]> Cypherpunks repositories - gostls13.git/commit
math/big: implement addVV in riscv64 assembly
authorJoel Sing <joel@sing.id.au>
Thu, 27 Jun 2024 09:52:30 +0000 (19:52 +1000)
committerJoel Sing <joel@sing.id.au>
Fri, 2 Aug 2024 14:14:14 +0000 (14:14 +0000)
commit9abd11440c5fa027304d6cda051fc0a30b6b430b
tree52dc6d6e5ab06e502008a0701a0e34f5446b6895
parent11dbbaffe1db00d8726215c3fa56e02d66e78de5
math/big: implement addVV in riscv64 assembly

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

On a StarFive VisionFive 2:

               │   addvv.1    │               addvv.2               │
               │    sec/op    │   sec/op     vs base                │
AddVV/1-4         73.45n ± 0%   48.08n ± 0%  -34.54% (p=0.000 n=10)
AddVV/2-4         88.14n ± 0%   58.76n ± 0%  -33.33% (p=0.000 n=10)
AddVV/3-4        102.80n ± 0%   69.44n ± 0%  -32.45% (p=0.000 n=10)
AddVV/4-4        117.50n ± 0%   72.18n ± 0%  -38.57% (p=0.000 n=10)
AddVV/5-4        132.20n ± 0%   82.79n ± 0%  -37.38% (p=0.000 n=10)
AddVV/10-4        216.3n ± 0%   126.8n ± 0%  -41.35% (p=0.000 n=10)
AddVV/100-4      1659.0n ± 0%   885.2n ± 0%  -46.64% (p=0.000 n=10)
AddVV/1000-4     16.089µ ± 0%   8.400µ ± 0%  -47.79% (p=0.000 n=10)
AddVV/10000-4     245.3µ ± 0%   176.9µ ± 0%  -27.88% (p=0.000 n=10)
AddVV/100000-4    2.537m ± 0%   1.873m ± 0%  -26.17% (p=0.000 n=10)
geomean           1.435µ        904.5n       -36.99%

               │   addvv.1    │                addvv.2                │
               │     B/s      │      B/s       vs base                │
AddVV/1-4        830.9Mi ± 0%   1269.5Mi ± 0%  +52.78% (p=0.000 n=10)
AddVV/2-4        1.353Gi ± 0%    2.029Gi ± 0%  +50.00% (p=0.000 n=10)
AddVV/3-4        1.739Gi ± 0%    2.575Gi ± 0%  +48.09% (p=0.000 n=10)
AddVV/4-4        2.029Gi ± 0%    3.303Gi ± 0%  +62.82% (p=0.000 n=10)
AddVV/5-4        2.254Gi ± 0%    3.600Gi ± 0%  +59.69% (p=0.000 n=10)
AddVV/10-4       2.755Gi ± 0%    4.699Gi ± 0%  +70.54% (p=0.000 n=10)
AddVV/100-4      3.594Gi ± 0%    6.734Gi ± 0%  +87.37% (p=0.000 n=10)
AddVV/1000-4     3.705Gi ± 0%    7.096Gi ± 0%  +91.54% (p=0.000 n=10)
AddVV/10000-4    2.430Gi ± 0%    3.369Gi ± 0%  +38.65% (p=0.000 n=10)
AddVV/100000-4   2.350Gi ± 0%    3.183Gi ± 0%  +35.44% (p=0.000 n=10)
geomean          2.119Gi         3.364Gi       +58.71%

Change-Id: I727b3d9f8ab01eada7270046480b1430d56d0a96
Reviewed-on: https://go-review.googlesource.com/c/go/+/595395
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: M Zhuo <mengzhuo1203@gmail.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Than McIntosh <thanm@google.com>
src/math/big/arith_riscv64.s