-
Ben Shi authored
MOVBS/MOVBU/MOVHS/MOVHU can be optimized with a single instruction on ARMv6 and ARMv7, instead of a pair of left/right shifts. The benchmark tests show big improvement in special cases and a little improvement in total. 1. A special case gets about 29% improvement. name old time/op new time/op delta TypePro-4 3.81ms ± 1% 2.71ms ± 1% -28.97% (p=0.000 n=26+25) The source code of this case can be found at https://github.com/benshi001/ugo1/blob/master/typepromotion_test.go 2. There is a little improvement in the go1 benchmark, excluding the noise. name old time/op new time/op delta BinaryTree17-4 42.1s ± 3% 42.1s ± 2% ~ (p=0.883 n=28+30) Fannkuch11-4 24.3s ± 4% 24.7s ± 7% +1.64% (p=0.026 n=30+30) FmtFprintfEmpty-4 833ns ± 2% 835ns ± 2% ~ (p=0.371 n=26+28) FmtFprintfString-4 1.36µs ± 3% 1.35µs ± 1% ~ (p=0.202 n=26+23) FmtFprintfInt-4 1.42µs ± 3% 1.43µs ± 1% +0.66% (p=0.000 n=26+27) FmtFprintfIntInt-4 2.10µs ± 1% 2.10µs ± 2% ~ (p=0.104 n=25+26) FmtFprintfPrefixedInt-4 2.37µs ± 2% 2.33µs ± 1% -1.75% (p=0.000 n=25+28) FmtFprintfFloat-4 4.50µs ± 0% 4.37µs ± 1% -2.81% (p=0.000 n=23+25) FmtManyArgs-4 8.08µs ± 0% 8.13µs ± 3% ~ (p=0.160 n=23+26) GobDecode-4 102ms ± 4% 103ms ± 4% +1.08% (p=0.001 n=28+26) GobEncode-4 96.0ms ± 2% 95.2ms ± 3% -0.81% (p=0.000 n=24+25) Gzip-4 4.17s ± 3% 4.11s ± 2% -1.45% (p=0.000 n=25+25) Gunzip-4 597ms ± 2% 594ms ± 2% -0.57% (p=0.000 n=24+26) HTTPClientServer-4 708µs ± 4% 708µs ± 4% ~ (p=0.852 n=28+28) JSONEncode-4 241ms ± 1% 245ms ± 3% +1.62% (p=0.000 n=27+28) JSONDecode-4 906ms ± 3% 889ms ± 3% -1.85% (p=0.000 n=23+24) Mandelbrot200-4 41.8ms ± 1% 41.8ms ± 1% ~ (p=0.929 n=25+24) GoParse-4 47.1ms ± 2% 45.3ms ± 4% -3.80% (p=0.000 n=28+24) RegexpMatchEasy0_32-4 1.27µs ± 2% 1.28µs ± 1% +0.77% (p=0.000 n=26+28) RegexpMatchEasy0_1K-4 8.08µs ± 9% 7.83µs ±10% -3.10% (p=0.012 n=26+26) RegexpMatchEasy1_32-4 1.29µs ± 5% 1.29µs ± 2% ~ (p=0.301 n=26+29) RegexpMatchEasy1_1K-4 10.5µs ± 4% 10.3µs ± 5% -1.95% (p=0.003 n=26+26) RegexpMatchMedium_32-4 1.94µs ± 1% 1.95µs ± 1% ~ (p=0.251 n=24+27) RegexpMatchMedium_1K-4 502µs ± 2% 502µs ± 2% ~ (p=0.336 n=25+28) RegexpMatchHard_32-4 26.7µs ± 1% 26.6µs ± 3% ~ (p=0.454 n=27+26) RegexpMatchHard_1K-4 801µs ± 3% 799µs ± 2% ~ (p=0.097 n=24+26) Revcomp-4 73.5ms ± 5% 73.2ms ± 3% ~ (p=0.240 n=26+26) Template-4 1.07s ± 2% 1.05s ± 1% -2.39% (p=0.000 n=26+24) TimeParse-4 6.87µs ± 1% 6.85µs ± 1% ~ (p=0.094 n=28+23) TimeFormat-4 13.4µs ± 1% 13.4µs ± 1% ~ (p=0.664 n=25+29) [Geo mean] 717µs 713µs -0.54% name old speed new speed delta GobDecode-4 7.52MB/s ± 4% 7.44MB/s ± 4% -1.10% (p=0.001 n=28+26) GobEncode-4 7.99MB/s ± 2% 8.06MB/s ± 3% +0.81% (p=0.000 n=24+25) Gzip-4 4.66MB/s ± 3% 4.72MB/s ± 2% +1.43% (p=0.000 n=25+25) Gunzip-4 32.5MB/s ± 2% 32.7MB/s ± 2% +0.56% (p=0.001 n=24+26) JSONEncode-4 8.04MB/s ± 1% 7.92MB/s ± 3% -1.59% (p=0.000 n=27+28) JSONDecode-4 2.14MB/s ± 3% 2.18MB/s ± 3% +1.90% (p=0.000 n=23+24) GoParse-4 1.23MB/s ± 3% 1.28MB/s ± 4% +4.23% (p=0.000 n=30+24) RegexpMatchEasy0_32-4 25.2MB/s ± 2% 25.0MB/s ± 1% -0.76% (p=0.000 n=26+28) RegexpMatchEasy0_1K-4 127MB/s ± 8% 131MB/s ± 9% +3.29% (p=0.012 n=26+26) RegexpMatchEasy1_32-4 24.8MB/s ± 5% 24.8MB/s ± 2% ~ (p=0.339 n=26+29) RegexpMatchEasy1_1K-4 97.9MB/s ± 4% 99.8MB/s ± 5% +1.98% (p=0.004 n=26+26) RegexpMatchMedium_32-4 514kB/s ± 3% 515kB/s ± 3% ~ (p=0.391 n=28+28) RegexpMatchMedium_1K-4 2.04MB/s ± 2% 2.04MB/s ± 2% ~ (p=0.517 n=25+28) RegexpMatchHard_32-4 1.20MB/s ± 3% 1.20MB/s ± 3% ~ (p=0.203 n=28+28) RegexpMatchHard_1K-4 1.28MB/s ± 3% 1.28MB/s ± 2% ~ (p=0.499 n=24+26) Revcomp-4 34.6MB/s ± 4% 34.7MB/s ± 3% ~ (p=0.245 n=26+26) Template-4 1.81MB/s ± 2% 1.85MB/s ± 3% +2.30% (p=0.000 n=26+25) [Geo mean] 6.82MB/s 6.88MB/s +0.84% fixes #20653 Change-Id: Ief0d6e726e517e51ae511325b21ee72598e759ff Reviewed-on: https://go-review.googlesource.com/71992Reviewed-by:
Cherry Zhang <cherryyz@google.com> Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
78ea9a71
