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

[Siva Chandra via Phabricator via libc-commits]

sivachandra accepted this revision.
sivachandra added inline comments.
This revision is now accepted and ready to land.


================
Comment at: libc/config/gpu/entrypoints.txt:84-90
+set(TARGET_LIBM_ENTRYPOINTS
+    # math.h entrypoints
+    libc.src.math.sin
+    libc.src.math.round
+    libc.src.math.roundf
+    libc.src.math.roundl
+)
----------------
lntue wrote:
> jhuber6 wrote:
> > lntue wrote:
> > > Can you also update the math implementation status table at https://github.com/llvm/llvm-project/blob/main/libc/docs/math/index.rst ?
> > > Thanks,
> > I was thinking of doing this in a separate patch, and we may want to have a different category for the GPU that indicates if it's implemented natively or if we rely on the vendor. Then we'd have some documentation about the considerations of GPU math, like reduced precision, ULP modes, linking to vendor documentation for those functions, etc.
> A separate patch is fine for me.  I plan to use this table for enabled entrypoints.  For extra information about how they are implemented and what are the options / tradeoffs, you can put them under https://libc.llvm.org/math/#algorithms-implementation-details , or add a separate page and add a link to it in that section.
Yes, I vote for a separate GPU math status page as it seems like is going to be different even wrt the math library promise we make about the CPU implementations.


Repository:
  rG LLVM Github Monorepo

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

https://reviews.llvm.org/D152486