[llvm] [NVPTX] Lower bfloat16 add/mul/sub as fma on SM80 (PR #121065)

Artem Belevich via llvm-commits llvm-commits at lists.llvm.org
Wed Jan 8 15:59:55 PST 2025

Previous message: [llvm] [DXIL] Define and generate `DXILAttribute` and `DXILProperty` (PR #117072)
Next message: [llvm] [NVPTX] Lower bfloat16 add/mul/sub as fma on SM80 (PR #121065)
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Artem-B wrote:

Could we just declare bfloat16 add/mul/sub legal when FMA is available, and just lower them as FMA via pattern-matching, without messing with the legalizer and combiner? I'm not quite sure why we want to convert those mul/add to FMA in the DAG itself.


https://github.com/llvm/llvm-project/pull/121065

Previous message: [llvm] [DXIL] Define and generate `DXILAttribute` and `DXILProperty` (PR #117072)
Next message: [llvm] [NVPTX] Lower bfloat16 add/mul/sub as fma on SM80 (PR #121065)
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

More information about the llvm-commits mailing list