[0035] Atomic vector accumulation missing from the spec.

[anupamachandra]

Which document does this relate to?
https://github.com/microsoft/hlsl-specs/blob/main/proposals/0035-linalg-matrix.md

Describe the issue you see with the spec
The LinAlg spec is missing the functionality to accumulate Fp16 elements of a vector atomically-elementwise to corresponding elements in memory.

Additional context
The earlier CoopVec spec had a vectoraccumulate function ( support set: Fp16), one important use case was accumulation of gradients in ML workloads. This piece is missing in the current spec. I am not sure if this functionality should be added to this spec, which is more matrix-centric, or as an independent feature.