Nigel Tao
72b9c23120
std/jpeg: decode RGB (not just YCbCr) JPEGs
...
The benchmarks don't exercise the RGB code paths per se, but it does
affect the numbers. Especially for gcc, this essentially undoes the
performance impact of the previous (trivial) commit. This is all
presumably another code alignment coincidence per
https://news.ycombinator.com/item?id=23633583
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 376MB/s ± 0% 376MB/s ± 0% ~ (p=1.000 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 338MB/s ± 0% 341MB/s ± 0% +0.98% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 1.01GB/s ± 0% 1.01GB/s ± 0% +0.35% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 841MB/s ± 0% 836MB/s ± 1% -0.65% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 913MB/s ± 0% 912MB/s ± 0% -0.18% (p=0.016 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 1.11GB/s ± 0% 1.10GB/s ± 0% -0.34% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 920MB/s ± 0% 916MB/s ± 0% -0.42% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 349MB/s ± 0% 368MB/s ± 0% +5.60% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 328MB/s ± 0% 279MB/s ± 0% -14.82% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 925MB/s ± 0% 947MB/s ± 1% +2.33% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 732MB/s ± 0% 770MB/s ± 0% +5.23% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 798MB/s ± 0% 838MB/s ± 0% +4.99% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 962MB/s ± 0% 996MB/s ± 0% +3.54% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 806MB/s ± 0% 847MB/s ± 0% +5.19% (p=0.008 n=5+5)
2023-08-21 16:12:21 +10:00
Nigel Tao
748e44fe71
base: skip bulk load/save on short slice length
...
This is about semantics (being easier to reason about if transforming
between u32 and u8 and the u8 slice length is short and not a multiple
of 4). It's not an optimization per se, but for the record, it does have
significant performance impact:
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 369MB/s ± 0% 376MB/s ± 0% +1.97% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 309MB/s ± 0% 338MB/s ± 0% +9.29% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 1.00GB/s ± 0% 1.01GB/s ± 0% +1.31% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 827MB/s ± 0% 841MB/s ± 0% +1.76% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 897MB/s ± 0% 913MB/s ± 0% +1.86% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 1.10GB/s ± 0% 1.11GB/s ± 0% +0.11% (p=0.024 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 900MB/s ± 0% 920MB/s ± 0% +2.20% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 369MB/s ± 0% 349MB/s ± 0% -5.45% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 283MB/s ± 0% 328MB/s ± 0% +15.76% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 945MB/s ± 0% 925MB/s ± 0% -2.12% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 769MB/s ± 0% 732MB/s ± 0% -4.88% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 842MB/s ± 0% 798MB/s ± 0% -5.25% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 1.01GB/s ± 0% 0.96GB/s ± 0% -4.77% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 848MB/s ± 0% 806MB/s ± 0% -4.96% (p=0.008 n=5+5)
2023-08-19 21:18:53 +10:00
Nigel Tao
14a8bd0feb
std/jpeg: allow fewer-components sequential scans
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 366MB/s ± 0% 369MB/s ± 0% +0.81% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 302MB/s ± 0% 309MB/s ± 0% +2.28% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 1.00GB/s ± 0% 1.00GB/s ± 0% -0.58% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 820MB/s ± 0% 827MB/s ± 0% +0.80% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 895MB/s ± 0% 897MB/s ± 0% +0.24% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 1.11GB/s ± 0% 1.10GB/s ± 0% -0.30% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 897MB/s ± 0% 900MB/s ± 0% +0.30% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 370MB/s ± 0% 369MB/s ± 0% -0.26% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 309MB/s ± 0% 283MB/s ± 0% -8.35% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 948MB/s ± 0% 945MB/s ± 0% -0.27% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 769MB/s ± 0% 769MB/s ± 0% ~ (p=0.310 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 841MB/s ± 0% 842MB/s ± 0% +0.13% (p=0.016 n=4+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 1.00GB/s ± 0% 1.01GB/s ± 0% +0.75% (p=0.016 n=4+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 846MB/s ± 0% 848MB/s ± 0% +0.22% (p=0.008 n=5+5)
2023-08-19 17:22:47 +10:00
Nigel Tao
2f5f203d1b
test/3pdata: regenerate nia-checksums-of-*.txt
2023-08-19 17:18:11 +10:00
Nigel Tao
41d0c095c8
std/jpeg: implement block smoothing
...
"script/confirm-matches-djpeg.sh test/data/*.jpeg" is now all matches.
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 369MB/s ± 0% 366MB/s ± 0% -0.79% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 335MB/s ± 0% 302MB/s ± 0% -9.93% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 1.01GB/s ± 0% 1.00GB/s ± 0% -1.07% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 829MB/s ± 0% 820MB/s ± 0% -1.03% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 903MB/s ± 0% 895MB/s ± 0% -0.93% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 1.12GB/s ± 0% 1.11GB/s ± 0% -1.28% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 906MB/s ± 0% 897MB/s ± 0% -0.93% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 350MB/s ± 0% 370MB/s ± 0% +5.54% (p=0.016 n=4+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 334MB/s ± 0% 309MB/s ± 0% -7.46% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 924MB/s ± 0% 948MB/s ± 0% +2.61% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 745MB/s ± 0% 769MB/s ± 0% +3.25% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 813MB/s ± 0% 841MB/s ± 0% +3.41% (p=0.029 n=4+4)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 1.00GB/s ± 0% 1.00GB/s ± 0% +0.25% (p=0.029 n=4+4)
wuffs_jpeg_decode_4002k_24bpp/gcc10 821MB/s ± 0% 846MB/s ± 0% +3.00% (p=0.008 n=5+5)
2023-08-19 17:16:51 +10:00
Nigel Tao
d60bfb05e8
check: ban bulk methods on slices of refined types
2023-08-06 15:30:42 +10:00
Nigel Tao
be9d3fcf0f
example/convert-to-nia: add -output-netpbm flag
2023-08-06 10:49:53 +10:00
Nigel Tao
9c3275658a
doc: add note about shipping in Google Chrome M93
2023-08-03 17:38:43 +10:00
Nigel Tao
fd5ab5f7b3
pixconv: tweak "if (triangle_filter_for_2to1)"
2023-07-31 16:05:10 +10:00
Nigel Tao
836a878861
pixconv: remove unused total_src_len? variables
2023-07-31 09:38:00 +10:00
Nigel Tao
5afaa8911c
pixconv: guard swizzle_ycc__etc__avx2 with #ifdef
2023-07-30 18:52:16 +10:00
Nigel Tao
e60966aa45
pixconv: remove some obsolete swizzle_ycc code
2023-07-30 17:49:19 +10:00
Nigel Tao
b3d0d7b855
pixconv: apply YCC->RGB SIMD code to 444 JPEGs
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 368MB/s ± 0% 368MB/s ± 0% ~ (p=0.421 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 333MB/s ± 0% 335MB/s ± 0% +0.54% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 1.01GB/s ± 0% 1.01GB/s ± 0% +0.38% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 829MB/s ± 0% 828MB/s ± 0% -0.16% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 902MB/s ± 0% 902MB/s ± 0% ~ (p=0.690 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 649MB/s ± 0% 1121MB/s ± 0% +72.72% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 905MB/s ± 0% 906MB/s ± 0% ~ (p=0.222 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 362MB/s ± 0% 351MB/s ± 0% -3.06% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 328MB/s ± 0% 330MB/s ± 0% +0.53% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 926MB/s ± 0% 921MB/s ± 0% -0.54% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 757MB/s ± 0% 743MB/s ± 0% -1.83% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 835MB/s ± 0% 814MB/s ± 0% -2.55% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 655MB/s ± 0% 1001MB/s ± 0% +52.74% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 843MB/s ± 0% 820MB/s ± 0% -2.70% (p=0.016 n=4+5)
2023-07-30 15:56:59 +10:00
Nigel Tao
3dbfb00c42
pixconv: add godbolt comment for upsample code
2023-07-29 11:49:08 +10:00
Nigel Tao
2687d1daaf
pixconv: fix -Werror=pedantic/overflow errors
2023-07-23 22:11:19 +10:00
Nigel Tao
bfc7fc2f45
std/jpeg: add upsample_inv_h2v2 GCC AVX2 comment
2023-07-23 22:04:02 +10:00
Nigel Tao
51bc60ef92
std/jpeg: avoid AVX2 upsample_inv_h2v2 for GCC
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 369MB/s ± 0% 368MB/s ± 0% ~ (p=1.000 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 332MB/s ± 0% 333MB/s ± 0% +0.46% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 1.01GB/s ± 0% 1.01GB/s ± 0% ~ (p=1.000 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 829MB/s ± 0% 829MB/s ± 0% ~ (p=0.905 n=4+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 902MB/s ± 0% 902MB/s ± 0% ~ (p=0.548 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 649MB/s ± 0% 649MB/s ± 0% ~ (p=0.841 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 904MB/s ± 0% 905MB/s ± 0% +0.14% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 322MB/s ± 0% 362MB/s ± 0% +12.60% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 284MB/s ± 0% 328MB/s ± 0% +15.63% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 908MB/s ± 0% 926MB/s ± 0% +2.04% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 730MB/s ± 0% 757MB/s ± 0% +3.76% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 785MB/s ± 0% 835MB/s ± 0% +6.40% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 625MB/s ± 0% 655MB/s ± 0% +4.76% (p=0.016 n=4+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 783MB/s ± 0% 843MB/s ± 0% +7.57% (p=0.016 n=5+4)
2023-07-23 21:59:52 +10:00
Nigel Tao
5dfc913cb0
base: let cpu_arch__have_x86_etc return true
2023-07-23 21:48:33 +10:00
Nigel Tao
e2b9566fc5
base: restrict x86_avx2 to 64-bit (not 32-bit) x86
...
32-bit C/C++ compilers don't always compile the AVX intrinsics. Per the
60773 bug below,
> clang++ -m32 etc etc fuzz/c/std/jpeg_fuzzer.c
> undefined reference to `_mm_cvtsi128_si64'
> undefined reference to `_mm256_extract_epi64'
See also the "#ifdef __x86_64__" guards in
https://github.com/gcc-mirror/gcc/blob/4a3895903c29ed85da6fcb886f31ff23d4c6e935/gcc/config/i386/emmintrin.h#L228-L231
https://github.com/llvm/llvm-project/blob/08fab9ebecf72a682279d75489dc2460121cbeed/clang/lib/Headers/emmintrin.h#L3502-L3515
Note that -m32 is not the same as -mx32
> $ touch empty.c
> $ gcc -dM -E empty.c | grep -E '__i386__|__x86_64__'
> #define __x86_64__ 1
> $ gcc -m32 -dM -E empty.c | grep -E '__i386__|__x86_64__'
> #define __i386__ 1
> $ gcc -mx32 -dM -E empty.c | grep -E '__i386__|__x86_64__'
> #define __x86_64__ 1
Updates https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=60773
2023-07-23 21:20:35 +10:00
Nigel Tao
605b10a492
pixconv: fix cpu_arch__have_x86_avx2 copy/pasto
...
Not sure why the benchmark numbers change, but they do.
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 376MB/s ± 0% 369MB/s ± 0% -2.06% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 311MB/s ± 0% 332MB/s ± 0% +6.69% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 1.02GB/s ± 0% 1.01GB/s ± 0% -1.35% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 842MB/s ± 0% 829MB/s ± 0% -1.54% (p=0.016 n=5+4)
wuffs_jpeg_decode_552k_24bpp_420/clang11 917MB/s ± 0% 902MB/s ± 0% -1.62% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 649MB/s ± 0% 649MB/s ± 0% ~ (p=0.548 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 921MB/s ± 0% 904MB/s ± 0% -1.82% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 326MB/s ± 0% 322MB/s ± 0% -1.26% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 292MB/s ± 0% 284MB/s ± 0% -2.69% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 920MB/s ± 0% 908MB/s ± 0% -1.31% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 742MB/s ± 0% 730MB/s ± 0% -1.62% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 792MB/s ± 0% 785MB/s ± 0% -0.94% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 627MB/s ± 0% 625MB/s ± 0% -0.23% (p=0.029 n=4+4)
wuffs_jpeg_decode_4002k_24bpp/gcc10 792MB/s ± 0% 783MB/s ± 0% -1.13% (p=0.008 n=5+5)
2023-07-23 21:13:43 +10:00
Nigel Tao
cd769a0cdf
pixconv: add upsample_inv_h2v2_triangle_x86_avx2
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 370MB/s ± 0% 376MB/s ± 0% +1.84% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 295MB/s ± 0% 311MB/s ± 0% +5.56% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 726MB/s ± 0% 1021MB/s ± 0% +40.62% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 622MB/s ± 0% 842MB/s ± 0% +35.49% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 659MB/s ± 0% 917MB/s ± 0% +39.15% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 649MB/s ± 0% 649MB/s ± 0% -0.09% (p=0.016 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 675MB/s ± 0% 921MB/s ± 0% +36.49% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 351MB/s ± 0% 326MB/s ± 0% -7.21% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 334MB/s ± 0% 292MB/s ± 0% -12.67% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 924MB/s ± 0% 920MB/s ± 0% -0.41% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 740MB/s ± 0% 742MB/s ± 0% +0.27% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 812MB/s ± 0% 792MB/s ± 0% -2.38% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 647MB/s ± 0% 627MB/s ± 0% -3.16% (p=0.016 n=5+4)
wuffs_jpeg_decode_4002k_24bpp/gcc10 817MB/s ± 0% 792MB/s ± 0% -3.01% (p=0.016 n=4+5)
2023-07-23 12:03:07 +10:00
Nigel Tao
fed3da528c
pixconv: add pixel swizzler x86_sse42 prototypes
2023-07-22 13:23:37 +10:00
Nigel Tao
35c1f76f86
pixconv: add swizzle_ycc__convert_rgbx_x86_avx2
...
Some of the benchmark numbers move around even though they don't use the
new etc__convert_rgbx_x86_avx2 code paths. It's possibly just a code
alignment coincidence per https://news.ycombinator.com/item?id=23633583
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 372MB/s ± 0% 370MB/s ± 0% -0.64% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 278MB/s ± 0% 295MB/s ± 0% +5.89% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 710MB/s ± 0% 726MB/s ± 0% +2.30% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 603MB/s ± 0% 622MB/s ± 0% +3.12% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 635MB/s ± 0% 659MB/s ± 0% +3.82% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 650MB/s ± 0% 649MB/s ± 0% -0.17% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 629MB/s ± 0% 675MB/s ± 0% +7.20% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 364MB/s ± 0% 351MB/s ± 0% -3.62% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 334MB/s ± 0% 334MB/s ± 0% -0.06% (p=0.032 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 944MB/s ± 0% 924MB/s ± 0% -2.19% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 769MB/s ± 0% 740MB/s ± 0% -3.75% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 842MB/s ± 0% 812MB/s ± 0% -3.60% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 661MB/s ± 0% 647MB/s ± 0% -2.13% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 849MB/s ± 0% 817MB/s ± 0% -3.77% (p=0.016 n=5+4)
2023-07-22 13:11:12 +10:00
Nigel Tao
4facef4e01
pixconv: add swizzle_ycc__convert_bgrx_x86_avx2
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 365MB/s ± 1% 372MB/s ± 0% +1.77% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 241MB/s ± 0% 278MB/s ± 0% +15.49% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 477MB/s ± 0% 710MB/s ± 0% +48.91% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 429MB/s ± 0% 603MB/s ± 0% +40.56% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 446MB/s ± 0% 635MB/s ± 0% +42.18% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 644MB/s ± 0% 650MB/s ± 0% +1.06% (p=0.016 n=4+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 458MB/s ± 0% 629MB/s ± 0% +37.46% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 352MB/s ± 0% 364MB/s ± 0% +3.62% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 295MB/s ± 0% 334MB/s ± 0% +13.19% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 669MB/s ± 0% 944MB/s ± 0% +41.07% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 567MB/s ± 0% 769MB/s ± 0% +35.59% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 605MB/s ± 0% 842MB/s ± 0% +39.07% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 647MB/s ± 0% 661MB/s ± 0% +2.19% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 613MB/s ± 0% 849MB/s ± 0% +38.61% (p=0.008 n=5+5)
2023-07-22 12:51:51 +10:00
Nigel Tao
10d4ff7532
pixconv: factor out swizzle_ycc__convert_func
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 368MB/s ± 0% 365MB/s ± 1% -0.75% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 231MB/s ± 0% 241MB/s ± 0% +4.47% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 470MB/s ± 0% 477MB/s ± 0% +1.35% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 416MB/s ± 0% 429MB/s ± 0% +3.14% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 433MB/s ± 0% 446MB/s ± 0% +3.02% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 646MB/s ± 0% 644MB/s ± 0% -0.39% (p=0.016 n=5+4)
wuffs_jpeg_decode_4002k_24bpp/clang11 439MB/s ± 0% 458MB/s ± 0% +4.35% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 364MB/s ± 0% 352MB/s ± 0% -3.49% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 294MB/s ± 0% 295MB/s ± 0% +0.38% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 680MB/s ± 0% 669MB/s ± 0% -1.58% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 578MB/s ± 0% 567MB/s ± 0% -1.92% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 622MB/s ± 0% 605MB/s ± 0% -2.67% (p=0.016 n=4+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 659MB/s ± 0% 647MB/s ± 0% -1.72% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 630MB/s ± 0% 613MB/s ± 0% -2.74% (p=0.008 n=5+5)
2023-07-20 11:22:06 +10:00
Nigel Tao
6f510552be
cgen: add type casts for some avx2 methods
2023-07-18 13:26:39 +10:00
Nigel Tao
44d89c8838
base: add some more x86_m256i methods
2023-07-18 12:48:52 +10:00
Nigel Tao
ecaf25ebf0
std/jpeg: "AC all zero" fast path for x86_avx2
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 370MB/s ± 0% 368MB/s ± 0% -0.47% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 231MB/s ± 0% 231MB/s ± 0% -0.26% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 470MB/s ± 0% 470MB/s ± 0% ~ (p=0.841 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 418MB/s ± 0% 416MB/s ± 0% -0.42% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 435MB/s ± 0% 433MB/s ± 0% -0.33% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 620MB/s ± 0% 646MB/s ± 0% +4.20% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 441MB/s ± 0% 439MB/s ± 0% -0.51% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 352MB/s ± 0% 364MB/s ± 0% +3.41% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 296MB/s ± 0% 294MB/s ± 0% -0.69% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 666MB/s ± 0% 680MB/s ± 0% +2.10% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 565MB/s ± 0% 578MB/s ± 0% +2.34% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 608MB/s ± 0% 622MB/s ± 0% +2.26% (p=0.016 n=5+4)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 624MB/s ± 0% 659MB/s ± 0% +5.47% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 616MB/s ± 0% 630MB/s ± 0% +2.29% (p=0.008 n=5+5)
2023-07-18 12:04:19 +10:00
Nigel Tao
2bff962cf1
std/jpeg: add decoder.decode_idct_x86_avx2
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 237MB/s ± 0% 370MB/s ± 0% +55.94% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 205MB/s ± 0% 231MB/s ± 0% +12.80% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 379MB/s ± 0% 470MB/s ± 0% +23.92% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 332MB/s ± 0% 418MB/s ± 0% +25.95% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 345MB/s ± 0% 435MB/s ± 0% +25.99% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 483MB/s ± 0% 620MB/s ± 0% +28.53% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 349MB/s ± 0% 441MB/s ± 0% +26.47% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 231MB/s ± 0% 352MB/s ± 0% +52.78% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 227MB/s ± 0% 296MB/s ± 0% +30.44% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 500MB/s ± 0% 666MB/s ± 0% +33.21% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 417MB/s ± 0% 565MB/s ± 0% +35.32% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 440MB/s ± 0% 608MB/s ± 0% +38.37% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 478MB/s ± 0% 624MB/s ± 0% +30.70% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 445MB/s ± 0% 616MB/s ± 0% +38.42% (p=0.008 n=5+5)
2023-07-18 09:17:08 +10:00
Nigel Tao
0af626b850
std/jpeg: grow quant_tables' elements, s/u8/u16/
2023-07-17 10:27:49 +10:00
Nigel Tao
e5b6f2563e
std/jpeg: prepare decode_idct_x86_avx2.wuffs
2023-07-16 19:54:36 +10:00
Nigel Tao
4dfd70928b
std: add "get_quirk(key: u32) u64"
2023-07-16 12:04:09 +10:00
Nigel Tao
84d1aa0d79
cmd/wuffs: show cmdArgs when "wuffs gen" fails
2023-07-16 11:44:08 +10:00
Nigel Tao
8954281a4e
pixconv: allow decoding to PIXEL_FORMAT__RGB
...
"wuffs genlib" binary size changes:
77936 clang-static/wuffs-base-pixconv.o # Before = 1.00x
79792 clang-static/wuffs-base-pixconv.o # After = 1.02x
173784 gcc-static/wuffs-base-pixconv.o # Before = 1.00x
177112 gcc-static/wuffs-base-pixconv.o # After = 1.02x
Some of the benchmark numbers move around even though they don't use the
new WUFFS_BASE__PIXEL_FORMAT__RGB code paths. It's possibly just a code
alignment coincidence per https://news.ycombinator.com/item?id=23633583
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 249MB/s ± 0% 251MB/s ± 0% +0.76% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 222MB/s ± 0% 225MB/s ± 0% +1.16% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 405MB/s ± 0% 406MB/s ± 0% ~ (p=1.000 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 357MB/s ± 0% 356MB/s ± 0% -0.24% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 370MB/s ± 0% 372MB/s ± 0% +0.46% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 504MB/s ± 0% 508MB/s ± 0% +0.83% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 375MB/s ± 0% 374MB/s ± 0% ~ (p=0.095 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 258MB/s ± 0% 254MB/s ± 0% -1.68% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 267MB/s ± 0% 240MB/s ± 0% -9.95% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 537MB/s ± 0% 537MB/s ± 0% ~ (p=0.556 n=4+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 458MB/s ± 0% 453MB/s ± 0% -1.18% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 482MB/s ± 0% 478MB/s ± 0% -0.78% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 508MB/s ± 0% 511MB/s ± 0% +0.49% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 489MB/s ± 0% 484MB/s ± 0% -1.11% (p=0.016 n=4+5)
wuffs_png_decode_image_19k_8bpp/clang11 269MB/s ± 0% 275MB/s ± 0% +2.25% (p=0.008 n=5+5)
wuffs_png_decode_image_40k_24bpp/clang11 328MB/s ± 0% 328MB/s ± 0% ~ (p=0.841 n=5+5)
wuffs_png_decode_image_77k_8bpp/clang11 949MB/s ± 0% 962MB/s ± 0% +1.29% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/clang11 854MB/s ± 0% 859MB/s ± 0% +0.54% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_verify_checksum/clang11 822MB/s ± 0% 826MB/s ± 0% +0.52% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/clang11 331MB/s ± 0% 332MB/s ± 0% +0.15% (p=0.008 n=5+5)
wuffs_png_decode_image_19k_8bpp/gcc10 287MB/s ± 0% 277MB/s ± 0% -3.46% (p=0.008 n=5+5)
wuffs_png_decode_image_40k_24bpp/gcc10 341MB/s ± 0% 346MB/s ± 0% +1.34% (p=0.008 n=5+5)
wuffs_png_decode_image_77k_8bpp/gcc10 1.02GB/s ± 0% 1.01GB/s ± 0% -0.41% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/gcc10 939MB/s ± 0% 916MB/s ± 0% -2.50% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_verify_checksum/gcc10 895MB/s ± 0% 874MB/s ± 0% -2.41% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/gcc10 352MB/s ± 0% 350MB/s ± 0% -0.44% (p=0.008 n=5+5)
Updates #131
2023-07-16 11:17:45 +10:00
Nigel Tao
129d8008a2
doc: add more "decode an image" docs/comments
2023-07-15 13:26:45 +10:00
Nigel Tao
2644620770
check: allow p[i] if p has type "ptr array[N] T"
2023-06-29 18:15:08 +10:00
Nigel Tao
b97560b292
cgen: emit "do { etc } while (0);" for some loops
2023-06-28 22:58:07 +10:00
Nigel Tao
99cc599f92
example/toy-aux-image: comment on config macros
2023-06-28 14:11:38 +10:00
Nigel Tao
dd29401137
std/jpeg: re-format generated IDCT code
2023-06-28 13:49:55 +10:00
Nigel Tao
97c4e85e0e
std/jpeg: add more-SIMD-like IDCT algorithm
...
This adds commented-out code that is equivalent to the existing IDCT
algorithm. It's slower as is (and hence commented out) but it is closer
to a follow-up commit's SIMD-optimized algorithm.
2023-06-28 13:45:25 +10:00
Nigel Tao
49f50cca91
std/jpeg: cap the implicit-zero bitstream padding
...
Updates https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=59966
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 241MB/s ± 0% 237MB/s ± 0% -1.57% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 213MB/s ± 0% 212MB/s ± 0% -0.43% (p=0.016 n=5+4)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 387MB/s ± 0% 386MB/s ± 0% -0.08% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 341MB/s ± 0% 340MB/s ± 0% -0.50% (p=0.016 n=4+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 356MB/s ± 0% 353MB/s ± 0% -0.89% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 484MB/s ± 0% 480MB/s ± 0% -0.96% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 364MB/s ± 0% 360MB/s ± 0% -1.12% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 244MB/s ± 0% 247MB/s ± 0% +0.88% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 254MB/s ± 0% 254MB/s ± 0% -0.20% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 508MB/s ± 0% 512MB/s ± 0% +0.61% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 434MB/s ± 0% 436MB/s ± 0% +0.39% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 456MB/s ± 0% 458MB/s ± 0% +0.41% (p=0.016 n=4+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 481MB/s ± 0% 483MB/s ± 0% +0.49% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 463MB/s ± 0% 466MB/s ± 0% +0.57% (p=0.016 n=4+5)
2023-06-23 21:38:01 +10:00
Nigel Tao
66ef485efa
cgen: emit "break/continue" not "goto", part 3
...
A selection of benchmarks below. Some are up, some are down.
name old speed new speed delta
wuffs_bzip2_decode_10k/clang11 59.5MB/s ± 0% 59.1MB/s ± 0% -0.72% (p=0.008 n=5+5)
wuffs_bzip2_decode_100k/clang11 49.6MB/s ± 1% 47.4MB/s ± 0% -4.48% (p=0.008 n=5+5)
wuffs_bzip2_decode_10k/gcc10 59.0MB/s ± 0% 59.8MB/s ± 0% +1.40% (p=0.008 n=5+5)
wuffs_bzip2_decode_100k/gcc10 49.8MB/s ± 2% 53.0MB/s ± 0% +6.51% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_full_init/clang11 184MB/s ± 0% 182MB/s ± 0% -1.06% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_part_init/clang11 213MB/s ± 0% 211MB/s ± 0% -0.65% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_full_init/clang11 399MB/s ± 0% 409MB/s ± 0% +2.68% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_part_init/clang11 409MB/s ± 0% 421MB/s ± 0% +2.96% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_just_one_read/clang11 518MB/s ± 0% 517MB/s ± 0% ~ (p=0.095 n=5+5)
wuffs_deflate_decode_100k_many_big_reads/clang11 322MB/s ± 0% 319MB/s ± 0% -0.93% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_full_init/gcc10 173MB/s ± 0% 175MB/s ± 0% +1.07% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_part_init/gcc10 205MB/s ± 0% 208MB/s ± 0% +1.26% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_full_init/gcc10 408MB/s ± 0% 396MB/s ± 0% -2.94% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_part_init/gcc10 421MB/s ± 0% 408MB/s ± 0% -2.88% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_just_one_read/gcc10 542MB/s ± 0% 519MB/s ± 0% -4.13% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_many_big_reads/gcc10 340MB/s ± 0% 327MB/s ± 0% -3.77% (p=0.008 n=5+5)
wuffs_json_decode_1k/clang11 1.26GB/s ± 0% 1.34GB/s ± 0% +6.90% (p=0.008 n=5+5)
wuffs_json_decode_21k_formatted/clang11 681MB/s ± 0% 684MB/s ± 0% +0.47% (p=0.008 n=5+5)
wuffs_json_decode_26k_compact/clang11 1.00GB/s ± 0% 1.14GB/s ± 0% +14.26% (p=0.008 n=5+5)
wuffs_json_decode_217k_stringy/clang11 786MB/s ± 0% 907MB/s ± 0% +15.40% (p=0.008 n=5+5)
wuffs_json_decode_1k/gcc10 1.20GB/s ± 0% 1.22GB/s ± 0% +1.41% (p=0.008 n=5+5)
wuffs_json_decode_21k_formatted/gcc10 681MB/s ± 0% 719MB/s ± 0% +5.59% (p=0.008 n=5+5)
wuffs_json_decode_26k_compact/gcc10 957MB/s ± 0% 976MB/s ± 0% +1.94% (p=0.008 n=5+5)
wuffs_json_decode_217k_stringy/gcc10 773MB/s ± 0% 781MB/s ± 0% +1.01% (p=0.008 n=5+5)
wuffs_png_decode_image_19k_8bpp/clang11 256MB/s ± 0% 256MB/s ± 0% ~ (p=0.421 n=5+5)
wuffs_png_decode_image_40k_24bpp/clang11 311MB/s ± 0% 313MB/s ± 0% +0.57% (p=0.008 n=5+5)
wuffs_png_decode_image_77k_8bpp/clang11 907MB/s ± 0% 904MB/s ± 0% -0.28% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/clang11 821MB/s ± 0% 812MB/s ± 0% -1.08% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_verify_checksum/clang11 790MB/s ± 0% 782MB/s ± 0% -0.94% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/clang11 315MB/s ± 0% 315MB/s ± 0% -0.09% (p=0.008 n=5+5)
wuffs_png_decode_image_19k_8bpp/gcc10 266MB/s ± 0% 273MB/s ± 0% +2.94% (p=0.008 n=5+5)
wuffs_png_decode_image_40k_24bpp/gcc10 329MB/s ± 0% 324MB/s ± 2% -1.65% (p=0.008 n=5+5)
wuffs_png_decode_image_77k_8bpp/gcc10 952MB/s ± 2% 967MB/s ± 0% +1.57% (p=0.016 n=5+4)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/gcc10 871MB/s ± 0% 894MB/s ± 0% +2.59% (p=0.029 n=4+4)
wuffs_png_decode_image_552k_32bpp_verify_checksum/gcc10 831MB/s ± 0% 851MB/s ± 0% +2.45% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/gcc10 336MB/s ± 0% 335MB/s ± 0% -0.53% (p=0.008 n=5+5)
2023-06-23 21:04:24 +10:00
Nigel Tao
de55ff7746
cgen: emit "break/continue" not "goto", part 2
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 241MB/s ± 0% 241MB/s ± 0% ~ (p=0.690 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 213MB/s ± 0% 213MB/s ± 0% ~ (p=0.690 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 386MB/s ± 0% 387MB/s ± 0% ~ (p=0.270 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 341MB/s ± 0% 341MB/s ± 0% ~ (p=0.556 n=5+4)
wuffs_jpeg_decode_552k_24bpp_420/clang11 356MB/s ± 0% 356MB/s ± 0% ~ (p=0.310 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 484MB/s ± 0% 484MB/s ± 0% ~ (p=0.095 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 362MB/s ± 0% 364MB/s ± 0% ~ (p=0.095 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 245MB/s ± 0% 244MB/s ± 0% -0.19% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 252MB/s ± 0% 254MB/s ± 0% +0.76% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 513MB/s ± 0% 508MB/s ± 0% -0.88% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 433MB/s ± 0% 434MB/s ± 0% +0.38% (p=0.016 n=4+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 458MB/s ± 0% 456MB/s ± 0% -0.41% (p=0.029 n=4+4)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 486MB/s ± 0% 481MB/s ± 0% -1.00% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 465MB/s ± 0% 463MB/s ± 0% -0.39% (p=0.029 n=4+4)
2023-06-23 18:29:20 +10:00
Nigel Tao
f155db4d5c
cgen: emit "break/continue" not "goto", part 1
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 229MB/s ± 0% 241MB/s ± 0% +5.29% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 205MB/s ± 0% 213MB/s ± 0% +3.89% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 375MB/s ± 0% 386MB/s ± 0% +3.01% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 326MB/s ± 0% 341MB/s ± 0% +4.80% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 339MB/s ± 0% 356MB/s ± 0% +5.14% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 475MB/s ± 0% 484MB/s ± 0% +2.04% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 343MB/s ± 0% 362MB/s ± 0% +5.77% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 245MB/s ± 0% 245MB/s ± 0% -0.07% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 254MB/s ± 0% 252MB/s ± 0% -0.52% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 514MB/s ± 0% 513MB/s ± 0% ~ (p=0.151 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 434MB/s ± 0% 433MB/s ± 0% -0.35% (p=0.016 n=5+4)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 459MB/s ± 0% 458MB/s ± 0% -0.12% (p=0.016 n=5+4)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 486MB/s ± 0% 486MB/s ± 0% -0.03% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 467MB/s ± 0% 465MB/s ± 0% -0.26% (p=0.016 n=5+4)
2023-06-23 17:46:40 +10:00
Nigel Tao
8d9d6c1292
Performance mystery: load-bearing "no-op continue"
...
There is no semantic difference in this commit. There's no further code
after the if-else chain in the loop body. Nonetheless, adding a "no-op
continue" impacts performance: bad for clang and good for gcc.
Profiling suggests we spend less than 5% overall in
wuffs_jpeg__decoder__fill_bitstream.
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 241MB/s ± 0% 229MB/s ± 0% -5.02% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 213MB/s ± 0% 205MB/s ± 0% -4.13% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 386MB/s ± 0% 375MB/s ± 0% -2.88% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 341MB/s ± 0% 326MB/s ± 0% -4.58% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 356MB/s ± 0% 339MB/s ± 0% -4.90% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 484MB/s ± 0% 475MB/s ± 0% -1.96% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 364MB/s ± 0% 343MB/s ± 0% -5.83% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 234MB/s ± 0% 245MB/s ± 0% +4.68% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 245MB/s ± 0% 254MB/s ± 0% +3.35% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 500MB/s ± 0% 514MB/s ± 0% +2.81% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 417MB/s ± 0% 434MB/s ± 0% +4.18% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 442MB/s ± 0% 459MB/s ± 0% +3.76% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 477MB/s ± 0% 486MB/s ± 0% +1.99% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 450MB/s ± 0% 467MB/s ± 0% +3.70% (p=0.008 n=5+5)
It is a mystery why this happens. Note though that the "continue" in the
Wuffs code becomes a "goto" in the generated C code, because (in the
general case) Wuffs has "labeled continue/break", which can jump out of
a not-inner-most loop, and C99 does not. Wuffs' "generate C code" thus
uses a C "goto". A follow-up commit may try a C "continue" instead.
To reproduce:
----
$ git clone https://github.com/google/wuffs.git
$ cd wuffs
$ gcc --version | head -n 1
gcc (Debian 10.2.1-6) 10.2.1 20210110
$ # Check out this commit.
$ git checkout --quiet THE_HASH_OF_THIS_COMMIT
$ gcc -O3 -Wall -std=c99 test/c/std/jpeg.c -DWUFFS_MIMIC -ljpeg
$ ./a.out -bench -focus=wuffs_jpeg_decode_77k_24bpp | grep MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 176988 ns/op 433.927 MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 176849 ns/op 434.268 MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 176800 ns/op 434.389 MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 176844 ns/op 434.279 MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 176869 ns/op 434.218 MB/s
$ # Check out the parent commit.
$ git checkout --quiet THE_HASH_OF_THIS_COMMIT^
$ gcc -O3 -Wall -std=c99 test/c/std/jpeg.c -DWUFFS_MIMIC -ljpeg
$ ./a.out -bench -focus=wuffs_jpeg_decode_77k_24bpp | grep MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 183848 ns/op 417.734 MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 183870 ns/op 417.684 MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 183833 ns/op 417.770 MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 183873 ns/op 417.679 MB/s
Benchmarkwuffs_jpeg_decode_77k_24bpp/gcc10 3000 183843 ns/op 417.746 MB/s
----
2023-06-23 17:02:14 +10:00
Nigel Tao
40fca0f03d
std/jpeg: roll back fill_bitstream inner loop
...
Commit 723c604a "std/jpeg: tweak decoder.fill_bitstream inner loop"
inadvertently changed "increment wi" behavior in the rare code paths.
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 237MB/s ± 0% 241MB/s ± 0% +1.79% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 213MB/s ± 0% 213MB/s ± 0% +0.36% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 384MB/s ± 0% 386MB/s ± 0% +0.50% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 338MB/s ± 0% 341MB/s ± 0% +0.95% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 353MB/s ± 0% 356MB/s ± 0% +0.99% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 481MB/s ± 0% 484MB/s ± 0% +0.61% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 359MB/s ± 0% 364MB/s ± 0% +1.40% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 245MB/s ± 0% 234MB/s ± 0% -4.45% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 247MB/s ± 0% 245MB/s ± 0% -0.59% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 510MB/s ± 0% 500MB/s ± 0% -1.94% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 432MB/s ± 0% 417MB/s ± 0% -3.62% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 456MB/s ± 0% 442MB/s ± 0% -3.14% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 484MB/s ± 0% 477MB/s ± 0% -1.49% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 465MB/s ± 0% 450MB/s ± 0% -3.28% (p=0.008 n=5+5)
2023-06-23 16:29:16 +10:00
Nigel Tao
a00150f62f
std/jpeg: tweak decode_mcu up-front r.length check
...
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 236MB/s ± 0% 237MB/s ± 0% +0.11% (p=0.032 n=4+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 210MB/s ± 0% 213MB/s ± 0% +1.20% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 385MB/s ± 0% 384MB/s ± 0% ~ (p=0.095 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 338MB/s ± 0% 338MB/s ± 0% ~ (p=0.151 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 353MB/s ± 0% 353MB/s ± 0% ~ (p=0.548 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 481MB/s ± 0% 481MB/s ± 0% ~ (p=0.690 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 360MB/s ± 0% 359MB/s ± 0% -0.35% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 244MB/s ± 0% 245MB/s ± 0% +0.44% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 251MB/s ± 0% 247MB/s ± 0% -1.46% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 509MB/s ± 0% 510MB/s ± 0% +0.13% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 431MB/s ± 0% 432MB/s ± 0% +0.24% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 455MB/s ± 0% 456MB/s ± 0% +0.26% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 481MB/s ± 0% 484MB/s ± 0% +0.56% (p=0.016 n=4+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10 463MB/s ± 0% 465MB/s ± 0% +0.36% (p=0.008 n=5+5)
2023-06-22 11:37:40 +10:00
Nigel Tao
255516bc21
std/jpeg: comments for "read Huffman LUT earlier"
...
This commit is the second of two. It just adds comments to the code
added by the previous commit.
2023-06-21 22:51:22 +10:00
Nigel Tao
d9c7740ea6
std/jpeg: read Huffman look-up table earlier
...
This commit is the first of two. It just changes code, and is the
minimal code change that leads to these performance improvements. The
follow-up commit will add comments that will also include a reference to
this one's commit hash. Even later commits may do some more refactoring.
name old speed new speed delta
wuffs_jpeg_decode_19k_8bpp/clang11 224MB/s ± 0% 236MB/s ± 0% +5.38% (p=0.016 n=5+4)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11 211MB/s ± 0% 210MB/s ± 0% -0.55% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11 377MB/s ± 0% 385MB/s ± 0% +1.91% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11 328MB/s ± 0% 338MB/s ± 0% +3.23% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11 341MB/s ± 0% 353MB/s ± 0% +3.47% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11 472MB/s ± 0% 481MB/s ± 0% +1.93% (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11 347MB/s ± 0% 360MB/s ± 0% +3.70% (p=0.008 n=5+5)
wuffs_jpeg_decode_19k_8bpp/gcc10 225MB/s ± 0% 244MB/s ± 0% +8.27% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10 246MB/s ± 0% 251MB/s ± 0% +1.67% (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10 498MB/s ± 0% 509MB/s ± 0% +2.18% (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10 408MB/s ± 0% 431MB/s ± 0% +5.86% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10 428MB/s ± 0% 455MB/s ± 0% +6.38% (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10 470MB/s ± 0% 481MB/s ± 0% +2.46% (p=0.016 n=5+4)
wuffs_jpeg_decode_4002k_24bpp/gcc10 433MB/s ± 0% 463MB/s ± 0% +7.03% (p=0.008 n=5+5)
2023-06-21 21:56:54 +10:00
Nigel Tao
bedf2bd7c3
lang: let skip_u32_fast imply can_undo_byte
2023-06-20 23:02:01 +10:00