[PATCH] D101292: [AMDGPU] Restrict immediate scratch offsets

[Jay Foad via Phabricator via llvm-commits]

foad added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:7321
+      FlatVariant == SIInstrFlags::FlatScratch && Offset < 0 &&
+      (Offset % 4) != 0) {
+    return false;
----------------
sebastian-ne wrote:
> foad wrote:
> > Is 4 always the correct value to use here? Doesn't it depend on the size of the access?
> It doesn’t depend on the size of the access. scratch_store_byte/short/dword/dwordx2/dwordx3/dwordx4 all behave the same.
> So yes, 4 is always correct here.
Please mention that in the description of FeatureNegativeUnalignedScratchOffsetBug.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D101292/new/

https://reviews.llvm.org/D101292