Emill
December 29th, 2009, 01:40 PM
Hello!
If I set up a 16-bit video mode in SDL, the SDL_Flip(screen) is very very slow. 32-bit video mode is much faster, because that is the default Xorg setting.
When using 16-bit mode, SDL creates a "shadow" 16-bit surface, and when calling SDL_Flip, it blits that shadow surface onto the real 32-bit surface.
Now to my question: If you look in the SDL_blit_N.c (http://wesnoth.repositoryhosting.com/trac/wesnoth_wesnoth/browser/trunk/Classes/SDL/src/video/SDL_blit_N.c) file, you see the blit_table normal_blit_2[] array, which contains a
{0x0000F800, 0x000007E0, 0x0000001F, 4, 0x00FF0000, 0x0000FF00,
0x000000FF,
0, Blit_RGB565_ARGB8888, SET_ALPHA}
but the expected blitting function is something with NO_ALPHA, because the shadow surface has the Amask set to 0. So it falls back to the slow BlitNtoN-function.
But, try to add this line into the blit-table:
{0x0000F800, 0x000007E0, 0x0000001F, 4, 0x00FF0000, 0x0000FF00,
0x000000FF,
0, Blit_RGB565_ARGB8888, NO_ALPHA}
The alpha, though, is set to 0xff on the 32-bit surface because of the RGB565_ARGB8888_LUT lookup-table, but that doesn't seem to matter...
Then everything seems to work properly and the blitting is now 3 times faster!
Why isn't something like this in the current version of SDL?
If I set up a 16-bit video mode in SDL, the SDL_Flip(screen) is very very slow. 32-bit video mode is much faster, because that is the default Xorg setting.
When using 16-bit mode, SDL creates a "shadow" 16-bit surface, and when calling SDL_Flip, it blits that shadow surface onto the real 32-bit surface.
Now to my question: If you look in the SDL_blit_N.c (http://wesnoth.repositoryhosting.com/trac/wesnoth_wesnoth/browser/trunk/Classes/SDL/src/video/SDL_blit_N.c) file, you see the blit_table normal_blit_2[] array, which contains a
{0x0000F800, 0x000007E0, 0x0000001F, 4, 0x00FF0000, 0x0000FF00,
0x000000FF,
0, Blit_RGB565_ARGB8888, SET_ALPHA}
but the expected blitting function is something with NO_ALPHA, because the shadow surface has the Amask set to 0. So it falls back to the slow BlitNtoN-function.
But, try to add this line into the blit-table:
{0x0000F800, 0x000007E0, 0x0000001F, 4, 0x00FF0000, 0x0000FF00,
0x000000FF,
0, Blit_RGB565_ARGB8888, NO_ALPHA}
The alpha, though, is set to 0xff on the 32-bit surface because of the RGB565_ARGB8888_LUT lookup-table, but that doesn't seem to matter...
Then everything seems to work properly and the blitting is now 3 times faster!
Why isn't something like this in the current version of SDL?