:00
|
|
|
nod(OADD, nod(OLEN, ns, nil), na))))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-03 20:48:00 -08:00
|
|
|
l = append(l, nx)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
nn := temp(types.Types[TINT])
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, nn, nod(OLEN, ns, nil))) // n = len(s)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
nx = nod(OSLICE, ns, nil) // ...s[:n+argc]
|
|
|
|
|
nx.SetSliceBounds(nil, nod(OADD, nn, na), nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
nx.Etype = 1
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, ns, nx)) // s = s[:n+argc]
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-09 12:39:36 -08:00
|
|
|
ls = n.List.Slice()[1:]
|
|
|
|
|
for i, n := range ls {
|
2016-09-16 11:00:54 +10:00
|
|
|
nx = nod(OINDEX, ns, nn) // s[n] ...
|
2017-02-27 19:56:38 +02:00
|
|
|
nx.SetBounded(true)
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, nx, n)) // s[n] = arg
|
2016-03-09 12:39:36 -08:00
|
|
|
if i+1 < len(ls) {
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, nn, nod(OADD, nn, nodintconst(1)))) // n = n + 1
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-19 17:02:01 -07:00
|
|
|
typecheckslice(l, Etop)
|
2015-02-13 14:40:36 -05:00
|
|
|
walkstmtlist(l)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(l...)
|
2015-02-13 14:40:36 -05:00
|
|
|
return ns
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Lower copy(a, b) to a memmove call or a runtime call.
|
|
|
|
|
//
|
|
|
|
|
// init {
|
|
|
|
|
// n := len(a)
|
|
|
|
|
// if n > len(b) { n = len(b) }
|
|
|
|
|
// memmove(a.ptr, b.ptr, n*sizeof(elem(a)))
|
|
|
|
|
// }
|
|
|
|
|
// n;
|
|
|
|
|
//
|
|
|
|
|
// Also works if b is a string.
|
|
|
|
|
//
|
2016-03-07 22:54:46 -08:00
|
|
|
func copyany(n *Node, init *Nodes, runtimecall bool) *Node {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
if types.Haspointers(n.Left.Type.Elem()) {
|
2015-02-23 16:07:24 -05:00
|
|
|
fn := writebarrierfn("typedslicecopy", n.Left.Type, n.Right.Type)
|
2016-03-30 10:57:47 -07:00
|
|
|
return mkcall1(fn, n.Type, init, typename(n.Left.Type.Elem()), n.Left, n.Right)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-10-20 10:00:07 -07:00
|
|
|
if runtimecall {
|
2015-02-23 16:07:24 -05:00
|
|
|
var fn *Node
|
2016-03-30 14:56:08 -07:00
|
|
|
if n.Right.Type.IsString() {
|
2016-03-04 15:19:06 -08:00
|
|
|
fn = syslook("slicestringcopy")
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-03-04 15:19:06 -08:00
|
|
|
fn = syslook("slicecopy")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
fn = substArgTypes(fn, n.Left.Type, n.Right.Type)
|
2016-09-15 14:34:20 +10:00
|
|
|
return mkcall1(fn, n.Type, init, n.Left, n.Right, nodintconst(n.Left.Type.Elem().Width))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n.Left = walkexpr(n.Left, init)
|
|
|
|
|
n.Right = walkexpr(n.Right, init)
|
2015-02-23 16:07:24 -05:00
|
|
|
nl := temp(n.Left.Type)
|
|
|
|
|
nr := temp(n.Right.Type)
|
2016-03-07 22:54:46 -08:00
|
|
|
var l []*Node
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, nl, n.Left))
|
|
|
|
|
l = append(l, nod(OAS, nr, n.Right))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
nfrm := nod(OSPTR, nr, nil)
|
|
|
|
|
nto := nod(OSPTR, nl, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
nlen := temp(types.Types[TINT])
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// n = len(to)
|
2016-09-16 11:00:54 +10:00
|
|
|
l = append(l, nod(OAS, nlen, nod(OLEN, nl, nil)))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// if n > len(frm) { n = len(frm) }
|
2016-09-16 11:00:54 +10:00
|
|
|
nif := nod(OIF, nil, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
nif.Left = nod(OGT, nlen, nod(OLEN, nr, nil))
|
|
|
|
|
nif.Nbody.Append(nod(OAS, nlen, nod(OLEN, nr, nil)))
|
2016-03-07 22:54:46 -08:00
|
|
|
l = append(l, nif)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Call memmove.
|
2016-03-04 15:19:06 -08:00
|
|
|
fn := syslook("memmove")
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-30 10:57:47 -07:00
|
|
|
fn = substArgTypes(fn, nl.Type.Elem(), nl.Type.Elem())
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
nwid := temp(types.Types[TUINTPTR])
|
|
|
|
|
l = append(l, nod(OAS, nwid, conv(nlen, types.Types[TUINTPTR])))
|
2016-09-16 11:00:54 +10:00
|
|
|
nwid = nod(OMUL, nwid, nodintconst(nl.Type.Elem().Width))
|
2016-03-07 22:54:46 -08:00
|
|
|
l = append(l, mkcall1(fn, nil, init, nto, nfrm, nwid))
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-19 17:02:01 -07:00
|
|
|
typecheckslice(l, Etop)
|
2015-02-13 14:40:36 -05:00
|
|
|
walkstmtlist(l)
|
2016-03-07 22:54:46 -08:00
|
|
|
init.Append(l...)
|
2015-02-13 14:40:36 -05:00
|
|
|
return nlen
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func eqfor(t *types.Type, needsize *int) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
// Should only arrive here with large memory or
|
|
|
|
|
// a struct/array containing a non-memory field/element.
|
|
|
|
|
// Small memory is handled inline, and single non-memory
|
|
|
|
|
// is handled during type check (OCMPSTR etc).
|
2016-04-01 11:22:03 -07:00
|
|
|
switch a, _ := algtype1(t); a {
|
|
|
|
|
case AMEM:
|
2016-03-04 15:19:06 -08:00
|
|
|
n := syslook("memequal")
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
n = substArgTypes(n, t, t)
|
2015-02-13 14:40:36 -05:00
|
|
|
*needsize = 1
|
|
|
|
|
return n
|
2016-04-01 11:22:03 -07:00
|
|
|
case ASPECIAL:
|
|
|
|
|
sym := typesymprefix(".eq", t)
|
|
|
|
|
n := newname(sym)
|
2017-04-25 18:14:12 -07:00
|
|
|
n.SetClass(PFUNC)
|
2016-09-16 11:00:54 +10:00
|
|
|
ntype := nod(OTFUNC, nil, nil)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
ntype.List.Append(anonfield(types.NewPtr(t)))
|
|
|
|
|
ntype.List.Append(anonfield(types.NewPtr(t)))
|
|
|
|
|
ntype.Rlist.Append(anonfield(types.Types[TBOOL]))
|
2016-04-01 11:22:03 -07:00
|
|
|
ntype = typecheck(ntype, Etype)
|
|
|
|
|
n.Type = ntype.Type
|
|
|
|
|
*needsize = 0
|
|
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-04-01 11:22:03 -07:00
|
|
|
Fatalf("eqfor %v", t)
|
|
|
|
|
return nil
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
// The result of walkcompare MUST be assigned back to n, e.g.
|
|
|
|
|
// n.Left = walkcompare(n.Left, init)
|
|
|
|
|
func walkcompare(n *Node, init *Nodes) *Node {
|
2015-02-13 14:40:36 -05:00
|
|
|
// Given interface value l and concrete value r, rewrite
|
|
|
|
|
// l == r
|
2016-06-06 13:01:48 -07:00
|
|
|
// into types-equal && data-equal.
|
|
|
|
|
// This is efficient, avoids allocations, and avoids runtime calls.
|
|
|
|
|
var l, r *Node
|
2016-03-30 14:45:47 -07:00
|
|
|
if n.Left.Type.IsInterface() && !n.Right.Type.IsInterface() {
|
2015-02-13 14:40:36 -05:00
|
|
|
l = n.Left
|
|
|
|
|
r = n.Right
|
2016-03-30 14:45:47 -07:00
|
|
|
} else if !n.Left.Type.IsInterface() && n.Right.Type.IsInterface() {
|
2015-02-13 14:40:36 -05:00
|
|
|
l = n.Right
|
|
|
|
|
r = n.Left
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if l != nil {
|
2016-06-06 13:01:48 -07:00
|
|
|
// Handle both == and !=.
|
|
|
|
|
eq := n.Op
|
|
|
|
|
var andor Op
|
|
|
|
|
if eq == OEQ {
|
|
|
|
|
andor = OANDAND
|
2015-02-13 14:40:36 -05:00
|
|
|
} else {
|
2016-06-06 13:01:48 -07:00
|
|
|
andor = OOROR
|
|
|
|
|
}
|
|
|
|
|
// Check for types equal.
|
|
|
|
|
// For empty interface, this is:
|
|
|
|
|
// l.tab == type(r)
|
|
|
|
|
// For non-empty interface, this is:
|
|
|
|
|
// l.tab != nil && l.tab._type == type(r)
|
|
|
|
|
var eqtype *Node
|
2016-09-16 11:00:54 +10:00
|
|
|
tab := nod(OITAB, l, nil)
|
2016-06-06 13:01:48 -07:00
|
|
|
rtyp := typename(r.Type)
|
|
|
|
|
if l.Type.IsEmptyInterface() {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
tab.Type = types.NewPtr(types.Types[TUINT8])
|
2017-04-25 18:02:43 -07:00
|
|
|
tab.SetTypecheck(1)
|
2016-09-16 11:00:54 +10:00
|
|
|
eqtype = nod(eq, tab, rtyp)
|
2016-06-06 13:01:48 -07:00
|
|
|
} else {
|
2016-09-16 11:00:54 +10:00
|
|
|
nonnil := nod(brcom(eq), nodnil(), tab)
|
|
|
|
|
match := nod(eq, itabType(tab), rtyp)
|
|
|
|
|
eqtype = nod(andor, nonnil, match)
|
2016-06-06 13:01:48 -07:00
|
|
|
}
|
|
|
|
|
// Check for data equal.
|
2016-09-16 11:00:54 +10:00
|
|
|
eqdata := nod(eq, ifaceData(l, r.Type), r)
|
2016-06-06 13:01:48 -07:00
|
|
|
// Put it all together.
|
2016-09-16 11:00:54 +10:00
|
|
|
expr := nod(andor, eqtype, eqdata)
|
2016-06-06 13:01:48 -07:00
|
|
|
n = finishcompare(n, expr, init)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return n
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Must be comparison of array or struct.
|
|
|
|
|
// Otherwise back end handles it.
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
// While we're here, decide whether to
|
|
|
|
|
// inline or call an eq alg.
|
2015-03-02 20:34:22 -05:00
|
|
|
t := n.Left.Type
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
var inline bool
|
2015-02-13 14:40:36 -05:00
|
|
|
switch t.Etype {
|
|
|
|
|
default:
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return n
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
case TARRAY:
|
|
|
|
|
inline = t.NumElem() <= 1 || (t.NumElem() <= 4 && issimple[t.Elem().Etype])
|
|
|
|
|
case TSTRUCT:
|
|
|
|
|
inline = t.NumFields() <= 4
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-03-02 20:34:22 -05:00
|
|
|
cmpl := n.Left
|
2015-02-13 14:40:36 -05:00
|
|
|
for cmpl != nil && cmpl.Op == OCONVNOP {
|
|
|
|
|
cmpl = cmpl.Left
|
|
|
|
|
}
|
2015-03-02 20:34:22 -05:00
|
|
|
cmpr := n.Right
|
2015-02-13 14:40:36 -05:00
|
|
|
for cmpr != nil && cmpr.Op == OCONVNOP {
|
|
|
|
|
cmpr = cmpr.Left
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
// Chose not to inline. Call equality function directly.
|
|
|
|
|
if !inline {
|
2017-01-16 12:37:11 -05:00
|
|
|
if isvaluelit(cmpl) {
|
|
|
|
|
var_ := temp(cmpl.Type)
|
|
|
|
|
anylit(cmpl, var_, init)
|
|
|
|
|
cmpl = var_
|
|
|
|
|
}
|
|
|
|
|
if isvaluelit(cmpr) {
|
|
|
|
|
var_ := temp(cmpr.Type)
|
|
|
|
|
anylit(cmpr, var_, init)
|
|
|
|
|
cmpr = var_
|
|
|
|
|
}
|
2017-01-14 21:40:16 -08:00
|
|
|
if !islvalue(cmpl) || !islvalue(cmpr) {
|
|
|
|
|
Fatalf("arguments of comparison must be lvalues - %v %v", cmpl, cmpr)
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
// eq algs take pointers
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
pl := temp(types.NewPtr(t))
|
2016-09-16 11:00:54 +10:00
|
|
|
al := nod(OAS, pl, nod(OADDR, cmpl, nil))
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
al.Right.Etype = 1 // addr does not escape
|
|
|
|
|
al = typecheck(al, Etop)
|
|
|
|
|
init.Append(al)
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
pr := temp(types.NewPtr(t))
|
2016-09-16 11:00:54 +10:00
|
|
|
ar := nod(OAS, pr, nod(OADDR, cmpr, nil))
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
ar.Right.Etype = 1 // addr does not escape
|
|
|
|
|
ar = typecheck(ar, Etop)
|
|
|
|
|
init.Append(ar)
|
|
|
|
|
|
|
|
|
|
var needsize int
|
2016-09-16 11:00:54 +10:00
|
|
|
call := nod(OCALL, eqfor(t, &needsize), nil)
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
call.List.Append(pl)
|
|
|
|
|
call.List.Append(pr)
|
|
|
|
|
if needsize != 0 {
|
2016-09-15 14:34:20 +10:00
|
|
|
call.List.Append(nodintconst(t.Width))
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
}
|
|
|
|
|
res := call
|
|
|
|
|
if n.Op != OEQ {
|
2016-09-16 11:00:54 +10:00
|
|
|
res = nod(ONOT, res, nil)
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
}
|
|
|
|
|
n = finishcompare(n, res, init)
|
|
|
|
|
return n
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
// inline: build boolean expression comparing element by element
|
|
|
|
|
andor := OANDAND
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Op == ONE {
|
|
|
|
|
andor = OOROR
|
|
|
|
|
}
|
2015-03-02 20:34:22 -05:00
|
|
|
var expr *Node
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
compare := func(el, er *Node) {
|
2016-09-16 11:00:54 +10:00
|
|
|
a := nod(n.Op, el, er)
|
2015-02-13 14:40:36 -05:00
|
|
|
if expr == nil {
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
expr = a
|
|
|
|
|
} else {
|
2016-09-16 11:00:54 +10:00
|
|
|
expr = nod(andor, expr, a)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
cmpl = safeexpr(cmpl, init)
|
|
|
|
|
cmpr = safeexpr(cmpr, init)
|
|
|
|
|
if t.IsStruct() {
|
|
|
|
|
for _, f := range t.Fields().Slice() {
|
|
|
|
|
sym := f.Sym
|
2017-04-21 07:51:41 -07:00
|
|
|
if sym.IsBlank() {
|
2015-02-13 14:40:36 -05:00
|
|
|
continue
|
|
|
|
|
}
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
compare(
|
2016-09-15 15:45:10 +10:00
|
|
|
nodSym(OXDOT, cmpl, sym),
|
|
|
|
|
nodSym(OXDOT, cmpr, sym),
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
} else {
|
|
|
|
|
for i := 0; int64(i) < t.NumElem(); i++ {
|
|
|
|
|
compare(
|
2016-09-16 11:00:54 +10:00
|
|
|
nod(OINDEX, cmpl, nodintconst(int64(i))),
|
|
|
|
|
nod(OINDEX, cmpr, nodintconst(int64(i))),
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
if expr == nil {
|
2016-09-15 15:45:10 +10:00
|
|
|
expr = nodbool(n.Op == OEQ)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: when inlining ==, don’t take the address of the values
This CL reworks walkcompare for clarity and concision.
It also makes one significant functional change.
(The functional change is hard to separate cleanly
from the cleanup, so I just did them together.)
When inlining and unrolling an equality comparison
for a small struct or array, compare the elements like:
a[0] == b[0] && a[1] == b[1]
rather than
pa := &a
pb := &b
pa[0] == pb[0] && pa[1] == pb[1]
The result is the same, but taking the address
and working through the indirect
forces the backends to generate less efficient code.
This is only an improvement with the SSA backend.
However, every port but s390x now has a working
SSA backend, and switching to the SSA backend
by default everywhere is a priority for Go 1.8.
It thus seems reasonable to start to prioritize
SSA performance over the old backend.
Updates #15303
Sample code:
type T struct {
a, b int8
}
func g(a T) bool {
return a == T{1, 2}
}
SSA before:
"".g t=1 size=80 args=0x10 locals=0x8
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $8-16
0x0000 00000 (badeq.go:7) SUBQ $8, SP
0x0004 00004 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0004 00004 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0004 00004 (badeq.go:8) MOVBLZX "".a+16(FP), AX
0x0009 00009 (badeq.go:8) MOVB AL, "".autotmp_0+6(SP)
0x000d 00013 (badeq.go:8) MOVBLZX "".a+17(FP), AX
0x0012 00018 (badeq.go:8) MOVB AL, "".autotmp_0+7(SP)
0x0016 00022 (badeq.go:8) MOVB $0, "".autotmp_1+4(SP)
0x001b 00027 (badeq.go:8) MOVB $1, "".autotmp_1+4(SP)
0x0020 00032 (badeq.go:8) MOVB $2, "".autotmp_1+5(SP)
0x0025 00037 (badeq.go:8) MOVBLZX "".autotmp_0+6(SP), AX
0x002a 00042 (badeq.go:8) MOVBLZX "".autotmp_1+4(SP), CX
0x002f 00047 (badeq.go:8) CMPB AL, CL
0x0031 00049 (badeq.go:8) JNE 70
0x0033 00051 (badeq.go:8) MOVBLZX "".autotmp_0+7(SP), AX
0x0038 00056 (badeq.go:8) CMPB AL, $2
0x003a 00058 (badeq.go:8) SETEQ AL
0x003d 00061 (badeq.go:8) MOVB AL, "".~r1+24(FP)
0x0041 00065 (badeq.go:8) ADDQ $8, SP
0x0045 00069 (badeq.go:8) RET
0x0046 00070 (badeq.go:8) MOVB $0, AL
0x0048 00072 (badeq.go:8) JMP 61
SSA after:
"".g t=1 size=32 args=0x10 locals=0x0
0x0000 00000 (badeq.go:7) TEXT "".g(SB), $0-16
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) NOP
0x0000 00000 (badeq.go:7) FUNCDATA $0, gclocals·23e8278e2b69a3a75fa59b23c49ed6ad(SB)
0x0000 00000 (badeq.go:7) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (badeq.go:8) MOVBLZX "".a+8(FP), AX
0x0005 00005 (badeq.go:8) CMPB AL, $1
0x0007 00007 (badeq.go:8) JNE 25
0x0009 00009 (badeq.go:8) MOVBLZX "".a+9(FP), CX
0x000e 00014 (badeq.go:8) CMPB CL, $2
0x0011 00017 (badeq.go:8) SETEQ AL
0x0014 00020 (badeq.go:8) MOVB AL, "".~r1+16(FP)
0x0018 00024 (badeq.go:8) RET
0x0019 00025 (badeq.go:8) MOVB $0, AL
0x001b 00027 (badeq.go:8) JMP 20
Change-Id: I120185d58012b7bbcdb1ec01225b5b08d0855d86
Reviewed-on: https://go-review.googlesource.com/22277
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-04-15 15:49:00 -07:00
|
|
|
n = finishcompare(n, expr, init)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return n
|
2015-03-02 20:34:22 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-23 16:43:17 -07:00
|
|
|
// The result of finishcompare MUST be assigned back to n, e.g.
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
// n.Left = finishcompare(n.Left, x, r, init)
|
2016-03-23 16:43:17 -07:00
|
|
|
func finishcompare(n, r *Node, init *Nodes) *Node {
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
// Use nn here to avoid passing r to typecheck.
|
2016-03-23 16:43:17 -07:00
|
|
|
nn := r
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
nn = typecheck(nn, Erv)
|
|
|
|
|
nn = walkexpr(nn, init)
|
|
|
|
|
r = nn
|
2015-02-13 14:40:36 -05:00
|
|
|
if r.Type != n.Type {
|
2016-09-16 11:00:54 +10:00
|
|
|
r = nod(OCONVNOP, r, nil)
|
2015-02-13 14:40:36 -05:00
|
|
|
r.Type = n.Type
|
2017-04-25 18:02:43 -07:00
|
|
|
r.SetTypecheck(1)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
nn = r
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
return nn
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-05-26 18:08:24 -07:00
|
|
|
// isIntOrdering reports whether n is a <, ≤, >, or ≥ ordering between integers.
|
|
|
|
|
func (n *Node) isIntOrdering() bool {
|
|
|
|
|
switch n.Op {
|
|
|
|
|
case OLE, OLT, OGE, OGT:
|
|
|
|
|
default:
|
|
|
|
|
return false
|
|
|
|
|
}
|
|
|
|
|
return n.Left.Type.IsInteger() && n.Right.Type.IsInteger()
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// walkinrange optimizes integer-in-range checks, such as 4 <= x && x < 10.
|
|
|
|
|
// n must be an OANDAND or OOROR node.
|
|
|
|
|
// The result of walkinrange MUST be assigned back to n, e.g.
|
|
|
|
|
// n.Left = walkinrange(n.Left)
|
|
|
|
|
func walkinrange(n *Node, init *Nodes) *Node {
|
|
|
|
|
// We are looking for something equivalent to a opl b OP b opr c, where:
|
|
|
|
|
// * a, b, and c have integer type
|
|
|
|
|
// * b is side-effect-free
|
|
|
|
|
// * opl and opr are each < or ≤
|
|
|
|
|
// * OP is &&
|
|
|
|
|
l := n.Left
|
|
|
|
|
r := n.Right
|
|
|
|
|
if !l.isIntOrdering() || !r.isIntOrdering() {
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Find b, if it exists, and rename appropriately.
|
|
|
|
|
// Input is: l.Left l.Op l.Right ANDAND/OROR r.Left r.Op r.Right
|
|
|
|
|
// Output is: a opl b(==x) ANDAND/OROR b(==x) opr c
|
|
|
|
|
a, opl, b := l.Left, l.Op, l.Right
|
|
|
|
|
x, opr, c := r.Left, r.Op, r.Right
|
|
|
|
|
for i := 0; ; i++ {
|
|
|
|
|
if samesafeexpr(b, x) {
|
|
|
|
|
break
|
|
|
|
|
}
|
|
|
|
|
if i == 3 {
|
|
|
|
|
// Tried all permutations and couldn't find an appropriate b == x.
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
if i&1 == 0 {
|
2016-09-15 15:45:10 +10:00
|
|
|
a, opl, b = b, brrev(opl), a
|
2016-05-26 18:08:24 -07:00
|
|
|
} else {
|
2016-09-15 15:45:10 +10:00
|
|
|
x, opr, c = c, brrev(opr), x
|
2016-05-26 18:08:24 -07:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// If n.Op is ||, apply de Morgan.
|
|
|
|
|
// Negate the internal ops now; we'll negate the top level op at the end.
|
|
|
|
|
// Henceforth assume &&.
|
|
|
|
|
negateResult := n.Op == OOROR
|
|
|
|
|
if negateResult {
|
2016-09-15 15:45:10 +10:00
|
|
|
opl = brcom(opl)
|
|
|
|
|
opr = brcom(opr)
|
2016-05-26 18:08:24 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cmpdir := func(o Op) int {
|
|
|
|
|
switch o {
|
|
|
|
|
case OLE, OLT:
|
|
|
|
|
return -1
|
|
|
|
|
case OGE, OGT:
|
|
|
|
|
return +1
|
|
|
|
|
}
|
|
|
|
|
Fatalf("walkinrange cmpdir %v", o)
|
|
|
|
|
return 0
|
|
|
|
|
}
|
|
|
|
|
if cmpdir(opl) != cmpdir(opr) {
|
|
|
|
|
// Not a range check; something like b < a && b < c.
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch opl {
|
|
|
|
|
case OGE, OGT:
|
|
|
|
|
// We have something like a > b && b ≥ c.
|
|
|
|
|
// Switch and reverse ops and rename constants,
|
|
|
|
|
// to make it look like a ≤ b && b < c.
|
|
|
|
|
a, c = c, a
|
2016-09-15 15:45:10 +10:00
|
|
|
opl, opr = brrev(opr), brrev(opl)
|
2016-05-26 18:08:24 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// We must ensure that c-a is non-negative.
|
|
|
|
|
// For now, require a and c to be constants.
|
|
|
|
|
// In the future, we could also support a == 0 and c == len/cap(...).
|
|
|
|
|
// Unfortunately, by this point, most len/cap expressions have been
|
|
|
|
|
// stored into temporary variables.
|
|
|
|
|
if !Isconst(a, CTINT) || !Isconst(c, CTINT) {
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if opl == OLT {
|
|
|
|
|
// We have a < b && ...
|
|
|
|
|
// We need a ≤ b && ... to safely use unsigned comparison tricks.
|
|
|
|
|
// If a is not the maximum constant for b's type,
|
|
|
|
|
// we can increment a and switch to ≤.
|
2016-09-16 00:33:29 +10:00
|
|
|
if a.Int64() >= maxintval[b.Type.Etype].Int64() {
|
2016-05-26 18:08:24 -07:00
|
|
|
return n
|
|
|
|
|
}
|
2016-09-15 14:34:20 +10:00
|
|
|
a = nodintconst(a.Int64() + 1)
|
2016-05-26 18:08:24 -07:00
|
|
|
opl = OLE
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
bound := c.Int64() - a.Int64()
|
|
|
|
|
if bound < 0 {
|
|
|
|
|
// Bad news. Something like 5 <= x && x < 3.
|
|
|
|
|
// Rare in practice, and we still need to generate side-effects,
|
|
|
|
|
// so just leave it alone.
|
|
|
|
|
return n
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// We have a ≤ b && b < c (or a ≤ b && b ≤ c).
|
|
|
|
|
// This is equivalent to (a-a) ≤ (b-a) && (b-a) < (c-a),
|
|
|
|
|
// which is equivalent to 0 ≤ (b-a) && (b-a) < (c-a),
|
|
|
|
|
// which is equivalent to uint(b-a) < uint(c-a).
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
ut := b.Type.ToUnsigned()
|
2016-09-16 11:00:54 +10:00
|
|
|
lhs := conv(nod(OSUB, b, a), ut)
|
2016-09-15 14:34:20 +10:00
|
|
|
rhs := nodintconst(bound)
|
2016-05-26 18:08:24 -07:00
|
|
|
if negateResult {
|
|
|
|
|
// Negate top level.
|
2016-09-15 15:45:10 +10:00
|
|
|
opr = brcom(opr)
|
2016-05-26 18:08:24 -07:00
|
|
|
}
|
2016-09-16 11:00:54 +10:00
|
|
|
cmp := nod(opr, lhs, rhs)
|
2016-12-07 17:40:46 -08:00
|
|
|
cmp.Pos = n.Pos
|
2016-05-26 18:08:24 -07:00
|
|
|
cmp = addinit(cmp, l.Ninit.Slice())
|
|
|
|
|
cmp = addinit(cmp, r.Ninit.Slice())
|
2016-10-24 15:36:26 -07:00
|
|
|
// Typecheck the AST rooted at cmp...
|
2016-05-26 18:08:24 -07:00
|
|
|
cmp = typecheck(cmp, Erv)
|
2016-10-24 15:36:26 -07:00
|
|
|
// ...but then reset cmp's type to match n's type.
|
|
|
|
|
cmp.Type = n.Type
|
2016-05-26 18:08:24 -07:00
|
|
|
cmp = walkexpr(cmp, init)
|
|
|
|
|
return cmp
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-13 14:40:36 -05:00
|
|
|
// return 1 if integer n must be in range [0, max), 0 otherwise
|
2015-02-17 22:13:49 -05:00
|
|
|
func bounded(n *Node, max int64) bool {
|
2016-03-30 15:09:25 -07:00
|
|
|
if n.Type == nil || !n.Type.IsInteger() {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-30 15:09:25 -07:00
|
|
|
sign := n.Type.IsSigned()
|
2015-02-23 16:07:24 -05:00
|
|
|
bits := int32(8 * n.Type.Width)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-15 14:34:20 +10:00
|
|
|
if smallintconst(n) {
|
2016-04-01 14:51:02 -07:00
|
|
|
v := n.Int64()
|
2015-02-17 22:13:49 -05:00
|
|
|
return 0 <= v && v < max
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch n.Op {
|
|
|
|
|
case OAND:
|
2015-02-23 16:07:24 -05:00
|
|
|
v := int64(-1)
|
2016-09-15 14:34:20 +10:00
|
|
|
if smallintconst(n.Left) {
|
2016-04-01 14:51:02 -07:00
|
|
|
v = n.Left.Int64()
|
2016-09-15 14:34:20 +10:00
|
|
|
} else if smallintconst(n.Right) {
|
2016-04-01 14:51:02 -07:00
|
|
|
v = n.Right.Int64()
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if 0 <= v && v < max {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
case OMOD:
|
2016-09-15 14:34:20 +10:00
|
|
|
if !sign && smallintconst(n.Right) {
|
2016-04-01 14:51:02 -07:00
|
|
|
v := n.Right.Int64()
|
2015-02-13 14:40:36 -05:00
|
|
|
if 0 <= v && v <= max {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
case ODIV:
|
2016-09-15 14:34:20 +10:00
|
|
|
if !sign && smallintconst(n.Right) {
|
2016-04-01 14:51:02 -07:00
|
|
|
v := n.Right.Int64()
|
2015-02-13 14:40:36 -05:00
|
|
|
for bits > 0 && v >= 2 {
|
|
|
|
|
bits--
|
|
|
|
|
v >>= 1
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
case ORSH:
|
2016-09-15 14:34:20 +10:00
|
|
|
if !sign && smallintconst(n.Right) {
|
2016-04-01 14:51:02 -07:00
|
|
|
v := n.Right.Int64()
|
2015-02-13 14:40:36 -05:00
|
|
|
if v > int64(bits) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
bits -= int32(v)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2015-03-01 07:54:01 +00:00
|
|
|
if !sign && bits <= 62 && 1<<uint(bits) <= max {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-10 16:15:26 -05:00
|
|
|
// usemethod check interface method calls for uses of reflect.Type.Method.
|
|
|
|
|
func usemethod(n *Node) {
|
|
|
|
|
t := n.Left.Type
|
|
|
|
|
|
|
|
|
|
// Looking for either of:
|
|
|
|
|
// Method(int) reflect.Method
|
|
|
|
|
// MethodByName(string) (reflect.Method, bool)
|
|
|
|
|
//
|
|
|
|
|
// TODO(crawshaw): improve precision of match by working out
|
|
|
|
|
// how to check the method name.
|
2016-03-17 13:26:08 -07:00
|
|
|
if n := t.Params().NumFields(); n != 1 {
|
2016-03-10 16:15:26 -05:00
|
|
|
return
|
|
|
|
|
}
|
2016-03-17 13:26:08 -07:00
|
|
|
if n := t.Results().NumFields(); n != 1 && n != 2 {
|
2016-03-10 16:15:26 -05:00
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
p0 := t.Params().Field(0)
|
|
|
|
|
res0 := t.Results().Field(0)
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
var res1 *types.Field
|
2016-03-17 13:26:08 -07:00
|
|
|
if t.Results().NumFields() == 2 {
|
2016-03-10 16:15:26 -05:00
|
|
|
res1 = t.Results().Field(1)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if res1 == nil {
|
|
|
|
|
if p0.Type.Etype != TINT {
|
|
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
} else {
|
2016-03-30 14:56:08 -07:00
|
|
|
if !p0.Type.IsString() {
|
2016-03-10 16:15:26 -05:00
|
|
|
return
|
|
|
|
|
}
|
2016-03-30 14:56:08 -07:00
|
|
|
if !res1.Type.IsBoolean() {
|
2016-03-10 16:15:26 -05:00
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
}
|
2016-08-31 10:32:40 -07:00
|
|
|
if res0.Type.String() != "reflect.Method" {
|
2016-03-10 16:15:26 -05:00
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
|
2017-02-27 19:56:38 +02:00
|
|
|
Curfn.Func.SetReflectMethod(true)
|
2016-03-10 16:15:26 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-13 14:40:36 -05:00
|
|
|
func usefield(n *Node) {
|
2017-04-18 12:53:25 -07:00
|
|
|
if objabi.Fieldtrack_enabled == 0 {
|
2015-02-13 14:40:36 -05:00
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch n.Op {
|
|
|
|
|
default:
|
2016-04-27 15:10:10 +10:00
|
|
|
Fatalf("usefield %v", n.Op)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-04-01 09:38:44 -07:00
|
|
|
case ODOT, ODOTPTR:
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
}
|
cmd/compile: change ODOT and friends to use Sym, not Right
The Node type ODOT and its variants all represent a selector, with a
simple name to the right of the dot. Before this change this was
represented by using an ONAME Node in the Right field. This ONAME node
served no useful purpose. This CL changes these Node types to store the
symbol in the Sym field instead, thus not requiring allocating a Node
for each selector.
When compiling x/tools/go/types this CL eliminates nearly 5000 calls to
newname and reduces the total number of Nodes allocated by about 6.6%.
It seems to cut compilation time by 1 to 2 percent.
Getting this right was somewhat subtle, and I added two dubious changes
to produce the exact same output as before. One is to ishairy in
inl.go: the ONAME node increased the cost of ODOT and friends by 1, and
I retained that, although really ODOT is not more expensive than any
other node. The other is to varexpr in walk.go: because the ONAME in
the Right field of an ODOT has no class, varexpr would always return
false for an ODOT, although in fact for some ODOT's it seemingly ought
to return true; I added an && false for now. I will send separate CLs,
that will break toolstash -cmp, to clean these up.
This CL passes toolstash -cmp.
Change-Id: I4af8a10cc59078c436130ce472f25abc3a9b2f80
Reviewed-on: https://go-review.googlesource.com/20890
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-03-18 16:52:30 -07:00
|
|
|
if n.Sym == nil {
|
2016-03-02 20:54:41 -08:00
|
|
|
// No field name. This DOTPTR was built by the compiler for access
|
|
|
|
|
// to runtime data structures. Ignore.
|
|
|
|
|
return
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2015-05-26 21:49:31 -04:00
|
|
|
t := n.Left.Type
|
2016-03-30 15:09:25 -07:00
|
|
|
if t.IsPtr() {
|
2016-03-30 10:57:47 -07:00
|
|
|
t = t.Elem()
|
2015-05-26 21:49:31 -04:00
|
|
|
}
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
field := dotField[typeSymKey{t.Orig, n.Sym}]
|
2015-02-13 14:40:36 -05:00
|
|
|
if field == nil {
|
cmd/compile: change ODOT and friends to use Sym, not Right
The Node type ODOT and its variants all represent a selector, with a
simple name to the right of the dot. Before this change this was
represented by using an ONAME Node in the Right field. This ONAME node
served no useful purpose. This CL changes these Node types to store the
symbol in the Sym field instead, thus not requiring allocating a Node
for each selector.
When compiling x/tools/go/types this CL eliminates nearly 5000 calls to
newname and reduces the total number of Nodes allocated by about 6.6%.
It seems to cut compilation time by 1 to 2 percent.
Getting this right was somewhat subtle, and I added two dubious changes
to produce the exact same output as before. One is to ishairy in
inl.go: the ONAME node increased the cost of ODOT and friends by 1, and
I retained that, although really ODOT is not more expensive than any
other node. The other is to varexpr in walk.go: because the ONAME in
the Right field of an ODOT has no class, varexpr would always return
false for an ODOT, although in fact for some ODOT's it seemingly ought
to return true; I added an && false for now. I will send separate CLs,
that will break toolstash -cmp, to clean these up.
This CL passes toolstash -cmp.
Change-Id: I4af8a10cc59078c436130ce472f25abc3a9b2f80
Reviewed-on: https://go-review.googlesource.com/20890
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-03-18 16:52:30 -07:00
|
|
|
Fatalf("usefield %v %v without paramfld", n.Left.Type, n.Sym)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-04-25 13:24:48 -07:00
|
|
|
if !strings.Contains(field.Note, "go:\"track\"") {
|
2015-02-13 14:40:36 -05:00
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
|
2016-03-10 16:15:44 -08:00
|
|
|
outer := n.Left.Type
|
2016-03-30 15:09:25 -07:00
|
|
|
if outer.IsPtr() {
|
2016-03-30 10:57:47 -07:00
|
|
|
outer = outer.Elem()
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2016-03-10 16:15:44 -08:00
|
|
|
if outer.Sym == nil {
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("tracked field must be in named struct type")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
if !exportname(field.Sym.Name) {
|
2016-09-15 15:45:10 +10:00
|
|
|
yyerror("tracked field must be exported (upper case)")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-10 16:15:44 -08:00
|
|
|
sym := tracksym(outer, field)
|
|
|
|
|
if Curfn.Func.FieldTrack == nil {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
Curfn.Func.FieldTrack = make(map[*types.Sym]struct{})
|
2016-03-10 16:15:44 -08:00
|
|
|
}
|
|
|
|
|
Curfn.Func.FieldTrack[sym] = struct{}{}
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-08 10:26:20 -08:00
|
|
|
func candiscardlist(l Nodes) bool {
|
2016-03-08 15:10:26 -08:00
|
|
|
for _, n := range l.Slice() {
|
|
|
|
|
if !candiscard(n) {
|
2016-02-27 14:31:33 -08:00
|
|
|
return false
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return true
|
|
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
func candiscard(n *Node) bool {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n == nil {
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch n.Op {
|
|
|
|
|
default:
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Discardable as long as the subpieces are.
|
|
|
|
|
case ONAME,
|
|
|
|
|
ONONAME,
|
|
|
|
|
OTYPE,
|
|
|
|
|
OPACK,
|
|
|
|
|
OLITERAL,
|
|
|
|
|
OADD,
|
|
|
|
|
OSUB,
|
|
|
|
|
OOR,
|
|
|
|
|
OXOR,
|
|
|
|
|
OADDSTR,
|
|
|
|
|
OADDR,
|
|
|
|
|
OANDAND,
|
|
|
|
|
OARRAYBYTESTR,
|
|
|
|
|
OARRAYRUNESTR,
|
|
|
|
|
OSTRARRAYBYTE,
|
|
|
|
|
OSTRARRAYRUNE,
|
|
|
|
|
OCAP,
|
|
|
|
|
OCMPIFACE,
|
|
|
|
|
OCMPSTR,
|
|
|
|
|
OCOMPLIT,
|
|
|
|
|
OMAPLIT,
|
|
|
|
|
OSTRUCTLIT,
|
|
|
|
|
OARRAYLIT,
|
2016-06-19 07:20:28 -07:00
|
|
|
OSLICELIT,
|
2015-02-13 14:40:36 -05:00
|
|
|
OPTRLIT,
|
|
|
|
|
OCONV,
|
|
|
|
|
OCONVIFACE,
|
|
|
|
|
OCONVNOP,
|
|
|
|
|
ODOT,
|
|
|
|
|
OEQ,
|
|
|
|
|
ONE,
|
|
|
|
|
OLT,
|
|
|
|
|
OLE,
|
|
|
|
|
OGT,
|
|
|
|
|
OGE,
|
|
|
|
|
OKEY,
|
2016-10-12 15:48:18 -07:00
|
|
|
OSTRUCTKEY,
|
2015-02-13 14:40:36 -05:00
|
|
|
OLEN,
|
|
|
|
|
OMUL,
|
|
|
|
|
OLSH,
|
|
|
|
|
ORSH,
|
|
|
|
|
OAND,
|
|
|
|
|
OANDNOT,
|
|
|
|
|
ONEW,
|
|
|
|
|
ONOT,
|
|
|
|
|
OCOM,
|
|
|
|
|
OPLUS,
|
|
|
|
|
OMINUS,
|
|
|
|
|
OOROR,
|
|
|
|
|
OPAREN,
|
|
|
|
|
ORUNESTR,
|
|
|
|
|
OREAL,
|
|
|
|
|
OIMAG,
|
|
|
|
|
OCOMPLEX:
|
|
|
|
|
break
|
|
|
|
|
|
|
|
|
|
// Discardable as long as we know it's not division by zero.
|
2015-04-01 09:38:44 -07:00
|
|
|
case ODIV, OMOD:
|
2016-03-20 13:55:42 -07:00
|
|
|
if Isconst(n.Right, CTINT) && n.Right.Val().U.(*Mpint).CmpInt64(0) != 0 {
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
}
|
2016-03-20 13:55:42 -07:00
|
|
|
if Isconst(n.Right, CTFLT) && n.Right.Val().U.(*Mpflt).CmpFloat64(0) != 0 {
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
}
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Discardable as long as we know it won't fail because of a bad size.
|
2015-04-01 09:38:44 -07:00
|
|
|
case OMAKECHAN, OMAKEMAP:
|
2016-03-20 13:55:42 -07:00
|
|
|
if Isconst(n.Left, CTINT) && n.Left.Val().U.(*Mpint).CmpInt64(0) == 0 {
|
2015-02-13 14:40:36 -05:00
|
|
|
break
|
|
|
|
|
}
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Difficult to tell what sizes are okay.
|
|
|
|
|
case OMAKESLICE:
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-03-03 20:48:00 -08:00
|
|
|
if !candiscard(n.Left) || !candiscard(n.Right) || !candiscardlist(n.Ninit) || !candiscardlist(n.Nbody) || !candiscardlist(n.List) || !candiscardlist(n.Rlist) {
|
2015-02-17 22:13:49 -05:00
|
|
|
return false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-17 22:13:49 -05:00
|
|
|
return true
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// rewrite
|
|
|
|
|
// print(x, y, z)
|
|
|
|
|
// into
|
|
|
|
|
// func(a1, a2, a3) {
|
|
|
|
|
// print(a1, a2, a3)
|
|
|
|
|
// }(x, y, z)
|
|
|
|
|
// and same for println.
|
|
|
|
|
|
|
|
|
|
var walkprintfunc_prgen int
|
|
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
// The result of walkprintfunc MUST be assigned back to n, e.g.
|
|
|
|
|
// n.Left = walkprintfunc(n.Left, init)
|
|
|
|
|
func walkprintfunc(n *Node, init *Nodes) *Node {
|
2016-03-08 15:10:26 -08:00
|
|
|
if n.Ninit.Len() != 0 {
|
2016-03-07 22:54:46 -08:00
|
|
|
walkstmtlist(n.Ninit.Slice())
|
|
|
|
|
init.AppendNodes(&n.Ninit)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
t := nod(OTFUNC, nil, nil)
|
2015-02-23 16:07:24 -05:00
|
|
|
num := 0
|
2016-03-03 20:48:00 -08:00
|
|
|
var printargs []*Node
|
2015-02-23 16:07:24 -05:00
|
|
|
var a *Node
|
|
|
|
|
var buf string
|
2016-03-08 15:10:26 -08:00
|
|
|
for _, n1 := range n.List.Slice() {
|
2015-02-13 14:40:36 -05:00
|
|
|
buf = fmt.Sprintf("a%d", num)
|
|
|
|
|
num++
|
2017-03-28 15:13:19 -07:00
|
|
|
a = namedfield(buf, n1.Type)
|
2016-03-08 15:10:26 -08:00
|
|
|
t.List.Append(a)
|
2016-03-03 20:48:00 -08:00
|
|
|
printargs = append(printargs, a.Left)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-02-23 16:07:24 -05:00
|
|
|
oldfn := Curfn
|
2015-02-13 14:40:36 -05:00
|
|
|
Curfn = nil
|
2017-04-10 13:03:14 -07:00
|
|
|
|
|
|
|
|
walkprintfunc_prgen++
|
|
|
|
|
sym := lookupN("print·%d", walkprintfunc_prgen)
|
|
|
|
|
fn := dclfunc(sym, t)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(n.Op, nil, nil)
|
2016-03-08 15:10:26 -08:00
|
|
|
a.List.Set(printargs)
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
a = typecheck(a, Etop)
|
|
|
|
|
a = walkstmt(a)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-03-10 10:13:42 -08:00
|
|
|
fn.Nbody.Set1(a)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2017-04-07 19:50:31 +00:00
|
|
|
funcbody(fn)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
fn = typecheck(fn, Etop)
|
2016-03-19 17:02:01 -07:00
|
|
|
typecheckslice(fn.Nbody.Slice(), Etop)
|
2016-03-09 20:29:21 -08:00
|
|
|
xtop = append(xtop, fn)
|
2015-02-13 14:40:36 -05:00
|
|
|
Curfn = oldfn
|
|
|
|
|
|
2016-09-16 11:00:54 +10:00
|
|
|
a = nod(OCALL, nil, nil)
|
2015-05-27 10:42:55 -04:00
|
|
|
a.Left = fn.Func.Nname
|
2016-03-08 15:10:26 -08:00
|
|
|
a.List.Set(n.List.Slice())
|
cmd/compile: reduce use of **Node parameters
Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.
This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.
Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.
This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39
For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.
Passes toolstash -cmp.
name old time/op new time/op delta
Template 335ms ± 5% 324ms ± 5% -3.35% (p=0.000 n=23+24)
Unicode 176ms ± 9% 165ms ± 6% -6.12% (p=0.000 n=23+24)
GoTypes 1.10s ± 4% 1.07s ± 2% -2.77% (p=0.000 n=24+24)
Compiler 5.31s ± 3% 5.15s ± 3% -2.95% (p=0.000 n=24+24)
MakeBash 41.6s ± 1% 41.7s ± 2% ~ (p=0.586 n=23+23)
name old alloc/op new alloc/op delta
Template 63.3MB ± 0% 62.4MB ± 0% -1.36% (p=0.000 n=25+23)
Unicode 42.4MB ± 0% 41.6MB ± 0% -1.99% (p=0.000 n=24+25)
GoTypes 220MB ± 0% 217MB ± 0% -1.11% (p=0.000 n=25+25)
Compiler 994MB ± 0% 973MB ± 0% -2.08% (p=0.000 n=24+25)
name old allocs/op new allocs/op delta
Template 681k ± 0% 574k ± 0% -15.71% (p=0.000 n=24+25)
Unicode 518k ± 0% 413k ± 0% -20.34% (p=0.000 n=25+24)
GoTypes 2.08M ± 0% 1.78M ± 0% -14.62% (p=0.000 n=25+25)
Compiler 9.26M ± 0% 7.64M ± 0% -17.48% (p=0.000 n=25+25)
name old text-bytes new text-bytes delta
HelloSize 578k ± 0% 578k ± 0% ~ (all samples are equal)
CmdGoSize 6.46M ± 0% 6.46M ± 0% ~ (all samples are equal)
name old data-bytes new data-bytes delta
HelloSize 128k ± 0% 128k ± 0% ~ (all samples are equal)
CmdGoSize 281k ± 0% 281k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 921k ± 0% 921k ± 0% ~ (all samples are equal)
CmdGoSize 9.86M ± 0% 9.86M ± 0% ~ (all samples are equal)
Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-03-20 08:03:31 -07:00
|
|
|
a = typecheck(a, Etop)
|
|
|
|
|
a = walkexpr(a, init)
|
|
|
|
|
return a
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
|
|
|
|
|
// substArgTypes substitutes the given list of types for
|
|
|
|
|
// successive occurrences of the "any" placeholder in the
|
|
|
|
|
// type syntax expression n.Type.
|
|
|
|
|
// The result of substArgTypes MUST be assigned back to old, e.g.
|
|
|
|
|
// n.Left = substArgTypes(n.Left, t1, t2)
|
|
|
|
|
func substArgTypes(old *Node, types_ ...*types.Type) *Node {
|
|
|
|
|
n := *old // make shallow copy
|
|
|
|
|
|
|
|
|
|
for _, t := range types_ {
|
|
|
|
|
dowidth(t)
|
|
|
|
|
}
|
|
|
|
|
n.Type = types.SubstAny(n.Type, &types_)
|
|
|
|
|
if len(types_) > 0 {
|
|
|
|
|
Fatalf("substArgTypes: too many argument types")
|
|
|
|
|
}
|
|
|
|
|
return &n
|
|
|
|
|
}
|