2015-08-18 10:26:28 -07:00
|
|
|
// Copyright 2015 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 ssa
|
|
|
|
|
|
2017-10-24 17:51:05 -04:00
|
|
|
import (
|
|
|
|
|
"cmd/compile/internal/types"
|
2020-08-17 16:57:22 -04:00
|
|
|
"sort"
|
2017-10-24 17:51:05 -04:00
|
|
|
)
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
|
2016-02-08 11:00:43 -08:00
|
|
|
// decompose converts phi ops on compound builtin types into phi
|
2017-10-24 17:51:05 -04:00
|
|
|
// ops on simple types, then invokes rewrite rules to decompose
|
|
|
|
|
// other ops on those types.
|
2016-02-08 11:00:43 -08:00
|
|
|
func decomposeBuiltIn(f *Func) {
|
2017-10-24 17:51:05 -04:00
|
|
|
// Decompose phis
|
2015-08-18 10:26:28 -07:00
|
|
|
for _, b := range f.Blocks {
|
|
|
|
|
for _, v := range b.Values {
|
|
|
|
|
if v.Op != OpPhi {
|
|
|
|
|
continue
|
|
|
|
|
}
|
2016-02-08 11:00:43 -08:00
|
|
|
decomposeBuiltInPhi(v)
|
2015-08-18 10:26:28 -07:00
|
|
|
}
|
|
|
|
|
}
|
2015-11-02 08:10:26 -08:00
|
|
|
|
2017-10-24 17:51:05 -04:00
|
|
|
// Decompose other values
|
2020-06-30 15:59:40 -07:00
|
|
|
// Note: deadcode is false because we need to keep the original
|
|
|
|
|
// values around so the name component resolution below can still work.
|
|
|
|
|
applyRewrite(f, rewriteBlockdec, rewriteValuedec, leaveDeadValues)
|
2017-10-24 17:51:05 -04:00
|
|
|
if f.Config.RegSize == 4 {
|
2020-06-30 15:59:40 -07:00
|
|
|
applyRewrite(f, rewriteBlockdec64, rewriteValuedec64, leaveDeadValues)
|
2017-10-24 17:51:05 -04:00
|
|
|
}
|
|
|
|
|
|
2015-11-02 08:10:26 -08:00
|
|
|
// Split up named values into their components.
|
2020-08-17 16:57:22 -04:00
|
|
|
// accumulate old names for aggregates (that are decomposed) in toDelete for efficient bulk deletion,
|
|
|
|
|
// accumulate new LocalSlots in newNames for addition after the iteration. This decomposition is for
|
|
|
|
|
// builtin types with leaf components, and thus there is no need to reprocess the newly create LocalSlots.
|
|
|
|
|
var toDelete []namedVal
|
2016-03-31 21:24:10 -07:00
|
|
|
var newNames []LocalSlot
|
2020-08-17 16:57:22 -04:00
|
|
|
for i, name := range f.Names {
|
2015-11-02 08:10:26 -08:00
|
|
|
t := name.Type
|
|
|
|
|
switch {
|
2017-04-28 00:19:49 +00:00
|
|
|
case t.IsInteger() && t.Size() > f.Config.RegSize:
|
2017-03-16 22:42:10 -07:00
|
|
|
hiName, loName := f.fe.SplitInt64(name)
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
newNames = append(newNames, hiName, loName)
|
2020-08-17 16:57:22 -04:00
|
|
|
for j, v := range f.NamedValues[name] {
|
2017-10-24 17:51:05 -04:00
|
|
|
if v.Op != OpInt64Make {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
f.NamedValues[hiName] = append(f.NamedValues[hiName], v.Args[0])
|
|
|
|
|
f.NamedValues[loName] = append(f.NamedValues[loName], v.Args[1])
|
2020-08-17 16:57:22 -04:00
|
|
|
toDelete = append(toDelete, namedVal{i, j})
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
}
|
2015-11-02 08:10:26 -08:00
|
|
|
case t.IsComplex():
|
2017-03-16 22:42:10 -07:00
|
|
|
rName, iName := f.fe.SplitComplex(name)
|
2016-03-31 21:24:10 -07:00
|
|
|
newNames = append(newNames, rName, iName)
|
2020-08-17 16:57:22 -04:00
|
|
|
for j, v := range f.NamedValues[name] {
|
2017-10-24 17:51:05 -04:00
|
|
|
if v.Op != OpComplexMake {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
f.NamedValues[rName] = append(f.NamedValues[rName], v.Args[0])
|
|
|
|
|
f.NamedValues[iName] = append(f.NamedValues[iName], v.Args[1])
|
2020-08-17 16:57:22 -04:00
|
|
|
toDelete = append(toDelete, namedVal{i, j})
|
2015-11-02 08:10:26 -08:00
|
|
|
}
|
|
|
|
|
case t.IsString():
|
2017-03-16 22:42:10 -07:00
|
|
|
ptrName, lenName := f.fe.SplitString(name)
|
2016-03-31 21:24:10 -07:00
|
|
|
newNames = append(newNames, ptrName, lenName)
|
2020-08-17 16:57:22 -04:00
|
|
|
for j, v := range f.NamedValues[name] {
|
2017-10-24 17:51:05 -04:00
|
|
|
if v.Op != OpStringMake {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
f.NamedValues[ptrName] = append(f.NamedValues[ptrName], v.Args[0])
|
|
|
|
|
f.NamedValues[lenName] = append(f.NamedValues[lenName], v.Args[1])
|
2020-08-17 16:57:22 -04:00
|
|
|
toDelete = append(toDelete, namedVal{i, j})
|
2015-11-02 08:10:26 -08:00
|
|
|
}
|
|
|
|
|
case t.IsSlice():
|
2017-03-16 22:42:10 -07:00
|
|
|
ptrName, lenName, capName := f.fe.SplitSlice(name)
|
2016-03-31 21:24:10 -07:00
|
|
|
newNames = append(newNames, ptrName, lenName, capName)
|
2020-08-17 16:57:22 -04:00
|
|
|
for j, v := range f.NamedValues[name] {
|
2017-10-24 17:51:05 -04:00
|
|
|
if v.Op != OpSliceMake {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
f.NamedValues[ptrName] = append(f.NamedValues[ptrName], v.Args[0])
|
|
|
|
|
f.NamedValues[lenName] = append(f.NamedValues[lenName], v.Args[1])
|
|
|
|
|
f.NamedValues[capName] = append(f.NamedValues[capName], v.Args[2])
|
2020-08-17 16:57:22 -04:00
|
|
|
toDelete = append(toDelete, namedVal{i, j})
|
2015-11-02 08:10:26 -08:00
|
|
|
}
|
|
|
|
|
case t.IsInterface():
|
2017-03-16 22:42:10 -07:00
|
|
|
typeName, dataName := f.fe.SplitInterface(name)
|
2016-03-31 21:24:10 -07:00
|
|
|
newNames = append(newNames, typeName, dataName)
|
2020-08-17 16:57:22 -04:00
|
|
|
for j, v := range f.NamedValues[name] {
|
2017-10-24 17:51:05 -04:00
|
|
|
if v.Op != OpIMake {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
f.NamedValues[typeName] = append(f.NamedValues[typeName], v.Args[0])
|
|
|
|
|
f.NamedValues[dataName] = append(f.NamedValues[dataName], v.Args[1])
|
2020-08-17 16:57:22 -04:00
|
|
|
toDelete = append(toDelete, namedVal{i, j})
|
2015-11-02 08:10:26 -08:00
|
|
|
}
|
2016-05-31 11:27:16 -04:00
|
|
|
case t.IsFloat():
|
2017-04-21 19:16:15 -07:00
|
|
|
// floats are never decomposed, even ones bigger than RegSize
|
2017-04-28 00:19:49 +00:00
|
|
|
case t.Size() > f.Config.RegSize:
|
2017-08-17 12:23:34 -07:00
|
|
|
f.Fatalf("undecomposed named type %s %v", name, t)
|
2015-11-02 08:10:26 -08:00
|
|
|
}
|
|
|
|
|
}
|
2020-08-17 16:57:22 -04:00
|
|
|
|
|
|
|
|
deleteNamedVals(f, toDelete)
|
|
|
|
|
f.Names = append(f.Names, newNames...)
|
2015-08-18 10:26:28 -07:00
|
|
|
}
|
|
|
|
|
|
2016-02-08 11:00:43 -08:00
|
|
|
func decomposeBuiltInPhi(v *Value) {
|
2016-01-11 21:05:33 -08:00
|
|
|
switch {
|
2017-04-28 00:19:49 +00:00
|
|
|
case v.Type.IsInteger() && v.Type.Size() > v.Block.Func.Config.RegSize:
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
decomposeInt64Phi(v)
|
2016-01-11 21:05:33 -08:00
|
|
|
case v.Type.IsComplex():
|
|
|
|
|
decomposeComplexPhi(v)
|
|
|
|
|
case v.Type.IsString():
|
|
|
|
|
decomposeStringPhi(v)
|
|
|
|
|
case v.Type.IsSlice():
|
|
|
|
|
decomposeSlicePhi(v)
|
|
|
|
|
case v.Type.IsInterface():
|
|
|
|
|
decomposeInterfacePhi(v)
|
2016-05-31 11:27:16 -04:00
|
|
|
case v.Type.IsFloat():
|
2017-04-21 19:16:15 -07:00
|
|
|
// floats are never decomposed, even ones bigger than RegSize
|
2017-04-28 00:19:49 +00:00
|
|
|
case v.Type.Size() > v.Block.Func.Config.RegSize:
|
2016-09-14 10:01:05 -07:00
|
|
|
v.Fatalf("undecomposed type %s", v.Type)
|
2016-01-11 21:05:33 -08:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2015-08-18 10:26:28 -07:00
|
|
|
func decomposeStringPhi(v *Value) {
|
2017-03-17 16:04:46 -07:00
|
|
|
types := &v.Block.Func.Config.Types
|
|
|
|
|
ptrType := types.BytePtr
|
|
|
|
|
lenType := types.Int
|
2015-08-18 10:26:28 -07:00
|
|
|
|
2016-12-07 18:14:35 -08:00
|
|
|
ptr := v.Block.NewValue0(v.Pos, OpPhi, ptrType)
|
|
|
|
|
len := v.Block.NewValue0(v.Pos, OpPhi, lenType)
|
2015-08-18 10:26:28 -07:00
|
|
|
for _, a := range v.Args {
|
2016-12-07 18:14:35 -08:00
|
|
|
ptr.AddArg(a.Block.NewValue1(v.Pos, OpStringPtr, ptrType, a))
|
|
|
|
|
len.AddArg(a.Block.NewValue1(v.Pos, OpStringLen, lenType, a))
|
2015-08-18 10:26:28 -07:00
|
|
|
}
|
2016-02-04 17:21:57 +01:00
|
|
|
v.reset(OpStringMake)
|
2015-08-18 10:26:28 -07:00
|
|
|
v.AddArg(ptr)
|
|
|
|
|
v.AddArg(len)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func decomposeSlicePhi(v *Value) {
|
2017-03-17 16:04:46 -07:00
|
|
|
types := &v.Block.Func.Config.Types
|
2020-08-21 20:20:12 -07:00
|
|
|
ptrType := v.Type.Elem().PtrTo()
|
2017-03-17 16:04:46 -07:00
|
|
|
lenType := types.Int
|
2015-08-18 10:26:28 -07:00
|
|
|
|
2016-12-07 18:14:35 -08:00
|
|
|
ptr := v.Block.NewValue0(v.Pos, OpPhi, ptrType)
|
|
|
|
|
len := v.Block.NewValue0(v.Pos, OpPhi, lenType)
|
|
|
|
|
cap := v.Block.NewValue0(v.Pos, OpPhi, lenType)
|
2015-08-18 10:26:28 -07:00
|
|
|
for _, a := range v.Args {
|
2016-12-07 18:14:35 -08:00
|
|
|
ptr.AddArg(a.Block.NewValue1(v.Pos, OpSlicePtr, ptrType, a))
|
|
|
|
|
len.AddArg(a.Block.NewValue1(v.Pos, OpSliceLen, lenType, a))
|
|
|
|
|
cap.AddArg(a.Block.NewValue1(v.Pos, OpSliceCap, lenType, a))
|
2015-08-18 10:26:28 -07:00
|
|
|
}
|
2016-02-04 17:21:57 +01:00
|
|
|
v.reset(OpSliceMake)
|
2015-08-18 10:26:28 -07:00
|
|
|
v.AddArg(ptr)
|
|
|
|
|
v.AddArg(len)
|
|
|
|
|
v.AddArg(cap)
|
|
|
|
|
}
|
|
|
|
|
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
func decomposeInt64Phi(v *Value) {
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
cfgtypes := &v.Block.Func.Config.Types
|
|
|
|
|
var partType *types.Type
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
if v.Type.IsSigned() {
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
partType = cfgtypes.Int32
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
} else {
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
partType = cfgtypes.UInt32
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
}
|
|
|
|
|
|
2016-12-07 18:14:35 -08:00
|
|
|
hi := v.Block.NewValue0(v.Pos, OpPhi, partType)
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
lo := v.Block.NewValue0(v.Pos, OpPhi, cfgtypes.UInt32)
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
for _, a := range v.Args {
|
2016-12-07 18:14:35 -08:00
|
|
|
hi.AddArg(a.Block.NewValue1(v.Pos, OpInt64Hi, partType, a))
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
lo.AddArg(a.Block.NewValue1(v.Pos, OpInt64Lo, cfgtypes.UInt32, a))
|
[dev.ssa] cmd/compile: decompose 64-bit integer on ARM
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
2016-05-18 18:14:36 -04:00
|
|
|
}
|
|
|
|
|
v.reset(OpInt64Make)
|
|
|
|
|
v.AddArg(hi)
|
|
|
|
|
v.AddArg(lo)
|
|
|
|
|
}
|
|
|
|
|
|
2015-08-28 14:24:10 -04:00
|
|
|
func decomposeComplexPhi(v *Value) {
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
cfgtypes := &v.Block.Func.Config.Types
|
|
|
|
|
var partType *types.Type
|
2017-04-28 00:19:49 +00:00
|
|
|
switch z := v.Type.Size(); z {
|
2015-08-28 14:24:10 -04:00
|
|
|
case 8:
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
partType = cfgtypes.Float32
|
2015-08-28 14:24:10 -04:00
|
|
|
case 16:
|
cmd/compile: change ssa.Type into *types.Type
When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.
In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.
Though this is a big CL, most of the changes are
mechanical and uninteresting.
Interesting bits:
* Add new singleton globals to package types for the special
SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
and TTUPLE, for SSA tuple types.
ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
to package types.
* We had picked the name "types" in our rules for the handy
list of types provided by ssa.Config. That conflicted with
the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
and probably also some other duplicated Type methods
designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
and they were not particularly careful about types in general.
Of necessity, this CL switches them to use *types.Type;
it does not make them more type-accurate.
Unfortunately, using types.Type means initializing a bit
of the types universe.
This is prime for refactoring and improvement.
This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.
name old alloc/op new alloc/op delta
Template 37.9MB ± 0% 37.7MB ± 0% -0.57% (p=0.000 n=10+8)
Unicode 28.9MB ± 0% 28.7MB ± 0% -0.52% (p=0.000 n=10+10)
GoTypes 110MB ± 0% 109MB ± 0% -0.88% (p=0.000 n=10+10)
Flate 24.7MB ± 0% 24.6MB ± 0% -0.66% (p=0.000 n=10+10)
GoParser 31.1MB ± 0% 30.9MB ± 0% -0.61% (p=0.000 n=10+9)
Reflect 73.9MB ± 0% 73.4MB ± 0% -0.62% (p=0.000 n=10+8)
Tar 25.8MB ± 0% 25.6MB ± 0% -0.77% (p=0.000 n=9+10)
XML 41.2MB ± 0% 40.9MB ± 0% -0.80% (p=0.000 n=10+10)
[Geo mean] 40.5MB 40.3MB -0.68%
name old allocs/op new allocs/op delta
Template 385k ± 0% 386k ± 0% ~ (p=0.356 n=10+9)
Unicode 343k ± 1% 344k ± 0% ~ (p=0.481 n=10+10)
GoTypes 1.16M ± 0% 1.16M ± 0% -0.16% (p=0.004 n=10+10)
Flate 238k ± 1% 238k ± 1% ~ (p=0.853 n=10+10)
GoParser 320k ± 0% 320k ± 0% ~ (p=0.720 n=10+9)
Reflect 957k ± 0% 957k ± 0% ~ (p=0.460 n=10+8)
Tar 252k ± 0% 252k ± 0% ~ (p=0.133 n=9+10)
XML 400k ± 0% 400k ± 0% ~ (p=0.796 n=10+10)
[Geo mean] 428k 428k -0.01%
Removing all the interface calls helps non-trivially with CPU, though.
name old time/op new time/op delta
Template 178ms ± 4% 173ms ± 3% -2.90% (p=0.000 n=94+96)
Unicode 85.0ms ± 4% 83.9ms ± 4% -1.23% (p=0.000 n=96+96)
GoTypes 543ms ± 3% 528ms ± 3% -2.73% (p=0.000 n=98+96)
Flate 116ms ± 3% 113ms ± 4% -2.34% (p=0.000 n=96+99)
GoParser 144ms ± 3% 140ms ± 4% -2.80% (p=0.000 n=99+97)
Reflect 344ms ± 3% 334ms ± 4% -3.02% (p=0.000 n=100+99)
Tar 106ms ± 5% 103ms ± 4% -3.30% (p=0.000 n=98+94)
XML 198ms ± 5% 192ms ± 4% -2.88% (p=0.000 n=92+95)
[Geo mean] 178ms 173ms -2.65%
name old user-time/op new user-time/op delta
Template 229ms ± 5% 224ms ± 5% -2.36% (p=0.000 n=95+99)
Unicode 107ms ± 6% 106ms ± 5% -1.13% (p=0.001 n=93+95)
GoTypes 696ms ± 4% 679ms ± 4% -2.45% (p=0.000 n=97+99)
Flate 137ms ± 4% 134ms ± 5% -2.66% (p=0.000 n=99+96)
GoParser 176ms ± 5% 172ms ± 8% -2.27% (p=0.000 n=98+100)
Reflect 430ms ± 6% 411ms ± 5% -4.46% (p=0.000 n=100+92)
Tar 128ms ±13% 123ms ±13% -4.21% (p=0.000 n=100+100)
XML 239ms ± 6% 233ms ± 6% -2.50% (p=0.000 n=95+97)
[Geo mean] 220ms 213ms -2.76%
Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2017-04-28 14:12:28 -07:00
|
|
|
partType = cfgtypes.Float64
|
2015-08-28 14:24:10 -04:00
|
|
|
default:
|
|
|
|
|
v.Fatalf("decomposeComplexPhi: bad complex size %d", z)
|
2015-08-28 14:24:10 -04:00
|
|
|
}
|
|
|
|
|
|
2016-12-07 18:14:35 -08:00
|
|
|
real := v.Block.NewValue0(v.Pos, OpPhi, partType)
|
|
|
|
|
imag := v.Block.NewValue0(v.Pos, OpPhi, partType)
|
2015-08-28 14:24:10 -04:00
|
|
|
for _, a := range v.Args {
|
2016-12-07 18:14:35 -08:00
|
|
|
real.AddArg(a.Block.NewValue1(v.Pos, OpComplexReal, partType, a))
|
|
|
|
|
imag.AddArg(a.Block.NewValue1(v.Pos, OpComplexImag, partType, a))
|
2015-08-28 14:24:10 -04:00
|
|
|
}
|
2016-02-04 17:21:57 +01:00
|
|
|
v.reset(OpComplexMake)
|
2015-08-28 14:24:10 -04:00
|
|
|
v.AddArg(real)
|
|
|
|
|
v.AddArg(imag)
|
|
|
|
|
}
|
|
|
|
|
|
2015-08-18 10:26:28 -07:00
|
|
|
func decomposeInterfacePhi(v *Value) {
|
2018-02-27 13:46:03 -08:00
|
|
|
uintptrType := v.Block.Func.Config.Types.Uintptr
|
2017-03-17 16:04:46 -07:00
|
|
|
ptrType := v.Block.Func.Config.Types.BytePtr
|
2015-08-18 10:26:28 -07:00
|
|
|
|
2018-02-27 13:46:03 -08:00
|
|
|
itab := v.Block.NewValue0(v.Pos, OpPhi, uintptrType)
|
2016-12-07 18:14:35 -08:00
|
|
|
data := v.Block.NewValue0(v.Pos, OpPhi, ptrType)
|
2015-08-18 10:26:28 -07:00
|
|
|
for _, a := range v.Args {
|
2018-02-27 13:46:03 -08:00
|
|
|
itab.AddArg(a.Block.NewValue1(v.Pos, OpITab, uintptrType, a))
|
2016-12-07 18:14:35 -08:00
|
|
|
data.AddArg(a.Block.NewValue1(v.Pos, OpIData, ptrType, a))
|
2015-08-18 10:26:28 -07:00
|
|
|
}
|
2016-02-04 17:21:57 +01:00
|
|
|
v.reset(OpIMake)
|
2015-08-18 10:26:28 -07:00
|
|
|
v.AddArg(itab)
|
|
|
|
|
v.AddArg(data)
|
|
|
|
|
}
|
2016-02-08 11:00:43 -08:00
|
|
|
|
2018-11-16 16:20:28 -05:00
|
|
|
func decomposeArgs(f *Func) {
|
2020-06-30 15:59:40 -07:00
|
|
|
applyRewrite(f, rewriteBlockdecArgs, rewriteValuedecArgs, removeDeadValues)
|
2018-11-16 16:20:28 -05:00
|
|
|
}
|
|
|
|
|
|
2016-02-08 11:00:43 -08:00
|
|
|
func decomposeUser(f *Func) {
|
|
|
|
|
for _, b := range f.Blocks {
|
|
|
|
|
for _, v := range b.Values {
|
|
|
|
|
if v.Op != OpPhi {
|
|
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
decomposeUserPhi(v)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
// Split up named values into their components.
|
|
|
|
|
i := 0
|
2016-03-31 21:24:10 -07:00
|
|
|
var newNames []LocalSlot
|
2016-02-08 11:00:43 -08:00
|
|
|
for _, name := range f.Names {
|
|
|
|
|
t := name.Type
|
|
|
|
|
switch {
|
|
|
|
|
case t.IsStruct():
|
2017-10-18 16:32:01 -04:00
|
|
|
newNames = decomposeUserStructInto(f, name, newNames)
|
2016-10-30 21:10:03 -07:00
|
|
|
case t.IsArray():
|
2017-10-18 16:32:01 -04:00
|
|
|
newNames = decomposeUserArrayInto(f, name, newNames)
|
2016-02-08 11:00:43 -08:00
|
|
|
default:
|
|
|
|
|
f.Names[i] = name
|
|
|
|
|
i++
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
f.Names = f.Names[:i]
|
2016-03-31 21:24:10 -07:00
|
|
|
f.Names = append(f.Names, newNames...)
|
2016-02-08 11:00:43 -08:00
|
|
|
}
|
|
|
|
|
|
2017-10-18 16:32:01 -04:00
|
|
|
// decomposeUserArrayInto creates names for the element(s) of arrays referenced
|
|
|
|
|
// by name where possible, and appends those new names to slots, which is then
|
|
|
|
|
// returned.
|
|
|
|
|
func decomposeUserArrayInto(f *Func, name LocalSlot, slots []LocalSlot) []LocalSlot {
|
|
|
|
|
t := name.Type
|
|
|
|
|
if t.NumElem() == 0 {
|
|
|
|
|
// TODO(khr): Not sure what to do here. Probably nothing.
|
|
|
|
|
// Names for empty arrays aren't important.
|
|
|
|
|
return slots
|
|
|
|
|
}
|
|
|
|
|
if t.NumElem() != 1 {
|
|
|
|
|
// shouldn't get here due to CanSSA
|
|
|
|
|
f.Fatalf("array not of size 1")
|
|
|
|
|
}
|
|
|
|
|
elemName := f.fe.SplitArray(name)
|
2020-08-17 16:57:22 -04:00
|
|
|
var keep []*Value
|
2017-10-18 16:32:01 -04:00
|
|
|
for _, v := range f.NamedValues[name] {
|
2017-10-24 17:51:05 -04:00
|
|
|
if v.Op != OpArrayMake1 {
|
2020-08-17 16:57:22 -04:00
|
|
|
keep = append(keep, v)
|
2017-10-24 17:51:05 -04:00
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
f.NamedValues[elemName] = append(f.NamedValues[elemName], v.Args[0])
|
2017-10-18 16:32:01 -04:00
|
|
|
}
|
2020-08-17 16:57:22 -04:00
|
|
|
if len(keep) == 0 {
|
|
|
|
|
// delete the name for the array as a whole
|
|
|
|
|
delete(f.NamedValues, name)
|
|
|
|
|
} else {
|
|
|
|
|
f.NamedValues[name] = keep
|
|
|
|
|
}
|
2017-10-18 16:32:01 -04:00
|
|
|
|
2018-04-24 13:39:51 -07:00
|
|
|
if t.Elem().IsArray() {
|
2017-10-18 16:32:01 -04:00
|
|
|
return decomposeUserArrayInto(f, elemName, slots)
|
2018-04-24 13:39:51 -07:00
|
|
|
} else if t.Elem().IsStruct() {
|
2017-10-18 16:32:01 -04:00
|
|
|
return decomposeUserStructInto(f, elemName, slots)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return append(slots, elemName)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// decomposeUserStructInto creates names for the fields(s) of structs referenced
|
|
|
|
|
// by name where possible, and appends those new names to slots, which is then
|
|
|
|
|
// returned.
|
|
|
|
|
func decomposeUserStructInto(f *Func, name LocalSlot, slots []LocalSlot) []LocalSlot {
|
|
|
|
|
fnames := []LocalSlot{} // slots for struct in name
|
|
|
|
|
t := name.Type
|
|
|
|
|
n := t.NumFields()
|
|
|
|
|
|
|
|
|
|
for i := 0; i < n; i++ {
|
|
|
|
|
fs := f.fe.SplitStruct(name, i)
|
|
|
|
|
fnames = append(fnames, fs)
|
|
|
|
|
// arrays and structs will be decomposed further, so
|
|
|
|
|
// there's no need to record a name
|
|
|
|
|
if !fs.Type.IsArray() && !fs.Type.IsStruct() {
|
|
|
|
|
slots = append(slots, fs)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2017-10-24 17:51:05 -04:00
|
|
|
makeOp := StructMakeOp(n)
|
2020-08-17 16:57:22 -04:00
|
|
|
var keep []*Value
|
2017-10-18 16:32:01 -04:00
|
|
|
// create named values for each struct field
|
|
|
|
|
for _, v := range f.NamedValues[name] {
|
2017-10-24 17:51:05 -04:00
|
|
|
if v.Op != makeOp {
|
2020-08-17 16:57:22 -04:00
|
|
|
keep = append(keep, v)
|
2017-10-24 17:51:05 -04:00
|
|
|
continue
|
|
|
|
|
}
|
2017-10-18 16:32:01 -04:00
|
|
|
for i := 0; i < len(fnames); i++ {
|
2017-10-24 17:51:05 -04:00
|
|
|
f.NamedValues[fnames[i]] = append(f.NamedValues[fnames[i]], v.Args[i])
|
2017-10-18 16:32:01 -04:00
|
|
|
}
|
|
|
|
|
}
|
2020-08-17 16:57:22 -04:00
|
|
|
if len(keep) == 0 {
|
|
|
|
|
// delete the name for the struct as a whole
|
|
|
|
|
delete(f.NamedValues, name)
|
|
|
|
|
} else {
|
|
|
|
|
f.NamedValues[name] = keep
|
|
|
|
|
}
|
2017-10-18 16:32:01 -04:00
|
|
|
|
|
|
|
|
// now that this f.NamedValues contains values for the struct
|
|
|
|
|
// fields, recurse into nested structs
|
|
|
|
|
for i := 0; i < n; i++ {
|
|
|
|
|
if name.Type.FieldType(i).IsStruct() {
|
|
|
|
|
slots = decomposeUserStructInto(f, fnames[i], slots)
|
|
|
|
|
delete(f.NamedValues, fnames[i])
|
|
|
|
|
} else if name.Type.FieldType(i).IsArray() {
|
|
|
|
|
slots = decomposeUserArrayInto(f, fnames[i], slots)
|
|
|
|
|
delete(f.NamedValues, fnames[i])
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return slots
|
|
|
|
|
}
|
2016-02-08 11:00:43 -08:00
|
|
|
func decomposeUserPhi(v *Value) {
|
|
|
|
|
switch {
|
|
|
|
|
case v.Type.IsStruct():
|
|
|
|
|
decomposeStructPhi(v)
|
2016-10-30 21:10:03 -07:00
|
|
|
case v.Type.IsArray():
|
|
|
|
|
decomposeArrayPhi(v)
|
2016-02-08 11:00:43 -08:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-10-30 21:10:03 -07:00
|
|
|
// decomposeStructPhi replaces phi-of-struct with structmake(phi-for-each-field),
|
|
|
|
|
// and then recursively decomposes the phis for each field.
|
2015-08-18 10:26:28 -07:00
|
|
|
func decomposeStructPhi(v *Value) {
|
2016-01-11 21:05:33 -08:00
|
|
|
t := v.Type
|
|
|
|
|
n := t.NumFields()
|
|
|
|
|
var fields [MaxStruct]*Value
|
2016-03-14 12:45:18 -07:00
|
|
|
for i := 0; i < n; i++ {
|
2016-12-07 18:14:35 -08:00
|
|
|
fields[i] = v.Block.NewValue0(v.Pos, OpPhi, t.FieldType(i))
|
2016-01-11 21:05:33 -08:00
|
|
|
}
|
|
|
|
|
for _, a := range v.Args {
|
2016-03-14 12:45:18 -07:00
|
|
|
for i := 0; i < n; i++ {
|
2016-12-07 18:14:35 -08:00
|
|
|
fields[i].AddArg(a.Block.NewValue1I(v.Pos, OpStructSelect, t.FieldType(i), int64(i), a))
|
2016-01-11 21:05:33 -08:00
|
|
|
}
|
|
|
|
|
}
|
2016-02-04 17:21:57 +01:00
|
|
|
v.reset(StructMakeOp(n))
|
2016-01-11 21:05:33 -08:00
|
|
|
v.AddArgs(fields[:n]...)
|
|
|
|
|
|
|
|
|
|
// Recursively decompose phis for each field.
|
|
|
|
|
for _, f := range fields[:n] {
|
2016-10-30 21:10:03 -07:00
|
|
|
decomposeUserPhi(f)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// decomposeArrayPhi replaces phi-of-array with arraymake(phi-of-array-element),
|
|
|
|
|
// and then recursively decomposes the element phi.
|
|
|
|
|
func decomposeArrayPhi(v *Value) {
|
|
|
|
|
t := v.Type
|
|
|
|
|
if t.NumElem() == 0 {
|
|
|
|
|
v.reset(OpArrayMake0)
|
|
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
if t.NumElem() != 1 {
|
|
|
|
|
v.Fatalf("SSAable array must have no more than 1 element")
|
2016-01-11 21:05:33 -08:00
|
|
|
}
|
2018-04-24 13:39:51 -07:00
|
|
|
elem := v.Block.NewValue0(v.Pos, OpPhi, t.Elem())
|
2016-10-30 21:10:03 -07:00
|
|
|
for _, a := range v.Args {
|
2018-04-24 13:39:51 -07:00
|
|
|
elem.AddArg(a.Block.NewValue1I(v.Pos, OpArraySelect, t.Elem(), 0, a))
|
2016-10-30 21:10:03 -07:00
|
|
|
}
|
|
|
|
|
v.reset(OpArrayMake1)
|
|
|
|
|
v.AddArg(elem)
|
|
|
|
|
|
|
|
|
|
// Recursively decompose elem phi.
|
|
|
|
|
decomposeUserPhi(elem)
|
2016-01-11 21:05:33 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// MaxStruct is the maximum number of fields a struct
|
|
|
|
|
// can have and still be SSAable.
|
|
|
|
|
const MaxStruct = 4
|
|
|
|
|
|
|
|
|
|
// StructMakeOp returns the opcode to construct a struct with the
|
|
|
|
|
// given number of fields.
|
2016-03-14 12:45:18 -07:00
|
|
|
func StructMakeOp(nf int) Op {
|
2016-01-11 21:05:33 -08:00
|
|
|
switch nf {
|
|
|
|
|
case 0:
|
|
|
|
|
return OpStructMake0
|
|
|
|
|
case 1:
|
|
|
|
|
return OpStructMake1
|
|
|
|
|
case 2:
|
|
|
|
|
return OpStructMake2
|
|
|
|
|
case 3:
|
|
|
|
|
return OpStructMake3
|
|
|
|
|
case 4:
|
|
|
|
|
return OpStructMake4
|
|
|
|
|
}
|
|
|
|
|
panic("too many fields in an SSAable struct")
|
2015-08-18 10:26:28 -07:00
|
|
|
}
|
2020-08-17 16:57:22 -04:00
|
|
|
|
|
|
|
|
type namedVal struct {
|
|
|
|
|
locIndex, valIndex int // f.NamedValues[f.Names[locIndex]][valIndex] = key
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// deleteNamedVals removes particular values with debugger names from f's naming data structures
|
|
|
|
|
func deleteNamedVals(f *Func, toDelete []namedVal) {
|
|
|
|
|
// Arrange to delete from larger indices to smaller, to ensure swap-with-end deletion does not invalid pending indices.
|
|
|
|
|
sort.Slice(toDelete, func(i, j int) bool {
|
|
|
|
|
if toDelete[i].locIndex != toDelete[j].locIndex {
|
|
|
|
|
return toDelete[i].locIndex > toDelete[j].locIndex
|
|
|
|
|
}
|
|
|
|
|
return toDelete[i].valIndex > toDelete[j].valIndex
|
|
|
|
|
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// Get rid of obsolete names
|
|
|
|
|
for _, d := range toDelete {
|
|
|
|
|
loc := f.Names[d.locIndex]
|
|
|
|
|
vals := f.NamedValues[loc]
|
|
|
|
|
l := len(vals) - 1
|
|
|
|
|
if l > 0 {
|
|
|
|
|
vals[d.valIndex] = vals[l]
|
|
|
|
|
f.NamedValues[loc] = vals[:l]
|
|
|
|
|
} else {
|
|
|
|
|
delete(f.NamedValues, loc)
|
|
|
|
|
l = len(f.Names) - 1
|
|
|
|
|
f.Names[d.locIndex] = f.Names[l]
|
|
|
|
|
f.Names = f.Names[:l]
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|