Vulkan CTS test ideas
-
Test the maximum workgroup size -
Test dual source blending (no CTS tests using it as far as I can tell) -
Test out of bounds vertex data with 0 stride -
Test bounds checking when the buffer size is not a multiple of stride -
Test mipmaps with linear surfaces (especially vkGetImageSubresourceLayout) -
Test reading prim ID in FS after tess -
Test spirv switch case that directly jumps to the merge block -
Test that image queries with null descriptors return 0 -
Test vertex shaders without exporting gl_Position (seems valid but undefined ?) -
Test that the predication value when conditional rendering is used is treated as 32-bit
-
Test GLSLstd450UnpackHalf2x16 with denorm flush to zero fp16 mode enabled (cf. shader float control modes) -
Test that various dynamic states are correctly ignored: -
Test that enabling sample shading with sample qualifier works if VS writes a constant to the varying that is marked per-sample (in Mesa, the linker had a bug and the per-sample qualifier was removed because the constant was moved to the next stage) See mesa/mesa!7892 (merged) -
Test clearing layered MSAA images with different render passes (see mesa/mesa#3710 (closed))