[all-commits] [llvm/llvm-project] 90daea: AMDGPU: Codegen for v_dual_dot2acc_f32_f16/bf16 fr...

[Petar Avramovic via All-commits]

  Branch: refs/heads/users/petar-avramovic/dualdot2acc
  Home:   https://github.com/llvm/llvm-project
  Commit: 90daea748636c5b7198ddef160dc00591301e215
      https://github.com/llvm/llvm-project/commit/90daea748636c5b7198ddef160dc00591301e215
  Author: Petar Avramovic <Petar.Avramovic at amd.com>
  Date:   2026-03-23 (Mon, 23 Mar 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/AMDGPU.td
    M llvm/lib/Target/AMDGPU/GCNVOPDUtils.cpp
    M llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
    M llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
    M llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
    M llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
    M llvm/lib/Target/AMDGPU/VOP3PInstructions.td
    M llvm/lib/Target/AMDGPU/VOPInstructions.td
    M llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fdot2.f32.bf16.ll
    M llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fdot2.ll

  Log Message:
  -----------
  AMDGPU: Codegen for v_dual_dot2acc_f32_f16/bf16 from VOP3

Codegen for v_dual_dot2acc_f32_f16/bf16 for targets that only have VOP3
version of the instruction.
Since there is no VOP2 version, instroduce temporary mir DOT2ACC pseudo
that is selected when there are no src_modifiers. This DOT2ACC pseudo
has src2 tied to dst (like the VOP2 version), PostRA pseudo expansion will
restore pseudo to VOP3 version of the instruction.
CreateVOPD will recoginize such VOP3 pseudo and generate v_dual_dot2acc.



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