mirror of
https://github.com/facebook/zstd
synced 2026-08-22 22:32:07 -04:00
Add more prefetching to decompression.
The benefit is hard to show in benchmarks, but observing the production environment, showed single digit improvement on AMD, and nothing but noise on Intel. Closest aproximation is benchmarks on AMD machines with disabled hardware prefetching and internal dataset: name old time/op new time/op delta BM_DecompressFlat/0/2 [html/level=2] 58.5µs ± 2% 57.8µs ± 1% -1.10% (p=0.000 n=18+19) BM_DecompressFlat/1/2 [urls/level=2] 621µs ± 1% 623µs ± 1% ~ (p=0.068 n=20+20) BM_DecompressFlat/2/2 [jpg/level=2] 3.38µs ± 1% 3.36µs ± 1% ~ (p=0.080 n=19+19) BM_DecompressFlat/3/2 [jpg_200/level=2] 2.81µs ± 1% 2.82µs ± 1% ~ (p=0.311 n=19+19) BM_DecompressFlat/4/2 [pdf/level=2] 11.8µs ± 1% 11.9µs ± 1% +0.60% (p=0.004 n=19+19) BM_DecompressFlat/5/2 [html4/level=2] 66.7µs ± 2% 66.6µs ± 2% ~ (p=0.512 n=20+20) BM_DecompressFlat/6/2 [txt1/level=2] 194µs ± 1% 190µs ± 1% -1.97% (p=0.000 n=19+17) BM_DecompressFlat/7/2 [txt2/level=2] 174µs ± 1% 170µs ± 1% -1.99% (p=0.000 n=19+20) BM_DecompressFlat/8/2 [txt3/level=2] 506µs ± 1% 497µs ± 1% -1.76% (p=0.000 n=18+19) BM_DecompressFlat/9/2 [txt4/level=2] 632µs ± 1% 621µs ± 1% -1.71% (p=0.000 n=18+20) BM_DecompressFlat/10/2 [pb/level=2] 53.7µs ± 2% 52.1µs ± 2% -3.02% (p=0.000 n=19+20) BM_DecompressFlat/11/2 [gaviota/level=2] 218µs ± 4% 218µs ± 1% ~ (p=0.369 n=20+20) BM_DecompressFlat/12/2 [cp/level=2] 28.7µs ± 1% 28.0µs ± 1% -2.29% (p=0.000 n=19+19) BM_DecompressFlat/13/2 [c/level=2] 17.8µs ± 1% 17.5µs ± 1% -1.97% (p=0.000 n=18+20) BM_DecompressFlat/14/2 [lsp/level=2] 7.15µs ± 1% 7.08µs ± 1% -0.96% (p=0.000 n=19+20) BM_DecompressFlat/15/2 [xls/level=2] 1.07ms ± 1% 1.09ms ± 1% +1.51% (p=0.000 n=18+18) BM_DecompressFlat/16/2 [xls_200/level=2] 2.21µs ± 1% 2.24µs ± 1% +1.20% (p=0.000 n=18+16) BM_DecompressFlat/17/2 [bin/level=2] 281µs ± 2% 286µs ± 1% +1.80% (p=0.000 n=19+19) BM_DecompressFlat/18/2 [bin_200/level=2] 268ns ± 3% 265ns ± 2% -1.09% (p=0.005 n=19+20) BM_DecompressFlat/19/2 [sum/level=2] 43.4µs ± 1% 41.9µs ± 2% -3.39% (p=0.000 n=20+19) BM_DecompressFlat/20/2 [man/level=2] 8.31µs ± 1% 8.17µs ± 2% -1.78% (p=0.000 n=20+18) BM_DecompressFlat/0/3 [html/level=3] 55.5µs ± 1% 53.7µs ± 1% -3.27% (p=0.000 n=20+19) BM_DecompressFlat/1/3 [urls/level=3] 631µs ± 2% 631µs ± 1% ~ (p=0.862 n=20+20) BM_DecompressFlat/2/3 [jpg/level=3] 3.38µs ± 2% 3.36µs ± 1% -0.65% (p=0.002 n=20+20) BM_DecompressFlat/3/3 [jpg_200/level=3] 2.83µs ± 2% 2.79µs ± 1% -1.27% (p=0.000 n=20+19) BM_DecompressFlat/4/3 [pdf/level=3] 11.6µs ± 2% 11.7µs ± 1% +0.70% (p=0.000 n=19+19) BM_DecompressFlat/5/3 [html4/level=3] 68.0µs ± 2% 72.1µs ± 3% +6.06% (p=0.000 n=20+20) BM_DecompressFlat/6/3 [txt1/level=3] 221µs ± 1% 216µs ± 1% -2.19% (p=0.000 n=20+18) BM_DecompressFlat/7/3 [txt2/level=3] 160µs ± 1% 157µs ± 2% -1.85% (p=0.000 n=20+20) BM_DecompressFlat/8/3 [txt3/level=3] 509µs ± 2% 496µs ± 1% -2.47% (p=0.000 n=20+18) BM_DecompressFlat/9/3 [txt4/level=3] 669µs ± 1% 655µs ± 1% -2.06% (p=0.000 n=18+19) BM_DecompressFlat/10/3 [pb/level=3] 51.8µs ± 2% 50.8µs ± 2% -1.86% (p=0.000 n=20+20) BM_DecompressFlat/11/3 [gaviota/level=3] 232µs ± 2% 231µs ± 1% ~ (p=0.879 n=20+19) BM_DecompressFlat/12/3 [cp/level=3] 27.5µs ± 1% 27.0µs ± 2% -1.67% (p=0.000 n=18+20) BM_DecompressFlat/13/3 [c/level=3] 15.0µs ± 1% 14.7µs ± 1% -2.51% (p=0.000 n=17+19) BM_DecompressFlat/14/3 [lsp/level=3] 6.85µs ± 1% 6.78µs ± 1% -1.07% (p=0.000 n=19+20) BM_DecompressFlat/15/3 [xls/level=3] 1.06ms ± 1% 1.07ms ± 1% +1.32% (p=0.000 n=20+19) BM_DecompressFlat/16/3 [xls_200/level=3] 2.21µs ± 2% 2.19µs ± 1% -0.55% (p=0.033 n=20+20) BM_DecompressFlat/17/3 [bin/level=3] 312µs ± 2% 315µs ± 1% +1.10% (p=0.000 n=20+18) BM_DecompressFlat/18/3 [bin_200/level=3] 266ns ± 2% 265ns ± 2% -0.65% (p=0.013 n=18+19) BM_DecompressFlat/19/3 [sum/level=3] 42.8µs ± 1% 41.0µs ± 2% -4.20% (p=0.000 n=16+20) BM_DecompressFlat/20/3 [man/level=3] 8.01µs ± 1% 7.89µs ± 1% -1.56% (p=0.000 n=19+20) BM_DecompressFlat/0/5 [html/level=5] 56.2µs ± 2% 55.7µs ± 1% -0.88% (p=0.000 n=20+20) BM_DecompressFlat/1/5 [urls/level=5] 633µs ± 1% 634µs ± 1% ~ (p=0.947 n=20+20) BM_DecompressFlat/2/5 [jpg/level=5] 3.37µs ± 1% 3.36µs ± 1% ~ (p=0.114 n=20+20) BM_DecompressFlat/3/5 [jpg_200/level=5] 2.85µs ± 1% 2.82µs ± 1% -1.22% (p=0.000 n=20+19) BM_DecompressFlat/4/5 [pdf/level=5] 14.8µs ± 2% 14.8µs ± 1% ~ (p=0.397 n=19+17) BM_DecompressFlat/5/5 [html4/level=5] 64.4µs ± 3% 67.5µs ± 3% +4.94% (p=0.000 n=20+19) BM_DecompressFlat/6/5 [txt1/level=5] 191µs ± 2% 188µs ± 1% -1.68% (p=0.000 n=20+19) BM_DecompressFlat/7/5 [txt2/level=5] 196µs ± 1% 192µs ± 2% -2.24% (p=0.000 n=20+20) BM_DecompressFlat/8/5 [txt3/level=5] 520µs ± 1% 510µs ± 1% -1.87% (p=0.000 n=20+20) BM_DecompressFlat/9/5 [txt4/level=5] 697µs ± 1% 682µs ± 1% -2.10% (p=0.000 n=20+19) BM_DecompressFlat/10/5 [pb/level=5] 51.3µs ± 4% 50.2µs ± 2% -2.15% (p=0.000 n=20+20) BM_DecompressFlat/11/5 [gaviota/level=5] 216µs ± 2% 218µs ± 1% +1.02% (p=0.001 n=20+19) BM_DecompressFlat/12/5 [cp/level=5] 29.9µs ± 2% 29.0µs ± 1% -3.01% (p=0.000 n=20+19) BM_DecompressFlat/13/5 [c/level=5] 14.2µs ± 1% 13.9µs ± 1% -1.97% (p=0.000 n=20+19) BM_DecompressFlat/14/5 [lsp/level=5] 6.62µs ± 1% 6.60µs ± 1% -0.43% (p=0.004 n=18+19) BM_DecompressFlat/15/5 [xls/level=5] 1.07ms ± 1% 1.09ms ± 1% +1.68% (p=0.000 n=19+20) BM_DecompressFlat/16/5 [xls_200/level=5] 2.65µs ± 1% 2.64µs ± 1% ~ (p=0.053 n=18+19) BM_DecompressFlat/17/5 [bin/level=5] 309µs ± 2% 312µs ± 1% +0.93% (p=0.004 n=20+20) BM_DecompressFlat/18/5 [bin_200/level=5] 268ns ± 2% 264ns ± 0% -1.66% (p=0.000 n=19+16) BM_DecompressFlat/19/5 [sum/level=5] 48.8µs ± 1% 46.8µs ± 2% -4.21% (p=0.000 n=19+20) BM_DecompressFlat/20/5 [man/level=5] 7.90µs ± 2% 7.73µs ± 1% -2.13% (p=0.000 n=19+20) BM_DecompressFlat/0/7 [html/level=7] 52.2µs ± 2% 51.9µs ± 2% -0.65% (p=0.007 n=18+20) BM_DecompressFlat/1/7 [urls/level=7] 605µs ± 2% 604µs ± 1% ~ (p=0.885 n=19+19) BM_DecompressFlat/2/7 [jpg/level=7] 3.37µs ± 1% 3.36µs ± 1% -0.38% (p=0.021 n=20+20) BM_DecompressFlat/3/7 [jpg_200/level=7] 2.85µs ± 1% 2.81µs ± 1% -1.22% (p=0.000 n=20+19) BM_DecompressFlat/4/7 [pdf/level=7] 13.2µs ± 1% 13.3µs ± 1% ~ (p=0.127 n=20+20) BM_DecompressFlat/5/7 [html4/level=7] 61.0µs ± 3% 64.1µs ± 4% +5.13% (p=0.000 n=17+20) BM_DecompressFlat/6/7 [txt1/level=7] 205µs ± 1% 202µs ± 1% -1.55% (p=0.000 n=18+20) BM_DecompressFlat/7/7 [txt2/level=7] 174µs ± 1% 170µs ± 2% -1.94% (p=0.000 n=19+20) BM_DecompressFlat/8/7 [txt3/level=7] 482µs ± 1% 473µs ± 1% -1.92% (p=0.000 n=20+20) BM_DecompressFlat/9/7 [txt4/level=7] 642µs ± 1% 631µs ± 1% -1.83% (p=0.000 n=18+20) BM_DecompressFlat/10/7 [pb/level=7] 47.1µs ± 2% 45.6µs ± 2% -3.16% (p=0.000 n=18+19) BM_DecompressFlat/11/7 [gaviota/level=7] 221µs ± 3% 222µs ± 1% ~ (p=0.217 n=20+18) BM_DecompressFlat/12/7 [cp/level=7] 28.3µs ± 2% 27.6µs ± 1% -2.38% (p=0.000 n=20+18) BM_DecompressFlat/13/7 [c/level=7] 13.5µs ± 2% 13.2µs ± 1% -2.56% (p=0.000 n=20+20) BM_DecompressFlat/14/7 [lsp/level=7] 6.54µs ± 1% 6.49µs ± 1% -0.85% (p=0.000 n=17+18) BM_DecompressFlat/15/7 [xls/level=7] 1.02ms ± 1% 1.04ms ± 1% +1.49% (p=0.000 n=20+19) BM_DecompressFlat/16/7 [xls_200/level=7] 2.65µs ± 1% 2.64µs ± 2% ~ (p=0.095 n=20+19) BM_DecompressFlat/17/7 [bin/level=7] 276µs ± 3% 280µs ± 3% +1.50% (p=0.000 n=19+20) BM_DecompressFlat/18/7 [bin_200/level=7] 267ns ± 1% 264ns ± 2% -1.13% (p=0.000 n=18+20) BM_DecompressFlat/19/7 [sum/level=7] 44.9µs ± 1% 43.3µs ± 1% -3.53% (p=0.000 n=20+20) BM_DecompressFlat/20/7 [man/level=7] 7.81µs ± 2% 7.63µs ± 1% -2.40% (p=0.000 n=19+19) BM_DecompressFlat/0/9 [html/level=9] 50.6µs ± 1% 49.6µs ± 1% -1.83% (p=0.000 n=19+20) BM_DecompressFlat/1/9 [urls/level=9] 592µs ± 2% 592µs ± 1% ~ (p=0.588 n=19+20) BM_DecompressFlat/2/9 [jpg/level=9] 3.37µs ± 1% 3.36µs ± 2% -0.26% (p=0.038 n=20+19) BM_DecompressFlat/3/9 [jpg_200/level=9] 2.85µs ± 1% 2.81µs ± 1% -1.22% (p=0.000 n=19+18) BM_DecompressFlat/4/9 [pdf/level=9] 12.9µs ± 1% 13.1µs ± 2% +0.87% (p=0.001 n=20+19) BM_DecompressFlat/5/9 [html4/level=9] 59.3µs ± 3% 61.4µs ± 2% +3.55% (p=0.000 n=20+19) BM_DecompressFlat/6/9 [txt1/level=9] 190µs ± 1% 187µs ± 1% -1.65% (p=0.000 n=20+19) BM_DecompressFlat/7/9 [txt2/level=9] 168µs ± 1% 164µs ± 1% -2.02% (p=0.000 n=20+20) BM_DecompressFlat/8/9 [txt3/level=9] 462µs ± 1% 454µs ± 1% -1.74% (p=0.000 n=20+20) BM_DecompressFlat/9/9 [txt4/level=9] 619µs ± 1% 605µs ± 1% -2.31% (p=0.000 n=20+17) BM_DecompressFlat/10/9 [pb/level=9] 45.6µs ± 2% 44.9µs ± 1% -1.42% (p=0.000 n=20+20) BM_DecompressFlat/11/9 [gaviota/level=9] 201µs ± 2% 202µs ± 1% ~ (p=0.192 n=20+20) BM_DecompressFlat/12/9 [cp/level=9] 27.8µs ± 1% 27.1µs ± 1% -2.55% (p=0.000 n=19+17) BM_DecompressFlat/13/9 [c/level=9] 13.5µs ± 2% 13.2µs ± 1% -2.21% (p=0.000 n=19+18) BM_DecompressFlat/14/9 [lsp/level=9] 6.53µs ± 1% 6.48µs ± 1% -0.73% (p=0.000 n=19+18) BM_DecompressFlat/15/9 [xls/level=9] 1.05ms ± 2% 1.06ms ± 1% +0.75% (p=0.009 n=20+19) BM_DecompressFlat/16/9 [xls_200/level=9] 2.66µs ± 2% 2.64µs ± 1% -0.64% (p=0.005 n=20+19) BM_DecompressFlat/17/9 [bin/level=9] 272µs ± 2% 275µs ± 2% +0.99% (p=0.003 n=20+19) BM_DecompressFlat/18/9 [bin_200/level=9] 266ns ± 2% 265ns ± 3% ~ (p=0.061 n=19+20) BM_DecompressFlat/19/9 [sum/level=9] 44.3µs ± 1% 42.7µs ± 2% -3.43% (p=0.000 n=20+18) BM_DecompressFlat/20/9 [man/level=9] 7.79µs ± 1% 7.61µs ± 1% -2.25% (p=0.000 n=19+19) BM_DecompressFlat/0/10 [html/level=10] 50.4µs ± 2% 49.3µs ± 2% -2.13% (p=0.000 n=16+20) BM_DecompressFlat/1/10 [urls/level=10] 589µs ± 2% 588µs ± 1% ~ (p=0.258 n=20+19) BM_DecompressFlat/2/10 [jpg/level=10] 3.37µs ± 1% 3.36µs ± 1% -0.47% (p=0.030 n=19+17) BM_DecompressFlat/3/10 [jpg_200/level=10] 2.85µs ± 1% 2.82µs ± 1% -1.00% (p=0.000 n=19+19) BM_DecompressFlat/4/10 [pdf/level=10] 13.0µs ± 2% 13.0µs ± 1% +0.42% (p=0.012 n=19+19) BM_DecompressFlat/5/10 [html4/level=10] 59.0µs ± 1% 62.3µs ± 5% +5.61% (p=0.000 n=18+20) BM_DecompressFlat/6/10 [txt1/level=10] 189µs ± 1% 184µs ± 1% -2.20% (p=0.000 n=20+17) BM_DecompressFlat/7/10 [txt2/level=10] 166µs ± 1% 163µs ± 1% -2.12% (p=0.000 n=18+20) BM_DecompressFlat/8/10 [txt3/level=10] 458µs ± 2% 448µs ± 1% -2.19% (p=0.000 n=20+20) BM_DecompressFlat/9/10 [txt4/level=10] 608µs ± 1% 597µs ± 1% -1.94% (p=0.000 n=19+19) BM_DecompressFlat/10/10 [pb/level=10] 45.3µs ± 2% 44.8µs ± 1% -1.16% (p=0.000 n=20+20) BM_DecompressFlat/11/10 [gaviota/level=10] 193µs ± 1% 194µs ± 1% ~ (p=0.084 n=20+19) BM_DecompressFlat/12/10 [cp/level=10] 27.7µs ± 2% 27.1µs ± 1% -2.25% (p=0.000 n=20+20) BM_DecompressFlat/13/10 [c/level=10] 13.5µs ± 1% 13.2µs ± 2% -1.98% (p=0.000 n=20+20) BM_DecompressFlat/14/10 [lsp/level=10] 6.56µs ± 2% 6.50µs ± 1% -0.94% (p=0.000 n=20+20) BM_DecompressFlat/15/10 [xls/level=10] 1.05ms ± 3% 1.05ms ± 1% ~ (p=0.149 n=19+20) BM_DecompressFlat/16/10 [xls_200/level=10] 2.65µs ± 2% 2.64µs ± 1% ~ (p=0.072 n=20+20) BM_DecompressFlat/17/10 [bin/level=10] 266µs ± 2% 269µs ± 1% +0.85% (p=0.001 n=20+18) BM_DecompressFlat/18/10 [bin_200/level=10] 267ns ± 2% 265ns ± 2% -0.93% (p=0.005 n=19+20) BM_DecompressFlat/19/10 [sum/level=10] 44.2µs ± 2% 42.5µs ± 1% -3.97% (p=0.000 n=20+18) BM_DecompressFlat/20/10 [man/level=10] 7.80µs ± 1% 7.63µs ± 2% -2.17% (p=0.000 n=20+19) BM_DecompressFlat/0/11 [html/level=11] 50.1µs ± 1% 49.0µs ± 1% -2.24% (p=0.000 n=20+20) BM_DecompressFlat/1/11 [urls/level=11] 586µs ± 3% 585µs ± 1% ~ (p=0.879 n=20+19) BM_DecompressFlat/2/11 [jpg/level=11] 3.37µs ± 1% 3.36µs ± 1% ~ (p=0.101 n=20+19) BM_DecompressFlat/3/11 [jpg_200/level=11] 2.85µs ± 1% 2.82µs ± 2% -1.01% (p=0.000 n=20+20) BM_DecompressFlat/4/11 [pdf/level=11] 13.0µs ± 1% 13.1µs ± 1% +0.48% (p=0.007 n=20+20) BM_DecompressFlat/5/11 [html4/level=11] 58.2µs ± 2% 63.1µs ± 5% +8.56% (p=0.000 n=18+20) BM_DecompressFlat/6/11 [txt1/level=11] 188µs ± 2% 184µs ± 1% -1.97% (p=0.000 n=20+19) BM_DecompressFlat/7/11 [txt2/level=11] 166µs ± 1% 163µs ± 1% -1.64% (p=0.000 n=19+19) BM_DecompressFlat/8/11 [txt3/level=11] 453µs ± 1% 443µs ± 1% -2.00% (p=0.000 n=20+20) BM_DecompressFlat/9/11 [txt4/level=11] 604µs ± 1% 591µs ± 1% -2.14% (p=0.000 n=20+20) BM_DecompressFlat/10/11 [pb/level=11] 45.1µs ± 2% 44.5µs ± 2% -1.35% (p=0.000 n=20+20) BM_DecompressFlat/11/11 [gaviota/level=11] 186µs ± 2% 186µs ± 1% ~ (p=0.444 n=20+19) BM_DecompressFlat/12/11 [cp/level=11] 27.7µs ± 1% 26.9µs ± 1% -2.79% (p=0.000 n=20+19) BM_DecompressFlat/13/11 [c/level=11] 13.9µs ± 1% 13.5µs ± 1% -2.79% (p=0.000 n=20+19) BM_DecompressFlat/14/11 [lsp/level=11] 6.61µs ± 2% 6.59µs ± 1% ~ (p=0.414 n=20+20) BM_DecompressFlat/15/11 [xls/level=11] 1.05ms ± 2% 1.05ms ± 1% ~ (p=0.411 n=20+19) BM_DecompressFlat/16/11 [xls_200/level=11] 2.68µs ± 2% 2.70µs ± 1% +0.49% (p=0.026 n=20+20) BM_DecompressFlat/17/11 [bin/level=11] 265µs ± 2% 267µs ± 1% +1.00% (p=0.001 n=20+20) BM_DecompressFlat/18/11 [bin_200/level=11] 266ns ± 2% 265ns ± 2% ~ (p=0.204 n=19+20) BM_DecompressFlat/19/11 [sum/level=11] 44.3µs ± 1% 42.5µs ± 1% -4.10% (p=0.000 n=20+20) BM_DecompressFlat/20/11 [man/level=11] 7.72µs ± 1% 7.60µs ± 2% -1.52% (p=0.000 n=19+20) BM_DecompressSink/0/2 [html/level=2] 58.2µs ± 1% 57.8µs ± 1% -0.75% (p=0.004 n=20+19) BM_DecompressSink/1/2 [urls/level=2] 652µs ± 2% 648µs ± 1% -0.64% (p=0.002 n=20+19) BM_DecompressSink/2/2 [jpg/level=2] 3.43µs ± 1% 3.42µs ± 1% -0.36% (p=0.031 n=19+18) BM_DecompressSink/3/2 [jpg_200/level=2] 2.86µs ± 1% 2.89µs ± 1% +1.08% (p=0.000 n=20+19) BM_DecompressSink/4/2 [pdf/level=2] 11.9µs ± 1% 11.9µs ± 1% +0.56% (p=0.001 n=19+19) BM_DecompressSink/5/2 [html4/level=2] 81.4µs ± 2% 81.1µs ± 2% ~ (p=0.065 n=20+19) BM_DecompressSink/6/2 [txt1/level=2] 198µs ± 1% 195µs ± 1% -1.83% (p=0.000 n=18+20) BM_DecompressSink/7/2 [txt2/level=2] 173µs ± 1% 170µs ± 1% -1.99% (p=0.000 n=19+18) BM_DecompressSink/8/2 [txt3/level=2] 526µs ± 1% 514µs ± 1% -2.18% (p=0.000 n=20+20) BM_DecompressSink/9/2 [txt4/level=2] 649µs ± 1% 639µs ± 1% -1.62% (p=0.000 n=19+20) BM_DecompressSink/10/2 [pb/level=2] 54.0µs ± 1% 52.1µs ± 2% -3.55% (p=0.000 n=20+19) BM_DecompressSink/11/2 [gaviota/level=2] 226µs ± 3% 223µs ± 1% -1.30% (p=0.000 n=19+18) BM_DecompressSink/12/2 [cp/level=2] 28.8µs ± 1% 28.0µs ± 1% -2.78% (p=0.000 n=19+20) BM_DecompressSink/13/2 [c/level=2] 17.9µs ± 1% 17.5µs ± 1% -2.31% (p=0.000 n=19+19) BM_DecompressSink/14/2 [lsp/level=2] 7.20µs ± 1% 7.11µs ± 1% -1.24% (p=0.000 n=18+19) BM_DecompressSink/15/2 [xls/level=2] 1.11ms ± 2% 1.13ms ± 2% +1.43% (p=0.000 n=18+20) BM_DecompressSink/16/2 [xls_200/level=2] 2.26µs ± 1% 2.24µs ± 1% -0.87% (p=0.000 n=18+19) BM_DecompressSink/17/2 [bin/level=2] 303µs ± 2% 304µs ± 1% ~ (p=0.201 n=20+20) BM_DecompressSink/18/2 [bin_200/level=2] 298ns ± 2% 301ns ± 1% +1.04% (p=0.000 n=19+20) BM_DecompressSink/19/2 [sum/level=2] 43.4µs ± 1% 41.9µs ± 1% -3.28% (p=0.000 n=19+19) BM_DecompressSink/20/2 [man/level=2] 8.35µs ± 1% 8.21µs ± 1% -1.62% (p=0.000 n=19+18) BM_DecompressSink/0/3 [html/level=3] 55.3µs ± 2% 53.7µs ± 2% -2.86% (p=0.000 n=20+20) BM_DecompressSink/1/3 [urls/level=3] 661µs ± 1% 655µs ± 1% -0.85% (p=0.000 n=19+20) BM_DecompressSink/2/3 [jpg/level=3] 3.43µs ± 1% 3.43µs ± 1% ~ (p=0.558 n=18+19) BM_DecompressSink/3/3 [jpg_200/level=3] 2.88µs ± 1% 2.84µs ± 1% -1.34% (p=0.000 n=19+19) BM_DecompressSink/4/3 [pdf/level=3] 11.6µs ± 1% 11.7µs ± 1% +0.56% (p=0.001 n=19+20) BM_DecompressSink/5/3 [html4/level=3] 83.0µs ± 1% 86.5µs ± 3% +4.14% (p=0.000 n=20+20) BM_DecompressSink/6/3 [txt1/level=3] 226µs ± 1% 221µs ± 1% -2.02% (p=0.000 n=19+20) BM_DecompressSink/7/3 [txt2/level=3] 159µs ± 1% 157µs ± 1% -1.43% (p=0.000 n=18+19) BM_DecompressSink/8/3 [txt3/level=3] 527µs ± 1% 515µs ± 1% -2.24% (p=0.000 n=19+18) BM_DecompressSink/9/3 [txt4/level=3] 686µs ± 0% 674µs ± 1% -1.69% (p=0.000 n=18+19) BM_DecompressSink/10/3 [pb/level=3] 51.7µs ± 2% 51.0µs ± 1% -1.36% (p=0.000 n=20+18) BM_DecompressSink/11/3 [gaviota/level=3] 238µs ± 1% 237µs ± 1% -0.69% (p=0.000 n=17+20) BM_DecompressSink/12/3 [cp/level=3] 27.5µs ± 1% 26.9µs ± 1% -2.26% (p=0.000 n=20+19) BM_DecompressSink/13/3 [c/level=3] 15.1µs ± 2% 14.7µs ± 1% -2.59% (p=0.000 n=20+19) BM_DecompressSink/14/3 [lsp/level=3] 6.90µs ± 1% 6.84µs ± 2% -0.95% (p=0.000 n=20+20) BM_DecompressSink/15/3 [xls/level=3] 1.09ms ± 1% 1.11ms ± 1% +1.66% (p=0.000 n=19+19) BM_DecompressSink/16/3 [xls_200/level=3] 2.21µs ± 2% 2.22µs ± 1% +0.61% (p=0.008 n=20+19) BM_DecompressSink/17/3 [bin/level=3] 334µs ± 1% 333µs ± 1% ~ (p=0.224 n=19+20) BM_DecompressSink/18/3 [bin_200/level=3] 298ns ± 1% 301ns ± 2% +0.95% (p=0.000 n=19+20) BM_DecompressSink/19/3 [sum/level=3] 42.7µs ± 1% 41.0µs ± 2% -3.94% (p=0.000 n=20+20) BM_DecompressSink/20/3 [man/level=3] 8.03µs ± 1% 7.89µs ± 1% -1.79% (p=0.000 n=17+19) BM_DecompressSink/0/5 [html/level=5] 56.0µs ± 1% 55.5µs ± 1% -0.87% (p=0.000 n=20+19) BM_DecompressSink/1/5 [urls/level=5] 662µs ± 1% 657µs ± 1% -0.74% (p=0.000 n=18+18) BM_DecompressSink/2/5 [jpg/level=5] 3.43µs ± 1% 3.43µs ± 1% ~ (p=0.247 n=20+19) BM_DecompressSink/3/5 [jpg_200/level=5] 2.87µs ± 1% 2.86µs ± 1% -0.37% (p=0.020 n=18+19) BM_DecompressSink/4/5 [pdf/level=5] 14.8µs ± 1% 14.9µs ± 1% +0.48% (p=0.022 n=19+19) BM_DecompressSink/5/5 [html4/level=5] 79.1µs ± 1% 81.1µs ± 6% +2.58% (p=0.002 n=18+20) BM_DecompressSink/6/5 [txt1/level=5] 196µs ± 1% 193µs ± 1% -1.75% (p=0.000 n=19+20) BM_DecompressSink/7/5 [txt2/level=5] 196µs ± 1% 191µs ± 1% -2.23% (p=0.000 n=20+19) BM_DecompressSink/8/5 [txt3/level=5] 537µs ± 1% 527µs ± 1% -1.96% (p=0.000 n=20+19) BM_DecompressSink/9/5 [txt4/level=5] 717µs ± 2% 700µs ± 1% -2.39% (p=0.000 n=20+20) BM_DecompressSink/10/5 [pb/level=5] 51.3µs ± 2% 50.1µs ± 2% -2.35% (p=0.000 n=19+20) BM_DecompressSink/11/5 [gaviota/level=5] 222µs ± 3% 224µs ± 1% +0.92% (p=0.002 n=20+18) BM_DecompressSink/12/5 [cp/level=5] 29.9µs ± 2% 29.0µs ± 1% -3.02% (p=0.000 n=20+20) BM_DecompressSink/13/5 [c/level=5] 14.3µs ± 2% 13.9µs ± 0% -2.54% (p=0.000 n=18+16) BM_DecompressSink/14/5 [lsp/level=5] 6.68µs ± 2% 6.62µs ± 1% -0.86% (p=0.000 n=19+19) BM_DecompressSink/15/5 [xls/level=5] 1.11ms ± 1% 1.12ms ± 1% +1.05% (p=0.000 n=20+19) BM_DecompressSink/16/5 [xls_200/level=5] 2.67µs ± 2% 2.66µs ± 1% ~ (p=0.718 n=20+20) BM_DecompressSink/17/5 [bin/level=5] 330µs ± 2% 330µs ± 1% ~ (p=0.529 n=20+20) BM_DecompressSink/18/5 [bin_200/level=5] 299ns ± 1% 302ns ± 1% +1.02% (p=0.000 n=19+20) BM_DecompressSink/19/5 [sum/level=5] 48.5µs ± 2% 46.6µs ± 1% -3.94% (p=0.000 n=18+20) BM_DecompressSink/20/5 [man/level=5] 7.91µs ± 2% 7.73µs ± 1% -2.27% (p=0.000 n=20+19) BM_DecompressSink/0/7 [html/level=7] 52.0µs ± 1% 51.5µs ± 1% -1.10% (p=0.000 n=19+20) BM_DecompressSink/1/7 [urls/level=7] 632µs ± 1% 628µs ± 1% -0.71% (p=0.000 n=20+19) BM_DecompressSink/2/7 [jpg/level=7] 3.43µs ± 1% 3.42µs ± 2% ~ (p=0.055 n=18+20) BM_DecompressSink/3/7 [jpg_200/level=7] 2.88µs ± 1% 2.86µs ± 1% -0.60% (p=0.003 n=18+19) BM_DecompressSink/4/7 [pdf/level=7] 13.2µs ± 1% 13.3µs ± 1% ~ (p=0.314 n=20+20) BM_DecompressSink/5/7 [html4/level=7] 75.4µs ± 2% 78.1µs ± 3% +3.63% (p=0.000 n=19+19) BM_DecompressSink/6/7 [txt1/level=7] 210µs ± 1% 206µs ± 1% -2.00% (p=0.000 n=20+20) BM_DecompressSink/7/7 [txt2/level=7] 174µs ± 1% 171µs ± 1% -1.92% (p=0.000 n=18+20) BM_DecompressSink/8/7 [txt3/level=7] 501µs ± 1% 492µs ± 1% -1.98% (p=0.000 n=20+20) BM_DecompressSink/9/7 [txt4/level=7] 663µs ± 1% 648µs ± 2% -2.26% (p=0.000 n=18+20) BM_DecompressSink/10/7 [pb/level=7] 46.9µs ± 2% 45.6µs ± 2% -2.84% (p=0.000 n=19+20) BM_DecompressSink/11/7 [gaviota/level=7] 227µs ± 2% 228µs ± 1% ~ (p=0.435 n=19+19) BM_DecompressSink/12/7 [cp/level=7] 28.4µs ± 1% 27.7µs ± 1% -2.20% (p=0.000 n=19+20) BM_DecompressSink/13/7 [c/level=7] 13.5µs ± 1% 13.2µs ± 1% -2.18% (p=0.000 n=19+20) BM_DecompressSink/14/7 [lsp/level=7] 6.58µs ± 1% 6.55µs ± 2% -0.42% (p=0.019 n=19+20) BM_DecompressSink/15/7 [xls/level=7] 1.05ms ± 1% 1.07ms ± 1% +1.72% (p=0.000 n=20+20) BM_DecompressSink/16/7 [xls_200/level=7] 2.66µs ± 1% 2.67µs ± 1% ~ (p=0.327 n=20+20) BM_DecompressSink/17/7 [bin/level=7] 298µs ± 1% 298µs ± 1% ~ (p=0.211 n=20+20) BM_DecompressSink/18/7 [bin_200/level=7] 298ns ± 1% 302ns ± 1% +1.48% (p=0.000 n=19+19) BM_DecompressSink/19/7 [sum/level=7] 44.7µs ± 1% 43.3µs ± 2% -3.14% (p=0.000 n=19+19) BM_DecompressSink/20/7 [man/level=7] 7.80µs ± 1% 7.62µs ± 1% -2.25% (p=0.000 n=19+19) BM_DecompressSink/0/9 [html/level=9] 50.6µs ± 1% 49.4µs ± 1% -2.35% (p=0.000 n=20+20) BM_DecompressSink/1/9 [urls/level=9] 620µs ± 1% 618µs ± 1% -0.33% (p=0.030 n=19+20) BM_DecompressSink/2/9 [jpg/level=9] 3.43µs ± 1% 3.42µs ± 1% ~ (p=0.199 n=18+19) BM_DecompressSink/3/9 [jpg_200/level=9] 2.88µs ± 1% 2.86µs ± 1% -0.62% (p=0.002 n=20+19) BM_DecompressSink/4/9 [pdf/level=9] 13.0µs ± 1% 13.0µs ± 1% ~ (p=0.196 n=20+18) BM_DecompressSink/5/9 [html4/level=9] 74.0µs ± 4% 76.0µs ± 5% +2.68% (p=0.000 n=20+20) BM_DecompressSink/6/9 [txt1/level=9] 196µs ± 2% 192µs ± 1% -1.99% (p=0.000 n=20+19) BM_DecompressSink/7/9 [txt2/level=9] 168µs ± 1% 164µs ± 1% -2.09% (p=0.000 n=20+18) BM_DecompressSink/8/9 [txt3/level=9] 482µs ± 2% 472µs ± 1% -2.05% (p=0.000 n=20+19) BM_DecompressSink/9/9 [txt4/level=9] 640µs ± 1% 624µs ± 1% -2.46% (p=0.000 n=20+20) BM_DecompressSink/10/9 [pb/level=9] 45.5µs ± 2% 44.8µs ± 2% -1.50% (p=0.000 n=19+20) BM_DecompressSink/11/9 [gaviota/level=9] 206µs ± 2% 207µs ± 2% ~ (p=0.134 n=19+20) BM_DecompressSink/12/9 [cp/level=9] 27.8µs ± 1% 27.1µs ± 1% -2.46% (p=0.000 n=17+19) BM_DecompressSink/13/9 [c/level=9] 13.5µs ± 1% 13.2µs ± 1% -2.06% (p=0.000 n=18+18) BM_DecompressSink/14/9 [lsp/level=9] 6.58µs ± 1% 6.55µs ± 1% -0.51% (p=0.011 n=20+20) BM_DecompressSink/15/9 [xls/level=9] 1.08ms ± 2% 1.09ms ± 1% +0.91% (p=0.004 n=20+18) BM_DecompressSink/16/9 [xls_200/level=9] 2.66µs ± 1% 2.67µs ± 1% ~ (p=0.061 n=19+19) BM_DecompressSink/17/9 [bin/level=9] 295µs ± 1% 293µs ± 2% -0.45% (p=0.017 n=18+20) BM_DecompressSink/18/9 [bin_200/level=9] 298ns ± 1% 301ns ± 1% +1.04% (p=0.000 n=19+19) BM_DecompressSink/19/9 [sum/level=9] 44.1µs ± 2% 42.6µs ± 1% -3.42% (p=0.000 n=19+19) BM_DecompressSink/20/9 [man/level=9] 7.78µs ± 1% 7.64µs ± 1% -1.77% (p=0.000 n=20+20) BM_DecompressSink/0/10 [html/level=10] 49.9µs ± 1% 49.3µs ± 2% -1.10% (p=0.000 n=20+19) BM_DecompressSink/1/10 [urls/level=10] 616µs ± 1% 613µs ± 1% -0.43% (p=0.013 n=18+19) BM_DecompressSink/2/10 [jpg/level=10] 3.43µs ± 1% 3.44µs ± 2% ~ (p=0.383 n=20+20) BM_DecompressSink/3/10 [jpg_200/level=10] 2.88µs ± 1% 2.86µs ± 2% -0.47% (p=0.028 n=20+20) BM_DecompressSink/4/10 [pdf/level=10] 13.0µs ± 1% 13.0µs ± 2% ~ (p=1.000 n=20+20) BM_DecompressSink/5/10 [html4/level=10] 73.5µs ± 2% 75.2µs ± 2% +2.30% (p=0.000 n=20+19) BM_DecompressSink/6/10 [txt1/level=10] 193µs ± 1% 189µs ± 1% -1.92% (p=0.000 n=18+18) BM_DecompressSink/7/10 [txt2/level=10] 166µs ± 1% 162µs ± 1% -2.10% (p=0.000 n=19+19) BM_DecompressSink/8/10 [txt3/level=10] 477µs ± 1% 465µs ± 1% -2.34% (p=0.000 n=20+18) BM_DecompressSink/9/10 [txt4/level=10] 629µs ± 1% 615µs ± 1% -2.16% (p=0.000 n=20+19) BM_DecompressSink/10/10 [pb/level=10] 45.5µs ± 1% 44.9µs ± 2% -1.32% (p=0.000 n=20+20) BM_DecompressSink/11/10 [gaviota/level=10] 199µs ± 2% 199µs ± 1% ~ (p=0.351 n=20+19) BM_DecompressSink/12/10 [cp/level=10] 27.6µs ± 1% 27.1µs ± 2% -2.11% (p=0.000 n=19+19) BM_DecompressSink/13/10 [c/level=10] 13.5µs ± 1% 13.3µs ± 2% -2.11% (p=0.000 n=19+20) BM_DecompressSink/14/10 [lsp/level=10] 6.58µs ± 1% 6.53µs ± 1% -0.71% (p=0.000 n=19+18) BM_DecompressSink/15/10 [xls/level=10] 1.07ms ± 2% 1.09ms ± 1% +1.73% (p=0.000 n=19+19) BM_DecompressSink/16/10 [xls_200/level=10] 2.66µs ± 1% 2.67µs ± 1% +0.43% (p=0.009 n=18+20) BM_DecompressSink/17/10 [bin/level=10] 288µs ± 1% 287µs ± 1% ~ (p=0.301 n=20+20) BM_DecompressSink/18/10 [bin_200/level=10] 298ns ± 1% 301ns ± 1% +1.07% (p=0.000 n=17+17) BM_DecompressSink/19/10 [sum/level=10] 44.0µs ± 1% 42.6µs ± 1% -3.24% (p=0.000 n=20+19) BM_DecompressSink/20/10 [man/level=10] 7.80µs ± 1% 7.64µs ± 1% -2.08% (p=0.000 n=19+20) BM_DecompressSink/0/11 [html/level=11] 50.1µs ± 2% 49.0µs ± 2% -2.18% (p=0.000 n=20+20) BM_DecompressSink/1/11 [urls/level=11] 613µs ± 1% 612µs ± 1% ~ (p=0.175 n=19+20) BM_DecompressSink/2/11 [jpg/level=11] 3.43µs ± 1% 3.42µs ± 1% -0.37% (p=0.049 n=19+18) BM_DecompressSink/3/11 [jpg_200/level=11] 2.89µs ± 1% 2.88µs ± 1% -0.43% (p=0.026 n=20+20) BM_DecompressSink/4/11 [pdf/level=11] 13.0µs ± 1% 13.1µs ± 1% ~ (p=0.194 n=19+20) BM_DecompressSink/5/11 [html4/level=11] 73.4µs ± 4% 78.0µs ± 1% +6.27% (p=0.000 n=20+16) BM_DecompressSink/6/11 [txt1/level=11] 193µs ± 2% 189µs ± 2% -2.17% (p=0.000 n=19+20) BM_DecompressSink/7/11 [txt2/level=11] 166µs ± 1% 163µs ± 1% -1.84% (p=0.000 n=19+20) BM_DecompressSink/8/11 [txt3/level=11] 471µs ± 1% 462µs ± 1% -1.90% (p=0.000 n=20+19) BM_DecompressSink/9/11 [txt4/level=11] 622µs ± 1% 608µs ± 1% -2.32% (p=0.000 n=20+19) BM_DecompressSink/10/11 [pb/level=11] 45.0µs ± 2% 44.5µs ± 2% -1.12% (p=0.002 n=20+19) BM_DecompressSink/11/11 [gaviota/level=11] 192µs ± 1% 191µs ± 1% -0.97% (p=0.000 n=19+18) BM_DecompressSink/12/11 [cp/level=11] 27.7µs ± 3% 27.0µs ± 1% -2.63% (p=0.000 n=20+20) BM_DecompressSink/13/11 [c/level=11] 13.9µs ± 2% 13.5µs ± 1% -2.27% (p=0.000 n=20+20) BM_DecompressSink/14/11 [lsp/level=11] 6.65µs ± 1% 6.64µs ± 1% ~ (p=0.309 n=20+19) BM_DecompressSink/15/11 [xls/level=11] 1.07ms ± 2% 1.09ms ± 1% +1.12% (p=0.000 n=20+19) BM_DecompressSink/16/11 [xls_200/level=11] 2.72µs ± 1% 2.73µs ± 1% +0.63% (p=0.003 n=19+19) BM_DecompressSink/17/11 [bin/level=11] 286µs ± 1% 285µs ± 1% -0.40% (p=0.033 n=20+18) BM_DecompressSink/18/11 [bin_200/level=11] 298ns ± 1% 301ns ± 1% +1.22% (p=0.000 n=19+20) BM_DecompressSink/19/11 [sum/level=11] 44.1µs ± 1% 42.5µs ± 1% -3.71% (p=0.000 n=20+20) BM_DecompressSink/20/11 [man/level=11] 7.74µs ± 1% 7.61µs ± 1% -1.70% (p=0.000 n=20+18)
This commit is contained in:
Ilya Tokar
parent
e19fe535f4
commit
4a455a1669
1 changed files with 13 additions and 0 deletions
|
|
@ -40,6 +40,7 @@
|
|||
#error "Cannot force the use of the short and the long ZSTD_decompressSequences variants!"
|
||||
#endif
|
||||
|
||||
#define PREFETCH_DISTANCE (8*CACHELINE_SIZE)
|
||||
|
||||
/*_*******************************************************
|
||||
* Memory operations
|
||||
|
|
@ -1414,6 +1415,8 @@ ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
|
|||
const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
|
||||
const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
|
||||
const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
|
||||
int counter = 0;
|
||||
const int kPrefetchRatio = 16;
|
||||
DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer (%i seqs)", nbSeq);
|
||||
|
||||
/* Literals are split between internal buffer & output buffer */
|
||||
|
|
@ -1501,6 +1504,12 @@ ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
|
|||
sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
|
||||
if (litPtr + sequence.litLength > dctx->litBufferEnd) break;
|
||||
{ size_t const oneSeqSize = ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence.litLength - WILDCOPY_OVERLENGTH, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
|
||||
/* Don't prefetch too often, because it is actually slower */
|
||||
counter++;
|
||||
if ((counter / kPrefetchRatio) == 0) {
|
||||
PREFETCH_L1(op+PREFETCH_DISTANCE);
|
||||
PREFETCH_L1(litPtr+PREFETCH_DISTANCE);
|
||||
}
|
||||
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
|
||||
assert(!ZSTD_isError(oneSeqSize));
|
||||
ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
|
||||
|
|
@ -1562,6 +1571,8 @@ ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
|
|||
for ( ; nbSeq ; nbSeq--) {
|
||||
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
|
||||
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
|
||||
PREFETCH_L1(op+PREFETCH_DISTANCE);
|
||||
PREFETCH_L1(litPtr+PREFETCH_DISTANCE);
|
||||
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
|
||||
assert(!ZSTD_isError(oneSeqSize));
|
||||
ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
|
||||
|
|
@ -1655,6 +1666,8 @@ ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
|
|||
for ( ; nbSeq ; nbSeq--) {
|
||||
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
|
||||
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
|
||||
PREFETCH_L1(op+PREFETCH_DISTANCE);
|
||||
PREFETCH_L1(litPtr+PREFETCH_DISTANCE);
|
||||
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
|
||||
assert(!ZSTD_isError(oneSeqSize));
|
||||
ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue