[PATCH] D53310: [X86] Match (cmp (and (shr X, C), mask), 0) to BEXTR+TEST.

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added inline comments.


================
Comment at: lib/Target/X86/X86ISelDAGToDAG.cpp:457
     bool foldLoadStoreIntoMemOperand(SDNode *Node);
-    bool matchBEXTRFromAndImm(SDNode *Node);
+    MachineSDNode *matchBEXTRFromAndImm(SDNode *Node);
     bool matchBEXTR(SDNode *Node);
----------------
Can probably be a separate preparatory cleanup commit.


================
Comment at: lib/Target/X86/X86ISelDAGToDAG.cpp:3564
+        SDValue BEXTR = SDValue(NewNode, 0);
+        NewNode = CurDAG->getMachineNode(TestOpc, dl, MVT::i32, BEXTR, BEXTR);
+        ReplaceUses(SDValue(Node, 0), SDValue(NewNode, 0));
----------------
Is this correct?
We will always use `MVT::i32`, even for `X86::TEST64rr`?
We shouldn't be using `CmpVT` here?


Repository:
  rL LLVM

https://reviews.llvm.org/D53310