-
- Downloads
math/big: implement mulAddVWW in riscv64 assembly
This provides an assembly implementation of mulAddVWW for riscv64, processing up to four words per loop, resulting in a significant performance gain. On a StarFive VisionFive 2: │ muladdvww.1 │ muladdvww.2 │ │ sec/op │ sec/op vs base │ MulAddVWW/1-4 68.18n ± 0% 65.49n ± 0% -3.95% (p=0.000 n=10) MulAddVWW/2-4 82.81n ± 0% 78.85n ± 0% -4.78% (p=0.000 n=10) MulAddVWW/3-4 97.49n ± 0% 72.18n ± 0% -25.96% (p=0.000 n=10) MulAddVWW/4-4 112.20n ± 0% 85.54n ± 0% -23.76% (p=0.000 n=10) MulAddVWW/5-4 126.90n ± 0% 98.90n ± 0% -22.06% (p=0.000 n=10) MulAddVWW/10-4 200.3n ± 0% 144.3n ± 0% -27.96% (p=0.000 n=10) MulAddVWW/100-4 1532.0n ± 0% 860.0n ± 0% -43.86% (p=0.000 n=10) MulAddVWW/1000-4 14.757µ ± 0% 8.076µ ± 0% -45.27% (p=0.000 n=10) MulAddVWW/10000-4 204.0µ ± 0% 137.1µ ± 0% -32.77% (p=0.000 n=10) MulAddVWW/100000-4 2.066m ± 0% 1.382m ± 0% -33.12% (p=0.000 n=10) geomean 1.311µ 950.0n -27.51% │ muladdvww.1 │ muladdvww.2 │ │ B/s │ B/s vs base │ MulAddVWW/1-4 895.1Mi ± 0% 932.0Mi ± 0% +4.11% (p=0.000 n=10) MulAddVWW/2-4 1.440Gi ± 0% 1.512Gi ± 0% +5.02% (p=0.000 n=10) MulAddVWW/3-4 1.834Gi ± 0% 2.477Gi ± 0% +35.07% (p=0.000 n=10) MulAddVWW/4-4 2.125Gi ± 0% 2.787Gi ± 0% +31.15% (p=0.000 n=10) MulAddVWW/5-4 2.349Gi ± 0% 3.013Gi ± 0% +28.28% (p=0.000 n=10) MulAddVWW/10-4 2.975Gi ± 0% 4.130Gi ± 0% +38.79% (p=0.000 n=10) MulAddVWW/100-4 3.891Gi ± 0% 6.930Gi ± 0% +78.11% (p=0.000 n=10) MulAddVWW/1000-4 4.039Gi ± 0% 7.380Gi ± 0% +82.72% (p=0.000 n=10) MulAddVWW/10000-4 2.922Gi ± 0% 4.346Gi ± 0% +48.74% (p=0.000 n=10) MulAddVWW/100000-4 2.884Gi ± 0% 4.313Gi ± 0% +49.52% (p=0.000 n=10) geomean 2.321Gi 3.202Gi +37.95% Change-Id: If08191607913ce5c7641f34bae8fa5c9dfb44777 Reviewed-on: https://go-review.googlesource.com/c/go/+/595399 Reviewed-by:Dmitri Shuralyov <dmitshur@google.com> Reviewed-by:
Mark Ryan <markdryan@rivosinc.com> Reviewed-by:
Cherry Mui <cherryyz@google.com> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by:
Meng Zhuo <mengzhuo1203@gmail.com>
Loading
Please register or sign in to comment