zstd: Optimize seqdeq amd64 asm #636
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
copyMemoryPrecise now generates a loop over 16-byte blocks with a single branchless 16-byte fixup after it. This is a tiny bit faster on the whole and quite a bit faster for some inputs. Benchmark results on Intel Core i7-3770K: name old speed new speed delta Decoder_DecoderSmall/kppkn.gtb.zst-8 369MB/s ± 0% 374MB/s ± 1% +1.56% (p=0.008 n=5+5) Decoder_DecoderSmall/geo.protodata.zst-8 977MB/s ± 0% 1056MB/s ± 1% +8.17% (p=0.008 n=5+5) Decoder_DecoderSmall/plrabn12.txt.zst-8 291MB/s ± 0% 289MB/s ± 0% -0.74% (p=0.008 n=5+5) Decoder_DecoderSmall/lcet10.txt.zst-8 329MB/s ± 1% 333MB/s ± 0% +1.23% (p=0.008 n=5+5) Decoder_DecoderSmall/asyoulik.txt.zst-8 310MB/s ± 0% 310MB/s ± 1% ~ (p=1.000 n=5+5) Decoder_DecoderSmall/alice29.txt.zst-8 291MB/s ± 0% 291MB/s ± 1% ~ (p=0.421 n=5+5) Decoder_DecoderSmall/html_x_4.zst-8 2.07GB/s ± 0% 2.15GB/s ± 2% +4.05% (p=0.008 n=5+5) Decoder_DecoderSmall/paper-100k.pdf.zst-8 3.58GB/s ± 3% 3.74GB/s ± 1% +4.31% (p=0.008 n=5+5) Decoder_DecoderSmall/fireworks.jpeg.zst-8 8.57GB/s ± 0% 8.60GB/s ± 0% ~ (p=0.056 n=5+5) Decoder_DecoderSmall/urls.10K.zst-8 474MB/s ± 1% 507MB/s ± 1% +6.80% (p=0.008 n=5+5) Decoder_DecoderSmall/html.zst-8 745MB/s ± 0% 803MB/s ± 0% +7.68% (p=0.008 n=5+5) Decoder_DecoderSmall/comp-data.bin.zst-8 399MB/s ± 1% 400MB/s ± 0% ~ (p=0.841 n=5+5) Decoder_DecodeAll/kppkn.gtb.zst-8 521MB/s ± 0% 521MB/s ± 0% ~ (p=0.841 n=5+5) Decoder_DecodeAll/geo.protodata.zst-8 1.27GB/s ± 1% 1.29GB/s ± 0% +1.19% (p=0.008 n=5+5) Decoder_DecodeAll/plrabn12.txt.zst-8 429MB/s ± 0% 427MB/s ± 0% -0.51% (p=0.032 n=5+5) Decoder_DecodeAll/lcet10.txt.zst-8 435MB/s ± 0% 439MB/s ± 0% +0.94% (p=0.008 n=5+5) Decoder_DecodeAll/asyoulik.txt.zst-8 438MB/s ± 0% 436MB/s ± 0% -0.39% (p=0.008 n=5+5) Decoder_DecodeAll/alice29.txt.zst-8 423MB/s ± 0% 420MB/s ± 1% -0.72% (p=0.008 n=5+5) Decoder_DecodeAll/html_x_4.zst-8 1.59GB/s ± 0% 1.59GB/s ± 1% +0.54% (p=0.032 n=5+5) Decoder_DecodeAll/paper-100k.pdf.zst-8 4.53GB/s ± 1% 4.54GB/s ± 1% ~ (p=0.310 n=5+5) Decoder_DecodeAll/fireworks.jpeg.zst-8 9.64GB/s ± 1% 9.57GB/s ± 0% ~ (p=0.151 n=5+5) Decoder_DecodeAll/urls.10K.zst-8 683MB/s ± 0% 681MB/s ± 0% ~ (p=0.056 n=5+5) Decoder_DecodeAll/html.zst-8 1.04GB/s ± 1% 1.06GB/s ± 0% +1.77% (p=0.008 n=5+5) Decoder_DecodeAll/comp-data.bin.zst-8 398MB/s ± 1% 399MB/s ± 1% ~ (p=1.000 n=5+5) Decoder_DecodeAllFiles/Mark.Twain-Tom.Sawyer.txt/fastest-8 439MB/s ± 0% 437MB/s ± 0% -0.39% (p=0.016 n=5+5) Decoder_DecodeAllFiles/Mark.Twain-Tom.Sawyer.txt/default-8 448MB/s ± 0% 448MB/s ± 0% ~ (p=0.841 n=5+5) Decoder_DecodeAllFiles/Mark.Twain-Tom.Sawyer.txt/better-8 478MB/s ± 0% 477MB/s ± 0% ~ (p=0.151 n=5+5) Decoder_DecodeAllFiles/Mark.Twain-Tom.Sawyer.txt/best-8 463MB/s ± 0% 460MB/s ± 0% -0.57% (p=0.008 n=5+5) Decoder_DecodeAllFiles/e.txt/fastest-8 9.62GB/s ± 3% 9.66GB/s ± 1% ~ (p=0.841 n=5+5) Decoder_DecodeAllFiles/e.txt/default-8 394MB/s ± 0% 395MB/s ± 0% ~ (p=0.056 n=5+5) Decoder_DecodeAllFiles/e.txt/better-8 438MB/s ± 0% 442MB/s ± 0% +0.82% (p=0.008 n=5+5) Decoder_DecodeAllFiles/e.txt/best-8 501MB/s ± 0% 506MB/s ± 0% +1.07% (p=0.008 n=5+5) Decoder_DecodeAllFiles/fse-artifact3.bin/fastest-8 1.04GB/s ± 0% 1.05GB/s ± 1% ~ (p=0.056 n=5+5) Decoder_DecodeAllFiles/fse-artifact3.bin/default-8 1.20GB/s ± 1% 1.20GB/s ± 1% ~ (p=0.095 n=5+5) Decoder_DecodeAllFiles/fse-artifact3.bin/better-8 1.01GB/s ± 0% 1.00GB/s ± 1% -0.82% (p=0.008 n=5+5) Decoder_DecodeAllFiles/fse-artifact3.bin/best-8 386MB/s ± 0% 383MB/s ± 0% -0.57% (p=0.008 n=5+5) Decoder_DecodeAllFiles/gettysburg.txt/fastest-8 271MB/s ± 1% 275MB/s ± 1% +1.59% (p=0.008 n=5+5) Decoder_DecodeAllFiles/gettysburg.txt/default-8 224MB/s ± 1% 223MB/s ± 1% ~ (p=0.222 n=5+5) Decoder_DecodeAllFiles/gettysburg.txt/better-8 228MB/s ± 0% 226MB/s ± 0% -0.89% (p=0.008 n=5+5) Decoder_DecodeAllFiles/gettysburg.txt/best-8 223MB/s ± 1% 221MB/s ± 1% -1.03% (p=0.016 n=5+5) Decoder_DecodeAllFiles/html.txt/fastest-8 592MB/s ± 1% 611MB/s ± 0% +3.20% (p=0.008 n=5+5) Decoder_DecodeAllFiles/html.txt/default-8 597MB/s ± 0% 607MB/s ± 0% +1.71% (p=0.008 n=5+5) Decoder_DecodeAllFiles/html.txt/better-8 623MB/s ± 0% 633MB/s ± 0% +1.57% (p=0.008 n=5+5) Decoder_DecodeAllFiles/html.txt/best-8 603MB/s ± 0% 610MB/s ± 0% +1.25% (p=0.008 n=5+5) Decoder_DecodeAllFiles/pi.txt/fastest-8 9.59GB/s ± 1% 9.70GB/s ± 1% +1.16% (p=0.032 n=5+5) Decoder_DecodeAllFiles/pi.txt/default-8 391MB/s ± 0% 393MB/s ± 0% +0.62% (p=0.008 n=5+5) Decoder_DecodeAllFiles/pi.txt/better-8 437MB/s ± 1% 441MB/s ± 2% ~ (p=0.087 n=5+5) Decoder_DecodeAllFiles/pi.txt/best-8 501MB/s ± 0% 507MB/s ± 0% +1.22% (p=0.008 n=5+5) Decoder_DecodeAllFiles/pngdata.bin/fastest-8 1.66GB/s ± 1% 1.70GB/s ± 0% +2.49% (p=0.008 n=5+5) Decoder_DecodeAllFiles/pngdata.bin/default-8 1.49GB/s ± 0% 1.51GB/s ± 0% +1.18% (p=0.008 n=5+5) Decoder_DecodeAllFiles/pngdata.bin/better-8 1.87GB/s ± 0% 1.90GB/s ± 1% ~ (p=0.056 n=5+5) Decoder_DecodeAllFiles/pngdata.bin/best-8 1.44GB/s ± 1% 1.46GB/s ± 0% +1.75% (p=0.008 n=5+5) Decoder_DecodeAllFiles/sharnd.out/fastest-8 9.64GB/s ± 1% 9.66GB/s ± 1% ~ (p=0.841 n=5+5) Decoder_DecodeAllFiles/sharnd.out/default-8 9.70GB/s ± 1% 9.70GB/s ± 2% ~ (p=1.000 n=5+5) Decoder_DecodeAllFiles/sharnd.out/better-8 9.71GB/s ± 1% 9.79GB/s ± 1% ~ (p=0.151 n=5+5) Decoder_DecodeAllFiles/sharnd.out/best-8 9.76GB/s ± 0% 9.80GB/s ± 0% ~ (p=0.056 n=5+5) Decoder_DecodeAllFilesP/Mark.Twain-Tom.Sawyer.txt/fastest-8 1.85GB/s ± 0% 1.85GB/s ± 0% -0.31% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/Mark.Twain-Tom.Sawyer.txt/default-8 1.86GB/s ± 0% 1.85GB/s ± 0% -0.47% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/Mark.Twain-Tom.Sawyer.txt/better-8 2.00GB/s ± 0% 2.00GB/s ± 0% -0.32% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/Mark.Twain-Tom.Sawyer.txt/best-8 1.93GB/s ± 0% 1.93GB/s ± 0% -0.22% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/e.txt/fastest-8 37.7GB/s ± 0% 37.5GB/s ± 0% -0.38% (p=0.016 n=5+5) Decoder_DecodeAllFilesP/e.txt/default-8 1.68GB/s ± 0% 1.69GB/s ± 0% +0.55% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/e.txt/better-8 1.91GB/s ± 0% 1.92GB/s ± 0% +0.96% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/e.txt/best-8 2.22GB/s ± 0% 2.25GB/s ± 0% +1.50% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/fse-artifact3.bin/fastest-8 5.18GB/s ± 0% 5.05GB/s ± 2% -2.50% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/fse-artifact3.bin/default-8 5.50GB/s ± 1% 5.34GB/s ± 1% -2.86% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/fse-artifact3.bin/better-8 5.11GB/s ± 0% 5.14GB/s ± 0% +0.57% (p=0.016 n=5+5) Decoder_DecodeAllFilesP/fse-artifact3.bin/best-8 2.36GB/s ± 0% 2.37GB/s ± 0% +0.20% (p=0.032 n=5+5) Decoder_DecodeAllFilesP/gettysburg.txt/fastest-8 1.16GB/s ± 0% 1.16GB/s ± 0% ~ (p=0.056 n=5+5) Decoder_DecodeAllFilesP/gettysburg.txt/default-8 1.09GB/s ± 0% 1.08GB/s ± 0% -1.19% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/gettysburg.txt/better-8 1.09GB/s ± 0% 1.08GB/s ± 1% -0.96% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/gettysburg.txt/best-8 1.03GB/s ± 3% 1.02GB/s ± 0% ~ (p=0.151 n=5+5) Decoder_DecodeAllFilesP/html.txt/fastest-8 2.50GB/s ± 1% 2.56GB/s ± 0% +2.39% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/html.txt/default-8 2.51GB/s ± 0% 2.55GB/s ± 0% +1.69% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/html.txt/better-8 2.61GB/s ± 0% 2.66GB/s ± 0% +1.93% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/html.txt/best-8 2.53GB/s ± 0% 2.56GB/s ± 0% +1.13% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/pi.txt/fastest-8 37.8GB/s ± 0% 37.6GB/s ± 0% -0.44% (p=0.016 n=5+5) Decoder_DecodeAllFilesP/pi.txt/default-8 1.67GB/s ± 0% 1.68GB/s ± 0% +0.61% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/pi.txt/better-8 1.91GB/s ± 0% 1.93GB/s ± 0% +0.82% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/pi.txt/best-8 2.23GB/s ± 0% 2.26GB/s ± 0% +1.35% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/pngdata.bin/fastest-8 6.99GB/s ± 0% 7.00GB/s ± 0% ~ (p=0.690 n=5+5) Decoder_DecodeAllFilesP/pngdata.bin/default-8 6.88GB/s ± 0% 6.87GB/s ± 0% ~ (p=0.222 n=5+5) Decoder_DecodeAllFilesP/pngdata.bin/better-8 8.49GB/s ± 0% 8.44GB/s ± 1% ~ (p=0.310 n=5+5) Decoder_DecodeAllFilesP/pngdata.bin/best-8 6.59GB/s ± 1% 6.53GB/s ± 1% -0.96% (p=0.032 n=5+5) Decoder_DecodeAllFilesP/sharnd.out/fastest-8 37.8GB/s ± 0% 37.5GB/s ± 0% -0.86% (p=0.008 n=5+5) Decoder_DecodeAllFilesP/sharnd.out/default-8 37.9GB/s ± 1% 38.0GB/s ± 1% ~ (p=0.310 n=5+5) Decoder_DecodeAllFilesP/sharnd.out/better-8 37.9GB/s ± 0% 37.8GB/s ± 2% ~ (p=0.841 n=5+5) Decoder_DecodeAllFilesP/sharnd.out/best-8 37.8GB/s ± 0% 38.0GB/s ± 1% ~ (p=0.310 n=5+5) Decoder_DecodeAllParallel/kppkn.gtb.zst-8 2.20GB/s ± 0% 2.20GB/s ± 0% ~ (p=1.000 n=5+5) Decoder_DecodeAllParallel/geo.protodata.zst-8 5.37GB/s ± 0% 5.39GB/s ± 0% +0.35% (p=0.008 n=5+5) Decoder_DecodeAllParallel/plrabn12.txt.zst-8 1.77GB/s ± 0% 1.76GB/s ± 0% -0.19% (p=0.008 n=5+5) Decoder_DecodeAllParallel/lcet10.txt.zst-8 1.90GB/s ± 0% 1.92GB/s ± 0% +0.80% (p=0.008 n=5+5) Decoder_DecodeAllParallel/asyoulik.txt.zst-8 1.83GB/s ± 0% 1.83GB/s ± 0% ~ (p=0.841 n=5+5) Decoder_DecodeAllParallel/alice29.txt.zst-8 1.74GB/s ± 0% 1.74GB/s ± 0% ~ (p=0.548 n=5+5) Decoder_DecodeAllParallel/html_x_4.zst-8 6.55GB/s ± 0% 6.49GB/s ± 0% -0.97% (p=0.008 n=5+5) Decoder_DecodeAllParallel/paper-100k.pdf.zst-8 18.3GB/s ± 0% 18.3GB/s ± 0% ~ (p=0.056 n=5+5) Decoder_DecodeAllParallel/fireworks.jpeg.zst-8 37.4GB/s ± 0% 37.2GB/s ± 1% -0.57% (p=0.016 n=4+5) Decoder_DecodeAllParallel/urls.10K.zst-8 2.97GB/s ± 0% 2.96GB/s ± 0% ~ (p=0.310 n=5+5) Decoder_DecodeAllParallel/html.zst-8 4.42GB/s ± 1% 4.43GB/s ± 0% ~ (p=0.556 n=5+4) Decoder_DecodeAllParallel/comp-data.bin.zst-8 1.69GB/s ± 1% 1.70GB/s ± 0% +0.84% (p=0.008 n=5+5) [Geo mean] 1.77GB/s 1.78GB/s +0.57%
greatroar
force-pushed
the
zstd-copymemoryprecise
branch
from
5d8f037
to
cc3f110
Compare
|
Great stuff. I will run a few tests tomorrow. |
|
@greatroar I can confirm your numbers. Some rather noticeable regressions, but with this I can port some of the s2 memcopy for an even bigger speedup and fixing some of the regression cases. Thanks for the contribution! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
copyMemoryPrecise now generates a loop over 16-byte blocks with a single branchless 16-byte fixup after it. This is a tiny bit faster on the whole and quite a bit faster for some inputs.
Benchmark results on Intel Core i7-3770K