Skip to content

disable alpha blending for textures with no transparency

Administrator requested to merge disable_alpha into master

This results in a pretty big speed-up in sw rendering, as BlitNtoNPixelAlpha was taking up something like ~85% of the total application time when I was hammering key presses.

Collected with: $ perf record --call-graph dwarf -F 99 ./_build/osk-sdl -d a -n a -v

I used the SDL2 sw rendering to collect this data, but there's a similar speed-up when using directfb too, it's just a lot easier to see in the perf report w/ SDL2 sw rendering (the dfb version has a lot of noise from internal dfb stuff.)

before:

99.75%     0.00%  osk-sdl   osk-sdl                    [.] main
        |
        ---main
           |
           |--86.03%--SDL_RenderPresent_REAL
           |          FlushRenderCommands
           |          SW_RunCommandQueue
           |          |
           |          |--84.49%--SDL_UpperBlit_REAL
           |          |          SDL_SoftBlit
           |          |          BlitNtoNPixelAlpha
           |          |
           |           --1.55%--SDL_FillRects_REAL
           |                     SDL_FillRect4

after:

99.31%     0.00%  osk-sdl   osk-sdl                    [.] main
        |
        ---main
           |
           |--47.90%--SDL_UpdateWindowSurface_REAL
           |          SDL_UpdateWindowTexture
           |          |
           |          |--46.91%--SDL_UpdateTexture_REAL
           |          |          GL_UpdateTexture
           |          |          0xffff893b3567
           |          |          |
           |          |           --46.74%--0xffff893ae997
           |          |          .....
           |          |
           |           --0.83%--KMSDRM_GLES_SwapWindow
           |                     eglSwapBuffers
           |
           |--47.71%--SDL_RenderPresent_REAL
           |          FlushRenderCommands
           |          |
           |           --47.56%--SW_RunCommandQueue
           |                     |
           |                     |--38.84%--SDL_UpperBlit_REAL
           |                     |          SDL_SoftBlit
           |                     |          |
           |                     |          |--23.86%--Blit_3or4_to_3or4__inversed_rgb
           |                     |          |
           |                     |           --14.97%--BlitNtoNPixelAlpha
           |                     |
           |                      --8.73%--SDL_FillRects_REAL
           |                                SDL_FillRect4

Merge request reports