Nigel Tao
|
62ee403d06
|
Add arm_neon u8x8, u16x4, etc types
|
2021-02-24 14:28:45 +11:00 |
|
Nigel Tao
|
3d6c609dc1
|
Add example/jsonptr/README.md
|
2021-02-22 21:05:53 +11:00 |
|
Nigel Tao
|
041267957c
|
Add example/json* -input-jwcc and -jwcc flags
|
2021-02-22 12:29:14 +11:00 |
|
Nigel Tao
|
773994c3f6
|
Have jsonptr collapse consecutive blank lines
|
2021-02-22 12:04:08 +11:00 |
|
Nigel Tao
|
d29cb7d304
|
Refactor cgen's writeBuiltinCPUArchX86
|
2021-02-18 23:49:51 +11:00 |
|
Nigel Tao
|
1765863601
|
Optimize make_m128i_slice128 pointer arg
|
2021-02-18 22:43:13 +11:00 |
|
Nigel Tao
|
fa774e647f
|
Add x86_sse42_utility.make_m128i_slice128
|
2021-02-18 22:32:21 +11:00 |
|
Nigel Tao
|
d4c23d091a
|
Add base.x86_sse42_utility.make_m128i_single_u32
|
2021-02-18 20:30:59 +11:00 |
|
Nigel Tao
|
8c8f534a5c
|
Add base.x86_sse42_utility
|
2021-02-18 20:02:58 +11:00 |
|
Nigel Tao
|
492068f3dc
|
Prepare to add cpu_arch_utility types
|
2021-02-18 16:16:48 +11:00 |
|
Nigel Tao
|
869be76fae
|
Rename some std/png c/p variables
|
2021-02-18 14:32:35 +11:00 |
|
Nigel Tao
|
fd071dfc90
|
Ban cpu_arch types from fields and func args/rets
|
2021-02-18 14:13:53 +11:00 |
|
Nigel Tao
|
8c4348b8f6
|
Rename some std etc_sse42 functions
|
2021-02-18 13:48:46 +11:00 |
|
Nigel Tao
|
ce97a7ccd9
|
Rename wuffs_base__cpu_arch__have_sse42
|
2021-02-18 13:44:33 +11:00 |
|
Nigel Tao
|
02e0955897
|
Rename std/png/decode_filter_sse42.wuffs
|
2021-02-18 13:40:40 +11:00 |
|
Nigel Tao
|
df9ec5bfb6
|
Check foo is choosy in "choose foo = etc"
|
2021-02-18 13:38:27 +11:00 |
|
Nigel Tao
|
26c7fb941f
|
Un-choosy-fy std/png decoder.filter_2
|
2021-02-18 13:32:25 +11:00 |
|
Nigel Tao
|
bbc720c261
|
Delete wuffs_base__cpu_arch__x86_64__capabilities
|
2021-02-18 13:30:28 +11:00 |
|
Nigel Tao
|
c8be16af3e
|
Add lib/armneonintrinsics
|
2021-02-18 10:33:14 +11:00 |
|
Nigel Tao
|
4d2df86a6d
|
Fix some -Werror=conversion false positives
|
2021-02-16 22:45:50 +11:00 |
|
Nigel Tao
|
b77137686d
|
Remove redundant arm_neon zero-initialization
|
2021-02-16 22:32:06 +11:00 |
|
Nigel Tao
|
9c1af68889
|
Factor out std/crc32/common_up_arm_crc32.wuffs
|
2021-02-16 22:27:33 +11:00 |
|
Nigel Tao
|
4c3f8ccfff
|
Factor out std/adler32/common_up_x86_sse42.wuffs
|
2021-02-16 22:23:54 +11:00 |
|
Nigel Tao
|
010b34ed38
|
Add std/adler32 hasher.up_arm_neon
On a Raspberry Pi 4 (32-bit armv7l) with -march=native and -mfpu=neon
("native" means "armv8-a+crc+simd"):
wuffs_adler32_10k/clang9 1.21GB/s ± 0% 6.58GB/s ± 1% +442.43% (p=0.008 n=5+5)
wuffs_adler32_100k/clang9 1.20GB/s ± 0% 6.58GB/s ± 0% +449.37% (p=0.008 n=5+5)
wuffs_adler32_10k/gcc8 1.24GB/s ± 0% 4.31GB/s ± 0% +248.87% (p=0.008 n=5+5)
wuffs_adler32_100k/gcc8 1.22GB/s ± 0% 4.29GB/s ± 0% +251.09% (p=0.008 n=5+5)
mimic_adler32_10k 1.40GB/s ± 0% 1.40GB/s ± 0% -0.04% (p=0.008 n=5+5)
mimic_adler32_100k 1.35GB/s ± 0% 1.38GB/s ± 0% +2.77% (p=0.008 n=5+5)
wuffs_zlib_decode_10k/clang9 85.3MB/s ± 0% 91.1MB/s ± 0% +6.83% (p=0.008 n=5+5)
wuffs_zlib_decode_100k/clang9 107MB/s ± 0% 113MB/s ± 0% +5.20% (p=0.008 n=5+5)
wuffs_zlib_decode_10k/gcc8 86.1MB/s ± 0% 91.0MB/s ± 0% +5.69% (p=0.008 n=5+5)
wuffs_zlib_decode_100k/gcc8 104MB/s ± 0% 111MB/s ± 0% +7.08% (p=0.016 n=5+4)
mimic_zlib_decode_10k 108MB/s ± 0% 108MB/s ± 0% ~ (p=0.548 n=5+5)
mimic_zlib_decode_100k 142MB/s ± 0% 141MB/s ± 1% ~ (p=0.310 n=5+5)
wuffs_png_decode_image_19k_8bpp/clang9 69.9MB/s ± 0% 74.3MB/s ± 0% +6.37% (p=0.016 n=5+4)
wuffs_png_decode_image_40k_24bpp/clang9 99.4MB/s ± 0% 104.3MB/s ± 0% +4.92% (p=0.008 n=5+5)
wuffs_png_decode_image_77k_8bpp/clang9 266MB/s ± 0% 280MB/s ± 0% +5.21% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/clang9 165MB/s ± 0% 169MB/s ± 0% +2.45% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_verify_checksum/clang9 146MB/s ± 0% 163MB/s ± 0% +11.86% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/clang9 93.2MB/s ± 0% 97.7MB/s ± 0% +4.78% (p=0.008 n=5+5)
wuffs_png_decode_image_19k_8bpp/gcc8 70.3MB/s ± 0% 74.3MB/s ± 0% +5.70% (p=0.008 n=5+5)
wuffs_png_decode_image_40k_24bpp/gcc8 101MB/s ± 0% 107MB/s ± 0% +5.67% (p=0.008 n=5+5)
wuffs_png_decode_image_77k_8bpp/gcc8 247MB/s ± 0% 259MB/s ± 0% +4.71% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/gcc8 162MB/s ± 0% 162MB/s ± 0% +0.09% (p=0.032 n=5+4)
wuffs_png_decode_image_552k_32bpp_verify_checksum/gcc8 140MB/s ± 0% 155MB/s ± 0% +10.67% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/gcc8 94.1MB/s ± 0% 99.4MB/s ± 0% +5.57% (p=0.008 n=5+5)
mimic_png_decode_image_19k_8bpp 43.6MB/s ± 0% 43.3MB/s ± 1% ~ (p=0.167 n=5+5)
mimic_png_decode_image_40k_24bpp 54.6MB/s ± 0% 54.6MB/s ± 0% ~ (p=1.000 n=5+5)
mimic_png_decode_image_77k_8bpp 125MB/s ± 0% 123MB/s ± 0% -1.03% (p=0.016 n=5+4)
mimic_png_decode_image_552k_32bpp_ignore_checksum skipped
mimic_png_decode_image_552k_32bpp_verify_checksum 102MB/s ± 0% 101MB/s ± 0% -0.43% (p=0.008 n=5+5)
mimic_png_decode_image_4002k_24bpp 81.3MB/s ± 0% 81.8MB/s ± 0% +0.62% (p=0.008 n=5+5)
|
2021-02-16 22:17:46 +11:00 |
|
Nigel Tao
|
d13039d96e
|
Add std/png decoder.filter_4_distance_?_arm_neon
On a Raspberry Pi 4 (32-bit armv7l) with -march=native and -mfpu=neon
("native" means "armv8-a+crc+simd"):
wuffs_png_decode_filt_4_dist_3/clang9 180MB/s ± 0% 217MB/s ± 0% +20.57% (p=0.008 n=5+5)
wuffs_png_decode_filt_4_dist_4/clang9 161MB/s ± 0% 287MB/s ± 0% +78.75% (p=0.008 n=5+5)
wuffs_png_decode_image_40k_24bpp/clang9 99.1MB/s ± 0% 99.4MB/s ± 0% +0.27% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/clang9 123MB/s ± 0% 165MB/s ± 0% +33.61% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_verify_checksum/clang9 112MB/s ± 0% 146MB/s ± 0% +29.64% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/clang9 93.1MB/s ± 0% 93.2MB/s ± 0% +0.15% (p=0.040 n=5+5)
wuffs_png_decode_filt_4_dist_3/gcc8 80.5MB/s ± 0% 211.7MB/s ± 0% +163.12% (p=0.008 n=5+5)
wuffs_png_decode_filt_4_dist_4/gcc8 77.3MB/s ± 0% 280.4MB/s ± 0% +262.98% (p=0.008 n=5+5)
wuffs_png_decode_image_40k_24bpp/gcc8 98.1MB/s ± 0% 100.8MB/s ± 0% +2.78% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/gcc8 114MB/s ± 0% 162MB/s ± 0% +41.95% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_verify_checksum/gcc8 103MB/s ± 0% 140MB/s ± 0% +36.07% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/gcc8 93.2MB/s ± 0% 94.1MB/s ± 0% +0.96% (p=0.008 n=5+5)
|
2021-02-14 23:53:36 +11:00 |
|
Nigel Tao
|
41825fcaae
|
Add std/png decoder.filter_3_distance_4_arm_neon
On a Raspberry Pi 4 (32-bit armv7l) with -march=native and -mfpu=neon
("native" means "armv8-a+crc+simd"):
wuffs_png_decode_filt_3_dist_4/clang9 639MB/s ± 0% 906MB/s ± 0% +41.71% (p=0.008 n=5+5)
wuffs_png_decode_filt_3_dist_4/gcc8 656MB/s ± 0% 714MB/s ± 0% +8.89% (p=0.008 n=5+5)
|
2021-02-14 22:12:48 +11:00 |
|
Nigel Tao
|
435330494e
|
Add std/png decoder.filter_1_distance_4_arm_neon
On a Raspberry Pi 4 (32-bit armv7l) with -march=native and -mfpu=neon
("native" means "armv8-a+crc+simd"):
wuffs_png_decode_filt_1_dist_4/clang9 1.09GB/s ± 1% 1.42GB/s ± 0% +30.50% (p=0.008 n=5+5)
wuffs_png_decode_filt_1_dist_4/gcc8 859MB/s ± 0% 891MB/s ± 0% +3.66% (p=0.008 n=5+5)
|
2021-02-14 21:43:56 +11:00 |
|
Nigel Tao
|
84f3592dbe
|
Add base.arm_crc32_u32 type
On a Raspberry Pi 4 (32-bit armv7l) with -march=native and -mfpu=neon
("native" means "armv8-a+crc+simd"):
wuffs_crc32_ieee_10k/clang9 825MB/s ± 0% 6802MB/s ± 0% +724.00% (p=0.008 n=5+5)
wuffs_crc32_ieee_100k/clang9 837MB/s ± 0% 6199MB/s ± 0% +640.25% (p=0.008 n=5+5)
wuffs_crc32_ieee_10k/gcc8 960MB/s ± 0% 4495MB/s ± 0% +368.35% (p=0.016 n=4+5)
wuffs_crc32_ieee_100k/gcc8 961MB/s ± 0% 4447MB/s ± 0% +363.01% (p=0.016 n=5+4)
mimic_crc32_ieee_10k 638MB/s ± 0% 631MB/s ± 0% -1.17% (p=0.008 n=5+5)
mimic_crc32_ieee_100k 639MB/s ± 0% 639MB/s ± 0% +0.02% (p=0.008 n=5+5)
wuffs_gzip_decode_10k/clang9 83.0MB/s ± 0% 90.7MB/s ± 0% +9.26% (p=0.008 n=5+5)
wuffs_gzip_decode_100k/clang9 100MB/s ± 0% 112MB/s ± 0% +11.91% (p=0.008 n=5+5)
wuffs_gzip_decode_10k/gcc8 86.1MB/s ± 0% 91.9MB/s ± 0% +6.72% (p=0.008 n=5+5)
wuffs_gzip_decode_100k/gcc8 102MB/s ± 0% 112MB/s ± 0% +9.34% (p=0.008 n=5+5)
mimic_gzip_decode_10k 98.8MB/s ± 0% 98.2MB/s ± 0% -0.60% (p=0.008 n=5+5)
mimic_gzip_decode_100k 128MB/s ± 0% 128MB/s ± 0% -0.08% (p=0.008 n=5+5)
wuffs_png_decode_image_19k_8bpp/clang9 67.6MB/s ± 0% 70.8MB/s ± 0% +4.72% (p=0.008 n=5+5)
wuffs_png_decode_image_40k_24bpp/clang9 92.8MB/s ± 0% 98.8MB/s ± 0% +6.49% (p=0.008 n=5+5)
wuffs_png_decode_image_77k_8bpp/clang9 255MB/s ± 0% 269MB/s ± 0% +5.44% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/clang9 123MB/s ± 0% 123MB/s ± 0% +0.13% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_verify_checksum/clang9 110MB/s ± 0% 112MB/s ± 0% +1.46% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/clang9 88.4MB/s ± 0% 92.8MB/s ± 0% +4.90% (p=0.008 n=5+5)
wuffs_png_decode_image_19k_8bpp/gcc8 67.2MB/s ± 0% 70.8MB/s ± 0% +5.37% (p=0.008 n=5+5)
wuffs_png_decode_image_40k_24bpp/gcc8 94.0MB/s ± 0% 98.2MB/s ± 0% +4.45% (p=0.008 n=5+5)
wuffs_png_decode_image_77k_8bpp/gcc8 236MB/s ± 0% 248MB/s ± 0% +4.84% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_ignore_checksum/gcc8 114MB/s ± 0% 115MB/s ± 0% +1.11% (p=0.008 n=5+5)
wuffs_png_decode_image_552k_32bpp_verify_checksum/gcc8 102MB/s ± 0% 104MB/s ± 0% +2.17% (p=0.008 n=5+5)
wuffs_png_decode_image_4002k_24bpp/gcc8 88.8MB/s ± 0% 94.3MB/s ± 0% +6.21% (p=0.008 n=5+5)
mimic_png_decode_image_19k_8bpp 43.7MB/s ± 1% 43.4MB/s ± 1% ~ (p=0.095 n=5+5)
mimic_png_decode_image_40k_24bpp 54.6MB/s ± 0% 53.9MB/s ± 0% -1.25% (p=0.008 n=5+5)
mimic_png_decode_image_77k_8bpp 125MB/s ± 0% 124MB/s ± 0% -0.73% (p=0.008 n=5+5)
mimic_png_decode_image_552k_32bpp_ignore_checksum skipped
mimic_png_decode_image_552k_32bpp_verify_checksum 102MB/s ± 0% 101MB/s ± 0% -0.95% (p=0.008 n=5+5)
mimic_png_decode_image_4002k_24bpp 81.3MB/s ± 0% 81.0MB/s ± 0% -0.46% (p=0.016 n=4+5)
|
2021-02-13 22:54:47 +11:00 |
|
Nigel Tao
|
81ba41e251
|
Add slice uintptr_low_12_bits method
|
2021-02-13 17:19:35 +11:00 |
|
Nigel Tao
|
0834b40d22
|
Add std/crc32 ieee_hasher.up_arm_crc32
|
2021-02-13 17:02:02 +11:00 |
|
Nigel Tao
|
3338b5eafa
|
Add arm_crc32,arm_neon cpu_arch conditions
|
2021-02-13 17:01:09 +11:00 |
|
Nigel Tao
|
3c627574a7
|
Fix and tweak hand-written decode_huffman_fast
|
2021-02-10 16:02:05 +11:00 |
|
Nigel Tao
|
8595a3b696
|
Add std/deflate decode_huffman_fast32
This undoes commit 1a63b53e "Have std/deflate read 64 bits once per
inner loop" for 32-bit CPUs.
On a Raspberry Pi 4 (32-bit armv7l):
wuffs_deflate_decode_1k_full_init/clang9 44.9MB/s ± 1% 50.1MB/s ± 1% +11.54% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_part_init/clang9 52.8MB/s ± 0% 61.1MB/s ± 0% +15.61% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_full_init/clang9 81.4MB/s ± 0% 84.5MB/s ± 0% +3.89% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_part_init/clang9 87.9MB/s ± 0% 91.1MB/s ± 0% +3.67% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_just_one_read/clang9 106MB/s ± 0% 113MB/s ± 0% +5.70% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_many_big_reads/clang9 59.6MB/s ± 0% 59.1MB/s ± 0% -0.81% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_full_init/gcc8 44.7MB/s ± 1% 50.9MB/s ± 1% +13.91% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_part_init/gcc8 52.6MB/s ± 0% 61.5MB/s ± 0% +17.06% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_full_init/gcc8 76.5MB/s ± 0% 85.6MB/s ± 1% +11.89% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_part_init/gcc8 82.3MB/s ± 0% 92.2MB/s ± 0% +12.07% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_just_one_read/gcc8 97.8MB/s ± 0% 112.1MB/s ± 0% +14.62% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_many_big_reads/gcc8 58.0MB/s ± 0% 61.6MB/s ± 0% +6.09% (p=0.008 n=5+5)
|
2021-02-08 23:19:00 +11:00 |
|
Nigel Tao
|
279d69bb38
|
Prepare to add std/deflate decode_huffman_fast32
|
2021-02-08 22:51:57 +11:00 |
|
Nigel Tao
|
4993f88a66
|
Add base.utility.cpu_arch_is_32_bit()
|
2021-02-08 22:45:51 +11:00 |
|
Nigel Tao
|
39e2044102
|
Rename std/deflate decode_huffman_fast function
|
2021-02-08 22:34:28 +11:00 |
|
Nigel Tao
|
97591ee3c9
|
Rename std/deflate/decode_huffman_fast.wuffs
|
2021-02-08 22:30:29 +11:00 |
|
Nigel Tao
|
77d5fa01e5
|
Have std/png decode 16-bit depth images
|
2021-02-06 23:41:37 +11:00 |
|
Nigel Tao
|
b69e6964f7
|
Add space before WUFFS_BASE__POTENTIALLY_UNUSED
|
2021-02-05 23:13:00 +11:00 |
|
Nigel Tao
|
d8d951b729
|
Remove std/png tricky field
|
2021-02-05 23:06:49 +11:00 |
|
Nigel Tao
|
0a1b8ddf78
|
Factor out std/png/decode_swizzle_fallback.wuffs
|
2021-02-05 22:59:26 +11:00 |
|
Nigel Tao
|
f1a8019f41
|
Drop std/png filter_and_swizzle_low_bit_depth
|
2021-02-05 22:56:55 +11:00 |
|
Nigel Tao
|
baad01d50f
|
Have std/png decode interlaced PNG images
|
2021-02-05 22:53:05 +11:00 |
|
Nigel Tao
|
4053c2abe6
|
Refactor some std/png internals
|
2021-02-03 22:08:26 +11:00 |
|
Nigel Tao
|
4942324f25
|
Swizzler support src_pixfmt = PIXEL_FORMAT__Y_16BE
Binary size, before:
30800 gen/lib/c/clang-9-dynamic/wuffs-base-pixconv.lo
32424 gen/lib/c/clang-9-static/wuffs-base-pixconv.o
56200 gen/lib/c/gcc-dynamic/wuffs-base-pixconv.lo
56152 gen/lib/c/gcc-static/wuffs-base-pixconv.o
After:
33416 gen/lib/c/clang-9-dynamic/wuffs-base-pixconv.lo
35120 gen/lib/c/clang-9-static/wuffs-base-pixconv.o
59504 gen/lib/c/gcc-dynamic/wuffs-base-pixconv.lo
59440 gen/lib/c/gcc-static/wuffs-base-pixconv.o
|
2021-02-03 13:11:22 +11:00 |
|
Nigel Tao
|
e77e16cd98
|
Add fact for "lhs = x[i .. j]" for constant i, j
|
2021-02-02 22:53:02 +11:00 |
|
Nigel Tao
|
10146b75dd
|
Drop old facts after an impure function call
|
2021-02-02 22:11:45 +11:00 |
|
Nigel Tao
|
4848859f97
|
Tweak std/deflate decode_huffman_fast asserts
wuffs_deflate_decode_1k_full_init/clang9 95.0MB/s ± 0% 92.9MB/s ± 0% -2.25% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_part_init/clang9 108MB/s ± 0% 108MB/s ± 0% +0.34% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_full_init/clang9 141MB/s ± 0% 165MB/s ± 1% +16.77% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_part_init/clang9 143MB/s ± 0% 169MB/s ± 0% +17.73% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_just_one_read/clang9 185MB/s ± 0% 181MB/s ± 0% -2.09% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_many_big_reads/clang9 156MB/s ± 1% 151MB/s ± 0% -3.12% (p=0.008 n=5+5)
wuffs_deflate_decode_1k_full_init/gcc10 98.6MB/s ± 0% 97.5MB/s ± 0% -1.05% (p=0.016 n=4+5)
wuffs_deflate_decode_1k_part_init/gcc10 112MB/s ± 0% 111MB/s ± 0% -1.05% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_full_init/gcc10 183MB/s ± 0% 181MB/s ± 1% -0.73% (p=0.008 n=5+5)
wuffs_deflate_decode_10k_part_init/gcc10 186MB/s ± 0% 185MB/s ± 0% -0.80% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_just_one_read/gcc10 213MB/s ± 0% 216MB/s ± 0% +1.55% (p=0.008 n=5+5)
wuffs_deflate_decode_100k_many_big_reads/gcc10 174MB/s ± 0% 179MB/s ± 0% +2.79% (p=0.008 n=5+5)
|
2021-02-02 13:08:21 +11:00 |
|
Nigel Tao
|
dc60821826
|
Roll back the previous "pixel swizzler" commit
See the previous commit for the rationale.
|
2021-02-02 11:32:23 +11:00 |
|
Nigel Tao
|
83a72f3d77
|
Performance mystery: unused pixel swizzler code
A recent commit (c392a632de, "Swizzler let
src_pixfmt = INDEXED__BGRA_NONPREMUL") introduced new pixel swizzler
functions. These code paths weren't taken by the std/gif codec, but for
some mysterious reason, they dramatically impact the std/gif benchmarks.
To record that quantitatively in the git log, this commit touches the
INDEXED__BGRA_NONPREMUL code (editing wuffs-unsupported-snapshot.c
directly), making it incorrect but affecting the std/gif benchmarks (for
gcc10 but not for clang9). To repeat, the new code is not exercised by
the std/gif code. Even though test_wuffs_pixel_swizzler_swizzle now
fails, the std/gif tests still pass with this incorrect commit.
It is a mystery why this happens. This small, self-contained commit
will be immediately followed by a roll back, but it is committed anyway
so that we can refer to these numbers in the git log. To reproduce:
----
$ git clone https://github.com/google/wuffs.git
$ cd wuffs
$ gcc --version | head -n 1
gcc (Debian 10.2.1-3) 10.2.1 20201224
$ # Check out this commit.
$ git checkout --quiet THE_HASH_OF_THIS_COMMIT
$ gcc -O3 -Wall -std=c99 test/c/std/gif.c -DWUFFS_MIMIC -lgif
$ ./a.out -bench -focus=wuffs_gif_decode_1000k_full_init | grep MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 8242690 ns/op 121.408 MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 8233370 ns/op 121.546 MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 8228690 ns/op 121.615 MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 8236640 ns/op 121.497 MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 8239820 ns/op 121.451 MB/s
$ # Check out the parent commit.
$ git checkout --quiet THE_HASH_OF_THIS_COMMIT^
$ gcc -O3 -Wall -std=c99 test/c/std/gif.c -DWUFFS_MIMIC -lgif
$ ./a.out -bench -focus=wuffs_gif_decode_1000k_full_init | grep MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 9614630 ns/op 104.084 MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 9556300 ns/op 104.719 MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 9522440 ns/op 105.092 MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 9438460 ns/op 106.027 MB/s
Benchmarkwuffs_gif_decode_1000k_full_init/gcc10 100 9477400 ns/op 105.591 MB/s
----
On a mid-range x86_64 laptop:
wuffs_gif_decode_1k_bw/clang9 367MB/s ± 0% 347MB/s ± 0% -5.37% (p=0.016 n=4+5)
wuffs_gif_decode_1k_color_full_init/clang9 89.8MB/s ± 0% 90.6MB/s ± 0% +0.83% (p=0.008 n=5+5)
wuffs_gif_decode_1k_color_part_init/clang9 107MB/s ± 0% 107MB/s ± 0% ~ (p=0.413 n=4+5)
wuffs_gif_decode_10k_bgra/clang9 421MB/s ± 0% 420MB/s ± 0% ~ (p=0.548 n=5+5)
wuffs_gif_decode_10k_indexed/clang9 111MB/s ± 0% 112MB/s ± 0% ~ (p=0.056 n=5+5)
wuffs_gif_decode_20k/clang9 134MB/s ± 0% 134MB/s ± 0% -0.33% (p=0.016 n=5+5)
wuffs_gif_decode_100k_artificial/clang9 301MB/s ± 0% 301MB/s ± 0% ~ (p=0.151 n=5+5)
wuffs_gif_decode_100k_realistic/clang9 115MB/s ± 0% 115MB/s ± 0% ~ (p=0.151 n=5+5)
wuffs_gif_decode_1000k_full_init/clang9 117MB/s ± 0% 117MB/s ± 0% -0.17% (p=0.016 n=5+5)
wuffs_gif_decode_1000k_part_init/clang9 117MB/s ± 1% 117MB/s ± 0% ~ (p=0.690 n=5+5)
wuffs_gif_decode_anim_screencap/clang9 675MB/s ± 0% 674MB/s ± 0% -0.20% (p=0.016 n=4+5)
wuffs_gif_decode_1k_bw/gcc10 309MB/s ± 0% 313MB/s ± 0% +1.18% (p=0.008 n=5+5)
wuffs_gif_decode_1k_color_full_init/gcc10 85.4MB/s ± 0% 104.1MB/s ± 0% +21.81% (p=0.008 n=5+5)
wuffs_gif_decode_1k_color_part_init/gcc10 97.9MB/s ± 0% 123.3MB/s ± 0% +25.92% (p=0.008 n=5+5)
wuffs_gif_decode_10k_bgra/gcc10 372MB/s ± 0% 447MB/s ± 0% +19.92% (p=0.008 n=5+5)
wuffs_gif_decode_10k_indexed/gcc10 97.9MB/s ± 0% 119.0MB/s ± 0% +21.54% (p=0.008 n=5+5)
wuffs_gif_decode_20k/gcc10 123MB/s ± 0% 139MB/s ± 0% +12.98% (p=0.008 n=5+5)
wuffs_gif_decode_100k_artificial/gcc10 271MB/s ± 0% 293MB/s ± 0% +8.16% (p=0.008 n=5+5)
wuffs_gif_decode_100k_realistic/gcc10 104MB/s ± 0% 119MB/s ± 0% +13.90% (p=0.008 n=5+5)
wuffs_gif_decode_1000k_full_init/gcc10 106MB/s ± 0% 121MB/s ± 0% +14.06% (p=0.008 n=5+5)
wuffs_gif_decode_1000k_part_init/gcc10 106MB/s ± 0% 121MB/s ± 0% +14.24% (p=0.008 n=5+5)
wuffs_gif_decode_anim_screencap/gcc10 633MB/s ± 0% 653MB/s ± 0% +3.24% (p=0.008 n=5+5)
mimic_gif_decode_1k_bw 84.3MB/s ± 0% 84.1MB/s ± 0% ~ (p=0.151 n=5+5)
mimic_gif_decode_1k_color 45.1MB/s ± 0% 45.0MB/s ± 1% ~ (p=0.421 n=5+5)
mimic_gif_decode_10k_indexed 52.4MB/s ± 0% 52.4MB/s ± 0% ~ (p=0.841 n=5+5)
mimic_gif_decode_20k 52.6MB/s ± 0% 52.6MB/s ± 0% ~ (p=0.310 n=5+5)
mimic_gif_decode_100k_artificial 82.9MB/s ± 0% 82.7MB/s ± 1% ~ (p=0.460 n=5+5)
mimic_gif_decode_100k_realistic 51.1MB/s ± 0% 51.0MB/s ± 0% -0.12% (p=0.016 n=5+5)
mimic_gif_decode_1000k 52.5MB/s ± 0% 52.4MB/s ± 0% ~ (p=0.548 n=5+5)
mimic_gif_decode_anim_screencap 98.2MB/s ± 0% 98.1MB/s ± 0% -0.11% (p=0.032 n=5+5)
|
2021-02-02 11:32:04 +11:00 |
|