[SM6.10] linalg::MultiplyAdd template: wrong Size check condition for InterpretedVector

[NealLou]

Description

DirectXShaderCompiler/tools/clang/lib/Headers/hlsl/dx/linalg.h

Lines 508 to 512 in 4ad9834

	typename hlsl::enable_if<
	InterpretedVector<InputElTy, VecK, InputInterp>::Size == K,
	vector<OutputElTy, M> >::type
	// clang-format on
	MultiplyAdd(Matrix<MatrixDT, M, K, MatrixUse::A, MatrixScope::Thread> MatrixA,

DirectXShaderCompiler/tools/clang/lib/Headers/hlsl/dx/linalg.h

Lines 544 to 548 in 4ad9834

	typename hlsl::enable_if<
	InterpretedVector<InputElTy, VecK, InputInterp>::Size == K,
	vector<OutputElTy, M> >::type
	// clang-format on
	MultiplyAdd(Matrix<MatrixDT, M, K, MatrixUse::A, MatrixScope::Thread> MatrixA,

After dxc fixed the linalg::Convert DstN value calculation in #8359,
this issue appears: for F16->FP8 case, when K = 15, InterpretedVector<uint, 4, ComponentEnum::F8_E4M3FN>::Size=ElementsPerScalar * N=4 * 4 =16.
I believe the correct logic should be to ensure Size >= K.

Steps to Reproduce

Actual Behavior

Environment

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

[tex3d]

I believe the correct logic should be to ensure Size >= K.

I think the check should be more exact than this. It could be by converting K to packed vector size (div by ElementsPerScalar rounded up) and comparing against VecK, or by comparing a range.