[PATCH] D152766: AMDGPU: Use correct lowering for llvm.log2.f32

Matt Arsenault via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Thu Jun 22 04:02:12 PDT 2023

arsenm added a comment.

In D152766#4440496 <https://reviews.llvm.org/D152766#4440496>, @rampitec wrote:

> To proceed I want to see a one single example of a computation giving a wrong result with 0.5ULP rounding, correct result with 0.51ULP rounding [whatever it is], and then again wrong result with 1ULP rounding.

This has nothing to do with the patch. There's no behavior change here for the f16 intrinsic. We continue to directly pass through to the hardware instructions, and the documentation is simply restating what the ISA manual states. The point of the documentation is stating the f16 case passes through directly and the f32 is a small expansion



More information about the llvm-commits mailing list