frasercrmck wrote:

@arsenm I've just updated this PR. It still moves the `sign` implementation to 
the CLC library but now also optimizes it for vector types. There's no control 
flow nor scalarizing for vectors. I trust this will be a good enough default 
for the scalar form (on GPUs) though I suppose CPUs may prefer the old control 
flow form. I doubt any would prefer the old vector form, though.

alive2 seems to be happy with the transformation (for `half2`): 
https://alive2.llvm.org/ce/z/RZ8Dc4

Let me know what you think.

https://github.com/llvm/llvm-project/pull/115699
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to