[all-commits] [llvm/llvm-project] 836201: Allow bf16 operands on new MFMAs (#144925)

[Umang Yadav via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 836201f1177c38f3ca0457de019bb179a04afe3c
      https://github.com/llvm/llvm-project/commit/836201f1177c38f3ca0457de019bb179a04afe3c
  Author: Umang Yadav <29876643+umangyadav at users.noreply.github.com>
  Date:   2025-06-19 (Thu, 19 Jun 2025)

  Changed paths:
    M mlir/lib/Conversion/AMDGPUToROCDL/AMDGPUToROCDL.cpp
    M mlir/test/Conversion/AMDGPUToROCDL/mfma-gfx950.mlir

  Log Message:
  -----------
  Allow bf16 operands on new MFMAs  (#144925)

New gfx950 MFMA allows bf16 operands. 


https://github.com/llvm/llvm-project/blob/c0cc81cdc03c97473ba771bbc3a2330bd22396bc/llvm/include/llvm/IR/IntrinsicsAMDGPU.td#L3434

When running `amdgpu-to-rocdl`, Current logic converts bf16 to i16
always which fails to compile for newer bf16 MFMA e.g.
`v_mfma_f32_16x16x32bf16`.
Backend expects bf16 type for the operands for those newer MFMAs. This
patch fixes it.

CC: @krzysz00  @dhernandez0  @giuseros  @antiagainst  @kuhar



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications