[X86] matchIndexRecursively - don't peek through multiuse sext(add_nsw(x,c)) (PR65895)

Fixes #65895

Author: RKSimon

llvm/lib/Target/X86/X86ISelDAGToDAG.cpp

@@ -2256,9 +2256,10 @@ SDValue X86DAGToDAGISel::matchIndexRecursively(SDValue N,
 
   // index: sext(add_nsw(x,c)) -> index: sext(x), disp + sext(c)
   // TODO: call matchIndexRecursively(AddSrc) if we won't corrupt sext?
-  if (Opc == ISD::SIGN_EXTEND && !VT.isVector()) {
+  if (Opc == ISD::SIGN_EXTEND && !VT.isVector() && N.hasOneUse()) {
     SDValue Src = N.getOperand(0);
-    if (Src.getOpcode() == ISD::ADD && Src->getFlags().hasNoSignedWrap()) {
+    if (Src.getOpcode() == ISD::ADD && Src->getFlags().hasNoSignedWrap() &&
+        Src.hasOneUse()) {
       if (CurDAG->isBaseWithConstantOffset(Src)) {
         SDValue AddSrc = Src.getOperand(0);
         auto *AddVal = cast<ConstantSDNode>(Src.getOperand(1));

llvm/test/CodeGen/X86/pr65895.ll

@@ -0,0 +1,60 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
+; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s
+
+@a = dso_local local_unnamed_addr global i16 0, align 2
+@c = internal unnamed_addr global i1 false, align 1
+@b = dso_local local_unnamed_addr global i8 0, align 1
+@d = dso_local local_unnamed_addr global i32 0, align 4
+@e = dso_local local_unnamed_addr global i64 0, align 8
+
+define i32 @PR65895() {
+; CHECK-LABEL: PR65895:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movb $2, %al
+; CHECK-NEXT:    subb c(%rip), %al
+; CHECK-NEXT:    cmpw $0, a(%rip)
+; CHECK-NEXT:    je .LBB0_3
+; CHECK-NEXT:  # %bb.1: # %for.body.lr.ph
+; CHECK-NEXT:    movb %al, b(%rip)
+; CHECK-NEXT:    .p2align 4, 0x90
+; CHECK-NEXT:  .LBB0_2: # %for.body
+; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:    jmp .LBB0_2
+; CHECK-NEXT:  .LBB0_3: # %for.end
+; CHECK-NEXT:    addb $-3, %al
+; CHECK-NEXT:    movsbl %al, %eax
+; CHECK-NEXT:    movl %eax, d(%rip)
+; CHECK-NEXT:    leal 247(%rax,%rax,2), %eax
+; CHECK-NEXT:    movb $1, c(%rip)
+; CHECK-NEXT:    movsbq %al, %rax
+; CHECK-NEXT:    movq %rax, e(%rip)
+; CHECK-NEXT:    xorl %eax, %eax
+; CHECK-NEXT:    retq
+entry:
+  %.pr = load i16, ptr @a, align 2
+  %tobool.not = icmp eq i16 %.pr, 0
+  %.b = load i1, ptr @c, align 1
+  %0 = select i1 %.b, i8 1, i8 2
+  br i1 %tobool.not, label %for.end, label %for.body.lr.ph
+
+for.body.lr.ph:
+  store i8 %0, ptr @b, align 1
+  br label %for.body
+
+for.body:
+  br label %for.body
+
+for.end:
+  %sub = add nsw i8 %0, -3
+  %conv2 = sext i8 %sub to i32
+  store i32 %conv2, ptr @d, align 4
+  %add = shl nsw i32 %conv2, 1
+  %sub7 = add nsw i32 %conv2, 247
+  %add8 = add nsw i32 %sub7, %add
+  %conv9 = zext i32 %add8 to i64
+  store i1 true, ptr @c, align 1
+  %sext = shl i64 %conv9, 56
+  %conv11 = ashr exact i64 %sext, 56
+  store i64 %conv11, ptr @e, align 8
+  ret i32 0
+}