shader-slang / slang-rhi / 20926626175

push

github

Cooperative Vector improvements (#620)

* Add enumToString for CooperativeVectorMatrixLayout

* Load VK_EXT_shader_float8 required for optimal float formats

* Do not align matrix sizes to 64 bytes on D3D12/Vulkan

* More exhaustive testing of getCooperativeVectorMatrixSize

* Only check offset alignment for optimal layouts

* Refactor matrix_view to use runtime info for rows/cols/majorness

* More conversion tests

* Extend conversion test to include Float16 type

* Align CUDA backend to D3D12/Vulkan when computing matrix sizes

3167 of 19752 branches covered (16.03%)

Branch coverage included in aggregate %.

20 of 39 new or added lines in 4 files covered. (51.28%)

9939 of 43109 relevant lines covered (23.06%)

73.71 hits per line