[all-commits] [llvm/llvm-project] 37c6cf: [AMDGPU] Fix incorrect MachineMemOperand offsets a...

[Abhinav Garg via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 37c6cfe1c3888d9af8c9b4b17eccbc8a94447903
      https://github.com/llvm/llvm-project/commit/37c6cfe1c3888d9af8c9b4b17eccbc8a94447903
  Author: Abhinav Garg <abhinav.garg at amd.com>
  Date:   2026-04-08 (Wed, 08 Apr 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
    M llvm/lib/Target/AMDGPU/SIISelLowering.cpp
    M llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.s.buffer.load.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.s.buffer.load.ll
    A llvm/test/CodeGen/AMDGPU/s-buffer-load-mmo-offsets.ll

  Log Message:
  -----------
  [AMDGPU] Fix incorrect MachineMemOperand offsets and sizes in wide s_buffer_load splits (#189890)

When G_AMDGPU_S_BUFFER_LOAD (or its SelectionDAG equivalent) falls back
to MUBUF due to a divergent offset, wide loads (256-bit, 512-bit) are
split into multiple 128-bit chunks. Both code paths that perform this
split had bugs in how they annotated MachineMemOperand (MMO) metadata on
each chunk instruction — reporting wrong offsets and wrong sizes. This
does not affect generated assembly correctness but degrades the analysis
if that MMO metadata is used.

---------

Co-authored-by: Abhinav Garg <abhigarg at amd.com>
Co-authored-by: Jay Foad <jay.foad at gmail.com>



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