You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Investigation results: Required MoltenVK fix for sampler_bias variant KhronosGroup/MoltenVK#2163 and suspicion that Metal may be incorrectly selecting the texture LOD for sampling when texture is 2D with height 1.
Reasoning:
Test creates a 2D texture with dimensions (max_width, 1, 1) for sampling and a 2D texture with dimentions (1, 1, 1) as render target. Explanation will focus on shader_bias since that's the one correctly working from MoltenVK's end. Without changes to the test, the sampled mip level will be the last one. However, if we increase the render target texture dimensions to be the same as the sampled texture, the test passes. Modifying the render target dimensions to other values in between for width and height, we see that the sampled mip will vary.
Proposed solution: I would like to test a few more things before discarding the issue as a Metal bug. I will revisit this later once all CTS failures are triaged.
Failing CTS tests:
Investigation results: Required MoltenVK fix for
sampler_bias
variant KhronosGroup/MoltenVK#2163 and suspicion that Metal may be incorrectly selecting the texture LOD for sampling when texture is 2D with height 1.Reasoning:
Test creates a 2D texture with dimensions (max_width, 1, 1) for sampling and a 2D texture with dimentions (1, 1, 1) as render target. Explanation will focus on
shader_bias
since that's the one correctly working from MoltenVK's end. Without changes to the test, the sampled mip level will be the last one. However, if we increase the render target texture dimensions to be the same as the sampled texture, the test passes. Modifying the render target dimensions to other values in between for width and height, we see that the sampled mip will vary.Proposed solution: I would like to test a few more things before discarding the issue as a Metal bug. I will revisit this later once all CTS failures are triaged.
cc/ @billhollings
The text was updated successfully, but these errors were encountered: