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

[Joseph Huber via Phabricator via libc-commits]

jhuber6 added inline comments.


================
Comment at: libc/src/math/gpu/vendor/amdgpu/platform.h:35
+extern const uint32_t __oclc_ISA_version = 7000;
+#elif defined(__gfx701__)
+extern const uint32_t __oclc_ISA_version = 7001;
----------------
JonChesterfield wrote:
> This oclc isa version thing is not great. We should patch clang to emit the value directly instead of linking in an IR file with the same name that defines that value, then this massive branch thing evaporates
I tried doing that at some point but gave up, primarily because we mix per-TU controls (fast math like above) and per-executable ones (like this). I don't see a point in only enabling this single one. I think @arsenm was looking at removing the need for the per-TU math controls which would make that patch viable again.


Repository:
  rG LLVM Github Monorepo

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

https://reviews.llvm.org/D152486