]> Cypherpunks repositories - gostls13.git/commit
internal/bytealg: correct alignment checks for compare/memequal on riscv64
authorJoel Sing <joel@sing.id.au>
Sat, 17 Sep 2022 06:55:57 +0000 (16:55 +1000)
committerJoel Sing <joel@sing.id.au>
Mon, 19 Sep 2022 18:59:50 +0000 (18:59 +0000)
commitfd82718e06a7b8a32becb1751592854d49904075
treebae05ec1b21f4ea2febeaa9f220ac1acc317a419
parentceffdc8545c3155b030de9e91d399dc34bd3c678
internal/bytealg: correct alignment checks for compare/memequal on riscv64

On riscv64 we need 8 byte alignment for 8 byte loads - the existing check
was only ensuring 4 byte alignment, which potentially results in unaligned
loads being performed. Unaligned loads incur a significant performance penality
due to the resulting kernel traps and fix ups.

Adjust BenchmarkCompareBytesBigUnaligned so that this issue would have been
more readily visible.

Updates #50615

name                                 old time/op    new time/op      delta
CompareBytesBigUnaligned/offset=1-4    6.98ms _ 5%      6.84ms _ 3%       ~     (p=0.319 n=5+5)
CompareBytesBigUnaligned/offset=2-4    6.75ms _ 1%      6.99ms _ 4%       ~     (p=0.063 n=5+5)
CompareBytesBigUnaligned/offset=3-4    6.84ms _ 1%      6.74ms _ 1%     -1.48%  (p=0.003 n=5+5)
CompareBytesBigUnaligned/offset=4-4     146ms _ 1%         7ms _ 6%    -95.08%  (p=0.000 n=5+5)
CompareBytesBigUnaligned/offset=5-4    7.05ms _ 5%      6.75ms _ 1%       ~     (p=0.079 n=5+5)
CompareBytesBigUnaligned/offset=6-4    7.11ms _ 5%      6.89ms _ 5%       ~     (p=0.177 n=5+5)
CompareBytesBigUnaligned/offset=7-4    7.14ms _ 5%      6.91ms _ 6%       ~     (p=0.165 n=5+5)

name                                 old speed      new speed        delta
CompareBytesBigUnaligned/offset=1-4   150MB/s _ 5%     153MB/s _ 3%       ~     (p=0.336 n=5+5)
CompareBytesBigUnaligned/offset=2-4   155MB/s _ 1%     150MB/s _ 4%       ~     (p=0.058 n=5+5)
CompareBytesBigUnaligned/offset=3-4   153MB/s _ 1%     156MB/s _ 1%     +1.51%  (p=0.004 n=5+5)
CompareBytesBigUnaligned/offset=4-4  7.16MB/s _ 1%  145.79MB/s _ 6%  +1936.23%  (p=0.000 n=5+5)
CompareBytesBigUnaligned/offset=5-4   149MB/s _ 5%     155MB/s _ 1%       ~     (p=0.078 n=5+5)
CompareBytesBigUnaligned/offset=6-4   148MB/s _ 5%     152MB/s _ 5%       ~     (p=0.175 n=5+5)
CompareBytesBigUnaligned/offset=7-4   147MB/s _ 5%     152MB/s _ 6%       ~     (p=0.160 n=5+5)

Change-Id: I2c859e061919db482318ce63b85b808aa973a9ba
Reviewed-on: https://go-review.googlesource.com/c/go/+/431099
Reviewed-by: Meng Zhuo <mzh@golangcn.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Joel Sing <joel@sing.id.au>
Reviewed-by: Bryan Mills <bcmills@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
src/bytes/compare_test.go
src/internal/bytealg/compare_riscv64.s
src/internal/bytealg/equal_riscv64.s