[DAGCombiner] (add (umax X, C), -C) --> (usubsat X, C) (PR40111)

Move the (add (umax X, C), -C) --> (usubsat X, C) X86 combine into generic DAGCombiner

First of a number of saturated arithmetic folds that can be moved out of X86-specific code for PR40111.

Differential Revision: https://reviews.llvm.org/D57754

llvm-svn: 353457

2 files changed, 12 insertions, 37 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index dabb8afd104..a9904a25783 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -2159,6 +2159,18 @@ SDValue DAGCombiner::visitADD(SDNode *N) {
                          DAG.getNode(ISD::ADD, SDLoc(N1), VT, N01, N11));
   }
 
+  // fold (add (umax X, C), -C) --> (usubsat X, C)
+  if (N0.getOpcode() == ISD::UMAX && hasOperation(ISD::USUBSAT, VT)) {
+    auto MatchUSUBSAT = [](ConstantSDNode *Max, ConstantSDNode *Op) {
+      return (!Max && !Op) ||
+             (Max && Op && Max->getAPIntValue() == (-Op->getAPIntValue()));
+    };
+    if (ISD::matchBinaryPredicate(N0.getOperand(1), N1, MatchUSUBSAT,
+                                  /*AllowUndefs*/ true))
+      return DAG.getNode(ISD::USUBSAT, DL, VT, N0.getOperand(0),
+                         N0.getOperand(1));
+  }
+
   if (SDValue V = foldAddSubBoolOfMaskedVal(N, DAG))
     return V;
 
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 8f2a6ef6264..ff69b387a8b 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -41383,40 +41383,6 @@ static SDValue matchPMADDWD(SelectionDAG &DAG, SDValue Op0, SDValue Op1,
                           PMADDBuilder);
 }
 
-// Try to turn (add (umax X, C), -C) into (psubus X, C)
-static SDValue combineAddToSUBUS(SDNode *N, SelectionDAG &DAG,
-                                 const X86Subtarget &Subtarget) {
-  if (!Subtarget.hasSSE2())
-    return SDValue();
-
-  EVT VT = N->getValueType(0);
-
-  // psubus is available in SSE2 for i8 and i16 vectors.
-  if (!VT.isVector() || VT.getVectorNumElements() < 2 ||
-      !isPowerOf2_32(VT.getVectorNumElements()) ||
-      !(VT.getVectorElementType() == MVT::i8 ||
-        VT.getVectorElementType() == MVT::i16))
-    return SDValue();
-
-  SDValue Op0 = N->getOperand(0);
-  SDValue Op1 = N->getOperand(1);
-  if (Op0.getOpcode() != ISD::UMAX)
-    return SDValue();
-
-  // The add should have a constant that is the negative of the max.
-  auto MatchUSUBSAT = [](ConstantSDNode *Max, ConstantSDNode *Op) {
-    return (!Max && !Op) ||
-           (Max && Op && Max->getAPIntValue() == (-Op->getAPIntValue()));
-  };
-  if (!ISD::matchBinaryPredicate(Op0.getOperand(1), Op1, MatchUSUBSAT,
-                                 /*AllowUndefs*/ true))
-    return SDValue();
-
-  SDLoc DL(N);
-  return DAG.getNode(ISD::USUBSAT, DL, VT, Op0.getOperand(0),
-                     Op0.getOperand(1));
-}
-
 // Attempt to turn this pattern into PMADDWD.
 // (mul (add (zext (build_vector)), (zext (build_vector))),
 //      (add (zext (build_vector)), (zext (build_vector)))
@@ -41572,9 +41538,6 @@ static SDValue combineAdd(SDNode *N, SelectionDAG &DAG,
   if (SDValue V = combineIncDecVector(N, DAG))
     return V;
 
-  if (SDValue V = combineAddToSUBUS(N, DAG, Subtarget))
-    return V;
-
   return combineAddOrSubToADCOrSBB(N, DAG);
 }