go/src/cmd/compile/internal/gc/const.go

1653 lines
31 KiB
Go
Raw Normal View History

// 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/internal/types"
cmd/compile: fix misleading "truncated to int" messages When defining an int const, the compiler tries to cast the RHS expression to int. The cast may fail for three reasons: 1. expr is an integer constant that overflows int 2. expr is a floating point constant 3. expr is a complex constant, or not a number In the second case, in order to print a sensible error message, we must distinguish between a floating point constant that should be included in the error message and a floating point constant that cannot be reasonably formatted for inclusion in an error message. For example, in: const a int = 1.1 const b int = 1 + 1e-100 a is in the former group, while b is in the latter, since the floating point value resulting from the evaluation of the rhs of the assignment (1.00...01) is too long to be fully printed in an error message, and cannot be shortened without making the error message misleading (rounding or truncating it would result in a "1", which looks like an integer constant, and it makes little sense in an error message about an invalid floating point expression). To fix this problem, we try to format the float value using fconv (which is used by the error reporting mechanism to format float arguments), and then parse the resulting string back to a big.Float. If the result is an integer, we assume that expr is a float value that cannot be reasonably be formatted as a string, and we emit an error message that does not include its string representation. Also, change the error message for overflows to a more conservative "integer too large", which does not mention overflows that are only caused by an internal implementation restriction. Also, change (*Mpint) SetFloat so that it returns a bool (instead of 0/-1 for success/failure). Fixes #11371 Change-Id: Ibbc73e2ed2eaf41f07827b0649d0eb637150ecaa Reviewed-on: https://go-review.googlesource.com/35411 Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com> Run-TryBot: Robert Griesemer <gri@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
2017-01-18 11:48:14 +01:00
"math/big"
"strings"
)
// Ctype describes the constant kind of an "ideal" (untyped) constant.
type Ctype uint8
const (
CTxxx Ctype = iota
CTINT
CTRUNE
CTFLT
CTCPLX
CTSTR
CTBOOL
CTNIL
)
type Val struct {
// U contains one of:
// bool bool when n.ValCtype() == CTBOOL
// *Mpint int when n.ValCtype() == CTINT, rune when n.ValCtype() == CTRUNE
// *Mpflt float when n.ValCtype() == CTFLT
// *Mpcplx pair of floats when n.ValCtype() == CTCPLX
// string string when n.ValCtype() == CTSTR
// *Nilval when n.ValCtype() == CTNIL
U interface{}
}
func (v Val) Ctype() Ctype {
switch x := v.U.(type) {
default:
Fatalf("unexpected Ctype for %T", v.U)
panic("not reached")
case nil:
return 0
case *NilVal:
return CTNIL
case bool:
return CTBOOL
case *Mpint:
if x.Rune {
return CTRUNE
}
return CTINT
case *Mpflt:
return CTFLT
case *Mpcplx:
return CTCPLX
case string:
return CTSTR
}
}
func eqval(a, b Val) bool {
if a.Ctype() != b.Ctype() {
return false
}
switch x := a.U.(type) {
default:
Fatalf("unexpected Ctype for %T", a.U)
panic("not reached")
case *NilVal:
return true
case bool:
y := b.U.(bool)
return x == y
case *Mpint:
y := b.U.(*Mpint)
return x.Cmp(y) == 0
case *Mpflt:
y := b.U.(*Mpflt)
return x.Cmp(y) == 0
case *Mpcplx:
y := b.U.(*Mpcplx)
return x.Real.Cmp(&y.Real) == 0 && x.Imag.Cmp(&y.Imag) == 0
case string:
y := b.U.(string)
return x == y
}
}
// Interface returns the constant value stored in v as an interface{}.
// It returns int64s for ints and runes, float64s for floats,
// complex128s for complex values, and nil for constant nils.
func (v Val) Interface() interface{} {
switch x := v.U.(type) {
default:
Fatalf("unexpected Interface for %T", v.U)
panic("not reached")
case *NilVal:
return nil
case bool, string:
return x
case *Mpint:
return x.Int64()
case *Mpflt:
return x.Float64()
case *Mpcplx:
return complex(x.Real.Float64(), x.Imag.Float64())
}
}
type NilVal struct{}
// Int64 returns n as an int64.
// n must be an integer or rune constant.
func (n *Node) Int64() int64 {
if !Isconst(n, CTINT) {
Fatalf("Int64(%v)", n)
}
return n.Val().U.(*Mpint).Int64()
}
// CanInt64 reports whether it is safe to call Int64() on n.
func (n *Node) CanInt64() bool {
if !Isconst(n, CTINT) {
return false
}
// if the value inside n cannot be represented as an int64, the
// return value of Int64 is undefined
return n.Val().U.(*Mpint).CmpInt64(n.Int64()) == 0
}
// Bool returns n as a bool.
// n must be a boolean constant.
func (n *Node) Bool() bool {
if !Isconst(n, CTBOOL) {
Fatalf("Bool(%v)", n)
}
return n.Val().U.(bool)
}
// truncate float literal fv to 32-bit or 64-bit precision
// according to type; return truncated value.
func truncfltlit(oldv *Mpflt, t *types.Type) *Mpflt {
if t == nil {
return oldv
}
if overflow(Val{oldv}, t) {
// If there was overflow, simply continuing would set the
// value to Inf which in turn would lead to spurious follow-on
// errors. Avoid this by returning the existing value.
return oldv
}
fv := newMpflt()
// convert large precision literal floating
// into limited precision (float64 or float32)
switch t.Etype {
case types.TFLOAT32:
fv.SetFloat64(oldv.Float32())
case types.TFLOAT64:
fv.SetFloat64(oldv.Float64())
default:
Fatalf("truncfltlit: unexpected Etype %v", t.Etype)
}
return fv
}
// truncate Real and Imag parts of Mpcplx to 32-bit or 64-bit
// precision, according to type; return truncated value. In case of
// overflow, calls yyerror but does not truncate the input value.
func trunccmplxlit(oldv *Mpcplx, t *types.Type) *Mpcplx {
if t == nil {
return oldv
}
if overflow(Val{oldv}, t) {
// If there was overflow, simply continuing would set the
// value to Inf which in turn would lead to spurious follow-on
// errors. Avoid this by returning the existing value.
return oldv
}
cv := newMpcmplx()
switch t.Etype {
case types.TCOMPLEX64:
cv.Real.SetFloat64(oldv.Real.Float32())
cv.Imag.SetFloat64(oldv.Imag.Float32())
case types.TCOMPLEX128:
cv.Real.SetFloat64(oldv.Real.Float64())
cv.Imag.SetFloat64(oldv.Imag.Float64())
default:
Fatalf("trunccplxlit: unexpected Etype %v", t.Etype)
}
return cv
}
2016-03-17 12:24:11 -07:00
// canReuseNode indicates whether it is known to be safe
// to reuse a Node.
type canReuseNode bool
const (
noReuse canReuseNode = false // not necessarily safe to reuse
reuseOK canReuseNode = true // safe to reuse
)
// convert n, if literal, to type t.
// implicit conversion.
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 convlit MUST be assigned back to n, e.g.
// n.Left = convlit(n.Left, t)
func convlit(n *Node, t *types.Type) *Node {
2016-03-17 12:24:11 -07:00
return convlit1(n, t, false, noReuse)
}
2016-03-17 12:24:11 -07:00
// convlit1 converts n, if literal, to type t.
// It returns a new node if necessary.
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 convlit1 MUST be assigned back to n, e.g.
2016-03-17 12:24:11 -07:00
// n.Left = convlit1(n.Left, t, explicit, reuse)
func convlit1(n *Node, t *types.Type, explicit bool, reuse canReuseNode) *Node {
if n == nil || t == nil || n.Type == nil || t.IsUntyped() || 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
return n
}
if !explicit && !n.Type.IsUntyped() {
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
}
2016-03-17 12:24:11 -07:00
if n.Op == OLITERAL && !reuse {
// Can't always set n.Type directly on OLITERAL nodes.
// See discussion on CL 20813.
n = n.copy()
2016-03-17 12:24:11 -07:00
reuse = true
}
switch n.Op {
default:
if n.Type == types.Idealbool {
if !t.IsBoolean() {
t = types.Types[TBOOL]
}
switch n.Op {
case ONOT:
n.Left = convlit(n.Left, t)
case OANDAND, OOROR:
n.Left = convlit(n.Left, t)
n.Right = convlit(n.Right, t)
}
n.Type = t
}
if n.Type.IsUntyped() {
if t.IsInterface() {
n.Left, n.Right = defaultlit2(n.Left, n.Right, true)
n.Type = n.Left.Type // same as n.Right.Type per defaultlit2
} else {
n.Left = convlit(n.Left, t)
n.Right = convlit(n.Right, t)
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
return n
// target is invalid type for a constant? leave alone.
case OLITERAL:
if !okforconst[t.Etype] && n.Type.Etype != TNIL {
2016-03-17 12:24:11 -07:00
return defaultlitreuse(n, nil, reuse)
}
case OLSH, ORSH:
n.Left = convlit1(n.Left, t, explicit && n.Left.Type.IsUntyped(), noReuse)
t = n.Left.Type
if t != nil && t.Etype == TIDEAL && n.Val().Ctype() != CTINT {
n.SetVal(toint(n.Val()))
}
if t != nil && !t.IsInteger() {
yyerror("invalid operation: %v (shift of type %v)", n, t)
t = nil
}
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
return n
case OCOMPLEX:
if n.Type.Etype == TIDEAL {
switch t.Etype {
default:
// If trying to convert to non-complex type,
// leave as complex128 and let typechecker complain.
t = types.Types[TCOMPLEX128]
fallthrough
case types.TCOMPLEX128:
n.Type = t
n.Left = convlit(n.Left, types.Types[TFLOAT64])
n.Right = convlit(n.Right, types.Types[TFLOAT64])
case TCOMPLEX64:
n.Type = t
n.Left = convlit(n.Left, types.Types[TFLOAT32])
n.Right = convlit(n.Right, types.Types[TFLOAT32])
}
}
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
}
// avoid repeated calculations, errors
if eqtype(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
return n
}
ct := consttype(n)
var et types.EType
if ct == 0 {
goto bad
}
et = t.Etype
if et == TINTER {
if ct == CTNIL && n.Type == types.Types[TNIL] {
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
return n
}
2016-03-17 12:24:11 -07:00
return defaultlitreuse(n, nil, reuse)
}
switch ct {
default:
goto bad
case CTNIL:
switch et {
default:
n.Type = nil
goto bad
// let normal conversion code handle it
case TSTRING:
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
case TARRAY:
goto bad
case TPTR32, TPTR64, TUNSAFEPTR:
n.SetVal(Val{new(Mpint)})
case TCHAN, TFUNC, TINTER, TMAP, TSLICE:
break
}
case CTSTR, CTBOOL:
if et != n.Type.Etype {
goto bad
}
case CTINT, CTRUNE, CTFLT, CTCPLX:
if n.Type.Etype == TUNSAFEPTR && t.Etype != TUINTPTR {
goto bad
}
ct := n.Val().Ctype()
if isInt[et] {
switch ct {
default:
goto bad
case CTCPLX, CTFLT, CTRUNE:
n.SetVal(toint(n.Val()))
fallthrough
case CTINT:
overflow(n.Val(), t)
}
} else if isFloat[et] {
switch ct {
default:
goto bad
case CTCPLX, CTINT, CTRUNE:
n.SetVal(toflt(n.Val()))
fallthrough
case CTFLT:
n.SetVal(Val{truncfltlit(n.Val().U.(*Mpflt), t)})
}
} else if isComplex[et] {
switch ct {
default:
goto bad
case CTFLT, CTINT, CTRUNE:
n.SetVal(tocplx(n.Val()))
fallthrough
case CTCPLX:
n.SetVal(Val{trunccmplxlit(n.Val().U.(*Mpcplx), t)})
}
} else if et == types.TSTRING && (ct == CTINT || ct == CTRUNE) && explicit {
n.SetVal(tostr(n.Val()))
} else {
goto bad
}
}
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
return n
bad:
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if !n.Diag() {
if !t.Broke() {
yyerror("cannot convert %L to type %v", n, t)
}
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
n.SetDiag(true)
}
if n.Type.IsUntyped() {
2016-03-17 12:24:11 -07:00
n = defaultlitreuse(n, nil, reuse)
}
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
}
func copyval(v Val) Val {
switch u := v.U.(type) {
case *Mpint:
i := new(Mpint)
i.Set(u)
i.Rune = u.Rune
v.U = i
case *Mpflt:
f := newMpflt()
f.Set(u)
v.U = f
case *Mpcplx:
c := new(Mpcplx)
c.Real.Set(&u.Real)
c.Imag.Set(&u.Imag)
v.U = c
}
return v
}
func tocplx(v Val) Val {
switch u := v.U.(type) {
case *Mpint:
c := new(Mpcplx)
c.Real.SetInt(u)
c.Imag.SetFloat64(0.0)
v.U = c
case *Mpflt:
c := new(Mpcplx)
c.Real.Set(u)
c.Imag.SetFloat64(0.0)
v.U = c
}
return v
}
func toflt(v Val) Val {
switch u := v.U.(type) {
case *Mpint:
f := newMpflt()
f.SetInt(u)
v.U = f
case *Mpcplx:
f := newMpflt()
f.Set(&u.Real)
if u.Imag.CmpFloat64(0) != 0 {
yyerror("constant %v truncated to real", u.GoString())
}
v.U = f
}
return v
}
func toint(v Val) Val {
switch u := v.U.(type) {
case *Mpint:
if u.Rune {
i := new(Mpint)
i.Set(u)
v.U = i
}
case *Mpflt:
i := new(Mpint)
cmd/compile: fix misleading "truncated to int" messages When defining an int const, the compiler tries to cast the RHS expression to int. The cast may fail for three reasons: 1. expr is an integer constant that overflows int 2. expr is a floating point constant 3. expr is a complex constant, or not a number In the second case, in order to print a sensible error message, we must distinguish between a floating point constant that should be included in the error message and a floating point constant that cannot be reasonably formatted for inclusion in an error message. For example, in: const a int = 1.1 const b int = 1 + 1e-100 a is in the former group, while b is in the latter, since the floating point value resulting from the evaluation of the rhs of the assignment (1.00...01) is too long to be fully printed in an error message, and cannot be shortened without making the error message misleading (rounding or truncating it would result in a "1", which looks like an integer constant, and it makes little sense in an error message about an invalid floating point expression). To fix this problem, we try to format the float value using fconv (which is used by the error reporting mechanism to format float arguments), and then parse the resulting string back to a big.Float. If the result is an integer, we assume that expr is a float value that cannot be reasonably be formatted as a string, and we emit an error message that does not include its string representation. Also, change the error message for overflows to a more conservative "integer too large", which does not mention overflows that are only caused by an internal implementation restriction. Also, change (*Mpint) SetFloat so that it returns a bool (instead of 0/-1 for success/failure). Fixes #11371 Change-Id: Ibbc73e2ed2eaf41f07827b0649d0eb637150ecaa Reviewed-on: https://go-review.googlesource.com/35411 Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com> Run-TryBot: Robert Griesemer <gri@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
2017-01-18 11:48:14 +01:00
if !i.SetFloat(u) {
if i.checkOverflow(0) {
cmd/compile: fix misleading "truncated to int" messages When defining an int const, the compiler tries to cast the RHS expression to int. The cast may fail for three reasons: 1. expr is an integer constant that overflows int 2. expr is a floating point constant 3. expr is a complex constant, or not a number In the second case, in order to print a sensible error message, we must distinguish between a floating point constant that should be included in the error message and a floating point constant that cannot be reasonably formatted for inclusion in an error message. For example, in: const a int = 1.1 const b int = 1 + 1e-100 a is in the former group, while b is in the latter, since the floating point value resulting from the evaluation of the rhs of the assignment (1.00...01) is too long to be fully printed in an error message, and cannot be shortened without making the error message misleading (rounding or truncating it would result in a "1", which looks like an integer constant, and it makes little sense in an error message about an invalid floating point expression). To fix this problem, we try to format the float value using fconv (which is used by the error reporting mechanism to format float arguments), and then parse the resulting string back to a big.Float. If the result is an integer, we assume that expr is a float value that cannot be reasonably be formatted as a string, and we emit an error message that does not include its string representation. Also, change the error message for overflows to a more conservative "integer too large", which does not mention overflows that are only caused by an internal implementation restriction. Also, change (*Mpint) SetFloat so that it returns a bool (instead of 0/-1 for success/failure). Fixes #11371 Change-Id: Ibbc73e2ed2eaf41f07827b0649d0eb637150ecaa Reviewed-on: https://go-review.googlesource.com/35411 Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com> Run-TryBot: Robert Griesemer <gri@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
2017-01-18 11:48:14 +01:00
yyerror("integer too large")
} else {
// The value of u cannot be represented as an integer;
// so we need to print an error message.
// Unfortunately some float values cannot be
// reasonably formatted for inclusion in an error
// message (example: 1 + 1e-100), so first we try to
// format the float; if the truncation resulted in
// something that looks like an integer we omit the
// value from the error message.
// (See issue #11371).
var t big.Float
t.Parse(u.GoString(), 10)
cmd/compile: fix misleading "truncated to int" messages When defining an int const, the compiler tries to cast the RHS expression to int. The cast may fail for three reasons: 1. expr is an integer constant that overflows int 2. expr is a floating point constant 3. expr is a complex constant, or not a number In the second case, in order to print a sensible error message, we must distinguish between a floating point constant that should be included in the error message and a floating point constant that cannot be reasonably formatted for inclusion in an error message. For example, in: const a int = 1.1 const b int = 1 + 1e-100 a is in the former group, while b is in the latter, since the floating point value resulting from the evaluation of the rhs of the assignment (1.00...01) is too long to be fully printed in an error message, and cannot be shortened without making the error message misleading (rounding or truncating it would result in a "1", which looks like an integer constant, and it makes little sense in an error message about an invalid floating point expression). To fix this problem, we try to format the float value using fconv (which is used by the error reporting mechanism to format float arguments), and then parse the resulting string back to a big.Float. If the result is an integer, we assume that expr is a float value that cannot be reasonably be formatted as a string, and we emit an error message that does not include its string representation. Also, change the error message for overflows to a more conservative "integer too large", which does not mention overflows that are only caused by an internal implementation restriction. Also, change (*Mpint) SetFloat so that it returns a bool (instead of 0/-1 for success/failure). Fixes #11371 Change-Id: Ibbc73e2ed2eaf41f07827b0649d0eb637150ecaa Reviewed-on: https://go-review.googlesource.com/35411 Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com> Run-TryBot: Robert Griesemer <gri@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
2017-01-18 11:48:14 +01:00
if t.IsInt() {
yyerror("constant truncated to integer")
} else {
yyerror("constant %v truncated to integer", u.GoString())
cmd/compile: fix misleading "truncated to int" messages When defining an int const, the compiler tries to cast the RHS expression to int. The cast may fail for three reasons: 1. expr is an integer constant that overflows int 2. expr is a floating point constant 3. expr is a complex constant, or not a number In the second case, in order to print a sensible error message, we must distinguish between a floating point constant that should be included in the error message and a floating point constant that cannot be reasonably formatted for inclusion in an error message. For example, in: const a int = 1.1 const b int = 1 + 1e-100 a is in the former group, while b is in the latter, since the floating point value resulting from the evaluation of the rhs of the assignment (1.00...01) is too long to be fully printed in an error message, and cannot be shortened without making the error message misleading (rounding or truncating it would result in a "1", which looks like an integer constant, and it makes little sense in an error message about an invalid floating point expression). To fix this problem, we try to format the float value using fconv (which is used by the error reporting mechanism to format float arguments), and then parse the resulting string back to a big.Float. If the result is an integer, we assume that expr is a float value that cannot be reasonably be formatted as a string, and we emit an error message that does not include its string representation. Also, change the error message for overflows to a more conservative "integer too large", which does not mention overflows that are only caused by an internal implementation restriction. Also, change (*Mpint) SetFloat so that it returns a bool (instead of 0/-1 for success/failure). Fixes #11371 Change-Id: Ibbc73e2ed2eaf41f07827b0649d0eb637150ecaa Reviewed-on: https://go-review.googlesource.com/35411 Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com> Run-TryBot: Robert Griesemer <gri@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
2017-01-18 11:48:14 +01:00
}
}
}
v.U = i
case *Mpcplx:
i := new(Mpint)
cmd/compile: fix misleading "truncated to int" messages When defining an int const, the compiler tries to cast the RHS expression to int. The cast may fail for three reasons: 1. expr is an integer constant that overflows int 2. expr is a floating point constant 3. expr is a complex constant, or not a number In the second case, in order to print a sensible error message, we must distinguish between a floating point constant that should be included in the error message and a floating point constant that cannot be reasonably formatted for inclusion in an error message. For example, in: const a int = 1.1 const b int = 1 + 1e-100 a is in the former group, while b is in the latter, since the floating point value resulting from the evaluation of the rhs of the assignment (1.00...01) is too long to be fully printed in an error message, and cannot be shortened without making the error message misleading (rounding or truncating it would result in a "1", which looks like an integer constant, and it makes little sense in an error message about an invalid floating point expression). To fix this problem, we try to format the float value using fconv (which is used by the error reporting mechanism to format float arguments), and then parse the resulting string back to a big.Float. If the result is an integer, we assume that expr is a float value that cannot be reasonably be formatted as a string, and we emit an error message that does not include its string representation. Also, change the error message for overflows to a more conservative "integer too large", which does not mention overflows that are only caused by an internal implementation restriction. Also, change (*Mpint) SetFloat so that it returns a bool (instead of 0/-1 for success/failure). Fixes #11371 Change-Id: Ibbc73e2ed2eaf41f07827b0649d0eb637150ecaa Reviewed-on: https://go-review.googlesource.com/35411 Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com> Run-TryBot: Robert Griesemer <gri@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
2017-01-18 11:48:14 +01:00
if !i.SetFloat(&u.Real) || u.Imag.CmpFloat64(0) != 0 {
yyerror("constant %v truncated to integer", u.GoString())
}
v.U = i
}
return v
}
func doesoverflow(v Val, t *types.Type) bool {
switch u := v.U.(type) {
case *Mpint:
if !t.IsInteger() {
Fatalf("overflow: %v integer constant", t)
}
return u.Cmp(minintval[t.Etype]) < 0 || u.Cmp(maxintval[t.Etype]) > 0
case *Mpflt:
if !t.IsFloat() {
Fatalf("overflow: %v floating-point constant", t)
}
return u.Cmp(minfltval[t.Etype]) <= 0 || u.Cmp(maxfltval[t.Etype]) >= 0
case *Mpcplx:
if !t.IsComplex() {
Fatalf("overflow: %v complex constant", t)
}
return u.Real.Cmp(minfltval[t.Etype]) <= 0 || u.Real.Cmp(maxfltval[t.Etype]) >= 0 ||
u.Imag.Cmp(minfltval[t.Etype]) <= 0 || u.Imag.Cmp(maxfltval[t.Etype]) >= 0
}
return false
}
func overflow(v Val, t *types.Type) bool {
// v has already been converted
// to appropriate form for t.
if t == nil || t.Etype == TIDEAL {
return false
}
// Only uintptrs may be converted to pointers, which cannot overflow.
if t.IsPtr() || t.IsUnsafePtr() {
return false
}
if doesoverflow(v, t) {
yyerror("constant %v overflows %v", v, t)
return true
}
return false
}
func tostr(v Val) Val {
switch u := v.U.(type) {
case *Mpint:
var i int64 = 0xFFFD
if u.Cmp(minintval[TUINT32]) >= 0 && u.Cmp(maxintval[TUINT32]) <= 0 {
i = u.Int64()
}
v.U = string(i)
case *NilVal:
// Can happen because of string([]byte(nil)).
v.U = ""
}
return v
}
func consttype(n *Node) Ctype {
if n == nil || n.Op != OLITERAL {
return 0
}
return n.Val().Ctype()
}
func Isconst(n *Node, ct Ctype) bool {
t := consttype(n)
// If the caller is asking for CTINT, allow CTRUNE too.
// Makes life easier for back ends.
return t == ct || (ct == CTINT && t == CTRUNE)
}
// if n is constant, rewrite as OLITERAL node.
func evconst(n *Node) {
// pick off just the opcodes that can be
// constant evaluated.
switch n.Op {
default:
return
case OADD,
OAND,
OANDAND,
OANDNOT,
OARRAYBYTESTR,
OCOM,
ODIV,
OEQ,
OGE,
OGT,
OLE,
OLSH,
OLT,
OMINUS,
OMOD,
OMUL,
ONE,
ONOT,
OOR,
OOROR,
OPLUS,
ORSH,
OSUB,
OXOR:
break
case OCONV:
if n.Type == nil {
return
}
if !okforconst[n.Type.Etype] && n.Type.Etype != TNIL {
return
}
// merge adjacent constants in the argument list.
case OADDSTR:
s := n.List.Slice()
for i1 := 0; i1 < len(s); i1++ {
if Isconst(s[i1], CTSTR) && i1+1 < len(s) && Isconst(s[i1+1], CTSTR) {
// merge from i1 up to but not including i2
var strs []string
i2 := i1
for i2 < len(s) && Isconst(s[i2], CTSTR) {
strs = append(strs, s[i2].Val().U.(string))
i2++
}
nl := *s[i1]
nl.Orig = &nl
nl.SetVal(Val{strings.Join(strs, "")})
s[i1] = &nl
s = append(s[:i1+1], s[i2:]...)
}
}
if len(s) == 1 && Isconst(s[0], CTSTR) {
n.Op = OLITERAL
n.SetVal(s[0].Val())
} else {
n.List.Set(s)
}
return
}
nl := n.Left
if nl == nil || nl.Type == nil {
return
}
if consttype(nl) == 0 {
return
}
wl := nl.Type.Etype
if isInt[wl] || isFloat[wl] || isComplex[wl] {
wl = TIDEAL
}
// avoid constant conversions in switches below
const (
CTINT_ = uint32(CTINT)
CTRUNE_ = uint32(CTRUNE)
CTFLT_ = uint32(CTFLT)
CTCPLX_ = uint32(CTCPLX)
CTSTR_ = uint32(CTSTR)
CTBOOL_ = uint32(CTBOOL)
CTNIL_ = uint32(CTNIL)
OCONV_ = uint32(OCONV) << 16
OARRAYBYTESTR_ = uint32(OARRAYBYTESTR) << 16
OPLUS_ = uint32(OPLUS) << 16
OMINUS_ = uint32(OMINUS) << 16
OCOM_ = uint32(OCOM) << 16
ONOT_ = uint32(ONOT) << 16
OLSH_ = uint32(OLSH) << 16
ORSH_ = uint32(ORSH) << 16
OADD_ = uint32(OADD) << 16
OSUB_ = uint32(OSUB) << 16
OMUL_ = uint32(OMUL) << 16
ODIV_ = uint32(ODIV) << 16
OMOD_ = uint32(OMOD) << 16
OOR_ = uint32(OOR) << 16
OAND_ = uint32(OAND) << 16
OANDNOT_ = uint32(OANDNOT) << 16
OXOR_ = uint32(OXOR) << 16
OEQ_ = uint32(OEQ) << 16
ONE_ = uint32(ONE) << 16
OLT_ = uint32(OLT) << 16
OLE_ = uint32(OLE) << 16
OGE_ = uint32(OGE) << 16
OGT_ = uint32(OGT) << 16
OOROR_ = uint32(OOROR) << 16
OANDAND_ = uint32(OANDAND) << 16
)
nr := n.Right
var rv Val
var wr types.EType
var ctype uint32
var v Val
if nr == nil {
// copy numeric value to avoid modifying
// nl, in case someone still refers to it (e.g. iota).
v = copyval(nl.Val())
// rune values are int values for the purpose of constant folding.
ctype = uint32(v.Ctype())
if ctype == CTRUNE_ {
ctype = CTINT_
}
switch uint32(n.Op)<<16 | ctype {
default:
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if !n.Diag() {
yyerror("illegal constant expression %v %v", n.Op, nl.Type)
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
n.SetDiag(true)
}
return
case OCONV_ | CTNIL_,
OARRAYBYTESTR_ | CTNIL_:
if n.Type.IsString() {
v = tostr(v)
nl.Type = n.Type
break
}
fallthrough
case OCONV_ | CTINT_,
OCONV_ | CTFLT_,
OCONV_ | CTCPLX_,
OCONV_ | CTSTR_,
OCONV_ | CTBOOL_:
2016-03-17 12:24:11 -07:00
nl = convlit1(nl, n.Type, true, false)
v = nl.Val()
case OPLUS_ | CTINT_:
break
case OMINUS_ | CTINT_:
v.U.(*Mpint).Neg()
case OCOM_ | CTINT_:
et := Txxx
if nl.Type != nil {
et = nl.Type.Etype
}
// calculate the mask in b
// result will be (a ^ mask)
var b Mpint
switch et {
// signed guys change sign
default:
b.SetInt64(-1)
// unsigned guys invert their bits
case TUINT8,
TUINT16,
TUINT32,
TUINT64,
TUINT,
TUINTPTR:
b.Set(maxintval[et])
}
v.U.(*Mpint).Xor(&b)
case OPLUS_ | CTFLT_:
break
case OMINUS_ | CTFLT_:
v.U.(*Mpflt).Neg()
case OPLUS_ | CTCPLX_:
break
case OMINUS_ | CTCPLX_:
v.U.(*Mpcplx).Real.Neg()
v.U.(*Mpcplx).Imag.Neg()
case ONOT_ | CTBOOL_:
if !v.U.(bool) {
goto settrue
}
goto setfalse
}
goto ret
}
if nr.Type == nil {
return
}
if consttype(nr) == 0 {
return
}
wr = nr.Type.Etype
if isInt[wr] || isFloat[wr] || isComplex[wr] {
wr = TIDEAL
}
// check for compatible general types (numeric, string, etc)
if wl != wr {
if wl == TINTER || wr == TINTER {
if n.Op == ONE {
goto settrue
}
goto setfalse
}
goto illegal
}
// check for compatible types.
switch n.Op {
// ideal const mixes with anything but otherwise must match.
default:
if nl.Type.Etype != TIDEAL {
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
nr = defaultlit(nr, nl.Type)
n.Right = nr
}
if nr.Type.Etype != TIDEAL {
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
nl = defaultlit(nl, nr.Type)
n.Left = nl
}
if nl.Type.Etype != nr.Type.Etype {
goto illegal
}
// right must be unsigned.
// left can be ideal.
case OLSH, ORSH:
nr = defaultlit(nr, types.Types[TUINT])
n.Right = nr
if nr.Type != nil && (nr.Type.IsSigned() || !nr.Type.IsInteger()) {
goto illegal
}
if nl.Val().Ctype() != CTRUNE {
nl.SetVal(toint(nl.Val()))
}
nr.SetVal(toint(nr.Val()))
}
// copy numeric value to avoid modifying
// n->left, in case someone still refers to it (e.g. iota).
v = copyval(nl.Val())
rv = nr.Val()
// convert to common ideal
if v.Ctype() == CTCPLX || rv.Ctype() == CTCPLX {
v = tocplx(v)
rv = tocplx(rv)
}
if v.Ctype() == CTFLT || rv.Ctype() == CTFLT {
v = toflt(v)
rv = toflt(rv)
}
// Rune and int turns into rune.
if v.Ctype() == CTRUNE && rv.Ctype() == CTINT {
i := new(Mpint)
i.Set(rv.U.(*Mpint))
i.Rune = true
rv.U = i
}
if v.Ctype() == CTINT && rv.Ctype() == CTRUNE {
if n.Op == OLSH || n.Op == ORSH {
i := new(Mpint)
i.Set(rv.U.(*Mpint))
rv.U = i
} else {
i := new(Mpint)
i.Set(v.U.(*Mpint))
i.Rune = true
v.U = i
}
}
if v.Ctype() != rv.Ctype() {
// Use of undefined name as constant?
if (v.Ctype() == 0 || rv.Ctype() == 0) && nerrors > 0 {
return
}
Fatalf("constant type mismatch %v(%d) %v(%d)", nl.Type, v.Ctype(), nr.Type, rv.Ctype())
}
// rune values are int values for the purpose of constant folding.
ctype = uint32(v.Ctype())
if ctype == CTRUNE_ {
ctype = CTINT_
}
// run op
switch uint32(n.Op)<<16 | ctype {
default:
goto illegal
case OADD_ | CTINT_:
v.U.(*Mpint).Add(rv.U.(*Mpint))
case OSUB_ | CTINT_:
v.U.(*Mpint).Sub(rv.U.(*Mpint))
case OMUL_ | CTINT_:
v.U.(*Mpint).Mul(rv.U.(*Mpint))
case ODIV_ | CTINT_:
if rv.U.(*Mpint).CmpInt64(0) == 0 {
yyerror("division by zero")
v.U.(*Mpint).SetOverflow()
break
}
v.U.(*Mpint).Quo(rv.U.(*Mpint))
case OMOD_ | CTINT_:
if rv.U.(*Mpint).CmpInt64(0) == 0 {
yyerror("division by zero")
v.U.(*Mpint).SetOverflow()
break
}
v.U.(*Mpint).Rem(rv.U.(*Mpint))
case OLSH_ | CTINT_:
v.U.(*Mpint).Lsh(rv.U.(*Mpint))
case ORSH_ | CTINT_:
v.U.(*Mpint).Rsh(rv.U.(*Mpint))
case OOR_ | CTINT_:
v.U.(*Mpint).Or(rv.U.(*Mpint))
case OAND_ | CTINT_:
v.U.(*Mpint).And(rv.U.(*Mpint))
case OANDNOT_ | CTINT_:
v.U.(*Mpint).AndNot(rv.U.(*Mpint))
case OXOR_ | CTINT_:
v.U.(*Mpint).Xor(rv.U.(*Mpint))
case OADD_ | CTFLT_:
v.U.(*Mpflt).Add(rv.U.(*Mpflt))
case OSUB_ | CTFLT_:
v.U.(*Mpflt).Sub(rv.U.(*Mpflt))
case OMUL_ | CTFLT_:
v.U.(*Mpflt).Mul(rv.U.(*Mpflt))
case ODIV_ | CTFLT_:
if rv.U.(*Mpflt).CmpFloat64(0) == 0 {
yyerror("division by zero")
v.U.(*Mpflt).SetFloat64(1.0)
break
}
v.U.(*Mpflt).Quo(rv.U.(*Mpflt))
// The default case above would print 'ideal % ideal',
// which is not quite an ideal error.
case OMOD_ | CTFLT_:
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if !n.Diag() {
yyerror("illegal constant expression: floating-point %% operation")
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
n.SetDiag(true)
}
return
case OADD_ | CTCPLX_:
v.U.(*Mpcplx).Real.Add(&rv.U.(*Mpcplx).Real)
v.U.(*Mpcplx).Imag.Add(&rv.U.(*Mpcplx).Imag)
case OSUB_ | CTCPLX_:
v.U.(*Mpcplx).Real.Sub(&rv.U.(*Mpcplx).Real)
v.U.(*Mpcplx).Imag.Sub(&rv.U.(*Mpcplx).Imag)
case OMUL_ | CTCPLX_:
v.U.(*Mpcplx).Mul(rv.U.(*Mpcplx))
case ODIV_ | CTCPLX_:
if !v.U.(*Mpcplx).Div(rv.U.(*Mpcplx)) {
yyerror("complex division by zero")
rv.U.(*Mpcplx).Real.SetFloat64(1.0)
rv.U.(*Mpcplx).Imag.SetFloat64(0.0)
break
}
case OEQ_ | CTNIL_:
goto settrue
case ONE_ | CTNIL_:
goto setfalse
case OEQ_ | CTINT_:
if v.U.(*Mpint).Cmp(rv.U.(*Mpint)) == 0 {
goto settrue
}
goto setfalse
case ONE_ | CTINT_:
if v.U.(*Mpint).Cmp(rv.U.(*Mpint)) != 0 {
goto settrue
}
goto setfalse
case OLT_ | CTINT_:
if v.U.(*Mpint).Cmp(rv.U.(*Mpint)) < 0 {
goto settrue
}
goto setfalse
case OLE_ | CTINT_:
if v.U.(*Mpint).Cmp(rv.U.(*Mpint)) <= 0 {
goto settrue
}
goto setfalse
case OGE_ | CTINT_:
if v.U.(*Mpint).Cmp(rv.U.(*Mpint)) >= 0 {
goto settrue
}
goto setfalse
case OGT_ | CTINT_:
if v.U.(*Mpint).Cmp(rv.U.(*Mpint)) > 0 {
goto settrue
}
goto setfalse
case OEQ_ | CTFLT_:
if v.U.(*Mpflt).Cmp(rv.U.(*Mpflt)) == 0 {
goto settrue
}
goto setfalse
case ONE_ | CTFLT_:
if v.U.(*Mpflt).Cmp(rv.U.(*Mpflt)) != 0 {
goto settrue
}
goto setfalse
case OLT_ | CTFLT_:
if v.U.(*Mpflt).Cmp(rv.U.(*Mpflt)) < 0 {
goto settrue
}
goto setfalse
case OLE_ | CTFLT_:
if v.U.(*Mpflt).Cmp(rv.U.(*Mpflt)) <= 0 {
goto settrue
}
goto setfalse
case OGE_ | CTFLT_:
if v.U.(*Mpflt).Cmp(rv.U.(*Mpflt)) >= 0 {
goto settrue
}
goto setfalse
case OGT_ | CTFLT_:
if v.U.(*Mpflt).Cmp(rv.U.(*Mpflt)) > 0 {
goto settrue
}
goto setfalse
case OEQ_ | CTCPLX_:
if v.U.(*Mpcplx).Real.Cmp(&rv.U.(*Mpcplx).Real) == 0 && v.U.(*Mpcplx).Imag.Cmp(&rv.U.(*Mpcplx).Imag) == 0 {
goto settrue
}
goto setfalse
case ONE_ | CTCPLX_:
if v.U.(*Mpcplx).Real.Cmp(&rv.U.(*Mpcplx).Real) != 0 || v.U.(*Mpcplx).Imag.Cmp(&rv.U.(*Mpcplx).Imag) != 0 {
goto settrue
}
goto setfalse
case OEQ_ | CTSTR_:
if strlit(nl) == strlit(nr) {
goto settrue
}
goto setfalse
case ONE_ | CTSTR_:
if strlit(nl) != strlit(nr) {
goto settrue
}
goto setfalse
case OLT_ | CTSTR_:
if strlit(nl) < strlit(nr) {
goto settrue
}
goto setfalse
case OLE_ | CTSTR_:
if strlit(nl) <= strlit(nr) {
goto settrue
}
goto setfalse
case OGE_ | CTSTR_:
if strlit(nl) >= strlit(nr) {
goto settrue
}
goto setfalse
case OGT_ | CTSTR_:
if strlit(nl) > strlit(nr) {
goto settrue
}
goto setfalse
case OOROR_ | CTBOOL_:
if v.U.(bool) || rv.U.(bool) {
goto settrue
}
goto setfalse
case OANDAND_ | CTBOOL_:
if v.U.(bool) && rv.U.(bool) {
goto settrue
}
goto setfalse
case OEQ_ | CTBOOL_:
if v.U.(bool) == rv.U.(bool) {
goto settrue
}
goto setfalse
case ONE_ | CTBOOL_:
if v.U.(bool) != rv.U.(bool) {
goto settrue
}
goto setfalse
}
ret:
setconst(n, v)
return
settrue:
setconst(n, Val{true})
return
setfalse:
setconst(n, Val{false})
return
illegal:
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if !n.Diag() {
yyerror("illegal constant expression: %v %v %v", nl.Type, n.Op, nr.Type)
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
n.SetDiag(true)
}
}
// setconst rewrites n as an OLITERAL with value v.
func setconst(n *Node, v Val) {
// Ensure n.Orig still points to a semantically-equivalent
// expression after we rewrite n into a constant.
if n.Orig == n {
n.Orig = n.copy()
n.Orig.Orig = n.Orig
}
*n = Node{
Op: OLITERAL,
Pos: n.Pos,
Orig: n.Orig,
Type: n.Type,
Xoffset: BADWIDTH,
}
n.SetVal(v)
// Check range.
lno := setlineno(n)
overflow(v, n.Type)
lineno = lno
// Truncate precision for non-ideal float.
if v.Ctype() == CTFLT && n.Type.Etype != TIDEAL {
n.SetVal(Val{truncfltlit(v.U.(*Mpflt), n.Type)})
}
}
func setintconst(n *Node, v int64) {
u := new(Mpint)
u.SetInt64(v)
setconst(n, Val{u})
}
// nodlit returns a new untyped constant with value v.
func nodlit(v Val) *Node {
n := nod(OLITERAL, nil, nil)
n.SetVal(v)
switch v.Ctype() {
default:
Fatalf("nodlit ctype %d", v.Ctype())
case CTSTR:
n.Type = types.Idealstring
case CTBOOL:
n.Type = types.Idealbool
case CTINT, CTRUNE, CTFLT, CTCPLX:
n.Type = types.Types[TIDEAL]
case CTNIL:
n.Type = types.Types[TNIL]
}
return n
}
// idealkind returns a constant kind like consttype
// but for an arbitrary "ideal" (untyped constant) expression.
func idealkind(n *Node) Ctype {
if n == nil || !n.Type.IsUntyped() {
return CTxxx
}
switch n.Op {
default:
return CTxxx
case OLITERAL:
return n.Val().Ctype()
// numeric kinds.
case OADD,
OAND,
OANDNOT,
OCOM,
ODIV,
OMINUS,
OMOD,
OMUL,
OSUB,
OXOR,
OOR,
OPLUS:
k1 := idealkind(n.Left)
k2 := idealkind(n.Right)
if k1 > k2 {
return k1
} else {
return k2
}
case OREAL, OIMAG:
return CTFLT
case OCOMPLEX:
return CTCPLX
case OADDSTR:
return CTSTR
case OANDAND,
OEQ,
OGE,
OGT,
OLE,
OLT,
ONE,
ONOT,
OOROR,
OCMPSTR,
OCMPIFACE:
return CTBOOL
// shifts (beware!).
case OLSH, ORSH:
return idealkind(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
// The result of defaultlit MUST be assigned back to n, e.g.
// n.Left = defaultlit(n.Left, t)
func defaultlit(n *Node, t *types.Type) *Node {
2016-03-17 12:24:11 -07:00
return defaultlitreuse(n, t, noReuse)
}
// The result of defaultlitreuse MUST be assigned back to n, e.g.
// n.Left = defaultlitreuse(n.Left, t, reuse)
func defaultlitreuse(n *Node, t *types.Type, reuse canReuseNode) *Node {
if n == nil || !n.Type.IsUntyped() {
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
}
2016-03-17 12:24:11 -07:00
if n.Op == OLITERAL && !reuse {
n = n.copy()
2016-03-17 12:24:11 -07:00
reuse = true
}
lno := setlineno(n)
ctype := idealkind(n)
var t1 *types.Type
switch ctype {
default:
if t != nil {
return convlit(n, t)
}
switch n.Val().Ctype() {
case CTNIL:
lineno = lno
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if !n.Diag() {
yyerror("use of untyped nil")
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
n.SetDiag(true)
}
n.Type = nil
case CTSTR:
t1 := types.Types[TSTRING]
2016-03-17 12:24:11 -07:00
n = convlit1(n, t1, false, reuse)
default:
yyerror("defaultlit: unknown literal: %v", n)
}
lineno = lno
return n
case CTxxx:
Fatalf("defaultlit: idealkind is CTxxx: %+v", n)
case CTBOOL:
t1 := types.Types[TBOOL]
if t != nil && t.IsBoolean() {
t1 = t
}
2016-03-17 12:24:11 -07:00
n = convlit1(n, t1, false, reuse)
lineno = lno
return n
case CTINT:
t1 = types.Types[TINT]
case CTRUNE:
t1 = types.Runetype
case CTFLT:
t1 = types.Types[TFLOAT64]
case CTCPLX:
t1 = types.Types[TCOMPLEX128]
}
// Note: n.Val().Ctype() can be CTxxx (not a constant) here
// in the case of an untyped non-constant value, like 1<<i.
v1 := n.Val()
if t != nil {
if t.IsInteger() {
t1 = t
v1 = toint(n.Val())
} else if t.IsFloat() {
t1 = t
v1 = toflt(n.Val())
} else if t.IsComplex() {
t1 = t
v1 = tocplx(n.Val())
}
if n.Val().Ctype() != CTxxx {
n.SetVal(v1)
}
}
if n.Val().Ctype() != CTxxx {
overflow(n.Val(), t1)
}
2016-03-17 12:24:11 -07:00
n = convlit1(n, t1, false, reuse)
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
}
// defaultlit on both nodes simultaneously;
// if they're both ideal going in they better
// get the same type going out.
// force means must assign concrete (non-ideal) 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
// The results of defaultlit2 MUST be assigned back to l and r, e.g.
// n.Left, n.Right = defaultlit2(n.Left, n.Right, force)
func defaultlit2(l *Node, r *Node, force bool) (*Node, *Node) {
if l.Type == nil || r.Type == 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 l, r
}
if !l.Type.IsUntyped() {
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 = convlit(r, l.Type)
return l, r
}
if !r.Type.IsUntyped() {
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 = convlit(l, r.Type)
return l, r
}
if !force {
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 l, r
}
if l.Type.IsBoolean() {
l = convlit(l, types.Types[TBOOL])
r = convlit(r, types.Types[TBOOL])
}
lkind := idealkind(l)
rkind := idealkind(r)
if lkind == CTCPLX || rkind == CTCPLX {
l = convlit(l, types.Types[TCOMPLEX128])
r = convlit(r, types.Types[TCOMPLEX128])
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 l, r
}
if lkind == CTFLT || rkind == CTFLT {
l = convlit(l, types.Types[TFLOAT64])
r = convlit(r, types.Types[TFLOAT64])
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 l, r
}
if lkind == CTRUNE || rkind == CTRUNE {
l = convlit(l, types.Runetype)
r = convlit(r, types.Runetype)
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 l, r
}
l = convlit(l, types.Types[TINT])
r = convlit(r, types.Types[TINT])
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 l, r
}
// strlit returns the value of a literal string Node as a string.
func strlit(n *Node) string {
return n.Val().U.(string)
}
func smallintconst(n *Node) bool {
if n.Op == OLITERAL && Isconst(n, CTINT) && n.Type != nil {
switch simtype[n.Type.Etype] {
case TINT8,
TUINT8,
TINT16,
TUINT16,
TINT32,
TUINT32,
TBOOL,
TPTR32:
return true
case TIDEAL, TINT64, TUINT64, TPTR64:
v, ok := n.Val().U.(*Mpint)
if ok && v.Cmp(minintval[TINT32]) > 0 && v.Cmp(maxintval[TINT32]) < 0 {
return true
}
}
}
return false
}
// nonnegintconst checks if Node n contains a constant expression
// representable as a non-negative small integer, and returns its
// (integer) value if that's the case. Otherwise, it returns -1.
func nonnegintconst(n *Node) int64 {
if n.Op != OLITERAL {
return -1
}
// toint will leave n.Val unchanged if it's not castable to an
// Mpint, so we still have to guard the conversion.
v := toint(n.Val())
vi, ok := v.U.(*Mpint)
if !ok || vi.CmpInt64(0) < 0 || vi.Cmp(maxintval[TINT32]) > 0 {
return -1
}
return vi.Int64()
}
// isGoConst reports whether n is a Go language constant (as opposed to a
// compile-time constant).
//
// Expressions derived from nil, like string([]byte(nil)), while they
// may be known at compile time, are not Go language constants.
// Only called for expressions known to evaluated to compile-time
// constants.
func (n *Node) isGoConst() bool {
if n.Orig != nil {
n = n.Orig
}
switch n.Op {
case OADD,
OADDSTR,
OAND,
OANDAND,
OANDNOT,
OCOM,
ODIV,
OEQ,
OGE,
OGT,
OLE,
OLSH,
OLT,
OMINUS,
OMOD,
OMUL,
ONE,
ONOT,
OOR,
OOROR,
OPLUS,
ORSH,
OSUB,
OXOR,
OIOTA,
OCOMPLEX,
OREAL,
OIMAG:
if n.Left.isGoConst() && (n.Right == nil || n.Right.isGoConst()) {
return true
}
case OCONV:
if okforconst[n.Type.Etype] && n.Left.isGoConst() {
return true
}
case OLEN, OCAP:
l := n.Left
if l.isGoConst() {
return true
}
// Special case: len/cap is constant when applied to array or
// pointer to array when the expression does not contain
// function calls or channel receive operations.
t := l.Type
if t != nil && t.IsPtr() {
t = t.Elem()
}
if t != nil && t.IsArray() && !hascallchan(l) {
return true
}
case OLITERAL:
if n.Val().Ctype() != CTNIL {
return true
}
case ONAME:
l := asNode(n.Sym.Def)
if l != nil && l.Op == OLITERAL && n.Val().Ctype() != CTNIL {
return true
}
case ONONAME:
if asNode(n.Sym.Def) != nil && asNode(n.Sym.Def).Op == OIOTA {
return true
}
case OALIGNOF, OOFFSETOF, OSIZEOF:
return true
}
//dump("nonconst", n);
return false
}
func hascallchan(n *Node) bool {
if n == nil {
return false
}
switch n.Op {
case OAPPEND,
OCALL,
OCALLFUNC,
OCALLINTER,
OCALLMETH,
OCAP,
OCLOSE,
OCOMPLEX,
OCOPY,
ODELETE,
OIMAG,
OLEN,
OMAKE,
ONEW,
OPANIC,
OPRINT,
OPRINTN,
OREAL,
ORECOVER,
ORECV:
return true
}
if hascallchan(n.Left) || hascallchan(n.Right) {
return true
}
for _, n1 := range n.List.Slice() {
if hascallchan(n1) {
return true
}
}
for _, n2 := range n.Rlist.Slice() {
if hascallchan(n2) {
return true
}
}
return false
}