Re: [PATCH v2 0/5] MR53: vkd3d-shader/hlsl: transpose(), smoothstep(), and properly convert args of element-wise intrinsics.

[Francisco Casas (＠fcasas)]

On Mon Dec 12 16:36:54 2022 +0000, Giovanni Mascellani wrote:
> With my iGPU (`Driver name: Intel open-source Mesa driver, driver info:
> Mesa 22.3.0.`) this fails with:
> shader_runner:551:Section [test], line 123: Test failed: Got
> {0.00000000e+00, 1.56250000e-01, 0.00000000e+00, 0.00000000e+00},
> expected {0.00000000e+00, 1.56250000e-01, 0.00000000e+00,
> 1.00000000e+00} at (0, 0).
>
> It's working with any other implementation I've access to (including
> NVidia proprietary, RADV and llvmpipe), so I guess it's a bug in the
> Intel driver. It's probably not enough to delay this patch, I'll try to
> debug it better as soon as I have some time. Just wanted to let you know.

Hmm, it seems that division by zero is indeed implementation-dependent.

https://stackoverflow.com/questions/5802351/what-happens-when-you-divide-by-...

I better just remove this test, lest we need to keep track of flaky tests as well.