[llvm] [RFC][WIP][AMDGPU] Use `bf16` instead of `i16` for bfloat (PR #80908)
Stanislav Mekhanoshin via llvm-commits
llvm-commits at lists.llvm.org
Tue Feb 6 14:50:48 PST 2024
================
@@ -2819,11 +2819,11 @@ def int_amdgcn_fdot2_f16_f16 :
def int_amdgcn_fdot2_bf16_bf16 :
ClangBuiltin<"__builtin_amdgcn_fdot2_bf16_bf16">,
DefaultAttrsIntrinsic<
- [llvm_i16_ty], // %r
+ [llvm_bfloat_ty], // %r
----------------
rampitec wrote:
This is the change in the existing interface, it will break programs already using it. It also will break clang builtin.
You can get away with changing only instruction's profile to use BF16 and tweak pattern with casts to bf16.
https://github.com/llvm/llvm-project/pull/80908
More information about the llvm-commits
mailing list