[llvm] [NVPTX] Support llvm.{exp2, log2} for f32/f16/bf16 and vectors (PR #120519)
Joseph Huber via llvm-commits
llvm-commits at lists.llvm.org
Thu Dec 19 16:11:26 PST 2024
jhuber6 wrote:
> @jhuber6 you've dealt with the math library recently. Do you know what are the expectations for the precision of `exp2f`?
The expected precision isn't really documented in the general sense, but OpenCL has some guidelines, or you could look at what GCC currently offers. OpenCL states that `exp2` should be Within 3 ULP https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_C.html#relative-error-as-ulps.
https://github.com/llvm/llvm-project/pull/120519
More information about the llvm-commits
mailing list