[PATCH] D130989: [AMDGPU][MC][GFX11] Correct v_dot2_f16_f16 and v_dot2_bf16_bf16
Stanislav Mekhanoshin via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Aug 2 11:25:01 PDT 2022
rampitec accepted this revision.
rampitec added a comment.
In D130989#3693679 <https://reviews.llvm.org/D130989#3693679>, @foad wrote:
>> For v_dot2_bf16_bf16 I am a bit unsure. Because we treat bf16 as int16, will ISel work correctly if we use the FP16InputMods? We don't have a codegen test using modifiers for v_dot2_bf16_bf16, one should probably be added to llvm.amdgcn.fdot2.bf16.bf16.ll . IntOpSelMods is a potential alternative SrcModifiers operand class to use.
>
> + @rampitec. This already came up on D129084 <https://reviews.llvm.org/D129084>.
At least from the encoding point of view modifiers are present, same as with f16. From the codegen point of view we cannot use these anyway because bf16 is not a legal type.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D130989/new/
https://reviews.llvm.org/D130989
More information about the llvm-commits
mailing list