[libc-commits] [PATCH] D152486: [libc] Begin implementing a 'libmgpu.a' for math on the GPU

[Jon Chesterfield via Phabricator via libc-commits]

JonChesterfield added a comment.

Math optimisations were a bit of a mess a few years ago. Some in clang, some in llvm, some in backends, some based on intrinsics and some based on name of libm functions. There's a decent chance ocml_sin(0) won't constant fold and sin(0) will for example. There's a decent chance we'd win overall by keeping the functions named after libm functions until at least some of the middle end has run, before translating them into the vendor ones latish. Phase ordering is always tricky.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D152486/new/

https://reviews.llvm.org/D152486