Skip to content

Latest commit

 

History

History

benches

README.md

Benchmarking is hard... (c)

Environment

  • All test were run on Gentoo Linux with AMD 3700X.
  • tiny-skia SSE2 is built with -Ctarget-cpu=x86-64
  • tiny-skia AVX is built with -Ctarget-cpu=haswell
  • Skia v85.
  • Skia SSE2 is built using clang with -march=x86-64
  • Skia AVX is built using clang with -march=haswell
  • cairo v1.16.0.
  • cairo is built using gcc with -march=native, because it's the system one
  • raqote version can be found at Cargo.lock
  • raqote is built with -Ctarget-cpu=x86-64. Testing with haswell doesn't change the results much.

Results

All measurement are in nanoseconds. Lower is better.

blending modes

blend.rs

Filling a shape with a solid color.

Mode/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
clear 51,020 41,785 45,898 50,203 62,331 1,043,659
source 51,467 41,287 46,195 50,949 62,548 1,114,946
destination 63 67 5,309 5,318 4 980,993
source_over 492,146 355,726 490,423 261,943 212,838 2,645,207
destination_over 582,450 427,992 583,660 306,542 236,970 3,446,445
source_in 588,029 413,983 577,464 291,008 1,104,817 1,163,684
destination_in 518,637 379,905 570,325 293,412 1,103,758 1,193,288
source_out 600,771 420,907 576,125 297,537 1,122,571 1,182,724
destination_out 519,200 379,237 576,521 299,481 556,088 1,193,241
source_atop 577,346 426,021 603,723 307,222 663,254 1,501,419
destination_atop 552,738 406,357 595,771 306,610 1,220,442 1,511,405
xor 581,027 435,475 601,201 310,439 671,439 1,634,236
plus 508,457 358,197 559,063 286,504 122,496 4,892,311
modulate 522,697 379,620 570,734 290,988 - -
screen 523,308 389,701 598,637 312,507 3,837,010 1,459,500
overlay 826,240 514,436 733,158 385,684 3,412,192 7,016,443
darken 631,307 434,548 610,493 317,215 3,579,931 5,230,384
lighten 633,910 434,123 621,340 319,867 3,584,409 5,255,202
color_dodge 1,794,356 1,002,832 780,556 679,996 5,151,488 9,634,447
color_burn 1,878,874 1,055,831 861,359 708,962 5,007,181 9,617,102
hard_light 804,040 508,038 734,878 370,307 3,442,584 7,034,417
soft_light 2,410,827 1,412,907 1,225,756 986,403 5,900,415 11,941,630
difference 650,915 440,554 632,797 326,285 3,936,284 5,718,776
exclusion 530,865 399,089 618,645 316,235 3,842,131 6,082,393
multiply 586,406 434,776 627,202 316,841 3,608,817 5,986,364
hue 2,675,939 2,008,062 1,705,745 1,413,791 7,517,902 13,716,827
saturation 2,825,041 2,006,886 1,676,187 1,411,117 7,443,261 13,752,382
color 2,350,240 1,655,401 1,431,804 1,115,800 6,070,058 10,537,391
luminosity 2,360,829 1,599,196 1,351,584 1,090,402 6,124,294 10,488,916

Destination is faster in tiny-skia, because we're exiting immediately, while Skia uses null blitter, so edges processing is still in place.

anti-aliased fill

fill_aa.rs

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
fill 663,415 542,188 495,359 348,326 538,842 1,520,830

memset fill

memset_fill.rs

Shape filling by overwritting original pixels. No blending.

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
opaque 56,173 45,190 45,392 47,662 45,804 2,432,875
source 56,542 50,158 50,433 50,663 48,839 678,795

rectangle fill

fill_rect.rs

Fills a rectangle with a solid solid color.

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
basic 699,878 499,512 711,992 348,918 203,921 3,672,295
with AA 744,711 541,858 735,547 371,368 192,178 2,087,047
with AA and transform 369,684 284,392 321,105 191,670 175,061 891,996

The last test simply fallbacks to path filling in Skia/tiny-skia.

Strangely, Skia is pretty slow in this task. Not sure why.

canvas fill

fill_all.rs

Filling the whole canvas with a color.

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
fill 64,213 64,916 1,112,417 554,113 285,316 46,560

spiral stroke

spiral.rs

A spiral has a lot of short horizontal strides which are not CPU-friendly when a rendering backend is designed for long horizontal strides.

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
stroke 1,658,247 1,602,847 1,205,407 1,174,229 3,161,763 5,707,383

hairline stroking

hairline.rs

Draws a large spiral using a subpixel stroke width.

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
no AA 1,391,101 1,345,418 1,022,641 1,027,160 2,469,764 -
with AA 3,363,682 3,084,936 1,846,240 1,868,304 13,457,819 -
  • raqote doesn't support hairline stroking.
  • Not sure why cairo is so slow with AA.

gradients

gradients.rs

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
linear, two stops, pad 1,417,980 1,223,505 993,781 571,583 2,458,368 3,379,723
linear, two stops, reflect 1,724,305 1,391,261 1,248,226 684,310 2,449,727 3,301,529
linear, two stops, repeat 1,633,253 1,294,505 1,043,289 599,061 2,445,711 3,115,416
linear, three stops, evenly spread 2,068,385 1,940,036 1,806,614 781,344 2,413,454 4,021,338
linear, three stops, unevenly spread 2,068,394 1,939,826 1,805,379 687,479 2,423,142 3,412,176
simple radial 2,293,883 2,205,554 2,050,437 805,376 4,704,141 5,531,178
two point radial 2,700,221 2,516,614 1,943,448 1,083,230 4,709,760 13,454,676

pattern

pattern.rs

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
plain (nearest, no ts) 2,633,745 2,263,839 1,315,079 1,122,982 785,550 1,865,327
lq (bilinear, with ts) 8,602,786 4,865,226 4,484,023 2,646,523 17,612,685 24,906,379
hq (bicubic/gauss, with ts) 27,967,739 14,760,398 12,386,848 9,364,356 162,771,632 -

Note that raqote doesn't support high quality filtering.

And yes, cairo is really that slow. Not sure why.

clipping

clip.rs

Test/Library tiny-skia SSE2 tiny-skia AVX Skia SSE2 Skia AVX cairo raqote
clip path 2,371,153 2,066,654 579,724 299,567 336,489 4,964,327
clip path AA 2,480,372 2,152,685 898,489 605,572 367,626 3,396,255

tiny-skia uses just a simple alpha mask for clipping, while Skia has a very complicated, but way faster algorithm.

png

png_io.rs

We are comparing raw png crate performance with tiny-skia premultiplying/demultiplying code to check how much overhead it adds.

Test/Library tiny-skia png
Decode RGB 128,367 67,293
Decode RGBA 109,752 90,793
Encode RGBA 302,154 275,042

RGB (without alpha) is slower, since we have to decode an image into a RGB buffer, then transform it into a RGBA buffer. Waiting for image-png/#239.

Running benchmarks

We support only Linux. The benchmark may work on other OS'es, but it will require a lot of preperation (building Skia and cairo).

You have to install cairo first and built Skia from sources (see below).

Run:

export SKIA_DIR="/path/to/skia"
export SKIA_LIB_DIR="/path/to/skia/out/Shared"
export LD_LIBRARY_PATH="/path/to/skia/out/Shared"
cargo bench

Building Skia

You will need git, clang, ninja and Python 2.

On Windows, use clang-cl and clang-cl++ for cc and cxx instead.

git clone https://skia.googlesource.com/skia.git
cd skia
git fetch --all
git checkout -b m85 origin/chrome/m85
python2 tools/git-sync-deps # this will download about 3 GiB of code
bin/gn gen out/Shared --args='
    is_official_build=false
    is_component_build=true
    is_debug=false
    cc="clang"
    cxx="clang++"
    extra_cflags_cc=["-march=native", "-DSK_FORCE_RASTER_PIPELINE_BLITTER"]
    werror=false
    paragraph_gms_enabled=false
    paragraph_tests_enabled=false
    skia_enable_android_utils=false
    skia_enable_discrete_gpu=false
    skia_enable_gpu=false
    skia_enable_nvpr=false
    skia_enable_particles=false
    skia_enable_pdf=false
    skia_enable_skottie=false
    skia_enable_skrive=false
    skia_enable_skshaper=false
    skia_enable_sksl_interpreter=false
    skia_enable_skvm_jit=false
    skia_enable_tools=false
    skia_use_expat=false
    skia_use_gl=false
    skia_use_harfbuzz=false
    skia_use_icu=false
    skia_use_libgifcodec=false
    skia_use_libheif=false
    skia_use_libjpeg_turbo_decode=false
    skia_use_libjpeg_turbo_encode=false
    skia_use_libwebp_decode=false
    skia_use_libwebp_encode=false
    skia_use_lua=false
    skia_use_piex=false'
ninja -C out/Shared