diff options
| author | Jay Kwak <82421531+jkwak-work@users.noreply.github.com> | 2025-09-25 10:06:29 -0700 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-09-25 17:06:29 +0000 |
| commit | f55f669d1badc94eb4eabf77a22a5994acca7a89 (patch) | |
| tree | db43b05492ddb64aea3b99f3f090228b76e30646 /source | |
| parent | 2e0fe3d0608c4de239fd6cd2ecdf0322951855f5 (diff) | |
Fix VVL errors on coopvec tests (#8541)
It appears that the inputType of the coopvec-mat-mul cannot be signed
int32.
It could be floating types or signed int32.
Changing the tests to use uint32 instead of int32.
The spec guarantees the following combinations and the rest should be
queried at the runtime if it is supported by the HW.
https://registry.khronos.org/vulkan/specs/latest/man/html/VkCooperativeVectorPropertiesNV.html#_description
inputType | inputInterpretation | matrixInterpretation |
biasInterpretation | resultType
-- | -- | -- | -- | --
FLOAT16 | FLOAT16 | FLOAT16 | FLOAT16 | FLOAT16
UINT32 | SINT8_PACKED | SINT8 | SINT32 | SINT32
SINT8 | SINT8 | SINT8 | SINT32 | SINT32
FLOAT32 | SINT8 | SINT8 | SINT32 | SINT32
FLOAT16 | FLOAT_E4M3 | FLOAT_E4M3 | FLOAT16 | FLOAT16
FLOAT16 | FLOAT_E5M2 | FLOAT_E5M2 | FLOAT16 | FLOAT16
Diffstat (limited to 'source')
0 files changed, 0 insertions, 0 deletions