[AArch64] When combining constant mul of power of 2 plus/minus 1, prefer shift

plus add.  The shift can be folded into the add.  This only effects codegen
when the constant is 3.

llvm-svn: 210445

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index e2c73ce..e2ed20d 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -6344,15 +6344,6 @@ static SDValue performMulCombine(SDNode *N, SelectionDAG &DAG,
   if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1))) {
     APInt Value = C->getAPIntValue();
     EVT VT = N->getValueType(0);
-    APInt VP1 = Value + 1;
-    if (VP1.isPowerOf2()) {
-      // Multiplying by one less than a power of two, replace with a shift
-      // and a subtract.
-      SDValue ShiftedVal =
-          DAG.getNode(ISD::SHL, SDLoc(N), VT, N->getOperand(0),
-                      DAG.getConstant(VP1.logBase2(), MVT::i64));
-      return DAG.getNode(ISD::SUB, SDLoc(N), VT, ShiftedVal, N->getOperand(0));
-    }
     APInt VM1 = Value - 1;
     if (VM1.isPowerOf2()) {
       // Multiplying by one more than a power of two, replace with a shift
@@ -6362,6 +6353,15 @@ static SDValue performMulCombine(SDNode *N, SelectionDAG &DAG,
                       DAG.getConstant(VM1.logBase2(), MVT::i64));
       return DAG.getNode(ISD::ADD, SDLoc(N), VT, ShiftedVal, N->getOperand(0));
     }
+    APInt VP1 = Value + 1;
+    if (VP1.isPowerOf2()) {
+      // Multiplying by one less than a power of two, replace with a shift
+      // and a subtract.
+      SDValue ShiftedVal =
+          DAG.getNode(ISD::SHL, SDLoc(N), VT, N->getOperand(0),
+                      DAG.getConstant(VP1.logBase2(), MVT::i64));
+      return DAG.getNode(ISD::SUB, SDLoc(N), VT, ShiftedVal, N->getOperand(0));
+    }
   }
   return SDValue();
 }

diff --git a/llvm/test/CodeGen/AArch64/arm64-arith.ll b/llvm/test/CodeGen/AArch64/arm64-arith.ll
index ed9b569..f36e706 100644 (file)
--- a/llvm/test/CodeGen/AArch64/arm64-arith.ll
+++ b/llvm/test/CodeGen/AArch64/arm64-arith.ll
@@ -260,3 +260,11 @@ define i64 @f3(i64 %a) nounwind readnone ssp {
   %res = mul nsw i64 %a, 17
   ret i64 %res
 }
+
+define i32 @f4(i32 %a) nounwind readnone ssp {
+; CHECK-LABEL: f4:
+; CHECK-NEXT: add w0, w0, w0, lsl #1
+; CHECK-NEXT: ret
+  %res = mul i32 %a, 3
+  ret i32 %res
+}