cmd/compile/internal/ssa: handle more cases in fuse pass
Currently fuseBlockIf handls four cases where s0 and s1 have only one predecessor
node. In fact, even if s0 and s1 have multiple predecessor nodes, it can be optimized
as well. This patch handles these cases.
This CL does not bring significant performance changes, it is more like an optimization
of dead code elimination. So it detects a lot of benchmarks whose functions are
completely optimized, such as BenchmarkNeIfaceConcrete, BenchmarkEqIfaceConcrete,
BenchmarkFullJapaneseRune, BenchmarkClearFat{8, 12, 16,...1024}, Benchmark{En|De}codeRune,
and BenchmarkCopyFat{8, 12, 16,...1024} and many others that I didn't check.
Test result of compilecmp on linux amd64, the data on linux arm64 is similar.
name old time/op new time/op delta
Template
466131893.640000ns +-16%
455546571.480000ns +-12% ~ (p=0.066 n=50+50)
Unicode
195858206.980000ns +- 9%
197387526.224490ns +- 9% ~ (p=0.231 n=50+49)
GoTypes
1581961264.260000ns +- 7%
1581828319.640000ns +- 5% ~ (p=0.964 n=50+50)
Compiler
7425890449.259998ns +- 2%
7424138343.632654ns +- 2% ~ (p=0.815 n=50+49)
SSA
17392652203.380009ns +- 1%
17248314507.854172ns +- 1% -0.83% (p=0.000 n=50+48)
Flate
298061504.940000ns +-12%
292773080.632653ns +-12% ~ (p=0.165 n=50+49)
GoParser
376046929.420000ns +- 9%
370746254.693878ns +- 9% ~ (p=0.231 n=50+49)
Reflect
995721330.160000ns +- 7%
985916024.620000ns +- 7% ~ (p=0.103 n=50+50)
Tar
398878287.040000ns +-10%
399186737.000000ns +-11% ~ (p=0.883 n=50+50)
XML
555898064.200000ns +- 8%
554877077.100000ns +-10% ~ (p=0.926 n=50+50)
LinkCompiler
757995424.632653ns +- 7%
758301900.420000ns +- 7% ~ (p=0.437 n=49+50)
ExternalLinkCompiler
2399985741.270834ns +- 4%
2396112274.630435ns +- 4% ~ (p=0.766 n=48+46)
LinkWithoutDebugCompiler
450472710.700000ns +- 9%
448935188.720000ns +- 7% ~ (p=0.668 n=50+50)
[Geo mean]
927951041.671984ns
922858781.007286ns -0.55%
name old user-time/op new user-time/op delta
Template
639484500.000000ns +-16%
630857140.000000ns +-13% ~ (p=0.363 n=50+50)
Unicode
370072081.632653ns +-11%
375108300.000000ns +-10% ~ (p=0.248 n=49+50)
GoTypes
2205838860.000000ns +- 5%
2206640900.000000ns +- 3% ~ (p=0.915 n=50+50)
Compiler
10337224272.727268ns +- 2%
10366187152.173916ns +- 2% ~ (p=0.324 n=44+46)
SSA
23843616104.166660ns +- 3%
23678281387.755112ns +- 3% -0.69% (p=0.002 n=48+49)
Flate
396446760.000000ns +-15%
391245333.333333ns +-10% ~ (p=0.313 n=50+48)
GoParser
509121660.000000ns +-10%
505938620.000000ns +- 9% ~ (p=0.436 n=50+50)
Reflect
1328136212.765958ns +- 6%
1312627440.000000ns +- 8% ~ (p=0.068 n=47+50)
Tar
563039480.000000ns +-10%
563064360.000000ns +-10% ~ (p=0.953 n=50+50)
XML
775596380.000000ns +- 9%
774736920.000000ns +-11% ~ (p=0.899 n=50+50)
LinkCompiler
1315155260.000000ns +- 9%
1312946860.000001ns +- 6% ~ (p=0.861 n=50+50)
ExternalLinkCompiler
2739635480.000000ns +- 6%
2732580380.000000ns +- 6% ~ (p=0.943 n=50+50)
LinkWithoutDebugCompiler
537418600.000000ns +-10%
535914620.000000ns +- 9% ~ (p=0.734 n=50+50)
[Geo mean]
1296231259.389525ns
1291929605.124197ns -0.33%
name old alloc/op new alloc/op delta
Template 35.0MB +- 0% 35.0MB +- 0% ~ (p=0.586 n=50+50)
Unicode 29.3MB +- 0% 29.3MB +- 0% +0.01% (p=0.004 n=50+50)
GoTypes 116MB +- 0% 116MB +- 0% ~ (p=0.102 n=50+50)
Compiler 555MB +- 0% 555MB +- 0% ~ (p=0.285 n=50+50)
SSA 1.37GB +- 0% 1.36GB +- 0% -0.98% (p=0.000 n=49+49)
Flate 21.8MB +- 0% 21.8MB +- 0% ~ (p=0.475 n=50+49)
GoParser 26.7MB +- 0% 26.7MB +- 0% ~ (p=0.385 n=50+50)
Reflect 74.2MB +- 0% 74.3MB +- 0% +0.01% (p=0.006 n=49+50)
Tar 32.7MB +- 0% 32.7MB +- 0% ~ (p=0.125 n=50+50)
XML 41.8MB +- 0% 41.8MB +- 0% ~ (p=0.649 n=49+50)
LinkCompiler 105MB +- 0% 105MB +- 0% -0.05% (p=0.000 n=50+48)
ExternalLinkCompiler 92.6MB +- 0% 92.6MB +- 0% -0.02% (p=0.000 n=50+50)
LinkWithoutDebugCompiler 63.9MB +- 0% 63.9MB +- 0% -0.05% (p=0.000 n=50+50)
[Geo mean] 77.0MB 76.9MB -0.08%
name old allocs/op new allocs/op delta
Template 344k +- 0% 344k +- 0% ~ (p=0.091 n=50+50)
Unicode 340k +- 0% 340k +- 0% ~ (p=0.084 n=50+50)
GoTypes 1.19M +- 0% 1.19M +- 0% ~ (p=0.634 n=50+50)
Compiler 5.05M +- 0% 5.05M +- 0% ~ (p=0.428 n=50+48)
SSA 12.9M +- 0% 12.8M +- 0% -0.94% (p=0.000 n=48+50)
Flate 216k +- 0% 216k +- 0% ~ (p=0.959 n=50+50)
GoParser 275k +- 0% 275k +- 0% ~ (p=0.705 n=50+50)
Reflect 887k +- 0% 887k +- 0% ~ (p=0.136 n=50+50)
Tar 318k +- 0% 318k +- 0% +0.01% (p=0.007 n=50+48)
XML 396k +- 0% 396k +- 0% ~ (p=0.577 n=48+50)
LinkCompiler 457k +- 0% 457k +- 0% -0.02% (p=0.000 n=50+48)
ExternalLinkCompiler 459k +- 0% 459k +- 0% -0.02% (p=0.000 n=50+49)
LinkWithoutDebugCompiler 111k +- 0% 111k +- 0% -0.09% (p=0.000 n=50+50)
[Geo mean] 603k 603k -0.08%
name old maxRSS/op new maxRSS/op delta
Template 33.7M +- 5% 33.7M +- 3% ~ (p=0.798 n=50+48)
Unicode 35.2M +- 5% 35.3M +- 5% ~ (p=0.586 n=50+50)
GoTypes 73.5M +- 5% 73.2M +- 5% ~ (p=0.436 n=50+50)
Compiler 315M +- 3% 315M +- 3% ~ (p=0.726 n=49+48)
SSA 705M +- 5% 691M +- 5% -2.01% (p=0.000 n=50+50)
Flate 25.2M +- 3% 25.0M +- 3% ~ (p=0.122 n=49+49)
GoParser 27.7M +- 5% 27.7M +- 3% ~ (p=0.967 n=50+50)
Reflect 54.6M +- 4% 54.8M +- 4% ~ (p=0.418 n=50+50)
Tar 32.2M +- 4% 32.1M +- 3% ~ (p=0.644 n=50+49)
XML 38.7M +- 5% 38.9M +- 5% ~ (p=0.612 n=50+50)
LinkCompiler 159M +- 1% 159M +- 1% ~ (p=0.302 n=49+50)
ExternalLinkCompiler 171M +- 1% 170M +- 1% -0.33% (p=0.000 n=50+50)
LinkWithoutDebugCompiler 133M +- 1% 132M +- 1% -0.44% (p=0.000 n=48+50)
[Geo mean] 76.9M 76.8M -0.25%
name old text-bytes new text-bytes delta
HelloSize 804kB +- 0% 804kB +- 0% ~ (all equal)
name old data-bytes new data-bytes delta
HelloSize 13.2kB +- 0% 13.2kB +- 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 206kB +- 0% 206kB +- 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.20MB +- 0% 1.20MB +- 0% ~ (all equal)
file before after Δ %
api
4904729 4904881 +152 +0.003%
asm
4849768 4849696 -72 -0.001%
buildid
2604760 2604768 +8 +0.000%
cgo
4505984 4506000 +16 +0.000%
compile
18936732 18864939 -71793 -0.379%
cover
4787385 4787377 -8 -0.000%
dist
3451964 3451988 +24 +0.001%
doc
3794929 3794913 -16 -0.000%
fix
3201863 3201847 -16 -0.000%
link
6467231 6467327 +96 +0.001%
objdump
4400076 4400132 +56 +0.001%
pprof
13354395 13354507 +112 +0.001%
trace
10186673 10187273 +600 +0.006%
vet
6727213 6727277 +64 +0.001%
total
105284348 105213571 -70777 -0.067%
Change-Id: I5020d112021f165a4ae18aa56402d8690330d8fb
Reviewed-on: https://go-review.googlesource.com/c/go/+/239457
Reviewed-by: eric fang <eric.fang@arm.com>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: eric fang <eric.fang@arm.com>
Run-TryBot: eric fang <eric.fang@arm.com>
TryBot-Result: Go Bot <gobot@golang.org>