8375b54d44
Rewrite the vector loop to process 64B per iteration, this greatly improves performance on POWER9/POWER10 for large sizes. Likewise, use a similar tricks for sizes >= 8 and <= 64. And, rewrite small comparisons, it's a little slower for 1 byte, but constant time for 1-7 bytes. Thus, it is increasingly faster for 2-7B. Benchmarks results below are from P8/P9 ppc64le (in that order), several additional testcases have been added to test interesting sizes. Likewise, the old variant was padded to the same code size of the new variant to minimize layout related noise: POWER8/ppc64le/linux: name old speed new speed delta Equal/1 110MB/s ± 0% 106MB/s ± 0% -3.26% Equal/2 202MB/s ± 0% 203MB/s ± 0% +0.18% Equal/3 280MB/s ± 0% 319MB/s ± 0% +13.89% Equal/4 350MB/s ± 0% 414MB/s ± 0% +18.27% Equal/5 412MB/s ± 0% 533MB/s ± 0% +29.19% Equal/6 462MB/s ± 0% 620MB/s ± 0% +34.11% Equal/7 507MB/s ± 0% 745MB/s ± 0% +47.02% Equal/8 913MB/s ± 0% 994MB/s ± 0% +8.84% Equal/9 909MB/s ± 0% 1117MB/s ± 0% +22.85% Equal/10 937MB/s ± 0% 1242MB/s ± 0% +32.59% Equal/11 962MB/s ± 0% 1370MB/s ± 0% +42.37% Equal/12 989MB/s ± 0% 1490MB/s ± 0% +50.60% Equal/13 1.01GB/s ± 0% 1.61GB/s ± 0% +60.27% Equal/14 1.02GB/s ± 0% 1.74GB/s ± 0% +71.22% Equal/15 1.03GB/s ± 0% 1.86GB/s ± 0% +81.45% Equal/16 1.60GB/s ± 0% 1.99GB/s ± 0% +24.21% Equal/17 1.54GB/s ± 0% 2.04GB/s ± 0% +32.28% Equal/20 1.48GB/s ± 0% 2.40GB/s ± 0% +62.64% Equal/32 3.58GB/s ± 0% 3.84GB/s ± 0% +7.18% Equal/63 3.74GB/s ± 0% 7.17GB/s ± 0% +91.79% Equal/64 6.35GB/s ± 0% 7.29GB/s ± 0% +14.75% Equal/65 5.85GB/s ± 0% 7.00GB/s ± 0% +19.66% Equal/127 6.74GB/s ± 0% 13.74GB/s ± 0% +103.77% Equal/128 10.6GB/s ± 0% 12.9GB/s ± 0% +21.98% Equal/129 9.66GB/s ± 0% 11.96GB/s ± 0% +23.85% Equal/191 9.12GB/s ± 0% 17.80GB/s ± 0% +95.26% Equal/192 13.4GB/s ± 0% 17.2GB/s ± 0% +28.66% Equal/4K 29.5GB/s ± 0% 37.3GB/s ± 0% +26.39% Equal/4M 22.6GB/s ± 0% 23.1GB/s ± 0% +2.40% Equal/64M 10.6GB/s ± 0% 11.2GB/s ± 0% +5.83% POWER9/ppc64le/linux: name old speed new speed delta Equal/1 122MB/s ± 0% 121MB/s ± 0% -0.94% Equal/2 223MB/s ± 0% 241MB/s ± 0% +8.29% Equal/3 289MB/s ± 0% 362MB/s ± 0% +24.90% Equal/4 366MB/s ± 0% 483MB/s ± 0% +31.82% Equal/5 427MB/s ± 0% 603MB/s ± 0% +41.28% Equal/6 462MB/s ± 0% 723MB/s ± 0% +56.65% Equal/7 509MB/s ± 0% 843MB/s ± 0% +65.57% Equal/8 974MB/s ± 0% 1066MB/s ± 0% +9.46% Equal/9 1.00GB/s ± 0% 1.20GB/s ± 0% +19.53% Equal/10 1.00GB/s ± 0% 1.33GB/s ± 0% +32.81% Equal/11 1.01GB/s ± 0% 1.47GB/s ± 0% +45.28% Equal/12 1.04GB/s ± 0% 1.60GB/s ± 0% +53.46% Equal/13 1.05GB/s ± 0% 1.73GB/s ± 0% +64.67% Equal/14 1.02GB/s ± 0% 1.87GB/s ± 0% +82.93% Equal/15 1.04GB/s ± 0% 2.00GB/s ± 0% +92.07% Equal/16 1.83GB/s ± 0% 2.13GB/s ± 0% +16.58% Equal/17 1.78GB/s ± 0% 2.18GB/s ± 0% +22.65% Equal/20 1.72GB/s ± 0% 2.57GB/s ± 0% +49.24% Equal/32 3.89GB/s ± 0% 4.10GB/s ± 0% +5.53% Equal/63 3.63GB/s ± 0% 7.63GB/s ± 0% +110.45% Equal/64 6.69GB/s ± 0% 7.75GB/s ± 0% +15.84% Equal/65 6.28GB/s ± 0% 7.07GB/s ± 0% +12.46% Equal/127 6.41GB/s ± 0% 13.65GB/s ± 0% +112.95% Equal/128 11.1GB/s ± 0% 14.1GB/s ± 0% +26.56% Equal/129 10.2GB/s ± 0% 11.2GB/s ± 0% +9.44% Equal/191 8.64GB/s ± 0% 16.39GB/s ± 0% +89.75% Equal/192 15.3GB/s ± 0% 17.8GB/s ± 0% +16.31% Equal/4K 24.6GB/s ± 0% 27.8GB/s ± 0% +13.12% Equal/4M 21.1GB/s ± 0% 22.7GB/s ± 0% +7.66% Equal/64M 20.8GB/s ± 0% 22.4GB/s ± 0% +8.06% Change-Id: Ie3c582133d526cc14e8846ef364c44c93eb7b9a2 Reviewed-on: https://go-review.googlesource.com/c/go/+/399976 Reviewed-by: Carlos Amedee <carlos@golang.org> Run-TryBot: Paul Murphy <murp@ibm.com> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Cherry Mui <cherryyz@google.com> Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com> |
||
|---|---|---|
| .github | ||
| api | ||
| doc | ||
| lib/time | ||
| misc | ||
| src | ||
| test | ||
| .gitattributes | ||
| .gitignore | ||
| AUTHORS | ||
| codereview.cfg | ||
| CONTRIBUTING.md | ||
| CONTRIBUTORS | ||
| LICENSE | ||
| PATENTS | ||
| README.md | ||
| SECURITY.md | ||
The Go Programming Language
Go is an open source programming language that makes it easy to build simple, reliable, and efficient software.
Gopher image by Renee French, licensed under Creative Commons 3.0 Attributions license.
Our canonical Git repository is located at https://go.googlesource.com/go. There is a mirror of the repository at https://github.com/golang/go.
Unless otherwise noted, the Go source files are distributed under the BSD-style license found in the LICENSE file.
Download and Install
Binary Distributions
Official binary distributions are available at https://go.dev/dl/.
After downloading a binary release, visit https://go.dev/doc/install for installation instructions.
Install From Source
If a binary distribution is not available for your combination of operating system and architecture, visit https://go.dev/doc/install/source for source installation instructions.
Contributing
Go is the work of thousands of contributors. We appreciate your help!
To contribute, please read the contribution guidelines at https://go.dev/doc/contribute.
Note that the Go project uses the issue tracker for bug reports and proposals only. See https://go.dev/wiki/Questions for a list of places to ask questions about the Go language.