[PATCH] D157511: GlobalISel: Add constant fold combine for zext/sext/anyext

[Pierre van Houtryve via Phabricator via llvm-commits]

Pierre-vh added inline comments.


================
Comment at: llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h:641-644
+  bool matchConstantFoldCastOp(MachineInstr &MI, APInt &MatchInfo);
+
+  /// Do constant folding when opportunities are exposed after MIR building.
+  bool matchConstantFoldBinOp(MachineInstr &MI, APInt &MatchInfo);
----------------
nit: could remove the second comment and just put the decls right next to each other.


================
Comment at: llvm/lib/CodeGen/GlobalISel/Utils.cpp:785
+  case TargetOpcode::G_SEXT:
+    return Val->sext(DstTy.getScalarSizeInBits());
+  case TargetOpcode::G_ZEXT:
----------------
nit: put `DstTy.getScalarSizeInBits()` in a variable


================
Comment at: llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-binop-same-val.mir:29
 
-    ; Fold: x and x -> x
 
----------------
lost comments in this file


================
Comment at: llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-bzero.mir:21
     liveins: $x0, $x1
-    ; Always use G_BZERO when the memset width is unknown on Darwin.
 
----------------
ditto


================
Comment at: llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-trivial-arith.mir:29
     liveins: $w0
-    ; Fold (x + 0) -> x
-    ;
----------------
ditto


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D157511/new/

https://reviews.llvm.org/D157511