-
bbc6de83
by Giovanni Mascellani at 2026-05-22T19:29:25+02:00
tests/d3d12: Skip test_hull_shader_relative_addressing() on Windows AMD.
Interestingly it fails in the same way as the RADV driver does,
so one might suspect there is some low level problem. I see the
test failing both with a RDNA and a RDNA2 GPU.
-
387d1d6c
by Giovanni Mascellani at 2026-05-22T19:32:57+02:00
tests/hlsl: Tweak a f16 test in arithmetic-float-uniform.shader_test.
The test currently fails on AMD, apparently because x / y is
implemented as x * (1 / y), and (1 / y) is denormal when y is
65504. It seems that the denorm is flushed to zero and then the
result as a whole is zero.
Since we don't want to test that behavior specifically in this
test, let's use less extreme numbers.
-
2f005b90
by Giovanni Mascellani at 2026-05-22T19:34:52+02:00
tests/hlsl: Skip a test in float-comparison.shader_test on Windows AMD.
Apparently getting NaN comparison right is quite tricky for GPU
vendors. Most shaders likely don't care too much anyway.
-
6e45d9aa
by Giovanni Mascellani at 2026-05-22T19:40:33+02:00
tests/hlsl: Account for a bug in switch.shader_test on Windows AMD.