Commit Graph
9 Commits
Author SHA1 Message Date
Nigel Tao 13731f989f Add a reader1.unread_u8 method.
Also re-enable a number of std/flate tests and benchmarks.

It's not an optimization per se, but as a data point, here are some
benchmark numbers between 36a93fa and this commit. 36a93fa is "36a93fa
Add a TODO about impure functions", which is one commit before "f968950
Read bits up front in decode_huffman_fast", and f968950 was the commit
that made decode_huffman_fast read more than one byte at a time.

name                             old speed      new speed       delta

puffs_adler32_10k/clang          2.45GB/s ± 0%   2.45GB/s ± 0%     ~     (p=0.684 n=10+10)
puffs_adler32_100k/clang         2.45GB/s ± 0%   2.46GB/s ± 0%   +0.07%  (p=0.019 n=10+10)
puffs_flate_decode_1k/clang       124MB/s ± 0%    132MB/s ± 0%   +6.06%  (p=0.000 n=10+10)
puffs_flate_decode_10k/clang      145MB/s ± 0%    174MB/s ± 0%  +20.19%  (p=0.000 n=9+10)
puffs_flate_decode_100k/clang     168MB/s ± 0%    208MB/s ± 0%  +23.31%  (p=0.000 n=10+10)
puffs_zlib_decode_10k/clang       136MB/s ± 0%    162MB/s ± 0%  +18.94%  (p=0.000 n=10+9)
puffs_zlib_decode_100k/clang      158MB/s ± 0%    192MB/s ± 0%  +21.58%  (p=0.000 n=10+9)

puffs_adler32_10k/gcc            3.26GB/s ± 0%   3.26GB/s ± 0%     ~     (p=0.631 n=10+10)
puffs_adler32_100k/gcc           3.27GB/s ± 0%   3.27GB/s ± 0%     ~     (p=0.237 n=10+8)
puffs_flate_decode_1k/gcc        95.3MB/s ± 0%  105.5MB/s ± 0%  +10.67%  (p=0.000 n=10+10)
puffs_flate_decode_10k/gcc        110MB/s ± 0%    140MB/s ± 0%  +27.32%  (p=0.000 n=9+9)
puffs_flate_decode_100k/gcc       121MB/s ± 0%    150MB/s ± 0%  +24.02%  (p=0.000 n=8+10)
puffs_zlib_decode_10k/gcc         112MB/s ± 0%    120MB/s ± 0%   +6.96%  (p=0.000 n=10+9)
puffs_zlib_decode_100k/gcc        123MB/s ± 0%    134MB/s ± 0%   +8.93%  (p=0.000 n=9+8)
2017-11-05 15:59:02 +11:00
Nigel Tao 325c36314a Prepare decode_huffman_fast to rewind in.src. 2017-11-05 14:13:27 +11:00
Nigel Tao b6a079fd3e Tweak the decode_huffman_fast loop condition more. 2017-11-05 13:47:28 +11:00
Nigel Tao f6e072fe77 Add some more "inconsistent n_bits" checks.
This uncovered a bug in the decode_huffman_fast function. A follow up commit
will actually fix the bug.

For what it's worth, some benchmark numbers are below, but they're
possibly bogus, as we might be decoding incorrect (and truncated) output
for some inputs. Everything since "f968950 Read bits up front in
decode_huffman_fast" landed possibly has bogus numbers too.

name                           old speed     new speed     delta
puffs_flate_decode_100k/clang  194MB/s ± 0%  204MB/s ± 1%   +5.02%  (p=0.008 n=5+5)
puffs_flate_decode_100k/gcc    180MB/s ± 3%  202MB/s ± 1%  +12.09%  (p=0.008 n=5+5)
2017-11-05 09:23:49 +11:00
Nigel Tao db57d8b707 Bump decode_huffman_fast's src available hurdle.
We will need this later, to prove that the other read_u8 calls in
decode_huffman_fast will not suspend.

It's not obvious why the clang benchmarks don't change but the gcc ones
(including the adler numbers, which are superficially unrelated) look
worse. If the change from 6 to 12 bytes is a genuine performance
regression, then a future commit can tighten the bound back to 6,
although it would probably make the decode_huffman_fast assertions and
proof requirements considerably more complicated.

name                           old speed      new speed      delta

puffs_adler32_10k/clang        2.40GB/s ± 1%  2.41GB/s ± 0%    ~     (p=0.690 n=5+5)
puffs_adler32_100k/clang       2.39GB/s ± 2%  2.40GB/s ± 1%    ~     (p=1.000 n=5+5)
puffs_flate_decode_1k/clang     128MB/s ± 0%   128MB/s ± 0%    ~     (p=1.000 n=5+5)
puffs_flate_decode_10k/clang    167MB/s ± 2%   168MB/s ± 0%    ~     (p=0.548 n=5+5)
puffs_flate_decode_100k/clang   193MB/s ± 2%   194MB/s ± 0%    ~     (p=0.310 n=5+5)
puffs_zlib_decode_10k/clang     157MB/s ± 0%   155MB/s ± 4%    ~     (p=0.421 n=5+5)
puffs_zlib_decode_100k/clang    180MB/s ± 1%   177MB/s ± 3%    ~     (p=0.222 n=5+5)

puffs_adler32_10k/gcc          3.26GB/s ± 0%  3.20GB/s ± 1%  -1.81%  (p=0.008 n=5+5)
puffs_adler32_100k/gcc         3.26GB/s ± 0%  3.21GB/s ± 0%  -1.63%  (p=0.008 n=5+5)
puffs_flate_decode_1k/gcc       129MB/s ± 0%   126MB/s ± 1%  -2.30%  (p=0.008 n=5+5)
puffs_flate_decode_10k/gcc      164MB/s ± 0%   161MB/s ± 0%  -1.86%  (p=0.008 n=5+5)
puffs_flate_decode_100k/gcc     185MB/s ± 0%   180MB/s ± 3%  -2.81%  (p=0.016 n=4+5)
puffs_zlib_decode_10k/gcc       157MB/s ± 0%   151MB/s ± 2%  -3.49%  (p=0.008 n=5+5)
puffs_zlib_decode_100k/gcc      175MB/s ± 0%   170MB/s ± 2%  -2.97%  (p=0.016 n=4+5)
2017-11-05 08:56:55 +11:00
Nigel Tao fe5e9c68b3 Remove a superfluous while loop.
It's not obvious why the adler benchmark numbers also change (for the
better), but it seems superficially repeatable for fresh before/after
measurements. I'll take it for now, but perhaps the adler benchmarks
need to run for longer.

name                           old speed      new speed      delta

puffs_adler32_10k/clang        2.38GB/s ± 2%  2.45GB/s ± 0%  +2.87%  (p=0.008 n=5+5)
puffs_adler32_100k/clang       2.39GB/s ± 1%  2.44GB/s ± 0%  +2.42%  (p=0.008 n=5+5)
puffs_flate_decode_1k/clang     129MB/s ± 2%   133MB/s ± 0%  +3.41%  (p=0.008 n=5+5)
puffs_flate_decode_10k/clang    161MB/s ± 1%   174MB/s ± 0%  +8.46%  (p=0.008 n=5+5)
puffs_flate_decode_100k/clang   187MB/s ± 3%   205MB/s ± 0%  +9.51%  (p=0.008 n=5+5)
puffs_zlib_decode_10k/clang     150MB/s ± 1%   162MB/s ± 0%  +8.09%  (p=0.008 n=5+5)
puffs_zlib_decode_100k/clang    174MB/s ± 1%   188MB/s ± 0%  +8.13%  (p=0.008 n=5+5)

puffs_adler32_10k/gcc          3.12GB/s ± 3%  3.25GB/s ± 0%  +4.30%  (p=0.016 n=5+4)
puffs_adler32_100k/gcc         3.12GB/s ± 3%  3.25GB/s ± 0%  +4.15%  (p=0.008 n=5+5)
puffs_flate_decode_1k/gcc       128MB/s ± 3%   132MB/s ± 0%  +3.11%  (p=0.008 n=5+5)
puffs_flate_decode_10k/gcc      172MB/s ± 4%   170MB/s ± 0%    ~     (p=0.151 n=5+5)
puffs_flate_decode_100k/gcc     196MB/s ± 2%   207MB/s ± 0%  +5.46%  (p=0.008 n=5+5)
puffs_zlib_decode_10k/gcc       164MB/s ± 1%   161MB/s ± 1%  -1.34%  (p=0.008 n=5+5)
puffs_zlib_decode_100k/gcc      186MB/s ± 0%   194MB/s ± 0%  +4.12%  (p=0.016 n=4+5)
2017-11-05 06:57:10 +11:00
Nigel Tao f968950345 Read bits up front in decode_huffman_fast.
name                           old speed      new speed       delta

puffs_adler32_10k/clang        2.36GB/s ± 4%   2.38GB/s ± 2%     ~     (p=1.000 n=5+5)
puffs_adler32_100k/clang       2.39GB/s ± 4%   2.39GB/s ± 1%     ~     (p=0.841 n=5+5)
puffs_flate_decode_1k/clang     121MB/s ± 1%    129MB/s ± 2%   +5.95%  (p=0.008 n=5+5)
puffs_flate_decode_10k/clang    138MB/s ± 3%    161MB/s ± 1%  +16.52%  (p=0.008 n=5+5)
puffs_flate_decode_100k/clang   163MB/s ± 4%    187MB/s ± 3%  +14.50%  (p=0.008 n=5+5)
puffs_zlib_decode_10k/clang     132MB/s ± 2%    150MB/s ± 1%  +13.92%  (p=0.008 n=5+5)
puffs_zlib_decode_100k/clang    152MB/s ± 6%    174MB/s ± 1%  +14.60%  (p=0.008 n=5+5)

puffs_adler32_10k/gcc          3.25GB/s ± 0%   3.12GB/s ± 3%   -4.06%  (p=0.008 n=5+5)
puffs_adler32_100k/gcc         3.25GB/s ± 0%   3.12GB/s ± 3%   -3.99%  (p=0.008 n=5+5)
puffs_flate_decode_1k/gcc      95.1MB/s ± 0%  127.7MB/s ± 3%  +34.25%  (p=0.008 n=5+5)
puffs_flate_decode_10k/gcc      110MB/s ± 0%    172MB/s ± 4%  +55.66%  (p=0.008 n=5+5)
puffs_flate_decode_100k/gcc     121MB/s ± 0%    196MB/s ± 2%  +61.45%  (p=0.008 n=5+5)
puffs_zlib_decode_10k/gcc       112MB/s ± 0%    164MB/s ± 1%  +46.14%  (p=0.008 n=5+5)
puffs_zlib_decode_100k/gcc      123MB/s ± 0%    186MB/s ± 0%  +51.22%  (p=0.016 n=5+4)
2017-11-04 17:37:54 +11:00
Nigel Tao 1c7a6792de Tweak the decode_huffman_fast loop condition. 2017-11-04 15:19:40 +11:00
Nigel Tao aa1f0078db Add decode_huffman_fast.
It's not fast yet, and the immediate impact is indeed a significant
performance loss, but follow-up commits should make it faster.

name                           old speed      new speed      delta

puffs_adler32_10k/clang        2.41GB/s ± 0%  2.40GB/s ± 1%     ~     (p=0.222 n=5+5)
puffs_adler32_100k/clang       2.41GB/s ± 1%  2.39GB/s ± 1%   -0.97%  (p=0.016 n=5+5)
puffs_flate_decode_1k/clang     116MB/s ± 5%   115MB/s ± 0%     ~     (p=0.310 n=5+5)
puffs_flate_decode_10k/clang    145MB/s ± 0%   139MB/s ± 0%   -4.18%  (p=0.016 n=4+5)
puffs_flate_decode_100k/clang   164MB/s ± 5%   158MB/s ± 1%     ~     (p=0.151 n=5+5)
puffs_zlib_decode_10k/clang     135MB/s ± 2%   131MB/s ± 0%   -2.84%  (p=0.008 n=5+5)
puffs_zlib_decode_100k/clang    154MB/s ± 1%   148MB/s ± 2%   -4.09%  (p=0.016 n=4+5)

puffs_adler32_10k/gcc          3.13GB/s ± 4%  3.19GB/s ± 0%   +1.94%  (p=0.032 n=5+5)
puffs_adler32_100k/gcc         3.14GB/s ± 3%  3.20GB/s ± 3%     ~     (p=0.056 n=5+5)
puffs_flate_decode_1k/gcc       122MB/s ± 1%    94MB/s ± 1%  -22.84%  (p=0.008 n=5+5)
puffs_flate_decode_10k/gcc      143MB/s ± 0%   110MB/s ± 0%  -23.24%  (p=0.008 n=5+5)
puffs_flate_decode_100k/gcc     158MB/s ± 0%   118MB/s ± 0%  -24.98%  (p=0.008 n=5+5)
puffs_zlib_decode_10k/gcc       136MB/s ± 0%   119MB/s ± 1%  -12.63%  (p=0.008 n=5+5)
puffs_zlib_decode_100k/gcc      150MB/s ± 1%   124MB/s ± 2%  -17.17%  (p=0.008 n=5+5)
2017-11-04 14:45:11 +11:00