[SM6.10] linAlgMatrixAccumulateToDescriptor: Incorrect Matrix definition in LinAlgTests

[NealLou]

Description
In 0035 Spec,
The validation rules for linAlgMatrixAccumulateToDescriptor state that:

• Layout is OuterProductOptimal for matrix with MatrixScope of Thread
• Layout is RowMajor or ColMajor for matrix with MatrixScope of Wave or ThreadGroup

However, there are at least 2 cases where a matrix is defined with Thread scope and RowMajor layout.
One of them is:

DirectXShaderCompiler/tools/clang/unittests/HLSLExec/LinAlgTests.cpp

Lines 622 to 623 in 4ad9834

	Params.Scope = MatrixScope::Thread;
	Params.Layout = LinalgMatrixLayout::RowMajor;

Please check LinAlgTests.

Steps to Reproduce

Actual Behavior

Environment

• DXC version
• Host Operating System <!--- Host operating system and version --->

[V-FEXrt]

Thanks for the report! I'll get on this right away

[V-FEXrt]

I'm still working on the fixes, but I wanted to give a quick update to say that it seems only 2 of the 4 Thread+RowMajor cases were out of spec.

Since the VecMul/VecMulAdd tests don't use AccumulateToDescriptor the usage on those tests are okay (they use LoadFromDescriptor/VecMul/VecMulAdd none of which require a particular layout for Thread matrixes per the spec)

[NealLou]

I'm still working on the fixes, but I wanted to give a quick update to say that it seems only 2 of the 4 Thread+RowMajor cases were out of spec.

Since the VecMul/VecMulAdd tests don't use AccumulateToDescriptor the usage on those tests are okay (they use LoadFromDescriptor/VecMul/VecMulAdd none of which require a particular layout for Thread matrixes per the spec)

Yes, agreed. I've modified the description.