Invalid SPIR-V generation for vector<X, 8>/vector<X, 16> usage

[jtytgat]

Slightly related to #5987 but the resolution might be different for the following:

When using vector<X, 8>/vector<X, 16>, Slang doesn't generate a warning nor error but generates invalid SPIR-V output instead.

StructuredBuffer<vector<float, 8>> bufferIn;
RWStructuredBuffer<float> resultOut;

[shader("compute")]
[numthreads(1,1,1)]
void computeMain(uint3 threadId : SV_DispatchThreadID)
{
	uint index = threadId.x;
	
	resultOut[index] = bufferIn[index][0] + bufferIn[index][4];
}

$ slangc test.slang -target spirv -o test.spv
$ spirv-val test.spv
error: line 37: Having 8 components for TypeVector requires the Vector16 capability
  %v8float = OpTypeVector %float 8

Marking the SPIR-V output with Vector16 capability is not an option as this is a kernel execution only capability.

Could one of the following be done?

• Generate an error for all output types where this is not supported.
• Split the vector<X,8>/vector<X,16> access into vector<X,4> access.

[jtytgat]

Looks like this issue is actually superseded by the proposal https://github.com/shader-slang/slang/blob/master/docs/proposals/014-extended-length-vectors.md
Am I right @expipiplus1 ?

[bmillsNV]

For now, let's just generate errors. Marking for current quarter.