[AArch64] Fix Fold of Compare with Right-shifted Value

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -25070,10 +25070,10 @@ static SDValue performSETCCCombine(SDNode *N,
   // setcc (srl x, imm), 0, ne ==> setcc (and x, (-1 << imm)), 0, ne
   if (Cond == ISD::SETNE && isNullConstant(RHS) &&
       LHS->getOpcode() == ISD::SRL && isa<ConstantSDNode>(LHS->getOperand(1)) &&
-      LHS->getConstantOperandVal(1) < VT.getScalarSizeInBits() &&
       LHS->hasOneUse()) {
     EVT TstVT = LHS->getValueType(0);
-    if (TstVT.isScalarInteger() && TstVT.getFixedSizeInBits() <= 64) {
+    if (TstVT.isScalarInteger() && TstVT.getFixedSizeInBits() <= 64 &&
+        LHS->getConstantOperandVal(1) < TstVT.getFixedSizeInBits()) {
       // this pattern will get better opt in emitComparison
       uint64_t TstImm = -1ULL << LHS->getConstantOperandVal(1);
       SDValue TST = DAG.getNode(ISD::AND, DL, TstVT, LHS->getOperand(0),

llvm/test/CodeGen/AArch64/shift-const-ne-0.ll

@@ -0,0 +1,122 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=aarch64-unknown-unknown < %s -o -| FileCheck %s
+
+define i1 @lsr_1_ne_0_16(i16 %x) {
+; CHECK-LABEL: lsr_1_ne_0_16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst w0, #0xfffe
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %shr = lshr i16 %x, 1
+  %cmp = icmp ne i16 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_1_ne_0_32(i32 %x) {
+; CHECK-LABEL: lsr_1_ne_0_32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst w0, #0xfffffffe
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %shr = lshr i32 %x, 1
+  %cmp = icmp ne i32 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_30_ne_0_32(i32 %x) {
+; CHECK-LABEL: lsr_30_ne_0_32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst w0, #0xc0000000
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %shr = lshr i32 %x, 30
+  %cmp = icmp ne i32 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_31_ne_0_32(i32 %x) {
+; CHECK-LABEL: lsr_31_ne_0_32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    lsr w0, w0, #31
+; CHECK-NEXT:    ret
+  %shr = lshr i32 %x, 31
+  %cmp = icmp ne i32 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_1_ne_0_64(i64 %x) {
+; CHECK-LABEL: lsr_1_ne_0_64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst x0, #0xfffffffffffffffe
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %shr = lshr i64 %x, 1
+  %cmp = icmp ne i64 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_31_ne_0_64(i64 %x) {
+; CHECK-LABEL: lsr_31_ne_0_64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst x0, #0xffffffff80000000
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %shr = lshr i64 %x, 31
+  %cmp = icmp ne i64 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_32_ne_0_64(i64 %x) {
+; CHECK-LABEL: lsr_32_ne_0_64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst x0, #0xffffffff00000000
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %shr = lshr i64 %x, 32
+  %cmp = icmp ne i64 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_33_ne_0_64(i64 %x) {
+; CHECK-LABEL: lsr_33_ne_0_64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst x0, #0xfffffffe00000000
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %shr = lshr i64 %x, 33
+  %cmp = icmp ne i64 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_62_ne_0_64(i64 %x) {
+; CHECK-LABEL: lsr_62_ne_0_64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst x0, #0xc000000000000000
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %shr = lshr i64 %x, 62
+  %cmp = icmp ne i64 %shr, 0
+  ret i1 %cmp
+}
+
+define i1 @lsr_63_ne_0_64(i64 %x) {
+; CHECK-LABEL: lsr_63_ne_0_64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    lsr x0, x0, #63
+; CHECK-NEXT:    // kill: def $w0 killed $w0 killed $x0
+; CHECK-NEXT:    ret
+  %shr = lshr i64 %x, 63
+  %cmp = icmp ne i64 %shr, 0
+  ret i1 %cmp
+}
+
+define <4 x i1> @lsr_1_ne_0_v4i16(<4 x i16> %x) {
+; CHECK-LABEL: lsr_1_ne_0_v4i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ushr v0.4h, v0.4h, #1
+; CHECK-NEXT:    cmtst v0.4h, v0.4h, v0.4h
+; CHECK-NEXT:    ret
+  %shr = lshr <4 x i16> %x, <i16 1, i16 1, i16 1, i16 1>
+  %cmp = icmp ne <4 x i16> %shr, <i16 0, i16 0, i16 0, i16 0>
+  ret <4 x i1> %cmp
+}