diff options
author | Piotr Sobczak <piotr.sobczak@amd.com> | 2025-03-07 08:56:44 +0100 |
---|---|---|
committer | GitHub <noreply@github.com> | 2025-03-07 08:56:44 +0100 |
commit | 170c0dac4488f9cfbc67e9593ebe6ad01cfa8f32 (patch) | |
tree | 3effcaf624f50da320ba86bfd9d51f6076f439c7 /llvm/lib/CodeGen/ModuloSchedule.cpp | |
parent | a21cfca320bddeef120618ceff9563778b5cbd94 (diff) | |
download | llvm-170c0dac4488f9cfbc67e9593ebe6ad01cfa8f32.zip llvm-170c0dac4488f9cfbc67e9593ebe6ad01cfa8f32.tar.gz llvm-170c0dac4488f9cfbc67e9593ebe6ad01cfa8f32.tar.bz2 |
[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.
Diffstat (limited to 'llvm/lib/CodeGen/ModuloSchedule.cpp')
0 files changed, 0 insertions, 0 deletions