[all-commits] [llvm/llvm-project] 170c0d: [AMDGPU] Fix edge case of buffer OOB handling (#11...

[Piotr Sobczak via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 170c0dac4488f9cfbc67e9593ebe6ad01cfa8f32
      https://github.com/llvm/llvm-project/commit/170c0dac4488f9cfbc67e9593ebe6ad01cfa8f32
  Author: Piotr Sobczak <piotr.sobczak at amd.com>
  Date:   2025-03-07 (Fri, 07 Mar 2025)

  Changed paths:
    M llvm/lib/Target/AMDGPU/AMDGPU.td
    M llvm/lib/Target/AMDGPU/GCNSubtarget.h
    M llvm/lib/Target/AMDGPU/SIISelLowering.cpp
    A llvm/test/CodeGen/AMDGPU/unaligned-buffer.ll
    M llvm/test/CodeGen/AMDGPU/vectorize-buffer-fat-pointer.ll
    M llvm/test/Transforms/LoadStoreVectorizer/AMDGPU/merge-vectors.ll
    A llvm/test/Transforms/LoadStoreVectorizer/AMDGPU/unaligned-buffer.ll

  Log Message:
  -----------
  [AMDGPU] Fix edge case of buffer OOB handling (#115479)

Strengthen out-of-bounds guarantees for buffer accesses by disallowing
buffer accesses with alignment lower than natural alignment.

This is needed to specifically address the edge case where an access
starts out-of-bounds and then enters in-bounds, as the hardware would
treat the entire access as being out-of-bounds. This is normally not
needed for most users, but at least one graphics device extension
(VK_EXT_robustness2) has very strict requirements - in-bounds accesses
must return correct value, and out-of-bounds accesses must return zero.

The direct consequence of the patch is that a buffer access at negative
address is not merged by load-store-vectorizer with one at a positive
address, which fixes a CTS test.

Targets that do not care about the new behavior are advised to use the
new target feature relaxed-buffer-oob-mode that maintains the state from
before the patch.



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