[Mlir-commits] [mlir] [OpenMP] Enable simd in non-reduction composite constructs (PR #146097)

[Tom Eccles]

tblah wrote:

I believe the reason for not applying the "not yet implemented" message to simd reduction for composite constructs is that if the reduction clause is used on the composite construct, that reduction is logically applied to most of the constructs comprising the composite construct. So disabling simd reduction would disable reduction for any composite construct containing simd.

SIMD is unusual because not implementing the reduction won't produce incorrect results (except some subtle effects around the re-presentable range of numbers) - it is just less likely to vectorize. I will post a patch implementing this soon so I don't think we need to worry.

https://github.com/llvm/llvm-project/pull/146097