[InstCombine] Allow KnownBits to be propagated

Bug #63699 shows a hang on arm in instcombine because we do not
propagate known bits for fshl/fshr rotates. We perform the propagation
and add regression test.

Differential Revision: https://reviews.llvm.org/D155307

llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp

@@ -918,16 +918,18 @@ Value *InstCombinerImpl::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
               SimplifyDemandedBits(I, 1, DemandedMaskRHS, RHSKnown, Depth + 1))
             return I;
         } else { // fshl is a rotate
-        // Avoid converting rotate into funnel shift. 
-        // Only simplify if one operand is constant.
-          KnownBits LHSKnown = computeKnownBits(I->getOperand(0), Depth + 1, I);
-          if (DemandedMaskLHS.isSubsetOf(LHSKnown.Zero | LHSKnown.One)) {
+          // Avoid converting rotate into funnel shift. 
+          // Only simplify if one operand is constant.
+          LHSKnown = computeKnownBits(I->getOperand(0), Depth + 1, I);
+          if (DemandedMaskLHS.isSubsetOf(LHSKnown.Zero | LHSKnown.One) &&
+              !match(I->getOperand(0), m_SpecificInt(LHSKnown.One))) {
             replaceOperand(*I, 0, Constant::getIntegerValue(VTy, LHSKnown.One));
             return I;
           }
 
-          KnownBits RHSKnown = computeKnownBits(I->getOperand(1), Depth + 1, I);
-          if (DemandedMaskRHS.isSubsetOf(RHSKnown.Zero | RHSKnown.One)) {
+          RHSKnown = computeKnownBits(I->getOperand(1), Depth + 1, I);
+          if (DemandedMaskRHS.isSubsetOf(RHSKnown.Zero | RHSKnown.One) &&
+              !match(I->getOperand(1), m_SpecificInt(RHSKnown.One))) {
             replaceOperand(*I, 1, Constant::getIntegerValue(VTy, RHSKnown.One));
             return I;
           }

llvm/test/Transforms/InstCombine/2023-07-13-arm-infiniteloop.ll

@@ -0,0 +1,41 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+; RUN: opt -S -passes=instcombine < %s | FileCheck %s
+
+declare i1 @llvm.is.constant.i32(i32)
+
+define void @test(ptr %bpf_prog_calc_tag___trans_tmp_3, i32 %0) {
+; CHECK-LABEL: define void @test
+; CHECK-SAME: (ptr [[BPF_PROG_CALC_TAG___TRANS_TMP_3:%.*]], i32 [[TMP0:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br i1 true, label [[IF_ELSE_I:%.*]], label [[IF_THEN_I:%.*]]
+; CHECK:       if.then.i:
+; CHECK-NEXT:    br label [[__FSWAB64_EXIT:%.*]]
+; CHECK:       if.else.i:
+; CHECK-NEXT:    br label [[__FSWAB64_EXIT]]
+; CHECK:       __fswab64.exit:
+; CHECK-NEXT:    store i32 0, ptr [[BPF_PROG_CALC_TAG___TRANS_TMP_3]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %conv = zext i32 %0 to i64
+  %1 = lshr i64 %conv, 32
+  %conv1.i = trunc i64 %1 to i32
+  %2 = call i1 @llvm.is.constant.i32(i32 %conv1.i)
+  br i1 %2, label %if.else.i, label %if.then.i
+
+if.then.i:                                        ; preds = %entry
+  %3 = load volatile i32, ptr null, align 2147483648
+  br label %__fswab64.exit
+
+if.else.i:                                        ; preds = %entry
+  %or.i = call i32 @llvm.fshl.i32(i32 %conv1.i, i32 0, i32 16)
+  br label %__fswab64.exit
+
+__fswab64.exit:                                   ; preds = %if.then.i, %if.else.i
+  %t.0.i = phi i32 [ %or.i, %if.else.i ], [ %3, %if.then.i ]
+  %shr2.i = lshr i32 %t.0.i, 1
+  store i32 %shr2.i, ptr %bpf_prog_calc_tag___trans_tmp_3, align 4
+  ret void
+}
+
+declare i32 @llvm.fshl.i32(i32, i32, i32)