]> Cypherpunks repositories - gostls13.git/commit
math/big: implement subVV in riscv64 assembly
authorJoel Sing <joel@sing.id.au>
Thu, 27 Jun 2024 10:02:04 +0000 (20:02 +1000)
committerJoel Sing <joel@sing.id.au>
Thu, 15 Aug 2024 12:06:59 +0000 (12:06 +0000)
commit1419d0e92511735adb319b310155a1965dc20eb8
treea94aea83d6024cff9875da40a285e5a83c99c390
parent49adb8e4602d1a525e5dcb841947c8645fd95c0b
math/big: implement subVV in riscv64 assembly

This provides an assembly implementation of subVV for riscv64,
processing up to four words per loop, resulting in a significant
performance gain.

On a StarFive VisionFive 2:

               │   subvv.1    │               subvv.2               │
               │    sec/op    │   sec/op     vs base                │
SubVV/1-4         73.46n ± 0%   48.08n ± 0%  -34.55% (p=0.000 n=10)
SubVV/2-4         88.13n ± 0%   58.76n ± 0%  -33.33% (p=0.000 n=10)
SubVV/3-4        102.80n ± 0%   69.45n ± 0%  -32.44% (p=0.000 n=10)
SubVV/4-4        117.50n ± 0%   72.11n ± 0%  -38.63% (p=0.000 n=10)
SubVV/5-4        132.20n ± 0%   82.80n ± 0%  -37.37% (p=0.000 n=10)
SubVV/10-4        216.3n ± 0%   126.9n ± 0%  -41.33% (p=0.000 n=10)
SubVV/100-4      1659.0n ± 0%   886.5n ± 0%  -46.56% (p=0.000 n=10)
SubVV/1000-4     16.089µ ± 0%   8.401µ ± 0%  -47.78% (p=0.000 n=10)
SubVV/10000-4     244.7µ ± 0%   176.8µ ± 0%  -27.74% (p=0.000 n=10)
SubVV/100000-4    2.562m ± 0%   1.871m ± 0%  -26.96% (p=0.000 n=10)
geomean           1.436µ        904.4n       -37.04%

               │   subvv.1    │                subvv.2                │
               │     B/s      │      B/s       vs base                │
SubVV/1-4        830.9Mi ± 0%   1269.5Mi ± 0%  +52.79% (p=0.000 n=10)
SubVV/2-4        1.353Gi ± 0%    2.029Gi ± 0%  +49.99% (p=0.000 n=10)
SubVV/3-4        1.739Gi ± 0%    2.575Gi ± 0%  +48.06% (p=0.000 n=10)
SubVV/4-4        2.029Gi ± 0%    3.306Gi ± 0%  +62.96% (p=0.000 n=10)
SubVV/5-4        2.254Gi ± 0%    3.600Gi ± 0%  +59.67% (p=0.000 n=10)
SubVV/10-4       2.755Gi ± 0%    4.699Gi ± 0%  +70.53% (p=0.000 n=10)
SubVV/100-4      3.594Gi ± 0%    6.723Gi ± 0%  +87.08% (p=0.000 n=10)
SubVV/1000-4     3.705Gi ± 0%    7.095Gi ± 0%  +91.52% (p=0.000 n=10)
SubVV/10000-4    2.436Gi ± 0%    3.372Gi ± 0%  +38.39% (p=0.000 n=10)
SubVV/100000-4   2.327Gi ± 0%    3.185Gi ± 0%  +36.91% (p=0.000 n=10)
geomean          2.118Gi         3.364Gi       +58.84%

Change-Id: I361cb3f4195b27a9f1e9486c9e1fdbeaa94d32b4
Reviewed-on: https://go-review.googlesource.com/c/go/+/595396
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Meng Zhuo <mengzhuo1203@gmail.com>
Reviewed-by: Mark Ryan <markdryan@rivosinc.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Carlos Amedee <carlos@golang.org>
src/math/big/arith_riscv64.s