This is a dumb question, but do we need this? Is there a reason an app can't just convert 4 ints to floats before clearing?

Is there a reason an app can't just convert 4 ints to floats before clearing?

This would work for Metal and Vulkan (with some bitcasting to preserve the integer bit pattern), but not for D3D. The reason is that you can have integer texture formats with more bits than can be precisely represented by a float.

E.g. if you have an R32_INT texture and you want to clear it to a value of 2^24 or above, you're out of luck. D3D will choose the closest integer value that it can represent as a float, which caps your maximum clear value. The only way to work around this is by clearing the target with a shader, which is what D3D's docs recommend.

Though, as the OP notes, this PR does not include the D3D shaders for clearing. @rabbit-ecl Could you write those up and provide a test program with integer texture clears in action? You can PR the test case to the SDL_gpu_examples repo.

I'm still skeptical this is worth adding for high-precision clearing, but if y'all think it's worth it I won't stand in the way.

I'm also presently of the opinion that this is not really worth it.

@rabbit-ecl Could you write those up and provide a test program with integer texture clears in action? You can PR the test case to the SDL_gpu_examples repo.

Yeah I'll look into it and we can always drop it if it ends up being too hacky. I think most of the hard work here has already been done for the blit emulation paths.