[0035] vector InterlockedAccumulate parameter ordering is inconsistent

[tex3d]

Which proposal does this relate to?

hlsl-specs/proposals/0035-linalg-matrix.md

Lines 287 to 290 in c2f44e3

	template <typename InputElTy, SIZE_TYPE M>
	typename hlsl::enable_if<hlsl::is_arithmetic<InputElTy>::value, void>::type
	InterlockedAccumulate(vector<InputElTy, M> Vec, RWByteAddressBuffer Res,
	uint StartOffset);

Describe the issue or outstanding question.
Parameter ordering for Interlocked operations typically starts with a parameter for the location into which you are accumulating. This new operation reverses the parameter ordering, starting with the vector to accumulate, followed by the target buffer.

I think it would make more sense to reverse the order of these parameters.

The earlier we do this, the less pain, since this operation is brand new and we haven't released a preview that includes it yet.

For the DXIL op (dx.op.vectorAccumulateToDescriptor), I think it would be more consistent with existing atomic operations to place the resource and offset before the vector.

While we're at it, I also think the DXIL op naming should be made more consistent with other atomic operations in DXIL as well, such as: AtomicVectorAccumulate. Though this could be considered an independent issue.

[pow2clk]

I don't object to any of this. We could take the chance to make the name more consistent as the rename didn't make it into my PR and we might consider adding an align argument while we're at it.

[V-FEXrt]

Sorry, I hadn't fully read this issue before our discussions today so I had a slightly different idea of its contents.

Specially I thought we were renaming VectorAccumulateToDescriptor to LinAlgVectorAccumulateToDescriptor and I think that effects some points below.

I think it would be more consistent with existing atomic operations to place the resource and offset before the vector.

Right now the ordering matches LinAlgMatrixAccumulateToDescriptor's DXIL Op and I'm tempted to keep that the same, we are currently positioning the operation as a "LinAlg Operation" and changing just this one thing makes it actually stick out more.

If we move forward with AtomicVectorAccumulate and say "actually this is a generic atomic vector operation" then I think the param swapping makes more sense though I did already do the aforementioned rename lol