mirror of
https://github.com/golang/go.git
synced 2025-12-08 06:10:04 +00:00
cmd/compile: optimize ARM code with NMULF/NMULD
NMULF and NMULD are efficient FP instructions, and the go compiler can use them to generate better code. The benchmark tests of my patch did not show general change, but big improvement in special cases. 1.A special test case improved 12.6%. https://github.com/benshi001/ugo1/blob/master/fpmul_test.go name old time/op new time/op delta FPMul-4 398µs ± 1% 348µs ± 1% -12.64% (p=0.000 n=40+40) 2. the compilecmp test showed little change. name old time/op new time/op delta Template 2.30s ± 1% 2.31s ± 1% ~ (p=0.754 n=17+19) Unicode 1.31s ± 3% 1.32s ± 5% ~ (p=0.265 n=20+20) GoTypes 7.73s ± 2% 7.73s ± 1% ~ (p=0.925 n=20+20) Compiler 37.0s ± 1% 37.3s ± 2% +0.79% (p=0.002 n=19+20) SSA 83.8s ± 4% 83.5s ± 2% ~ (p=0.964 n=20+17) Flate 1.43s ± 2% 1.44s ± 1% ~ (p=0.602 n=20+20) GoParser 1.82s ± 2% 1.81s ± 2% ~ (p=0.141 n=19+20) Reflect 5.08s ± 2% 5.08s ± 3% ~ (p=0.835 n=20+19) Tar 2.36s ± 1% 2.35s ± 1% ~ (p=0.195 n=18+17) XML 2.57s ± 2% 2.56s ± 1% ~ (p=0.283 n=20+17) [Geo mean] 4.74s 4.75s +0.05% name old user-time/op new user-time/op delta Template 2.75s ± 2% 2.75s ± 0% ~ (p=0.620 n=20+15) Unicode 1.59s ± 4% 1.60s ± 4% ~ (p=0.479 n=20+19) GoTypes 9.48s ± 1% 9.47s ± 1% ~ (p=0.743 n=20+20) Compiler 45.7s ± 1% 45.7s ± 1% ~ (p=0.482 n=19+20) SSA 109s ± 1% 109s ± 2% ~ (p=0.800 n=18+20) Flate 1.67s ± 3% 1.67s ± 3% ~ (p=0.598 n=19+18) GoParser 2.15s ± 4% 2.13s ± 3% ~ (p=0.153 n=20+20) Reflect 5.95s ± 2% 5.95s ± 2% ~ (p=0.961 n=19+20) Tar 2.93s ± 2% 2.92s ± 3% ~ (p=0.242 n=20+19) XML 3.02s ± 3% 3.04s ± 3% ~ (p=0.233 n=19+18) [Geo mean] 5.74s 5.74s -0.04% name old text-bytes new text-bytes delta HelloSize 588kB ± 0% 588kB ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.46kB ± 0% 5.46kB ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 72.9kB ± 0% 72.9kB ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 1.03MB ± 0% 1.03MB ± 0% ~ (all equal) 3. The go1 benchmark showed little change in total. name old time/op new time/op delta BinaryTree17-4 41.8s ± 1% 41.8s ± 1% ~ (p=0.388 n=40+39) Fannkuch11-4 24.1s ± 1% 24.1s ± 1% ~ (p=0.077 n=40+40) FmtFprintfEmpty-4 834ns ± 1% 831ns ± 1% -0.31% (p=0.002 n=40+37) FmtFprintfString-4 1.34µs ± 1% 1.34µs ± 0% ~ (p=0.387 n=40+40) FmtFprintfInt-4 1.44µs ± 1% 1.44µs ± 1% ~ (p=0.421 n=40+40) FmtFprintfIntInt-4 2.09µs ± 0% 2.09µs ± 1% ~ (p=0.589 n=40+39) FmtFprintfPrefixedInt-4 2.32µs ± 1% 2.33µs ± 1% +0.15% (p=0.001 n=40+40) FmtFprintfFloat-4 4.51µs ± 0% 4.44µs ± 1% -1.50% (p=0.000 n=40+40) FmtManyArgs-4 7.94µs ± 0% 7.97µs ± 0% +0.36% (p=0.001 n=32+40) GobDecode-4 104ms ± 1% 102ms ± 2% -1.27% (p=0.000 n=39+37) GobEncode-4 90.5ms ± 1% 90.9ms ± 2% +0.40% (p=0.006 n=37+40) Gzip-4 4.10s ± 2% 4.08s ± 1% -0.30% (p=0.004 n=40+40) Gunzip-4 603ms ± 0% 602ms ± 1% ~ (p=0.303 n=37+40) HTTPClientServer-4 672µs ± 3% 658µs ± 2% -2.08% (p=0.000 n=39+37) JSONEncode-4 238ms ± 1% 239ms ± 0% +0.26% (p=0.001 n=40+25) JSONDecode-4 884ms ± 1% 885ms ± 1% +0.16% (p=0.012 n=40+40) Mandelbrot200-4 49.3ms ± 0% 49.3ms ± 0% ~ (p=0.588 n=40+38) GoParse-4 46.3ms ± 1% 46.4ms ± 2% ~ (p=0.487 n=40+40) RegexpMatchEasy0_32-4 1.28µs ± 1% 1.28µs ± 0% +0.12% (p=0.003 n=40+40) RegexpMatchEasy0_1K-4 7.78µs ± 5% 7.78µs ± 4% ~ (p=0.825 n=40+40) RegexpMatchEasy1_32-4 1.29µs ± 1% 1.29µs ± 0% ~ (p=0.659 n=40+40) RegexpMatchEasy1_1K-4 10.3µs ± 3% 10.4µs ± 2% ~ (p=0.266 n=40+40) RegexpMatchMedium_32-4 2.05µs ± 1% 2.05µs ± 0% -0.18% (p=0.002 n=40+28) RegexpMatchMedium_1K-4 533µs ± 1% 534µs ± 1% ~ (p=0.397 n=37+40) RegexpMatchHard_32-4 28.9µs ± 1% 28.9µs ± 1% -0.22% (p=0.002 n=40+40) RegexpMatchHard_1K-4 868µs ± 1% 870µs ± 1% +0.21% (p=0.015 n=40+40) Revcomp-4 67.3ms ± 1% 67.2ms ± 2% ~ (p=0.262 n=38+39) Template-4 1.07s ± 1% 1.07s ± 1% ~ (p=0.276 n=40+40) TimeParse-4 7.16µs ± 1% 7.16µs ± 1% ~ (p=0.610 n=39+40) TimeFormat-4 13.3µs ± 1% 13.3µs ± 1% ~ (p=0.617 n=38+40) [Geo mean] 720µs 719µs -0.13% name old speed new speed delta GobDecode-4 7.39MB/s ± 1% 7.49MB/s ± 2% +1.25% (p=0.000 n=39+38) GobEncode-4 8.48MB/s ± 1% 8.45MB/s ± 2% -0.40% (p=0.005 n=37+40) Gzip-4 4.74MB/s ± 2% 4.75MB/s ± 1% +0.30% (p=0.018 n=40+40) Gunzip-4 32.2MB/s ± 0% 32.2MB/s ± 1% ~ (p=0.272 n=36+40) JSONEncode-4 8.15MB/s ± 1% 8.13MB/s ± 0% -0.26% (p=0.003 n=40+25) JSONDecode-4 2.19MB/s ± 1% 2.19MB/s ± 1% ~ (p=0.676 n=40+40) GoParse-4 1.25MB/s ± 2% 1.25MB/s ± 2% ~ (p=0.823 n=40+40) RegexpMatchEasy0_32-4 25.1MB/s ± 1% 25.1MB/s ± 0% -0.12% (p=0.006 n=40+40) RegexpMatchEasy0_1K-4 132MB/s ± 5% 132MB/s ± 5% ~ (p=0.821 n=40+40) RegexpMatchEasy1_32-4 24.7MB/s ± 1% 24.7MB/s ± 0% ~ (p=0.630 n=40+40) RegexpMatchEasy1_1K-4 99.1MB/s ± 3% 98.8MB/s ± 2% ~ (p=0.268 n=40+40) RegexpMatchMedium_32-4 487kB/s ± 2% 490kB/s ± 0% +0.51% (p=0.001 n=40+40) RegexpMatchMedium_1K-4 1.92MB/s ± 1% 1.92MB/s ± 1% ~ (p=0.208 n=39+40) RegexpMatchHard_32-4 1.11MB/s ± 1% 1.11MB/s ± 0% +0.36% (p=0.000 n=40+33) RegexpMatchHard_1K-4 1.18MB/s ± 1% 1.18MB/s ± 1% ~ (p=0.207 n=40+37) Revcomp-4 37.8MB/s ± 1% 37.8MB/s ± 2% ~ (p=0.276 n=38+39) Template-4 1.82MB/s ± 1% 1.81MB/s ± 1% ~ (p=0.122 n=38+40) [Geo mean] 6.81MB/s 6.81MB/s +0.06% fixes #19843 Change-Id: Ief3a0c2b15f59d40c7b40f2784eeb71196685b59 Reviewed-on: https://go-review.googlesource.com/61150 Reviewed-by: Cherry Zhang <cherryyz@google.com> Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
This commit is contained in:
Ben Shi
Cherry Zhang
parent
538b4bab3d
commit
2899c3e8cb
5 changed files with 285 additions and 9 deletions
|
|
@ -184,6 +184,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
|
|||
ssa.OpARMSUBD,
|
||||
ssa.OpARMMULF,
|
||||
ssa.OpARMMULD,
|
||||
ssa.OpARMNMULF,
|
||||
ssa.OpARMNMULD,
|
||||
ssa.OpARMDIVF,
|
||||
ssa.OpARMDIVD:
|
||||
r := v.Reg()
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue