Commit Graph
18 Commits
Author SHA1 Message Date
Nigel Tao f9cea5c17c Fix the adler.update implementation. 2017-10-19 20:45:43 +11:00
Nigel Tao d4372cb5b8 Change license to Apache 2.
Also remove explicit PATENTS file. The new LICENSE file already
discusses patents.

Note that the Puffs project has not yet been publicly released.
2017-10-12 11:17:41 +11:00
Nigel Tao cf6b1be5b5 Re-organize flate Huffman decoder table bits.
Benchmarks are mixed, but it seems conceptually cleaner to arrange the
bits in the order that they're examined.

name                           old speed     new speed     delta
puffs_flate_decode_1k/clang    125MB/s ± 0%  123MB/s ± 1%  -1.91%  (p=0.008 n=5+5)
puffs_flate_decode_10k/clang   163MB/s ± 1%  157MB/s ± 0%  -3.55%  (p=0.008 n=5+5)
puffs_flate_decode_100k/clang  175MB/s ± 0%  173MB/s ± 0%  -0.75%  (p=0.016 n=4+5)
puffs_flate_decode_1k/gcc      127MB/s ± 0%  130MB/s ± 0%  +2.14%  (p=0.008 n=5+5)
puffs_flate_decode_10k/gcc     142MB/s ± 1%  149MB/s ± 0%  +4.99%  (p=0.016 n=5+4)
puffs_flate_decode_100k/gcc    162MB/s ± 2%  168MB/s ± 0%  +3.66%  (p=0.016 n=5+4)
2017-10-06 13:07:49 +11:00
Nigel Tao 2e6b364dcc Do 8- instead of 1-byte back-copies.
name                           old speed     new speed     delta
puffs_flate_decode_1k/clang    191MB/s ± 1%  171MB/s ±12%     ~     (p=0.095 n=5+5)
puffs_flate_decode_10k/clang   277MB/s ± 1%  325MB/s ± 7%  +17.47%  (p=0.008 n=5+5)
puffs_flate_decode_100k/clang  315MB/s ± 1%  416MB/s ± 1%  +31.88%  (p=0.008 n=5+5)
puffs_flate_decode_1k/gcc      194MB/s ± 2%  197MB/s ± 0%   +1.98%  (p=0.016 n=5+4)
puffs_flate_decode_10k/gcc     279MB/s ± 1%  361MB/s ± 1%  +29.19%  (p=0.016 n=4+5)
puffs_flate_decode_100k/gcc    315MB/s ± 0%  437MB/s ± 0%  +38.77%  (p=0.008 n=5+5)

Note that these are benchmark numbers for an 'ideal', hand-written C
implementation, not for Puffs' generated C code.
2017-08-03 17:45:47 +10:00
Nigel Tao c3aa366c66 Clean up #ifdef style; remove unnecessary ()s. 2017-08-03 11:05:33 +10:00
Nigel Tao 6af3852fa8 Change the back-copy unroll count from 8 to 3.
We're possibly overfitting now on the very small benchmark corpus (3
files), but we can test on larger corpuses later. Algorithmically, it
does match what zlib does.

name                           old speed     new speed     delta
puffs_flate_decode_1k/clang    188MB/s ± 0%  191MB/s ± 1%   +1.55%  (p=0.008 n=5+5)
puffs_flate_decode_10k/clang   242MB/s ± 2%  277MB/s ± 1%  +14.72%  (p=0.008 n=5+5)
puffs_flate_decode_100k/clang  273MB/s ± 1%  315MB/s ± 1%  +15.63%  (p=0.008 n=5+5)
puffs_flate_decode_1k/gcc      187MB/s ± 0%  194MB/s ± 2%   +3.56%  (p=0.008 n=5+5)
puffs_flate_decode_10k/gcc     242MB/s ± 0%  279MB/s ± 1%  +15.29%  (p=0.016 n=5+4)
puffs_flate_decode_100k/gcc    288MB/s ± 1%  315MB/s ± 0%   +9.29%  (p=0.008 n=5+5)

Note that these are benchmark numbers for an 'ideal', hand-written C
implementation, not for Puffs' generated C code.
2017-08-03 10:53:41 +10:00
Nigel Tao fa89c397b0 Do one 8-byte load instead of six 1-byte loads.
name                           old speed     new speed     delta
puffs_flate_decode_1k/clang    178MB/s ± 2%  188MB/s ± 0%  +5.98%  (p=0.008 n=5+5)
puffs_flate_decode_10k/clang   235MB/s ± 1%  242MB/s ± 2%  +2.69%  (p=0.032 n=5+5)
puffs_flate_decode_100k/clang  274MB/s ± 0%  273MB/s ± 1%    ~     (p=0.690 n=5+5)
puffs_flate_decode_1k/gcc      177MB/s ± 0%  187MB/s ± 0%  +5.44%  (p=0.008 n=5+5)
puffs_flate_decode_10k/gcc     234MB/s ± 0%  242MB/s ± 0%  +3.64%  (p=0.008 n=5+5)
puffs_flate_decode_100k/gcc    275MB/s ± 0%  288MB/s ± 1%  +4.68%  (p=0.008 n=5+5)

Note that these are benchmark numbers for an 'ideal', hand-written C
implementation, not for Puffs' generated C code.
2017-08-03 10:47:44 +10:00
Nigel Tao bc92bf4ff8 Fix C decode_huffman's exit with excess high bits. 2017-08-03 10:34:37 +10:00
Nigel Tao 265d3a5a1a Add a hand-written C impl of decode_huffman.
This gives a rough upper bound on how fast Puffs' generated C code can
be, with a sufficiently smart Puffs compiler.

A comparison of the 'old', safe, generated C code and the 'new', unsafe,
hand-written C code:

name                           old speed     new speed     delta
puffs_flate_decode_1k/clang    127MB/s ± 2%  181MB/s ± 2%  +42.27%  (p=0.000 n=83+5)
puffs_flate_decode_10k/clang   151MB/s ± 1%  235MB/s ± 2%  +56.04%  (p=0.000 n=82+5)
puffs_flate_decode_100k/clang  165MB/s ± 1%  275MB/s ± 3%  +66.69%  (p=0.000 n=89+5)
puffs_flate_decode_1k/gcc      155MB/s ± 2%  181MB/s ± 1%  +16.96%  (p=0.000 n=91+5)
puffs_flate_decode_10k/gcc     146MB/s ± 2%  241MB/s ± 2%  +64.78%  (p=0.000 n=86+5)
puffs_flate_decode_100k/gcc    166MB/s ± 1%  277MB/s ± 3%  +66.78%  (p=0.000 n=88+5)

While we're discussing benchmark numbers, below is an overall snapshot.

std/flate

puffs_flate_decode_1k/clang       127MB/s ± 2%
puffs_flate_decode_10k/clang      151MB/s ± 1%
puffs_flate_decode_100k/clang     165MB/s ± 1%

puffs_flate_decode_1k/gcc         155MB/s ± 2%
puffs_flate_decode_10k/gcc        146MB/s ± 2%
puffs_flate_decode_100k/gcc       166MB/s ± 1%

mimic_adler32_10k                2.98GB/s ± 1%
mimic_adler32_100k               2.93GB/s ± 1%
mimic_crc32_10k                  1.38GB/s ± 1%
mimic_crc32_100k                 1.37GB/s ± 1%

mimic_flate_decode_1k             220MB/s ± 2%
mimic_flate_decode_10k            262MB/s ± 1%
mimic_flate_decode_100k           281MB/s ± 1%
mimic_gzip_decode_10k             221MB/s ± 1%
mimic_gzip_decode_100k            259MB/s ± 1%
mimic_zlib_decode_10k             242MB/s ± 2%
mimic_zlib_decode_100k            289MB/s ± 1%

std/gif

puffs_gif_lzw_decode_20k/clang    125MB/s ± 1%
puffs_gif_lzw_decode_100k/clang   173MB/s ± 1%
puffs_gif_decode_1k/clang         376MB/s ± 3%
puffs_gif_decode_10k/clang        128MB/s ± 2%
puffs_gif_decode_100k/clang       120MB/s ± 2%
puffs_gif_decode_1000k/clang      121MB/s ± 3%

puffs_gif_lzw_decode_20k/gcc      153MB/s ± 1%
puffs_gif_lzw_decode_100k/gcc     208MB/s ± 1%
puffs_gif_decode_1k/gcc           429MB/s ± 1%
puffs_gif_decode_10k/gcc          161MB/s ± 1%
puffs_gif_decode_100k/gcc         140MB/s ± 1%
puffs_gif_decode_1000k/gcc        143MB/s ± 1%

mimic_gif_decode_1k               158MB/s ± 1%
mimic_gif_decode_10k             93.9MB/s ± 1%
mimic_gif_decode_100k             100MB/s ± 1%
mimic_gif_decode_1000k            102MB/s ± 1%
2017-07-30 11:13:40 +10:00
Nigel Tao 89d4f68036 Fill array-of-u32 Huffman table for base+extra. 2017-07-27 17:20:31 +10:00
Nigel Tao adadb0fd5a Add a worked example of the flate file format. 2017-07-20 15:36:59 +10:00
Nigel Tao c463f4956f Add script/print-bits.go. 2017-07-17 17:28:19 +10:00
Nigel Tao 547351cfec Add testdata/romeo.txt.{flate,zlib}. 2017-07-17 16:53:45 +10:00
Nigel Tao 03572374dd Add zlib (not just raw flate) benchmark. 2017-07-14 12:00:00 +10:00
Nigel Tao 4efce30bfc Add more testdata/*.gz files. 2017-07-06 16:35:18 +10:00
Nigel Tao 30ca6d42b2 Rename some testdata files for consistency. 2017-06-30 10:47:00 +10:00
Nigel Tao d9b1602557 Add testdata/pi.txt* data. 2017-06-23 18:17:57 +10:00
Nigel Tao 2e35b4cfd0 Rename scripts to script, consistent with cmd. 2017-06-04 10:23:25 +10:00