[DAGCombine][X86][AArch64][AMDGPU] (x - y) + -1 -> add (xor y, -1), x fold

Summary:
This prevents regressions in next patch,
and somewhat recovers from the regression to AMDGPU test in D62223.

It is indeed not great that we leave vector decrement,
don't transform it into vector add all-ones..

https://rise4fun.com/Alive/ZRl

Reviewers: RKSimon, craig.topper, spatel, arsenm

Reviewed By: RKSimon, arsenm

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62263

llvm-svn: 361855

1 files changed, 14 insertions, 0 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index df842c27e27..43899ecb8e6 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -2303,6 +2303,13 @@ SDValue DAGCombiner::visitADDLike(SDNode *N) {
     }
   }
 
+  // (x - y) + -1  ->  add (xor y, -1), x
+  if (N0.hasOneUse() && N0.getOpcode() == ISD::SUB &&
+      isAllOnesOrAllOnesSplat(N1)) {
+    SDValue Xor = DAG.getNode(ISD::XOR, DL, VT, N0.getOperand(1), N1);
+    return DAG.getNode(ISD::ADD, DL, VT, Xor, N0.getOperand(0));
+  }
+
   if (SDValue Combined = visitADDLikeCommutative(N0, N1, N))
     return Combined;
 
@@ -2931,6 +2938,13 @@ SDValue DAGCombiner::visitSUB(SDNode *N) {
   if (SDValue V = foldAddSubMasked1(false, N0, N1, DAG, SDLoc(N)))
     return V;
 
+  // (x - y) - 1  ->  add (xor y, -1), x
+  if (N0.hasOneUse() && N0.getOpcode() == ISD::SUB && isOneOrOneSplat(N1)) {
+    SDValue Xor = DAG.getNode(ISD::XOR, DL, VT, N0.getOperand(1),
+                              DAG.getAllOnesConstant(DL, VT));
+    return DAG.getNode(ISD::ADD, DL, VT, Xor, N0.getOperand(0));
+  }
+
   // Hoist one-use addition by constant:  (x + C) - y  ->  (x - y) + C
   if (N0.hasOneUse() && N0.getOpcode() == ISD::ADD &&
       isConstantOrConstantVector(N0.getOperand(1))) {