[AArch64] Implement target hook function to decide folding (mul (add x, c1), c2)

Prevent the folding if it leads to worse code.

Reviewed By: dmgreen

Differential Revision: https://reviews.llvm.org/D108871

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -12189,6 +12189,33 @@ bool AArch64TargetLowering::isLegalAddImmediate(int64_t Immed) const {
   return IsLegal;
 }
 
+// Return false to prevent folding
+// (mul (add x, c1), c2) -> (add (mul x, c2), c2*c1) in DAGCombine,
+// if the folding leads to worse code.
+bool AArch64TargetLowering::isMulAddWithConstProfitable(
+    const SDValue &AddNode, const SDValue &ConstNode) const {
+  // Let the DAGCombiner decide for vector types and large types.
+  const EVT VT = AddNode.getValueType();
+  if (VT.isVector() || VT.getScalarSizeInBits() > 64)
+    return true;
+
+  // It is worse if c1 is legal add immediate, while c1*c2 is not
+  // and has to be composed by at least two instructions.
+  const ConstantSDNode *C1Node = cast<ConstantSDNode>(AddNode.getOperand(1));
+  const ConstantSDNode *C2Node = cast<ConstantSDNode>(ConstNode);
+  const int64_t C1 = C1Node->getSExtValue();
+  const int64_t C2 = C2Node->getSExtValue();
+  if (!isLegalAddImmediate(C1) || isLegalAddImmediate(C1 * C2))
+    return true;
+  SmallVector<AArch64_IMM::ImmInsnModel, 4> Insn;
+  AArch64_IMM::expandMOVImm(C1 * C2, VT.getSizeInBits(),Insn);
+  if (Insn.size() > 1)
+    return false;
+
+  // Default to true and let the DAGCombiner decide.
+  return true;
+}
+
 // Integer comparisons are implemented with ADDS/SUBS, so the range of valid
 // immediates is the same as for an add or a sub.
 bool AArch64TargetLowering::isLegalICmpImmediate(int64_t Immed) const {

llvm/lib/Target/AArch64/AArch64ISelLowering.h

@@ -595,6 +595,9 @@ public:
   bool isLegalAddImmediate(int64_t) const override;
   bool isLegalICmpImmediate(int64_t) const override;
 
+  bool isMulAddWithConstProfitable(const SDValue &AddNode,
+                                   const SDValue &ConstNode) const override;
+
   bool shouldConsiderGEPOffsetSplit() const override;
 
   EVT getOptimalMemOpType(const MemOp &Op,

llvm/test/CodeGen/AArch64/addimm-mulimm.ll

@@ -104,10 +104,9 @@ define signext i32 @addimm_mulimm_accept_13(i32 signext %a) {
 define i64 @addimm_mulimm_reject_00(i64 %a) {
 ; CHECK-LABEL: addimm_mulimm_reject_00:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w9, #1200
-; CHECK-NEXT:    mov w8, #3700
-; CHECK-NEXT:    movk w9, #175, lsl #16
-; CHECK-NEXT:    madd x0, x0, x8, x9
+; CHECK-NEXT:    add x8, x0, #3100
+; CHECK-NEXT:    mov w9, #3700
+; CHECK-NEXT:    mul x0, x8, x9
 ; CHECK-NEXT:    ret
   %tmp0 = add i64 %a, 3100
   %tmp1 = mul i64 %tmp0, 3700
@@ -117,10 +116,9 @@ define i64 @addimm_mulimm_reject_00(i64 %a) {
 define i64 @addimm_mulimm_reject_01(i64 %a) {
 ; CHECK-LABEL: addimm_mulimm_reject_01:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov x9, #-1200
-; CHECK-NEXT:    mov w8, #3700
-; CHECK-NEXT:    movk x9, #65360, lsl #16
-; CHECK-NEXT:    madd x0, x0, x8, x9
+; CHECK-NEXT:    sub x8, x0, #3100
+; CHECK-NEXT:    mov w9, #3700
+; CHECK-NEXT:    mul x0, x8, x9
 ; CHECK-NEXT:    ret
   %tmp0 = add i64 %a, -3100
   %tmp1 = mul i64 %tmp0, 3700
@@ -130,10 +128,9 @@ define i64 @addimm_mulimm_reject_01(i64 %a) {
 define signext i32 @addimm_mulimm_reject_02(i32 signext %a) {
 ; CHECK-LABEL: addimm_mulimm_reject_02:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w9, #1200
-; CHECK-NEXT:    mov w8, #3700
-; CHECK-NEXT:    movk w9, #175, lsl #16
-; CHECK-NEXT:    madd w0, w0, w8, w9
+; CHECK-NEXT:    add w8, w0, #3100
+; CHECK-NEXT:    mov w9, #3700
+; CHECK-NEXT:    mul w0, w8, w9
 ; CHECK-NEXT:    ret
   %tmp0 = add i32 %a, 3100
   %tmp1 = mul i32 %tmp0, 3700
@@ -143,10 +140,9 @@ define signext i32 @addimm_mulimm_reject_02(i32 signext %a) {
 define signext i32 @addimm_mulimm_reject_03(i32 signext %a) {
 ; CHECK-LABEL: addimm_mulimm_reject_03:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w9, #64336
-; CHECK-NEXT:    mov w8, #3700
-; CHECK-NEXT:    movk w9, #65360, lsl #16
-; CHECK-NEXT:    madd w0, w0, w8, w9
+; CHECK-NEXT:    sub w8, w0, #3100
+; CHECK-NEXT:    mov w9, #3700
+; CHECK-NEXT:    mul w0, w8, w9
 ; CHECK-NEXT:    ret
   %tmp0 = add i32 %a, -3100
   %tmp1 = mul i32 %tmp0, 3700

llvm/test/CodeGen/AArch64/urem-seteq-nonzero.ll

@@ -154,11 +154,10 @@ define i1 @t32_6_3(i32 %X) nounwind {
 define i1 @t32_6_4(i32 %X) nounwind {
 ; CHECK-LABEL: t32_6_4:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #43691
-; CHECK-NEXT:    mov w9, #21844
-; CHECK-NEXT:    movk w8, #43690, lsl #16
-; CHECK-NEXT:    movk w9, #21845, lsl #16
-; CHECK-NEXT:    madd w8, w0, w8, w9
+; CHECK-NEXT:    mov w9, #43691
+; CHECK-NEXT:    sub w8, w0, #4
+; CHECK-NEXT:    movk w9, #43690, lsl #16
+; CHECK-NEXT:    mul w8, w8, w9
 ; CHECK-NEXT:    mov w9, #43690
 ; CHECK-NEXT:    ror w8, w8, #1
 ; CHECK-NEXT:    movk w9, #10922, lsl #16
@@ -173,11 +172,10 @@ define i1 @t32_6_4(i32 %X) nounwind {
 define i1 @t32_6_5(i32 %X) nounwind {
 ; CHECK-LABEL: t32_6_5:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #43691
-; CHECK-NEXT:    mov w9, #43689
-; CHECK-NEXT:    movk w8, #43690, lsl #16
+; CHECK-NEXT:    mov w9, #43691
+; CHECK-NEXT:    sub w8, w0, #5
 ; CHECK-NEXT:    movk w9, #43690, lsl #16
-; CHECK-NEXT:    madd w8, w0, w8, w9
+; CHECK-NEXT:    mul w8, w8, w9
 ; CHECK-NEXT:    mov w9, #43690
 ; CHECK-NEXT:    ror w8, w8, #1
 ; CHECK-NEXT:    movk w9, #10922, lsl #16