[PATCH] D110567: [ARM] Delay reverting WLS in arm-block-placement

[Sjoerd Meijer via Phabricator via llvm-commits]

SjoerdMeijer accepted this revision.
SjoerdMeijer added a comment.
This revision is now accepted and ready to land.

Looks like a good fix to me.

The test is indeed big, but I guess it is what it is if it's difficult to reduce.



================
Comment at: llvm/lib/Target/ARM/ARMBlockPlacement.cpp:34
   MachineLoopInfo *MLI = nullptr;
+  SmallVector<MachineInstr *> RevertedWhileLoops;
 
----------------
Nit: perhaps add a comment why we are adding/doing this bookkeeping (i.e. the description of this patch).


================
Comment at: llvm/lib/Target/ARM/ARMBlockPlacement.cpp:88
 // because of the branches this requires an extra block to be created.
-bool ARMBlockPlacement::revertWhileToDo(MachineInstr *WLS, MachineLoop *ML) {
+bool ARMBlockPlacement::revertWhileToDo(MachineInstr *WLS) {
   //   lr = t2WhileLoopStartTP r0, r1, TgtBB
----------------
Unrelated nit: perhaps time to rename this to `revertWhileToDoLoop` while we are at it?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D110567/new/

https://reviews.llvm.org/D110567