[PATCH] D137689: [AArch64] Select BFI/BFXIL to ORR with shifted operand when one operand is the left or right shift of another operand

[Mingming Liu via Phabricator via llvm-commits]

mingmingl added inline comments.


================
Comment at: llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp:2945
+      SDValue Ops[] = {OrOpd1, OrOpd1,
+                       CurDAG->getTargetConstant(ShlImm, DL, VT)};
+      CurDAG->SelectNodeTo(N, OrrOpc, VT, Ops);
----------------
dmgreen wrote:
> Should this go through getShifterImm, for consistency even if LSL == 0?
Done in https://reviews.llvm.org/D137797.


================
Comment at: llvm/test/CodeGen/AArch64/bitfield-insert.ll:641
 ; For or operation, one operand is a left shift of another operand.
 ; Use orr with left-shifted operand is better than bfi.
 define i64 @test_orr_not_bfi_i64(i64 %0) {
----------------
khchen wrote:
> seem we could update the comments as well.
I realize it could be confusing to say "orr is better than a instruction that  wouldn't be generated going forward", so rephrased to 'orr is generated, not bfi/bfxil'. Let me know if I miss what should be updated, thanks!


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D137689/new/

https://reviews.llvm.org/D137689