[Attributor][Not for commit] Make call site specific deduction default.

[kuter]

Summary

Interesting, not as many tests as I expected where effected.
I will look for tests that should have been effected but are not.

I have tried increasing the default depth argument

[kuter]

Some Attributes that have CBContext doesn't seem to be querying the CB specific attributes.
I will update this once I have a solution.

llvm/test/Transforms/Attributor/nocapture-2.ll

319

Is this normal ?

[kuter]

add auto context injection.

[jdoerfert]

I looked at some and we need to check what is happening.

llvm/test/Transforms/Attributor/ArgumentPromotion/control-flow.ll

36 ↗

(On Diff #281521)

Something is off.

llvm/test/Transforms/Attributor/IPConstantProp/solve-after-each-resolving-undefs-for-function.ll

65 ↗

(On Diff #281521)

I don't see this.

llvm/test/Transforms/Attributor/memory_locations.ll

614 ↗

(On Diff #281521)

This should be 1, shouldn't it?

llvm/test/Transforms/Attributor/nocapture-2.ll

319

It can make sense, haven't verified it though.

llvm/lib/Transforms/IPO/Attributor.cpp

static cl::list<std::string>
     SeedAllowList("attributor-seed-allow-list", cl::Hidden,
                   cl::desc("Comma seperated list of attrbute names that are "
                            "allowed to be seeded."),
                   cl::ZeroOrMore, cl::CommaSeparated);
 
 static cl::opt<bool> EnableCallSiteSpecific(
     "attributor-enable-call-site-specific-deduction", cl::Hidden,
     cl::desc("Allow the Attributor to do call site specific analysis"),
-    cl::init(false));
+    cl::init(true));
 
 static cl::opt<unsigned>
     CallSiteSpecificDepth("attributor-call-site-specifc-deduction-depth",
                           cl::Hidden,
                           cl::desc("Maximum depth of values that call site "
                                    "specific analysis context will reach."),
                           cl::init(6));
 
void Attributor::runTillFixpoint() {
   }
 
   LLVM_DEBUG({
     if (!Visited.empty())
       dbgs() << "\n[Attributor] Finalized " << Visited.size()
              << " abstract attributes.\n";
   });
 
-  if (VerifyMaxFixpointIterations &&
-      IterationCounter != MaxFixpointIterations) {
+  // FIXME
+  if (false && IterationCounter != MaxFixpointIterations) {
     errs() << "\n[Attributor] Fixpoint iteration done after: "
            << IterationCounter << "/" << MaxFixpointIterations
            << " iterations\n";
     llvm_unreachable("The fixpoint was not reached with exactly the number of "
                      "specified iterations!");
   }
 }
 

llvm/lib/Transforms/IPO/AttributorAttributes.cpp

ChangeStatus updateImpl(Attributor &A) override {
     // dead, we can take the known information from the state T.
     return clampStateAndIndicateChange<StateType>(this->getState(), S);
   }
 };
 
 /// Helper class for generic replication: function returned -> cs returned.
 template <typename AAType, typename BaseType,
           typename StateType = typename BaseType::StateType,
-          bool IntroduceCallBaseContext = false>
+          bool IntroduceCallBaseContext = true>
 struct AACallSiteReturnedFromReturned : public BaseType {
   AACallSiteReturnedFromReturned(const IRPosition &IRP, Attributor &A)
       : BaseType(IRP, A) {}
 
   /// See AbstractAttribute::updateImpl(...).
   ChangeStatus updateImpl(Attributor &A) override {
     assert(this->getIRPosition().getPositionKind() ==
                IRPosition::IRP_CALL_SITE_RETURNED &&

llvm/test/Transforms/Attributor/ArgumentPromotion/byval.ll

 ; IS__TUNIT_NPM: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@main()
 ; IS__TUNIT_NPM-NEXT:  entry:
 ; IS__TUNIT_NPM-NEXT:    [[S:%.*]] = alloca [[STRUCT_SS:%.*]], align 4
 ; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 0
 ; IS__TUNIT_NPM-NEXT:    store i32 1, i32* [[TMP1]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[TMP4:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
 ; IS__TUNIT_NPM-NEXT:    store i64 2, i64* [[TMP4]], align 4
-; IS__TUNIT_NPM-NEXT:    [[S_CAST:%.*]] = bitcast %struct.ss* [[S]] to i32*
-; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i32, i32* [[S_CAST]], align 8
-; IS__TUNIT_NPM-NEXT:    [[S_0_1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i64, i64* [[S_0_1]], align 8
-; IS__TUNIT_NPM-NEXT:    [[C0:%.*]] = call i32 @f(i32 [[TMP0]], i64 [[TMP1]])
 ; IS__TUNIT_NPM-NEXT:    [[S_CAST1:%.*]] = bitcast %struct.ss* [[S]] to i32*
-; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i32, i32* [[S_CAST1]], align 32
+; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i32, i32* [[S_CAST1]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[S_0_12:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i64, i64* [[S_0_12]], align 32
+; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i64, i64* [[S_0_12]], align 8
+; IS__TUNIT_NPM-NEXT:    [[C0:%.*]] = call i32 @f(i32 [[TMP0]], i64 [[TMP1]])
+; IS__TUNIT_NPM-NEXT:    [[S_CAST:%.*]] = bitcast %struct.ss* [[S]] to i32*
+; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i32, i32* [[S_CAST]], align 32
+; IS__TUNIT_NPM-NEXT:    [[S_0_1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
+; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i64, i64* [[S_0_1]], align 32
 ; IS__TUNIT_NPM-NEXT:    [[C1:%.*]] = call i32 @g(i32 [[TMP2]], i64 [[TMP3]])
 ; IS__TUNIT_NPM-NEXT:    [[A:%.*]] = add i32 [[C0]], [[C1]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[A]]
 ;
 ; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@main()
 ; IS__CGSCC_OPM-NEXT:  entry:
 ; IS__CGSCC_OPM-NEXT:    [[S:%.*]] = alloca [[STRUCT_SS:%.*]], align 4

llvm/test/Transforms/Attributor/IPConstantProp/PR16052.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --scrub-attributes --check-attributes
 ; RUN: opt -attributor -attributor-manifest-internal  -attributor-max-iterations-verify -attributor-annotate-decl-cs -attributor-max-iterations=2 -S < %s | FileCheck %s --check-prefixes=CHECK,NOT_CGSCC_NPM,NOT_CGSCC_OPM,NOT_TUNIT_NPM,IS__TUNIT____,IS________OPM,IS__TUNIT_OPM
 ; RUN: opt -aa-pipeline=basic-aa -passes=attributor -attributor-manifest-internal  -attributor-max-iterations-verify -attributor-annotate-decl-cs -attributor-max-iterations=2 -S < %s | FileCheck %s --check-prefixes=CHECK,NOT_CGSCC_OPM,NOT_CGSCC_NPM,NOT_TUNIT_OPM,IS__TUNIT____,IS________NPM,IS__TUNIT_NPM
 ; RUN: opt -attributor-cgscc -attributor-manifest-internal  -attributor-annotate-decl-cs -S < %s | FileCheck %s --check-prefixes=CHECK,NOT_TUNIT_NPM,NOT_TUNIT_OPM,NOT_CGSCC_NPM,IS__CGSCC____,IS________OPM,IS__CGSCC_OPM
 ; RUN: opt -aa-pipeline=basic-aa -passes=attributor-cgscc -attributor-manifest-internal  -attributor-annotate-decl-cs -S < %s | FileCheck %s --check-prefixes=CHECK,NOT_TUNIT_NPM,NOT_TUNIT_OPM,NOT_CGSCC_OPM,IS__CGSCC____,IS________NPM,IS__CGSCC_NPM
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 define i64 @fn2() {
 ; IS__TUNIT____: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT____-LABEL: define {{[^@]+}}@fn2()
 ; IS__TUNIT____-NEXT:  entry:
-; IS__TUNIT____-NEXT:    [[CALL2:%.*]] = call i64 @fn1(i64 undef) #0, !range !0
+; IS__TUNIT____-NEXT:    [[CALL2:%.*]] = call i64 @fn1(i64 undef)
 ; IS__TUNIT____-NEXT:    ret i64 [[CALL2]]
 ;
 ; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@fn2()
 ; IS__CGSCC____-NEXT:  entry:
 ; IS__CGSCC____-NEXT:    [[CALL2:%.*]] = call i64 @fn1(i64 undef)
 ; IS__CGSCC____-NEXT:    ret i64 [[CALL2]]
 ;
entry:
   %call2 = call i64 @fn1(i64 %div)
   ret i64 %call2
 }
 
 define i64 @fn2c() {
 ; IS__TUNIT____: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT____-LABEL: define {{[^@]+}}@fn2c()
 ; IS__TUNIT____-NEXT:  entry:
-; IS__TUNIT____-NEXT:    [[CALL2:%.*]] = call i64 @fn1(i64 42) #0, !range !0
+; IS__TUNIT____-NEXT:    [[CALL2:%.*]] = call i64 @fn1(i64 42)
 ; IS__TUNIT____-NEXT:    ret i64 [[CALL2]]
 ;
 ; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@fn2c()
 ; IS__CGSCC____-NEXT:  entry:
 ; IS__CGSCC____-NEXT:    ret i64 42
 ;
 entry:

llvm/test/Transforms/Attributor/nocapture-2.ll

 ; CHECK-NEXT:    [[TOBOOL:%.*]] = icmp ne i8* [[BC2]], null
 ; CHECK-NEXT:    br i1 [[TOBOOL]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 ; CHECK:       cond.true:
 ; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i16* [[A]] to double*
 ; CHECK-NEXT:    [[CALL1:%.*]] = call dereferenceable_or_null(8) i64* @scc_B(double* noalias nofree nonnull readnone dereferenceable(8) "no-capture-maybe-returned" [[TMP0]])
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[CALL1]] to i8*
 ; CHECK-NEXT:    br label [[COND_END:%.*]]
 ; CHECK:       cond.false:
-; CHECK-NEXT:    [[CALL2:%.*]] = call dereferenceable_or_null(4) i8* @scc_C(i16* noalias nofree nonnull readnone dereferenceable(4) "no-capture-maybe-returned" [[A]])
+; CHECK-NEXT:    [[CALL2:%.*]] = call nonnull dereferenceable(4294967295) i8* @scc_C(i16* noalias nofree nonnull readnone dereferenceable(4) "no-capture-maybe-returned" [[A]])
 ; CHECK-NEXT:    br label [[COND_END]]
 ; CHECK:       cond.end:
 ; CHECK-NEXT:    [[COND:%.*]] = phi i8* [ [[TMP1]], [[COND_TRUE]] ], [ [[CALL2]], [[COND_FALSE]] ]
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[COND]] to i32*
-; CHECK-NEXT:    [[CALL3:%.*]] = call float* @scc_A(i32* noalias nofree nonnull readnone dereferenceable(4) "no-capture-maybe-returned" [[TMP2]])
+; CHECK-NEXT:    [[CALL3:%.*]] = call float* @scc_A(i32* noalias nofree nonnull readnone dereferenceable(8) "no-capture-maybe-returned" [[TMP2]])
 ; CHECK-NEXT:    [[TMP3:%.*]] = bitcast float* [[CALL3]] to i8*
 ; CHECK-NEXT:    ret i8* [[TMP3]]
 ;
 entry:
   %bc = bitcast i16* %a to i32*
   %call = call float* @scc_A(i32* %bc)
   %bc2 = bitcast float* %call to i8*
   %tobool = icmp ne i8* %bc2, null

llvm/test/Transforms/Attributor/returned.ll

 ;
 entry:
   %call = call i32 @sink_r0(i32 %r)
   %call1 = call i32 @scc_r2(i32 %r, i32 %a, i32 %call)
   ret i32 %call1
 }
 
 define i32 @scc_r2(i32 %a, i32 %b, i32 %r) #0 {
-; CHECK: Function Attrs: nofree noinline nosync nounwind readnone uwtable
-; CHECK-LABEL: define {{[^@]+}}@scc_r2
-; CHECK-SAME: (i32 [[A:%.*]], i32 [[B:%.*]], i32 returned [[R:%.*]])
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[A]], [[B]]
-; CHECK-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
-; CHECK:       if.then:
-; CHECK-NEXT:    [[CALL:%.*]] = call i32 @sink_r0(i32 [[R]])
-; CHECK-NEXT:    [[CALL1:%.*]] = call i32 @scc_r2(i32 [[B]], i32 [[A]], i32 [[CALL]])
-; CHECK-NEXT:    br label [[RETURN:%.*]]
-; CHECK:       if.end:
-; CHECK-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[A]], [[B]]
-; CHECK-NEXT:    br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END12:%.*]]
-; CHECK:       if.then3:
-; CHECK-NEXT:    [[CALL4:%.*]] = call i32 @sink_r0(i32 [[B]])
-; CHECK-NEXT:    [[CALL5:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef)
-; CHECK-NEXT:    [[CALL6:%.*]] = call i32 @scc_r2(i32 [[R]], i32 [[R]], i32 [[R]])
-; CHECK-NEXT:    [[CALL7:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[CALL6]], i32 undef)
-; CHECK-NEXT:    [[CALL8:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
-; CHECK-NEXT:    [[CALL9:%.*]] = call i32 @scc_r2(i32 [[CALL5]], i32 [[CALL7]], i32 [[CALL8]])
-; CHECK-NEXT:    [[CALL11:%.*]] = call i32 @scc_r1(i32 [[CALL4]], i32 [[CALL9]], i32 undef)
-; CHECK-NEXT:    br label [[RETURN]]
-; CHECK:       if.end12:
-; CHECK-NEXT:    [[CMP13:%.*]] = icmp eq i32 [[A]], [[B]]
-; CHECK-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
-; CHECK:       cond.true:
-; CHECK-NEXT:    br label [[COND_END:%.*]]
-; CHECK:       cond.false:
-; CHECK-NEXT:    [[CALL14:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
-; CHECK-NEXT:    br label [[COND_END]]
-; CHECK:       cond.end:
-; CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[R]], [[COND_TRUE]] ], [ [[CALL14]], [[COND_FALSE]] ]
-; CHECK-NEXT:    br label [[RETURN]]
-; CHECK:       return:
-; CHECK-NEXT:    [[RETVAL_0:%.*]] = phi i32 [ [[CALL1]], [[IF_THEN]] ], [ [[CALL11]], [[IF_THEN3]] ], [ [[COND]], [[COND_END]] ]
-; CHECK-NEXT:    ret i32 [[RETVAL_0]]
+; IS________OPM: Function Attrs: nofree noinline nosync nounwind readnone uwtable
+; IS________OPM-LABEL: define {{[^@]+}}@scc_r2
+; IS________OPM-SAME: (i32 [[A:%.*]], i32 [[B:%.*]], i32 returned [[R:%.*]])
+; IS________OPM-NEXT:  entry:
+; IS________OPM-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[A]], [[B]]
+; IS________OPM-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; IS________OPM:       if.then:
+; IS________OPM-NEXT:    [[CALL:%.*]] = call i32 @sink_r0(i32 [[R]])
+; IS________OPM-NEXT:    [[CALL1:%.*]] = call i32 @scc_r2(i32 [[B]], i32 [[A]], i32 [[CALL]])
+; IS________OPM-NEXT:    br label [[RETURN:%.*]]
+; IS________OPM:       if.end:
+; IS________OPM-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[A]], [[B]]
+; IS________OPM-NEXT:    br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END12:%.*]]
+; IS________OPM:       if.then3:
+; IS________OPM-NEXT:    [[CALL4:%.*]] = call i32 @sink_r0(i32 [[B]])
+; IS________OPM-NEXT:    [[CALL5:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef)
+; IS________OPM-NEXT:    [[CALL6:%.*]] = call i32 @scc_r2(i32 [[R]], i32 [[R]], i32 [[R]])
+; IS________OPM-NEXT:    [[CALL7:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[CALL6]], i32 undef)
+; IS________OPM-NEXT:    [[CALL8:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
+; IS________OPM-NEXT:    [[CALL9:%.*]] = call i32 @scc_r2(i32 [[CALL5]], i32 [[CALL7]], i32 [[CALL8]])
+; IS________OPM-NEXT:    [[CALL11:%.*]] = call i32 @scc_r1(i32 [[CALL4]], i32 [[CALL9]], i32 undef)
+; IS________OPM-NEXT:    br label [[RETURN]]
+; IS________OPM:       if.end12:
+; IS________OPM-NEXT:    [[CMP13:%.*]] = icmp eq i32 [[A]], [[B]]
+; IS________OPM-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+; IS________OPM:       cond.true:
+; IS________OPM-NEXT:    br label [[COND_END:%.*]]
+; IS________OPM:       cond.false:
+; IS________OPM-NEXT:    [[CALL14:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
+; IS________OPM-NEXT:    br label [[COND_END]]
+; IS________OPM:       cond.end:
+; IS________OPM-NEXT:    [[COND:%.*]] = phi i32 [ [[R]], [[COND_TRUE]] ], [ [[CALL14]], [[COND_FALSE]] ]
+; IS________OPM-NEXT:    br label [[RETURN]]
+; IS________OPM:       return:
+; IS________OPM-NEXT:    [[RETVAL_0:%.*]] = phi i32 [ [[CALL1]], [[IF_THEN]] ], [ [[CALL11]], [[IF_THEN3]] ], [ [[COND]], [[COND_END]] ]
+; IS________OPM-NEXT:    ret i32 [[RETVAL_0]]
+;
+; IS__TUNIT_NPM: Function Attrs: nofree noinline nosync nounwind readnone uwtable
+; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@scc_r2
+; IS__TUNIT_NPM-SAME: (i32 [[A:%.*]], i32 [[B:%.*]], i32 returned [[R:%.*]])
+; IS__TUNIT_NPM-NEXT:  entry:
+; IS__TUNIT_NPM-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[A]], [[B]]
+; IS__TUNIT_NPM-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; IS__TUNIT_NPM:       if.then:
+; IS__TUNIT_NPM-NEXT:    [[CALL:%.*]] = call i32 @sink_r0(i32 [[R]])
+; IS__TUNIT_NPM-NEXT:    [[CALL1:%.*]] = call i32 @scc_r2(i32 [[B]], i32 [[A]], i32 [[CALL]])
+; IS__TUNIT_NPM-NEXT:    br label [[RETURN:%.*]]
+; IS__TUNIT_NPM:       if.end:
+; IS__TUNIT_NPM-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[A]], [[B]]
+; IS__TUNIT_NPM-NEXT:    br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END12:%.*]]
+; IS__TUNIT_NPM:       if.then3:
+; IS__TUNIT_NPM-NEXT:    [[CALL4:%.*]] = call i32 @sink_r0(i32 [[B]]) #6, !range !0
+; IS__TUNIT_NPM-NEXT:    [[CALL5:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef) #6, !range !0
+; IS__TUNIT_NPM-NEXT:    [[CALL6:%.*]] = call i32 @scc_r2(i32 [[R]], i32 [[R]], i32 [[R]])
+; IS__TUNIT_NPM-NEXT:    [[CALL7:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[CALL6]], i32 undef)
+; IS__TUNIT_NPM-NEXT:    [[CALL8:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
+; IS__TUNIT_NPM-NEXT:    [[CALL9:%.*]] = call i32 @scc_r2(i32 [[CALL5]], i32 [[CALL7]], i32 [[CALL8]])
+; IS__TUNIT_NPM-NEXT:    [[CALL11:%.*]] = call i32 @scc_r1(i32 [[CALL4]], i32 [[CALL9]], i32 undef)
+; IS__TUNIT_NPM-NEXT:    br label [[RETURN]]
+; IS__TUNIT_NPM:       if.end12:
+; IS__TUNIT_NPM-NEXT:    [[CMP13:%.*]] = icmp eq i32 [[A]], [[B]]
+; IS__TUNIT_NPM-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+; IS__TUNIT_NPM:       cond.true:
+; IS__TUNIT_NPM-NEXT:    br label [[COND_END:%.*]]
+; IS__TUNIT_NPM:       cond.false:
+; IS__TUNIT_NPM-NEXT:    [[CALL14:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
+; IS__TUNIT_NPM-NEXT:    br label [[COND_END]]
+; IS__TUNIT_NPM:       cond.end:
+; IS__TUNIT_NPM-NEXT:    [[COND:%.*]] = phi i32 [ [[R]], [[COND_TRUE]] ], [ [[CALL14]], [[COND_FALSE]] ]
+; IS__TUNIT_NPM-NEXT:    br label [[RETURN]]
+; IS__TUNIT_NPM:       return:
+; IS__TUNIT_NPM-NEXT:    [[RETVAL_0:%.*]] = phi i32 [ [[CALL1]], [[IF_THEN]] ], [ [[CALL11]], [[IF_THEN3]] ], [ [[COND]], [[COND_END]] ]
+; IS__TUNIT_NPM-NEXT:    ret i32 [[RETVAL_0]]
+;
+; IS__CGSCC_NPM: Function Attrs: nofree noinline nosync nounwind readnone uwtable
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@scc_r2
+; IS__CGSCC_NPM-SAME: (i32 [[A:%.*]], i32 [[B:%.*]], i32 returned [[R:%.*]])
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[A]], [[B]]
+; IS__CGSCC_NPM-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; IS__CGSCC_NPM:       if.then:
+; IS__CGSCC_NPM-NEXT:    [[CALL:%.*]] = call i32 @sink_r0(i32 [[R]])
+; IS__CGSCC_NPM-NEXT:    [[CALL1:%.*]] = call i32 @scc_r2(i32 [[B]], i32 [[A]], i32 [[CALL]])
+; IS__CGSCC_NPM-NEXT:    br label [[RETURN:%.*]]
+; IS__CGSCC_NPM:       if.end:
+; IS__CGSCC_NPM-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[A]], [[B]]
+; IS__CGSCC_NPM-NEXT:    br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END12:%.*]]
+; IS__CGSCC_NPM:       if.then3:
+; IS__CGSCC_NPM-NEXT:    [[CALL4:%.*]] = call i32 @sink_r0(i32 [[B]])
+; IS__CGSCC_NPM-NEXT:    [[CALL5:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef) #7, !range !0
+; IS__CGSCC_NPM-NEXT:    [[CALL6:%.*]] = call i32 @scc_r2(i32 [[R]], i32 [[R]], i32 [[R]])
+; IS__CGSCC_NPM-NEXT:    [[CALL7:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[CALL6]], i32 undef)
+; IS__CGSCC_NPM-NEXT:    [[CALL8:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
+; IS__CGSCC_NPM-NEXT:    [[CALL9:%.*]] = call i32 @scc_r2(i32 [[CALL5]], i32 [[CALL7]], i32 [[CALL8]])
+; IS__CGSCC_NPM-NEXT:    [[CALL11:%.*]] = call i32 @scc_r1(i32 [[CALL4]], i32 [[CALL9]], i32 undef)
+; IS__CGSCC_NPM-NEXT:    br label [[RETURN]]
+; IS__CGSCC_NPM:       if.end12:
+; IS__CGSCC_NPM-NEXT:    [[CMP13:%.*]] = icmp eq i32 [[A]], [[B]]
+; IS__CGSCC_NPM-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+; IS__CGSCC_NPM:       cond.true:
+; IS__CGSCC_NPM-NEXT:    br label [[COND_END:%.*]]
+; IS__CGSCC_NPM:       cond.false:
+; IS__CGSCC_NPM-NEXT:    [[CALL14:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
+; IS__CGSCC_NPM-NEXT:    br label [[COND_END]]
+; IS__CGSCC_NPM:       cond.end:
+; IS__CGSCC_NPM-NEXT:    [[COND:%.*]] = phi i32 [ [[R]], [[COND_TRUE]] ], [ [[CALL14]], [[COND_FALSE]] ]
+; IS__CGSCC_NPM-NEXT:    br label [[RETURN]]
+; IS__CGSCC_NPM:       return:
+; IS__CGSCC_NPM-NEXT:    [[RETVAL_0:%.*]] = phi i32 [ [[CALL1]], [[IF_THEN]] ], [ [[CALL11]], [[IF_THEN3]] ], [ [[COND]], [[COND_END]] ]
+; IS__CGSCC_NPM-NEXT:    ret i32 [[RETVAL_0]]
 ;
 entry:
   %cmp = icmp sgt i32 %a, %b
   br i1 %cmp, label %if.then, label %if.end
 
 if.then:                                          ; preds = %entry
   %call = call i32 @sink_r0(i32 %r)
   %call1 = call i32 @scc_r2(i32 %b, i32 %a, i32 %call)
cond.end:                                         ; preds = %cond.false, %cond.true
   br label %return
 
 return:                                           ; preds = %cond.end, %if.then3, %if.then
   %retval.0 = phi i32 [ %call1, %if.then ], [ %call11, %if.then3 ], [ %cond, %cond.end ]
   ret i32 %retval.0
 }
 
 define i32 @scc_rX(i32 %a, i32 %b, i32 %r) #0 {
-; IS__TUNIT____: Function Attrs: nofree noinline nosync nounwind readnone uwtable
-; IS__TUNIT____-LABEL: define {{[^@]+}}@scc_rX
-; IS__TUNIT____-SAME: (i32 [[A:%.*]], i32 [[B:%.*]], i32 [[R:%.*]])
-; IS__TUNIT____-NEXT:  entry:
-; IS__TUNIT____-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[A]], [[B]]
-; IS__TUNIT____-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
-; IS__TUNIT____:       if.then:
-; IS__TUNIT____-NEXT:    [[CALL:%.*]] = call i32 @sink_r0(i32 [[R]])
-; IS__TUNIT____-NEXT:    [[CALL1:%.*]] = call i32 @scc_r2(i32 [[B]], i32 [[A]], i32 [[CALL]])
-; IS__TUNIT____-NEXT:    br label [[RETURN:%.*]]
-; IS__TUNIT____:       if.end:
-; IS__TUNIT____-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[A]], [[B]]
-; IS__TUNIT____-NEXT:    br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END12:%.*]]
-; IS__TUNIT____:       if.then3:
-; IS__TUNIT____-NEXT:    [[CALL4:%.*]] = call i32 @sink_r0(i32 [[B]])
-; IS__TUNIT____-NEXT:    [[CALL5:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef)
-; IS__TUNIT____-NEXT:    [[CALL6:%.*]] = call i32 @scc_r2(i32 [[R]], i32 [[R]], i32 [[R]])
-; IS__TUNIT____-NEXT:    [[CALL7:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[CALL6]], i32 undef)
-; IS__TUNIT____-NEXT:    [[CALL8:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef)
-; IS__TUNIT____-NEXT:    [[CALL9:%.*]] = call i32 @scc_r2(i32 [[CALL5]], i32 [[CALL7]], i32 [[CALL8]])
-; IS__TUNIT____-NEXT:    [[CALL11:%.*]] = call i32 @scc_r1(i32 [[CALL4]], i32 [[CALL9]], i32 undef)
-; IS__TUNIT____-NEXT:    br label [[RETURN]]
-; IS__TUNIT____:       if.end12:
-; IS__TUNIT____-NEXT:    [[CMP13:%.*]] = icmp eq i32 [[A]], [[B]]
-; IS__TUNIT____-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
-; IS__TUNIT____:       cond.true:
-; IS__TUNIT____-NEXT:    br label [[COND_END:%.*]]
-; IS__TUNIT____:       cond.false:
-; IS__TUNIT____-NEXT:    [[CALL14:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
-; IS__TUNIT____-NEXT:    br label [[COND_END]]
-; IS__TUNIT____:       cond.end:
-; IS__TUNIT____-NEXT:    [[COND:%.*]] = phi i32 [ [[R]], [[COND_TRUE]] ], [ [[CALL14]], [[COND_FALSE]] ]
-; IS__TUNIT____-NEXT:    br label [[RETURN]]
-; IS__TUNIT____:       return:
-; IS__TUNIT____-NEXT:    [[RETVAL_0:%.*]] = phi i32 [ [[CALL1]], [[IF_THEN]] ], [ [[CALL11]], [[IF_THEN3]] ], [ [[COND]], [[COND_END]] ]
-; IS__TUNIT____-NEXT:    ret i32 [[RETVAL_0]]
+; IS__TUNIT_OPM: Function Attrs: nofree noinline nosync nounwind readnone uwtable
+; IS__TUNIT_OPM-LABEL: define {{[^@]+}}@scc_rX
+; IS__TUNIT_OPM-SAME: (i32 [[A:%.*]], i32 [[B:%.*]], i32 [[R:%.*]])
+; IS__TUNIT_OPM-NEXT:  entry:
+; IS__TUNIT_OPM-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[A]], [[B]]
+; IS__TUNIT_OPM-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; IS__TUNIT_OPM:       if.then:
+; IS__TUNIT_OPM-NEXT:    [[CALL:%.*]] = call i32 @sink_r0(i32 [[R]])
+; IS__TUNIT_OPM-NEXT:    [[CALL1:%.*]] = call i32 @scc_r2(i32 [[B]], i32 [[A]], i32 [[CALL]])
+; IS__TUNIT_OPM-NEXT:    br label [[RETURN:%.*]]
+; IS__TUNIT_OPM:       if.end:
+; IS__TUNIT_OPM-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[A]], [[B]]
+; IS__TUNIT_OPM-NEXT:    br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END12:%.*]]
+; IS__TUNIT_OPM:       if.then3:
+; IS__TUNIT_OPM-NEXT:    [[CALL4:%.*]] = call i32 @sink_r0(i32 [[B]])
+; IS__TUNIT_OPM-NEXT:    [[CALL5:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef)
+; IS__TUNIT_OPM-NEXT:    [[CALL6:%.*]] = call i32 @scc_r2(i32 [[R]], i32 [[R]], i32 [[R]])
+; IS__TUNIT_OPM-NEXT:    [[CALL7:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[CALL6]], i32 undef)
+; IS__TUNIT_OPM-NEXT:    [[CALL8:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef)
+; IS__TUNIT_OPM-NEXT:    [[CALL9:%.*]] = call i32 @scc_r2(i32 [[CALL5]], i32 [[CALL7]], i32 [[CALL8]])
+; IS__TUNIT_OPM-NEXT:    [[CALL11:%.*]] = call i32 @scc_r1(i32 [[CALL4]], i32 [[CALL9]], i32 undef)
+; IS__TUNIT_OPM-NEXT:    br label [[RETURN]]
+; IS__TUNIT_OPM:       if.end12:
+; IS__TUNIT_OPM-NEXT:    [[CMP13:%.*]] = icmp eq i32 [[A]], [[B]]
+; IS__TUNIT_OPM-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+; IS__TUNIT_OPM:       cond.true:
+; IS__TUNIT_OPM-NEXT:    br label [[COND_END:%.*]]
+; IS__TUNIT_OPM:       cond.false:
+; IS__TUNIT_OPM-NEXT:    [[CALL14:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
+; IS__TUNIT_OPM-NEXT:    br label [[COND_END]]
+; IS__TUNIT_OPM:       cond.end:
+; IS__TUNIT_OPM-NEXT:    [[COND:%.*]] = phi i32 [ [[R]], [[COND_TRUE]] ], [ [[CALL14]], [[COND_FALSE]] ]
+; IS__TUNIT_OPM-NEXT:    br label [[RETURN]]
+; IS__TUNIT_OPM:       return:
+; IS__TUNIT_OPM-NEXT:    [[RETVAL_0:%.*]] = phi i32 [ [[CALL1]], [[IF_THEN]] ], [ [[CALL11]], [[IF_THEN3]] ], [ [[COND]], [[COND_END]] ]
+; IS__TUNIT_OPM-NEXT:    ret i32 [[RETVAL_0]]
+;
+; IS__TUNIT_NPM: Function Attrs: nofree noinline nosync nounwind readnone uwtable
+; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@scc_rX
+; IS__TUNIT_NPM-SAME: (i32 [[A:%.*]], i32 [[B:%.*]], i32 [[R:%.*]])
+; IS__TUNIT_NPM-NEXT:  entry:
+; IS__TUNIT_NPM-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[A]], [[B]]
+; IS__TUNIT_NPM-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; IS__TUNIT_NPM:       if.then:
+; IS__TUNIT_NPM-NEXT:    [[CALL:%.*]] = call i32 @sink_r0(i32 [[R]])
+; IS__TUNIT_NPM-NEXT:    [[CALL1:%.*]] = call i32 @scc_r2(i32 [[B]], i32 [[A]], i32 [[CALL]])
+; IS__TUNIT_NPM-NEXT:    br label [[RETURN:%.*]]
+; IS__TUNIT_NPM:       if.end:
+; IS__TUNIT_NPM-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[A]], [[B]]
+; IS__TUNIT_NPM-NEXT:    br i1 [[CMP2]], label [[IF_THEN3:%.*]], label [[IF_END12:%.*]]
+; IS__TUNIT_NPM:       if.then3:
+; IS__TUNIT_NPM-NEXT:    [[CALL4:%.*]] = call i32 @sink_r0(i32 [[B]]) #6, !range !0
+; IS__TUNIT_NPM-NEXT:    [[CALL5:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef) #6, !range !0
+; IS__TUNIT_NPM-NEXT:    [[CALL6:%.*]] = call i32 @scc_r2(i32 [[R]], i32 [[R]], i32 [[R]])
+; IS__TUNIT_NPM-NEXT:    [[CALL7:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[CALL6]], i32 undef)
+; IS__TUNIT_NPM-NEXT:    [[CALL8:%.*]] = call i32 @scc_r1(i32 [[A]], i32 [[B]], i32 undef) #6, !range !0
+; IS__TUNIT_NPM-NEXT:    [[CALL9:%.*]] = call i32 @scc_r2(i32 [[CALL5]], i32 [[CALL7]], i32 [[CALL8]])
+; IS__TUNIT_NPM-NEXT:    [[CALL11:%.*]] = call i32 @scc_r1(i32 [[CALL4]], i32 [[CALL9]], i32 undef)
+; IS__TUNIT_NPM-NEXT:    br label [[RETURN]]
+; IS__TUNIT_NPM:       if.end12:
+; IS__TUNIT_NPM-NEXT:    [[CMP13:%.*]] = icmp eq i32 [[A]], [[B]]
+; IS__TUNIT_NPM-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+; IS__TUNIT_NPM:       cond.true:
+; IS__TUNIT_NPM-NEXT:    br label [[COND_END:%.*]]
+; IS__TUNIT_NPM:       cond.false:
+; IS__TUNIT_NPM-NEXT:    [[CALL14:%.*]] = call i32 @scc_r2(i32 [[A]], i32 [[B]], i32 [[R]])
+; IS__TUNIT_NPM-NEXT:    br label [[COND_END]]
+; IS__TUNIT_NPM:       cond.end:
+; IS__TUNIT_NPM-NEXT:    [[COND:%.*]] = phi i32 [ [[R]], [[COND_TRUE]] ], [ [[CALL14]], [[COND_FALSE]] ]
+; IS__TUNIT_NPM-NEXT:    br label [[RETURN]]
+; IS__TUNIT_NPM:       return:
+; IS__TUNIT_NPM-NEXT:    [[RETVAL_0:%.*]] = phi i32 [ [[CALL1]], [[IF_THEN]] ], [ [[CALL11]], [[IF_THEN3]] ], [ [[COND]], [[COND_END]] ]
+; IS__TUNIT_NPM-NEXT:    ret i32 [[RETVAL_0]]
 ;
 ; IS__CGSCC____: Function Attrs: nofree noinline nosync nounwind readnone uwtable
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@scc_rX
 ; IS__CGSCC____-SAME: (i32 [[A:%.*]], i32 [[B:%.*]], i32 [[R:%.*]])
 ; IS__CGSCC____-NEXT:  entry:
 ; IS__CGSCC____-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[A]], [[B]]
 ; IS__CGSCC____-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
 ; IS__CGSCC____:       if.then: