Question: Washed out colors using SDL_CreateColorCursor

[klei1984]

Recently I installed 2 new (latest) nvidia graphics drivers on Windows 10 64 bit. Since then my SDL3 v3.2.28 application's color hardware cursors created by SDL_CreateColorCursor from SDL surface with SDL_PIXELFORMAT_ARGB8888 data all look washed out. Their brightness or gamma is off.

Setting color space for surface with SDL_SetSurfaceColorspace has no visible effect on the outcome. Converting my sRGB data to linear RGB fixes the issue. But there is absolutely no way to tell what format the OS or whatever driver expects.

I had to add a configuration switch to the application that conditionally enables sRGB to linear RGB conversion on the cursor raw pixel data:

float srgb_to_linear(float c)
{
    if (c <= 0.04045f)
        return c / 12.92f;
    else
        return powf((c + 0.055f) / 1.055f, 2.4f);
}

The above result is packed back into the ARGB8888 surface.

Did anyone else bump into this issue? How am I supposed to solve this?

[slouken]

This seems like a driver bug. Cursors should be in sRGB colorspace.

[slouken]

Can you attach the cursor you're using so we can try to reproduce here?

[klei1984]

It is a game asset using custom format and palette, converted to SDL_surface with SDL_PIXELFORMAT_ARGB8888. I will try to export the raw ARGB888 pixel data and dimensions. Or as a BMP.

[slouken]

A BMP or PNG would be easiest. Can you reproduce the issue using testcustomcursor and your asset?

[klei1984]

So I exported the cursor to a bmp file.

HANDPTR_0_0.bmp

I loaded the bmp and created a color cursor from it. Black is the transparent color.

testcustomcursor.zip

I tried to take a screenshot, but hw cursor is not captured. Then I recorded a video of the screen, which captured the cursor with PROPER colors 😲

then i put the video still (left) and the actual application (right) next to each other and took a photo with a camera.

It is bad quality, but the issue is, i hope clearly, visible.

[slouken]

Interesting, I think this is an NVIDIA driver bug, given that it works fine here on AMD, it worked for you on an earlier NVIDIA driver, and the video capture tool doesn't observe the issue.