[PATCH] D101357: [CodeGen][AArch64][SVE] Fold [rdffr, ptest] => rdffrs; bugfix for optimizePTestInstr

[Peter Waller via Phabricator via llvm-commits]

peterwaller-arm marked 2 inline comments as done.
peterwaller-arm added a comment.

Found and fixed additional bug here.



================
Comment at: llvm/lib/Target/AArch64/AArch64InstrInfo.cpp:1313-1321
+  const TargetRegisterInfo *TRI = &getRegisterInfo();
+
+  // If another instruction between the propagation and test sets the
+  // flags, don't remove the ptest.
+  MachineBasicBlock::iterator I = Pred, E = PTest;
+  ++I; // Skip past the predicate op itself.
+  for (; I != E; ++I)
----------------
paulwalker-arm wrote:
> I have no proof on this so feel free to ignore if you disagree.
> 
> Hoisting the `Pred->getParent() != PTest->getParent()` part offers a faster route to termination so I'm good with that.
> 
> However, I feel iterating across the BB's instructions may not and so that part might be better to keep until after we've decided `Mask` and `Pred` are interesting.
Opted to leave the code as is for now and change it in next patch.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D101357/new/

https://reviews.llvm.org/D101357