tree 851799e70599bb974d7a28a928e73cf9e42aed34
parent 37a5ead3b7b70350ca10fe0783a763b58768750d
author Nigel Tao <nigeltao@golang.org> 1549794973 +1100
committer Nigel Tao <nigeltao@golang.org> 1550019286 +1100

Move some struct fields to private_data

This means that they are not zero'ed by the initialize methods, which
improves some benchmark numbers. There is a trade-off here, as bugs in
the *.wuffs code could then mean that we are potentially reading
uninitialized memory (bulk data fields). Nonetheless, on some
micro-benchmarks, the relative performance gain on skipping this fixed
cost is stark, up to +96% throughput, or equivalently, -49% in time per
operation.

Callers of the Wuffs libraries have the option to zero-initialize the
entire state (and then pass WUFFS_INITIALIZE__ALREADY_ZEROED) or to
zero-initialize private_impl state but not private_data state (i.e. skip
the bulk data fields).

The numbers below are in two sections: throughput (e.g. MB/s) and time
per operation. Relatively large gains in throughput (e.g. +96%) can
correspond to small absolute gains (e.g. it's the difference between an
operation taking 2.02µs instead of 3.98µs).

--------

name                                             old speed      new speed      delta

wuffs_adler32_10k/clang5                         2.43GB/s ± 0%  2.44GB/s ± 0%   +0.13%  (p=0.000 n=47+48)
wuffs_adler32_100k/clang5                        2.44GB/s ± 0%  2.44GB/s ± 1%   +0.10%  (p=0.001 n=45+47)

wuffs_adler32_10k/gcc7                           3.24GB/s ± 0%  3.24GB/s ± 0%     ~     (p=0.933 n=47+46)
wuffs_adler32_100k/gcc7                          3.25GB/s ± 0%  3.25GB/s ± 1%     ~     (p=0.551 n=42+47)

wuffs_crc32_ieee_10k/clang5                      2.87GB/s ± 0%  2.87GB/s ± 0%     ~     (p=0.519 n=45+46)
wuffs_crc32_ieee_100k/clang5                     2.89GB/s ± 0%  2.89GB/s ± 1%     ~     (p=0.479 n=45+45)

wuffs_crc32_ieee_10k/gcc7                        3.35GB/s ± 0%  3.38GB/s ± 0%   +0.95%  (p=0.000 n=44+49)
wuffs_crc32_ieee_100k/gcc7                       3.39GB/s ± 1%  3.42GB/s ± 0%   +0.96%  (p=0.000 n=47+47)

wuffs_deflate_decode_1k/clang5                    139MB/s ± 1%   170MB/s ± 1%  +22.72%  (p=0.000 n=48+47)
wuffs_deflate_decode_10k/clang5                   232MB/s ± 1%   241MB/s ± 1%   +4.13%  (p=0.000 n=48+45)
wuffs_deflate_decode_100k_just_one_read/clang5    274MB/s ± 0%   277MB/s ± 1%   +1.14%  (p=0.000 n=47+45)
wuffs_deflate_decode_100k_many_big_reads/clang5   226MB/s ± 0%   226MB/s ± 1%     ~     (p=0.057 n=48+47)

wuffs_deflate_decode_1k/gcc7                      157MB/s ± 1%   198MB/s ± 1%  +25.61%  (p=0.000 n=45+45)
wuffs_deflate_decode_10k/gcc7                     265MB/s ± 0%   272MB/s ± 1%   +2.62%  (p=0.000 n=47+47)
wuffs_deflate_decode_100k_just_one_read/gcc7      322MB/s ± 0%   323MB/s ± 1%   +0.40%  (p=0.000 n=46+45)
wuffs_deflate_decode_100k_many_big_reads/gcc7     254MB/s ± 0%   254MB/s ± 0%   +0.23%  (p=0.000 n=43+46)

wuffs_gif_decode_1k_bw/clang5                     246MB/s ± 1%   448MB/s ± 0%  +82.21%  (p=0.000 n=48+44)
wuffs_gif_decode_1k_color/clang5                  145MB/s ± 0%   194MB/s ± 1%  +33.70%  (p=0.000 n=48+48)
wuffs_gif_decode_10k_bgra/clang5                  740MB/s ± 1%   747MB/s ± 0%   +0.97%  (p=0.000 n=49+45)
wuffs_gif_decode_10k_indexed/clang5               199MB/s ± 1%   201MB/s ± 0%   +1.30%  (p=0.000 n=46+45)
wuffs_gif_decode_20k/clang5                       251MB/s ± 0%   251MB/s ± 0%   +0.23%  (p=0.000 n=44+44)
wuffs_gif_decode_100k_artificial/clang5           557MB/s ± 1%   557MB/s ± 1%     ~     (p=0.653 n=45+47)
wuffs_gif_decode_100k_realistic/clang5            230MB/s ± 0%   229MB/s ± 0%   -0.45%  (p=0.000 n=45+46)
wuffs_gif_decode_1000k/clang5                     236MB/s ± 0%   233MB/s ± 0%   -0.91%  (p=0.000 n=46+48)
wuffs_gif_decode_anim_screencap/clang5           1.12GB/s ± 1%  1.11GB/s ± 0%   -0.74%  (p=0.000 n=47+45)

wuffs_gif_decode_1k_bw/gcc7                       257MB/s ± 1%   506MB/s ± 0%  +96.52%  (p=0.000 n=50+46)
wuffs_gif_decode_1k_color/gcc7                    146MB/s ± 1%   199MB/s ± 1%  +36.35%  (p=0.000 n=48+47)
wuffs_gif_decode_10k_bgra/gcc7                    632MB/s ± 1%   649MB/s ± 0%   +2.80%  (p=0.000 n=47+47)
wuffs_gif_decode_10k_indexed/gcc7                 198MB/s ± 1%   204MB/s ± 1%   +3.32%  (p=0.000 n=48+47)
wuffs_gif_decode_20k/gcc7                         242MB/s ± 0%   246MB/s ± 0%   +1.59%  (p=0.000 n=47+48)
wuffs_gif_decode_100k_artificial/gcc7             517MB/s ± 1%   521MB/s ± 0%   +0.71%  (p=0.000 n=46+46)
wuffs_gif_decode_100k_realistic/gcc7              214MB/s ± 0%   215MB/s ± 0%   +0.34%  (p=0.000 n=48+44)
wuffs_gif_decode_1000k/gcc7                       218MB/s ± 1%   218MB/s ± 0%     ~     (p=0.238 n=44+49)
wuffs_gif_decode_anim_screencap/gcc7             1.08GB/s ± 0%  1.07GB/s ± 1%   -0.27%  (p=0.000 n=45+45)

wuffs_gzip_decode_10k/clang5                      213MB/s ± 0%   222MB/s ± 0%   +4.08%  (p=0.000 n=43+46)
wuffs_gzip_decode_100k/clang5                     249MB/s ± 1%   253MB/s ± 0%   +1.50%  (p=0.000 n=47+46)

wuffs_gzip_decode_10k/gcc7                        246MB/s ± 1%   258MB/s ± 0%   +4.76%  (p=0.000 n=48+47)
wuffs_gzip_decode_100k/gcc7                       291MB/s ± 0%   291MB/s ± 0%     ~     (p=0.679 n=48+48)

wuffs_lzw_decode_20k/clang5                       274MB/s ± 0%   280MB/s ± 0%   +2.51%  (p=0.000 n=46+47)
wuffs_lzw_decode_100k/clang5                      484MB/s ± 0%   487MB/s ± 0%   +0.58%  (p=0.000 n=46+49)

wuffs_lzw_decode_20k/gcc7                         267MB/s ± 0%   274MB/s ± 2%   +2.31%  (p=0.000 n=41+43)
wuffs_lzw_decode_100k/gcc7                        489MB/s ± 1%   486MB/s ± 2%   -0.67%  (p=0.000 n=47+42)

wuffs_zlib_decode_10k/clang5                      212MB/s ± 1%   219MB/s ± 0%   +3.66%  (p=0.000 n=47+47)
wuffs_zlib_decode_100k/clang5                     246MB/s ± 1%   248MB/s ± 0%   +0.69%  (p=0.000 n=47+46)

wuffs_zlib_decode_10k/gcc7                        251MB/s ± 0%   255MB/s ± 0%   +1.83%  (p=0.000 n=45+44)
wuffs_zlib_decode_100k/gcc7                       294MB/s ± 1%   290MB/s ± 0%   -1.33%  (p=0.000 n=43+42)

--------

name                                             old time/op    new time/op    delta

wuffs_adler32_10k/clang5                           4.54µs ± 1%    4.54µs ± 0%   -0.15%  (p=0.000 n=48+48)
wuffs_adler32_100k/clang5                          41.0µs ± 0%    41.0µs ± 1%   -0.10%  (p=0.001 n=45+47)

wuffs_adler32_10k/gcc7                             3.41µs ± 0%    3.41µs ± 0%     ~     (p=0.890 n=47+46)
wuffs_adler32_100k/gcc7                            30.8µs ± 0%    30.8µs ± 1%     ~     (p=0.557 n=42+47)

wuffs_crc32_ieee_10k/clang5                        3.86µs ± 0%    3.86µs ± 0%     ~     (p=0.602 n=45+46)
wuffs_crc32_ieee_100k/clang5                       34.6µs ± 0%    34.6µs ± 1%     ~     (p=0.475 n=45+45)

wuffs_crc32_ieee_10k/gcc7                          3.30µs ± 0%    3.27µs ± 0%   -0.94%  (p=0.000 n=45+49)
wuffs_crc32_ieee_100k/gcc7                         29.5µs ± 1%    29.3µs ± 0%   -0.95%  (p=0.000 n=47+47)

wuffs_deflate_decode_1k/clang5                     6.79µs ± 1%    5.54µs ± 1%  -18.52%  (p=0.000 n=48+47)
wuffs_deflate_decode_10k/clang5                    47.8µs ± 1%    45.9µs ± 1%   -3.96%  (p=0.000 n=48+45)
wuffs_deflate_decode_100k_just_one_read/clang5      365µs ± 0%     361µs ± 1%   -1.12%  (p=0.000 n=47+45)
wuffs_deflate_decode_100k_many_big_reads/clang5     442µs ± 0%     443µs ± 1%     ~     (p=0.057 n=48+47)

wuffs_deflate_decode_1k/gcc7                       5.99µs ± 1%    4.77µs ± 1%  -20.39%  (p=0.000 n=45+45)
wuffs_deflate_decode_10k/gcc7                      41.7µs ± 0%    40.6µs ± 1%   -2.56%  (p=0.000 n=47+47)
wuffs_deflate_decode_100k_just_one_read/gcc7        311µs ± 0%     310µs ± 1%   -0.40%  (p=0.000 n=46+45)
wuffs_deflate_decode_100k_many_big_reads/gcc7       394µs ± 0%     393µs ± 0%   -0.23%  (p=0.000 n=43+46)

wuffs_gif_decode_1k_bw/clang5                      4.17µs ± 1%    2.29µs ± 0%  -45.12%  (p=0.000 n=46+44)
wuffs_gif_decode_1k_color/clang5                   6.95µs ± 0%    5.20µs ± 1%  -25.21%  (p=0.000 n=48+48)
wuffs_gif_decode_10k_bgra/clang5                   54.5µs ± 1%    54.0µs ± 0%   -0.96%  (p=0.000 n=49+45)
wuffs_gif_decode_10k_indexed/clang5                50.7µs ± 1%    50.1µs ± 0%   -1.28%  (p=0.000 n=46+45)
wuffs_gif_decode_20k/clang5                        76.6µs ± 0%    76.4µs ± 0%   -0.23%  (p=0.000 n=44+44)
wuffs_gif_decode_100k_artificial/clang5             248µs ± 1%     248µs ± 1%     ~     (p=0.653 n=45+47)
wuffs_gif_decode_100k_realistic/clang5              599µs ± 0%     602µs ± 0%   +0.45%  (p=0.000 n=45+46)
wuffs_gif_decode_1000k/clang5                      4.25ms ± 0%    4.29ms ± 0%   +0.92%  (p=0.000 n=46+48)
wuffs_gif_decode_anim_screencap/clang5             4.17ms ± 1%    4.20ms ± 0%   +0.74%  (p=0.000 n=47+45)

wuffs_gif_decode_1k_bw/gcc7                        3.98µs ± 1%    2.02µs ± 0%  -49.12%  (p=0.000 n=50+46)
wuffs_gif_decode_1k_color/gcc7                     6.90µs ± 1%    5.06µs ± 0%  -26.67%  (p=0.000 n=48+46)
wuffs_gif_decode_10k_bgra/gcc7                     63.8µs ± 1%    62.1µs ± 0%   -2.73%  (p=0.000 n=47+47)
wuffs_gif_decode_10k_indexed/gcc7                  50.9µs ± 1%    49.3µs ± 1%   -3.22%  (p=0.000 n=48+47)
wuffs_gif_decode_20k/gcc7                          79.3µs ± 0%    78.1µs ± 0%   -1.57%  (p=0.000 n=47+48)
wuffs_gif_decode_100k_artificial/gcc7               267µs ± 0%     265µs ± 0%   -0.69%  (p=0.000 n=45+46)
wuffs_gif_decode_100k_realistic/gcc7                643µs ± 0%     641µs ± 0%   -0.34%  (p=0.000 n=48+44)
wuffs_gif_decode_1000k/gcc7                        4.59ms ± 1%    4.58ms ± 0%     ~     (p=0.238 n=44+49)
wuffs_gif_decode_anim_screencap/gcc7               4.32ms ± 0%    4.33ms ± 1%   +0.27%  (p=0.000 n=45+45)

wuffs_gzip_decode_10k/clang5                       51.9µs ± 0%    49.8µs ± 0%   -3.92%  (p=0.000 n=43+46)
wuffs_gzip_decode_100k/clang5                       401µs ± 1%     395µs ± 0%   -1.48%  (p=0.000 n=47+46)

wuffs_gzip_decode_10k/gcc7                         44.9µs ± 1%    42.9µs ± 0%   -4.54%  (p=0.000 n=48+47)
wuffs_gzip_decode_100k/gcc7                         343µs ± 0%     343µs ± 0%     ~     (p=0.679 n=48+48)

wuffs_lzw_decode_20k/clang5                        70.2µs ± 0%    68.5µs ± 0%   -2.45%  (p=0.000 n=46+47)
wuffs_lzw_decode_100k/clang5                        207µs ± 0%     206µs ± 0%   -0.58%  (p=0.000 n=46+49)

wuffs_lzw_decode_20k/gcc7                          71.8µs ± 0%    70.2µs ± 2%   -2.25%  (p=0.000 n=41+43)
wuffs_lzw_decode_100k/gcc7                          204µs ± 1%     206µs ± 2%   +0.67%  (p=0.000 n=47+42)

wuffs_zlib_decode_10k/clang5                       52.3µs ± 1%    50.4µs ± 0%   -3.53%  (p=0.000 n=47+47)
wuffs_zlib_decode_100k/clang5                       406µs ± 1%     404µs ± 0%   -0.68%  (p=0.000 n=47+46)

wuffs_zlib_decode_10k/gcc7                         44.2µs ± 0%    43.4µs ± 0%   -1.79%  (p=0.000 n=45+44)
wuffs_zlib_decode_100k/gcc7                         340µs ± 1%     345µs ± 0%   +1.35%  (p=0.000 n=43+42)

--------