[PATCH] D121491: [AMDGPU] Restrict machine copy propagation from creating unaligned classes

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:4798
+    if (Size > 32) {
+      const SIRegisterInfo *RI = static_cast<const SIRegisterInfo*>(TRI);
+      if (SIRegisterInfo::isVGPRClass(RC))
----------------
RI is already a class member


================
Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:4799-4806
+      if (SIRegisterInfo::isVGPRClass(RC))
+        RC = RI->getVGPRClassForBitWidth(Size);
+      else if (SIRegisterInfo::isAGPRClass(RC))
+        RC = RI->getAGPRClassForBitWidth(Size);
+      else if (RI->isVectorSuperClass(RC))
+        RC = RI->getVectorSuperClassForBitWidth(Size);
+    }
----------------
Why fix this up here when adjustAllocatableRegClass can return the aligned version in the first place?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D121491/new/

https://reviews.llvm.org/D121491