[FPEnv] Use strictfp metadata in casting nodes

[kpn]

Summary

The strictfp metadata was added to the casting AST nodes in D85960, but we aren't using that metadata yet. This patch adds that support.

In order to avoid lots of ad-hoc passing around of the strictfp bits I updated the IRBuilder when moving from a function that has the Expr* to a function that lacks it. I believe we should switch to this pattern to keep the strictfp support from being overly invasive.

For the purpose of testing that we're picking up the right metadata, I also made my tests use a pragma to make the AST's strictfp metadata not match the global strictfp metadata. This exposes issues that we need to deal with in subsequent patches, and I believe this is the right method for most all of our clang strictfp tests.

[sepavloff]

clang/lib/CodeGen/CodeGenFunction.cpp

145

Actually this is not correct, because subclass relations are not preserved in this case. For instance, CompounsAssignmentOperator is not handled, as it does not return BinaryOperatorClass.

I would recommend using Expr::getFPFeaturesInEffect or adding similar method.

clang/lib/Sema/SemaExpr.cpp

700

Can cast of kind CK_NullToPointer or CK_LValueToRValue depend on FP options?

[rjmccall]

clang/lib/Sema/SemaExpr.cpp

700

The first never involves an FP type.

The second can be a load of an FP type, but that should never be options-sensitive. I think it actually *can* do a FP operation because of half->float promotion, but I don't believe there are any options that ever affect that, as it's always perfectly value-preserving.

[kpn]

clang/lib/CodeGen/CodeGenFunction.cpp

145

That's much cleaner. Sorry I missed that earlier.

[kpn]

Update/simplify for review comments.

[sepavloff]

clang/lib/CodeGen/CodeGenFunction.cpp

143–145

The local variable FPFeatures can be eliminated here.

153–154

This check can be lifted to just after the definition of OldFPFeatures.

clang/lib/CodeGen/CodeGenFunction.h

638–639

Is it possible to merge this variables into OldFPFeatures? In what cases state of Builder may differ from the state of CGF.CurFPFeatures?

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics-constrained.c

6

Why did you change exception behavior to maytrap?

26

Why they are wrong?

[kpn]

clang/lib/CodeGen/CodeGenFunction.h

638–639

Without this patch they will differ if the AST changes CGF.CurFPFeatures and the Builder hasn't been updated yet. Say, because we're going back up the tree.

The concern is that we'll miss places that need to be updated to respect the AST. But if we don't make a point of resetting the Builder then we'll end up with stale data in the Builder that "just happens" to work. We need tests to actually test what they are supposed to test without relying on stale data.

The comments in the tests about "maytrap" being wrong show us that we absolutely have exactly this bug in multiple places. The Builder isn't getting properly updated and this change makes that visible and clear.

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics-constrained.c

6

Because "maytrap" is not the IRBuilder's default of "strict" and it isn't clang's default of "ignore".

The #pragma sets the exception handling to "strict". Having the command line arguments set a global "maytrap" with a #pragma that sets it to "strict" at the top of the file means that places in clang that need to be updated will become visible.

26

Because the #pragma covers the entire file and sets exception handling to "strict". Thus all constrained intrinsic calls should be "strict", and if they are "maytrap" or "ignore" then we have a bug.

[kpn]

Updates from review.

[sepavloff]

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics-constrained.c

26

What is the reason for that? Does #pragma float_control work incorrectly? Why in clang/test/CodeGen/complex-math-strictfp.c exception handling is correct?

[kpn]

clang/lib/CodeGen/CodeGenFunction.cpp

153–154

Actually, no, because this leaves OldExcept and OldRounding uninitialized when they get used by the destructor.

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics-constrained.c

26

The #pragma works just fine. The problem is that we need to get the strictfp bits from the AST to the IRBuilder, and we haven't finished implementing that. So sometimes it works, like in complex-math-strictfp.c, and sometimes it doesn't. As you can see in the tests in this patch.

[kpn]

Fix a bug introduced in the last round, and eliminate the pass manager argument from one of the tests.

[kpn]

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics-constrained.c

26

I forgot to mention that complex-math-strictfp.c gets a correct BinOpInfo passed down to get the IRBuilder set correctly. But there are other places that don't correctly set BinOpInfo and so we get inconsistent behavior despite the call to the IRBuilder being wrapped in FPOptsRAII. And _that's_ why I structured this patch the way I did.

[sepavloff]

Generally the patch looks good. But the need to expect incorrect values in tests is a concern. Maybe this is a consequence of storing exception behavior in a separate field of CGFPOptionsRAII. This misbehavior should be fixed.

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics-constrained.c

26

The problem is that we need to get the strictfp bits from the AST to the IRBuilder

What is the status of this problem? Is it on track?

[kpn]

In D88913#2353379, @sepavloff wrote:

Generally the patch looks good. But the need to expect incorrect values in tests is a concern. Maybe this is a consequence of storing exception behavior in a separate field of CGFPOptionsRAII. This misbehavior should be fixed.

In this patch? Because that's going to be a huge patch. Fixing all the issues with strictfp AST->IRBuilder is going to be large. Fixing them incrementally seems better to me. Eliminating the incorrect values in the tests marked FIXME would sweep the problem under the rug.

It's true that incorrect values in tests is _not_ the desired end state. But it seems to me that as a _temporary_ incremental step it beats the alternatives.

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics-constrained.c

26

This is a larger problem that we need to tackle incrementally. I'd like to get a signoff on the approach taken in this patch before going too far down the road of writing fixes for all the issues we have here. But it is at the top of my list.

[sepavloff]

In D88913#2356152, @kpn wrote:

In D88913#2353379, @sepavloff wrote:

Generally the patch looks good. But the need to expect incorrect values in tests is a concern. Maybe this is a consequence of storing exception behavior in a separate field of CGFPOptionsRAII. This misbehavior should be fixed.

In this patch? Because that's going to be a huge patch. Fixing all the issues with strictfp AST->IRBuilder is going to be large. Fixing them incrementally seems better to me. Eliminating the incorrect values in the tests marked FIXME would sweep the problem under the rug.

It's true that incorrect values in tests is _not_ the desired end state. But it seems to me that as a _temporary_ incremental step it beats the alternatives.

OK. Users that called compiler with -ffp-exception-behavior=strict shouldn't notice any changes, right? Other non-default FP mode combinations anyway doesn't work now. So this patch shouldn't break existing code.

The only missing thing is tests that check non-default rounding mode. #pragma STDC FENV_ROUND can be used to prepare them, as it is done in D90026. Could you please add such test, to check if rounding mode is correctly processed in cast nodes?

[kpn]

Add some tests for changing the rounding metadata.

These tests brought out some places that needed to set the strictfp metadata, so I've added those as well.

[kpn]

Fix test, and eliminate a debugging "tee" command accidentally left in.

[sepavloff]

This patch contains many tests and they are valuable. However many of them are likely to be compatible with unpatched compiler, so are not related to casting nodes. I would propose to move such tests into a separate patch and make this one dependent on it. Leave here only tests that depend on the changes made by this patch.

Also tests should be as compact as possible. Test assertions should avoid checking unrelated nodes, as it increases the change of breaking the test and complicated its analysis.

clang/test/CodeGen/builtin_float_strictfp.c

2

Part of this test, namely functions test_floats, test_doubles and tests_halves must pass on unpatched compiler, because they don't contain conversion nodes. Maybe they can be extracted into separate patch, on which this one would depend?

clang/test/CodeGen/complex-math-strictfp.c

1 ↗

(On Diff #301695)

Throughout this test the only operation that involves complex is construction of complex value from real part. It does not use any conversion node, so this test must with unpatched compiler.

clang/test/CodeGen/complex-strictfp.c

93–96

This function generated complicated code, which is hard to check and easy to break. Can this function be split into smaller functions which would test only one operation? It could allow to use more compact checks. I think these tests should pass on unpatched compiler.

132

This function does not produce constrained intrinsics at all.

240

Can it be split into smaller functions?

361

It seems only this statement involves floating point arithmetic. Others are integer only.

398

Integer-only arithmetic.

422

No floating point operations, only moves.

470–473

Only these operations involve FP operations, but none produces constrained intrinsics. And they don't produce cast nodes. Besides, these inc/dec operations are represented in IR as add/sub, does it makes sense to test them separately?

clang/test/CodeGen/exprs-strictfp.c

2

No FP casts.

clang/test/CodeGen/incdec-strictfp.c

22 ↗

(On Diff #301695)

This line and corresponding function declaration seem unnecessary?

clang/test/CodeGen/ubsan-conditional-strictfp.c

1 ↗

(On Diff #301695)

What is the effect of -fsanitize=float-divide-by-zero? If it absents does the test change behavior?

clang/test/CodeGen/zvector-strictfp.c

9–23 ↗

(On Diff #301695)

These are integer values, they cannot produce constrained intrinsics. Why they are tested here?

[kpn]

I'll see how much I can simplify of the tests.

clang/test/CodeGen/exprs-strictfp.c

2

There's an implied cast in the AST, and I'm pretty sure the bits from that are used when simplifying this into the fcmp. I'll doublecheck.

clang/test/CodeGen/ubsan-conditional-strictfp.c

1 ↗

(On Diff #301695)

An earlier version of this patch did try passing bits all the way down to the calls to the IRBuilder. In this version of the patch the sanitizer was required to be involved to test one of the places that was changed. I had 100% test coverage for that version of the code.

I'll check and see if it is still needed.

clang/test/CodeGen/zvector-strictfp.c

9–23 ↗

(On Diff #301695)

This test is also a holdover from the first version of the patch. It was also needed to achieve 100% test coverage with that first version. I'll check and see if it is still needed in this v2 version of the patch.

[kpn]

Update for review comments.

clang/test/CodeGen/builtin_float_strictfp.c

2

I deleted test_floats and test_doubles because of that lack of conversion nodes. But test_half does have implicit conversions and I added checks for more of them.

clang/test/CodeGen/complex-math-strictfp.c

1 ↗

(On Diff #301695)

That's valid. The conversion from float to "_Complex float" does get an ImplicitCastExpr, but it shouldn't result in constrained intrinsics. So the test isn't needed. I'll nuke it.

clang/test/CodeGen/complex-strictfp.c

93–96

I'll pull this out and put it in a subsequent patch. It shows we have work to do, but isn't needed for this patch.

132

Eliminated

240

Sure, can do.

398

Eliminated

422

Eliminated

470–473

Yes, a separate patch for those would be better.

clang/test/CodeGen/exprs-strictfp.c

2

Confirmed. In CodeGenFunction::EvaluateExprAsBool() we set the constrained metadata and then call down eventually to ScalarExprEmitter::EmitFloatToBoolConversion().

The node we get the metadata from is a "ImplicitCastExpr 0x80fe2b338 'double' <LValueToRValue> FPExceptionMode=2".

clang/test/CodeGen/incdec-strictfp.c

22 ↗

(On Diff #301695)

This whole test is better left to a subsequent patch. It's another holdover from my previous attempt at the code changes.

clang/test/CodeGen/ubsan-conditional-strictfp.c

1 ↗

(On Diff #301695)

Right, the test was meant to test a change to the ubsan block towards the top of ScalarExprEmitter::EmitDiv(). The change isn't present in this particular patch. We still have a problem here, but it doesn't involve casts, so this can wait for a future patch.

clang/test/CodeGen/zvector-strictfp.c

9–23 ↗

(On Diff #301695)

It was needed to test one of the CreateFAdd() calls in ScalarExprEmitter::EmitScalarPrePostIncDec() but this is another case where there are no casts involved so this should go in a later patch.

[kpn]

Update for review comments.

[sepavloff]

LGTM.

clang/lib/CodeGen/CGExpr.cpp

llvm::Value *CodeGenFunction::EvaluateExprAsBool(const Expr *E) {
   PGO.setCurrentStmt(E);
   if (const MemberPointerType *MPT = E->getType()->getAs<MemberPointerType>()) {
     llvm::Value *MemPtr = EmitScalarExpr(E);
     return CGM.getCXXABI().EmitMemberPointerIsNotNull(*this, MemPtr, MPT);
   }
 
   QualType BoolTy = getContext().BoolTy;
   SourceLocation Loc = E->getExprLoc();
+  CGFPOptionsRAII FPOptsRAII(*this, E);
   if (!E->getType()->isAnyComplexType())
     return EmitScalarConversion(EmitScalarExpr(E), E->getType(), BoolTy, Loc);
 
   return EmitComplexToScalarConversion(EmitComplexExpr(E), E->getType(), BoolTy,
                                        Loc);
 }
 
 /// EmitIgnoredExpr - Emit code to compute the specified expression,

clang/lib/CodeGen/CGExprComplex.cpp

ComplexPairTy ComplexExprEmitter::EmitCast(CastKind CK, Expr *Op,
   case CK_FixedPointToFloating:
   case CK_FixedPointCast:
   case CK_FixedPointToBoolean:
   case CK_FixedPointToIntegral:
   case CK_IntegralToFixedPoint:
     llvm_unreachable("invalid cast kind for complex value");
 
   case CK_FloatingRealToComplex:
-  case CK_IntegralRealToComplex:
+  case CK_IntegralRealToComplex: {
+    CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, Op);
     return EmitScalarToComplexCast(CGF.EmitScalarExpr(Op), Op->getType(),
                                    DestTy, Op->getExprLoc());
+  }
 
   case CK_FloatingComplexCast:
   case CK_FloatingComplexToIntegralComplex:
   case CK_IntegralComplexCast:
-  case CK_IntegralComplexToFloatingComplex:
+  case CK_IntegralComplexToFloatingComplex: {
+    CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, Op);
     return EmitComplexToComplexCast(Visit(Op), Op->getType(), DestTy,
                                     Op->getExprLoc());
   }
+  }
 
   llvm_unreachable("unknown cast resulting in complex value");
 }
 
 ComplexPairTy ComplexExprEmitter::VisitUnaryMinus(const UnaryOperator *E) {
   TestAndClearIgnoreReal();
   TestAndClearIgnoreImag();
   ComplexPairTy Op = Visit(E->getSubExpr());
EmitCompoundAssignLValue(const CompoundAssignOperator *E,
           ComplexPairTy (ComplexExprEmitter::*Func)(const BinOpInfo&),
                          RValue &Val) {
   TestAndClearIgnoreReal();
   TestAndClearIgnoreImag();
   QualType LHSTy = E->getLHS()->getType();
   if (const AtomicType *AT = LHSTy->getAs<AtomicType>())
     LHSTy = AT->getValueType();
 
+  CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
   BinOpInfo OpInfo;
 
   // Load the RHS and LHS operands.
   // __block variables need to have the rhs evaluated first, plus this should
   // improve codegen a little.
   OpInfo.Ty = E->getComputationResultType();
   QualType ComplexElementTy = cast<ComplexType>(OpInfo.Ty)->getElementType();
 

clang/lib/CodeGen/CGExprScalar.cpp

case CK_IntegralCast: {
     }
     return EmitScalarConversion(Visit(E), E->getType(), DestTy,
                                 CE->getExprLoc(), Opts);
   }
   case CK_IntegralToFloating:
   case CK_FloatingToIntegral:
   case CK_FloatingCast:
   case CK_FixedPointToFloating:
-  case CK_FloatingToFixedPoint:
+  case CK_FloatingToFixedPoint: {
+    CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, CE);
     return EmitScalarConversion(Visit(E), E->getType(), DestTy,
                                 CE->getExprLoc());
+  }
   case CK_BooleanToSignedIntegral: {
     ScalarConversionOpts Opts;
     Opts.TreatBooleanAsSigned = true;
     return EmitScalarConversion(Visit(E), E->getType(), DestTy,
                                 CE->getExprLoc(), Opts);
   }
   case CK_IntegralToBoolean:
     return EmitIntToBoolConversion(Visit(E));
   case CK_PointerToBoolean:
     return EmitPointerToBoolConversion(Visit(E), E->getType());
-  case CK_FloatingToBoolean:
+  case CK_FloatingToBoolean: {
+    CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, CE);
     return EmitFloatToBoolConversion(Visit(E));
+  }
   case CK_MemberPointerToBoolean: {
     llvm::Value *MemPtr = Visit(E);
     const MemberPointerType *MPT = E->getType()->getAs<MemberPointerType>();
     return CGF.CGM.getCXXABI().EmitMemberPointerIsNotNull(CGF, MemPtr, MPT);
   }
 
   case CK_FloatingComplexToReal:
   case CK_IntegralComplexToReal:

clang/lib/CodeGen/CodeGenFunction.h

public:
   template <class T>
   typename DominatingValue<T>::saved_type saveValueInCond(T value) {
     return DominatingValue<T>::save(*this, value);
   }
 
   class CGFPOptionsRAII {
   public:
     CGFPOptionsRAII(CodeGenFunction &CGF, FPOptions FPFeatures);
+    CGFPOptionsRAII(CodeGenFunction &CGF, const Expr *E);
     ~CGFPOptionsRAII();
 
   private:
+    void ConstructorHelper(FPOptions FPFeatures);
     CodeGenFunction &CGF;
     FPOptions OldFPFeatures;
+    llvm::fp::ExceptionBehavior OldExcept;
+    llvm::RoundingMode OldRounding;
     Optional<CGBuilderTy::FastMathFlagGuard> FMFGuard;
   };
   FPOptions CurFPFeatures;
 
 public:
   /// ObjCEHValueStack - Stack of Objective-C exception values, used for
   /// rethrows.
   SmallVector<llvm::Value*, 8> ObjCEHValueStack;

clang/lib/CodeGen/CodeGenFunction.cpp

 #include "CodeGenModule.h"
 #include "CodeGenPGO.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/ASTLambda.h"
 #include "clang/AST/Attr.h"
 #include "clang/AST/Decl.h"
 #include "clang/AST/DeclCXX.h"
+#include "clang/AST/Expr.h"
 #include "clang/AST/StmtCXX.h"
 #include "clang/AST/StmtObjC.h"
 #include "clang/Basic/Builtins.h"
 #include "clang/Basic/CodeGenOptions.h"
 #include "clang/Basic/TargetInfo.h"
 #include "clang/CodeGen/CGFunctionInfo.h"
 #include "clang/Frontend/FrontendDiagnostic.h"
 #include "llvm/ADT/ArrayRef.h"
void CodeGenFunction::SetFastMathFlags(FPOptions FPFeatures) {
   FMF.setNoSignedZeros(FPFeatures.getNoSignedZero());
   FMF.setAllowReciprocal(FPFeatures.getAllowReciprocal());
   FMF.setApproxFunc(FPFeatures.getAllowApproxFunc());
   FMF.setAllowContract(FPFeatures.allowFPContractAcrossStatement());
   Builder.setFastMathFlags(FMF);
 }
 
 CodeGenFunction::CGFPOptionsRAII::CGFPOptionsRAII(CodeGenFunction &CGF,
+                                                  const Expr *E)
+    : CGF(CGF) {
+  ConstructorHelper(E->getFPFeaturesInEffect(CGF.getLangOpts()));
+}
+
+CodeGenFunction::CGFPOptionsRAII::CGFPOptionsRAII(CodeGenFunction &CGF,
                                                   FPOptions FPFeatures)
-    : CGF(CGF), OldFPFeatures(CGF.CurFPFeatures) {
+    : CGF(CGF) {
+  ConstructorHelper(FPFeatures);
+}
+
+void CodeGenFunction::CGFPOptionsRAII::ConstructorHelper(FPOptions FPFeatures) {
+  OldFPFeatures = CGF.CurFPFeatures;
   CGF.CurFPFeatures = FPFeatures;
 
+  OldExcept = CGF.Builder.getDefaultConstrainedExcept();
+  OldRounding = CGF.Builder.getDefaultConstrainedRounding();
+
   if (OldFPFeatures == FPFeatures)
     return;
 
   FMFGuard.emplace(CGF.Builder);
 
   llvm::RoundingMode NewRoundingBehavior =
       static_cast<llvm::RoundingMode>(FPFeatures.getRoundingMode());
   CGF.Builder.setDefaultConstrainedRounding(NewRoundingBehavior);
   auto NewExceptionBehavior =
       ToConstrainedExceptMD(static_cast<LangOptions::FPExceptionModeKind>(
void CodeGenFunction::CGFPOptionsRAII::ConstructorHelper(FPOptions FPFeatures) {
   mergeFnAttrValue("unsafe-fp-math", FPFeatures.getAllowFPReassociate() &&
                                          FPFeatures.getAllowReciprocal() &&
                                          FPFeatures.getAllowApproxFunc() &&
                                          FPFeatures.getNoSignedZero());
 }
 
 CodeGenFunction::CGFPOptionsRAII::~CGFPOptionsRAII() {
   CGF.CurFPFeatures = OldFPFeatures;
+  CGF.Builder.setDefaultConstrainedExcept(OldExcept);
+  CGF.Builder.setDefaultConstrainedRounding(OldRounding);
 }
 
 LValue CodeGenFunction::MakeNaturalAlignAddrLValue(llvm::Value *V, QualType T) {
   LValueBaseInfo BaseInfo;
   TBAAAccessInfo TBAAInfo;
   CharUnits Alignment = CGM.getNaturalTypeAlignment(T, &BaseInfo, &TBAAInfo);
   return LValue::MakeAddr(Address(V, Alignment), T, getContext(), BaseInfo,
                           TBAAInfo);

clang/lib/Sema/SemaExpr.cpp

ExprResult Sema::DefaultLvalueConversion(Expr *E) {
 
   if (E->getType().isDestructedType() == QualType::DK_nontrivial_c_struct)
     Cleanup.setExprNeedsCleanups(true);
 
   // C++ [conv.lval]p3:
   //   If T is cv std::nullptr_t, the result is a null pointer constant.
   CastKind CK = T->isNullPtrType() ? CK_NullToPointer : CK_LValueToRValue;
   Res = ImplicitCastExpr::Create(Context, T, CK, E, nullptr, VK_RValue,
-                                 FPOptionsOverride());
+                                 CurFPFeatureOverrides());
 
   // C11 6.3.2.1p2:
   //   ... if the lvalue has atomic type, the value has the non-atomic version
   //   of the type of the lvalue ...
   if (const AtomicType *Atomic = T->getAs<AtomicType>()) {
     T = Atomic->getValueType().getUnqualifiedType();
     Res = ImplicitCastExpr::Create(Context, T, CK_AtomicToNonAtomic, Res.get(),
                                    nullptr, VK_RValue, FPOptionsOverride());

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics-constrained.c

 // RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a\
 // RUN: -fallow-half-arguments-and-returns -flax-vector-conversions=none -S -disable-O0-optnone -emit-llvm -o - %s \
 // RUN: | opt -S -mem2reg \
 // RUN: | FileCheck --check-prefix=COMMON --check-prefix=COMMONIR --check-prefix=UNCONSTRAINED %s
 // RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a\
-// RUN: -ffp-exception-behavior=strict \
+// RUN: -ffp-exception-behavior=maytrap -DEXCEPT=1 \
 // RUN: -fexperimental-strict-floating-point \
 // RUN: -fallow-half-arguments-and-returns -flax-vector-conversions=none -S -disable-O0-optnone -emit-llvm -o - %s \
 // RUN: | opt -S -mem2reg \
 // RUN: | FileCheck --check-prefix=COMMON --check-prefix=COMMONIR --check-prefix=CONSTRAINED %s
 // RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a\
 // RUN: -fallow-half-arguments-and-returns -flax-vector-conversions=none -S -disable-O0-optnone -emit-llvm -o - %s \
 // RUN: | opt -S -mem2reg | llc -o=- - \
 // RUN: | FileCheck --check-prefix=COMMON --check-prefix=CHECK-ASM %s
 // RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a\
-// RUN: -ffp-exception-behavior=strict \
+// RUN: -ffp-exception-behavior=maytrap -DEXCEPT=1 \
 // RUN: -fexperimental-strict-floating-point \
 // RUN: -fallow-half-arguments-and-returns -flax-vector-conversions=none -S -disable-O0-optnone -emit-llvm -o - %s \
 // RUN: | opt -S -mem2reg | llc -o=- - \
 // RUN: | FileCheck --check-prefix=COMMON --check-prefix=CHECK-ASM %s
-
+//
 // REQUIRES: aarch64-registered-target
 
+// Test that the constrained intrinsics are picking up the exception
+// metadata from the AST instead of the global default from the command line.
+// FIXME: All cases of "fpexcept.maytrap" in this test are wrong.
+
+#if EXCEPT
+#pragma float_control(except, on)
+#endif
+
 #include <arm_neon.h>
 
 // COMMON-LABEL: test_vsqrt_f16
 // UNCONSTRAINED:  [[SQR:%.*]] = call <4 x half> @llvm.sqrt.v4f16(<4 x half> %a)
-// CONSTRAINED:    [[SQR:%.*]] = call <4 x half> @llvm.experimental.constrained.sqrt.v4f16(<4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:    [[SQR:%.*]] = call <4 x half> @llvm.experimental.constrained.sqrt.v4f16(<4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:      fsqrt v{{[0-9]+}}.4h, v{{[0-9]+}}.4h
 // COMMONIR:       ret <4 x half> [[SQR]]
 float16x4_t test_vsqrt_f16(float16x4_t a) {
   return vsqrt_f16(a);
 }
 
 // COMMON-LABEL: test_vsqrtq_f16
 // UNCONSTRAINED:  [[SQR:%.*]] = call <8 x half> @llvm.sqrt.v8f16(<8 x half> %a)
-// CONSTRAINED:    [[SQR:%.*]] = call <8 x half> @llvm.experimental.constrained.sqrt.v8f16(<8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:    [[SQR:%.*]] = call <8 x half> @llvm.experimental.constrained.sqrt.v8f16(<8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:      fsqrt v{{[0-9]+}}.8h, v{{[0-9]+}}.8h
 // COMMONIR:       ret <8 x half> [[SQR]]
 float16x8_t test_vsqrtq_f16(float16x8_t a) {
   return vsqrtq_f16(a);
 }
 
 // COMMON-LABEL: test_vfma_f16
 // UNCONSTRAINED:  [[ADD:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> %b, <4 x half> %c, <4 x half> %a)
-// CONSTRAINED:    [[ADD:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> %b, <4 x half> %c, <4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:    [[ADD:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> %b, <4 x half> %c, <4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:      fmla v{{[0-9]+}}.4h, v{{[0-9]+}}.4h, v{{[0-9]+}}.4h
 // COMMONIR:       ret <4 x half> [[ADD]]
 float16x4_t test_vfma_f16(float16x4_t a, float16x4_t b, float16x4_t c) {
   return vfma_f16(a, b, c);
 }
 
 // COMMON-LABEL: test_vfmaq_f16
 // UNCONSTRAINED:  [[ADD:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> %b, <8 x half> %c, <8 x half> %a)
-// CONSTRAINED:    [[ADD:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> %b, <8 x half> %c, <8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:    [[ADD:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> %b, <8 x half> %c, <8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:      fmla v{{[0-9]+}}.8h, v{{[0-9]+}}.8h, v{{[0-9]+}}.8h
 // COMMONIR:       ret <8 x half> [[ADD]]
 float16x8_t test_vfmaq_f16(float16x8_t a, float16x8_t b, float16x8_t c) {
   return vfmaq_f16(a, b, c);
 }
 
 // COMMON-LABEL: test_vfms_f16
 // COMMONIR:       [[SUB:%.*]] = fneg <4 x half> %b
 // CHECK-ASM:      fneg v{{[0-9]+}}.4h, v{{[0-9]+}}.4h
 // UNCONSTRAINED:  [[ADD:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[SUB]], <4 x half> %c, <4 x half> %a)
-// CONSTRAINED:    [[ADD:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[SUB]], <4 x half> %c, <4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:    [[ADD:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[SUB]], <4 x half> %c, <4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:      fmla v{{[0-9]+}}.4h, v{{[0-9]+}}.4h, v{{[0-9]+}}.4h
 // COMMONIR:       ret <4 x half> [[ADD]]
 float16x4_t test_vfms_f16(float16x4_t a, float16x4_t b, float16x4_t c) {
   return vfms_f16(a, b, c);
 }
 
 // COMMON-LABEL: test_vfmsq_f16
 // COMMONIR:       [[SUB:%.*]] = fneg <8 x half> %b
 // CHECK-ASM:      fneg v{{[0-9]+}}.8h, v{{[0-9]+}}.8h
 // UNCONSTRAINED:  [[ADD:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[SUB]], <8 x half> %c, <8 x half> %a)
-// CONSTRAINED:    [[ADD:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[SUB]], <8 x half> %c, <8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:    [[ADD:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[SUB]], <8 x half> %c, <8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:      fmla v{{[0-9]+}}.8h, v{{[0-9]+}}.8h, v{{[0-9]+}}.8h
 // COMMONIR:       ret <8 x half> [[ADD]]
 float16x8_t test_vfmsq_f16(float16x8_t a, float16x8_t b, float16x8_t c) {
   return vfmsq_f16(a, b, c);
 }
 
 // COMMON-LABEL: test_vfma_lane_f16
 // COMMONIR:      [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8>
 // COMMONIR:      [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8>
 // COMMONIR:      [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8>
 // COMMONIR:      [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half>
 // COMMONIR:      [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
 // COMMONIR:      [[TMP4:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half>
 // COMMONIR:      [[TMP5:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
 // UNCONSTRAINED: [[FMLA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[TMP4]], <4 x half> [[LANE]], <4 x half> [[TMP5]])
-// CONSTRAINED:   [[FMLA:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[TMP4]], <4 x half> [[LANE]], <4 x half> [[TMP5]], metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMLA:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[TMP4]], <4 x half> [[LANE]], <4 x half> [[TMP5]], metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla v{{[0-9]+}}.4h, v{{[0-9]+}}.4h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <4 x half> [[FMLA]]
 float16x4_t test_vfma_lane_f16(float16x4_t a, float16x4_t b, float16x4_t c) {
   return vfma_lane_f16(a, b, c, 3);
 }
 
 // COMMON-LABEL: test_vfmaq_lane_f16
 // COMMONIR:      [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8>
 // COMMONIR:      [[TMP1:%.*]] = bitcast <8 x half> %b to <16 x i8>
 // COMMONIR:      [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8>
 // COMMONIR:      [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half>
 // COMMONIR:      [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
 // COMMONIR:      [[TMP4:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half>
 // COMMONIR:      [[TMP5:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half>
 // UNCONSTRAINED: [[FMLA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[TMP4]], <8 x half> [[LANE]], <8 x half> [[TMP5]])
-// CONSTRAINED:   [[FMLA:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[TMP4]], <8 x half> [[LANE]], <8 x half> [[TMP5]], metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMLA:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[TMP4]], <8 x half> [[LANE]], <8 x half> [[TMP5]], metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla v{{[0-9]+}}.8h, v{{[0-9]+}}.8h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <8 x half> [[FMLA]]
 float16x8_t test_vfmaq_lane_f16(float16x8_t a, float16x8_t b, float16x4_t c) {
   return vfmaq_lane_f16(a, b, c, 3);
 }
 
 // COMMON-LABEL: test_vfma_laneq_f16
 // COMMONIR:      [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8>
 // COMMONIR:      [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8>
 // COMMONIR:      [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8>
 // COMMONIR:      [[TMP3:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
 // COMMONIR:      [[TMP4:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half>
 // COMMONIR:      [[TMP5:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half>
 // COMMONIR:      [[LANE:%.*]] = shufflevector <8 x half> [[TMP5]], <8 x half> [[TMP5]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
 // UNCONSTRAINED: [[FMLA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[LANE]], <4 x half> [[TMP4]], <4 x half> [[TMP3]])
-// CONSTRAINED:   [[FMLA:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[LANE]], <4 x half> [[TMP4]], <4 x half> [[TMP3]], metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMLA:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[LANE]], <4 x half> [[TMP4]], <4 x half> [[TMP3]], metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla v{{[0-9]+}}.4h, v{{[0-9]+}}.4h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <4 x half> [[FMLA]]
 float16x4_t test_vfma_laneq_f16(float16x4_t a, float16x4_t b, float16x8_t c) {
   return vfma_laneq_f16(a, b, c, 7);
 }
 
 // COMMON-LABEL: test_vfmaq_laneq_f16
 // COMMONIR:      [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8>
 // COMMONIR:      [[TMP1:%.*]] = bitcast <8 x half> %b to <16 x i8>
 // COMMONIR:      [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8>
 // COMMONIR:      [[TMP3:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half>
 // COMMONIR:      [[TMP4:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half>
 // COMMONIR:      [[TMP5:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half>
 // COMMONIR:      [[LANE:%.*]] = shufflevector <8 x half> [[TMP5]], <8 x half> [[TMP5]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
 // UNCONSTRAINED: [[FMLA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[LANE]], <8 x half> [[TMP4]], <8 x half> [[TMP3]])
-// CONSTRAINED:   [[FMLA:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[LANE]], <8 x half> [[TMP4]], <8 x half> [[TMP3]], metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMLA:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[LANE]], <8 x half> [[TMP4]], <8 x half> [[TMP3]], metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla v{{[0-9]+}}.8h, v{{[0-9]+}}.8h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <8 x half> [[FMLA]]
 float16x8_t test_vfmaq_laneq_f16(float16x8_t a, float16x8_t b, float16x8_t c) {
   return vfmaq_laneq_f16(a, b, c, 7);
 }
 
 // COMMON-LABEL: test_vfma_n_f16
 // COMMONIR:      [[TMP0:%.*]] = insertelement <4 x half> undef, half %c, i32 0
 // COMMONIR:      [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %c, i32 1
 // COMMONIR:      [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %c, i32 2
 // COMMONIR:      [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half %c, i32 3
 // UNCONSTRAINED: [[FMA:%.*]]  = call <4 x half> @llvm.fma.v4f16(<4 x half> %b, <4 x half> [[TMP3]], <4 x half> %a)
-// CONSTRAINED:   [[FMA:%.*]]  = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> %b, <4 x half> [[TMP3]], <4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]]  = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> %b, <4 x half> [[TMP3]], <4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla v{{[0-9]+}}.4h, v{{[0-9]+}}.4h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <4 x half> [[FMA]]
 float16x4_t test_vfma_n_f16(float16x4_t a, float16x4_t b, float16_t c) {
   return vfma_n_f16(a, b, c);
 }
 
 // COMMON-LABEL: test_vfmaq_n_f16
 // COMMONIR:      [[TMP0:%.*]] = insertelement <8 x half> undef, half %c, i32 0
 // COMMONIR:      [[TMP1:%.*]] = insertelement <8 x half> [[TMP0]], half %c, i32 1
 // COMMONIR:      [[TMP2:%.*]] = insertelement <8 x half> [[TMP1]], half %c, i32 2
 // COMMONIR:      [[TMP3:%.*]] = insertelement <8 x half> [[TMP2]], half %c, i32 3
 // COMMONIR:      [[TMP4:%.*]] = insertelement <8 x half> [[TMP3]], half %c, i32 4
 // COMMONIR:      [[TMP5:%.*]] = insertelement <8 x half> [[TMP4]], half %c, i32 5
 // COMMONIR:      [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %c, i32 6
 // COMMONIR:      [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %c, i32 7
 // UNCONSTRAINED: [[FMA:%.*]]  = call <8 x half> @llvm.fma.v8f16(<8 x half> %b, <8 x half> [[TMP7]], <8 x half> %a)
-// CONSTRAINED:   [[FMA:%.*]]  = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> %b, <8 x half> [[TMP7]], <8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]]  = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> %b, <8 x half> [[TMP7]], <8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla v{{[0-9]+}}.8h, v{{[0-9]+}}.8h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <8 x half> [[FMA]]
 float16x8_t test_vfmaq_n_f16(float16x8_t a, float16x8_t b, float16_t c) {
   return vfmaq_n_f16(a, b, c);
 }
 
 // COMMON-LABEL: test_vfmah_lane_f16
 // COMMONIR:      [[EXTR:%.*]] = extractelement <4 x half> %c, i32 3
 // UNCONSTRAINED: [[FMA:%.*]]  = call half @llvm.fma.f16(half %b, half [[EXTR]], half %a)
-// CONSTRAINED:   [[FMA:%.*]]  = call half @llvm.experimental.constrained.fma.f16(half %b, half [[EXTR]], half %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]]  = call half @llvm.experimental.constrained.fma.f16(half %b, half [[EXTR]], half %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla h{{[0-9]+}}, h{{[0-9]+}}, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret half [[FMA]]
 float16_t test_vfmah_lane_f16(float16_t a, float16_t b, float16x4_t c) {
   return vfmah_lane_f16(a, b, c, 3);
 }
 
 // COMMON-LABEL: test_vfmah_laneq_f16
 // COMMONIR:      [[EXTR:%.*]] = extractelement <8 x half> %c, i32 7
 // UNCONSTRAINED: [[FMA:%.*]]  = call half @llvm.fma.f16(half %b, half [[EXTR]], half %a)
-// CONSTRAINED:   [[FMA:%.*]]  = call half @llvm.experimental.constrained.fma.f16(half %b, half [[EXTR]], half %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]]  = call half @llvm.experimental.constrained.fma.f16(half %b, half [[EXTR]], half %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla h{{[0-9]+}}, h{{[0-9]+}}, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret half [[FMA]]
 float16_t test_vfmah_laneq_f16(float16_t a, float16_t b, float16x8_t c) {
   return vfmah_laneq_f16(a, b, c, 7);
 }
 
 // COMMON-LABEL: test_vfms_lane_f16
 // COMMONIR:      [[SUB:%.*]]  = fneg <4 x half> %b
 // COMMONIR:      [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8>
 // COMMONIR:      [[TMP1:%.*]] = bitcast <4 x half> [[SUB]] to <8 x i8>
 // COMMONIR:      [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8>
 // COMMONIR:      [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half>
 // COMMONIR:      [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
 // COMMONIR:      [[TMP4:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half>
 // COMMONIR:      [[TMP5:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
 // UNCONSTRAINED: [[FMA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[TMP4]], <4 x half> [[LANE]], <4 x half> [[TMP5]])
-// CONSTRAINED:   [[FMA:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[TMP4]], <4 x half> [[LANE]], <4 x half> [[TMP5]], metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[TMP4]], <4 x half> [[LANE]], <4 x half> [[TMP5]], metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmls v{{[0-9]+}}.4h, v{{[0-9]+}}.4h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <4 x half> [[FMA]]
 float16x4_t test_vfms_lane_f16(float16x4_t a, float16x4_t b, float16x4_t c) {
   return vfms_lane_f16(a, b, c, 3);
 }
 
 // COMMON-LABEL: test_vfmsq_lane_f16
 // COMMONIR:      [[SUB:%.*]]  = fneg <8 x half> %b
 // COMMONIR:      [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8>
 // COMMONIR:      [[TMP1:%.*]] = bitcast <8 x half> [[SUB]] to <16 x i8>
 // COMMONIR:      [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8>
 // COMMONIR:      [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half>
 // COMMONIR:      [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
 // COMMONIR:      [[TMP4:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half>
 // COMMONIR:      [[TMP5:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half>
 // UNCONSTRAINED: [[FMLA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[TMP4]], <8 x half> [[LANE]], <8 x half> [[TMP5]])
-// CONSTRAINED:   [[FMLA:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[TMP4]], <8 x half> [[LANE]], <8 x half> [[TMP5]], metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMLA:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[TMP4]], <8 x half> [[LANE]], <8 x half> [[TMP5]], metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmls v{{[0-9]+}}.8h, v{{[0-9]+}}.8h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <8 x half> [[FMLA]]
 float16x8_t test_vfmsq_lane_f16(float16x8_t a, float16x8_t b, float16x4_t c) {
   return vfmsq_lane_f16(a, b, c, 3);
 }
 
 // COMMON-LABEL: test_vfms_laneq_f16
 // COMMONIR:      [[SUB:%.*]]  = fneg <4 x half> %b
 // CHECK-ASM-NOT: fneg
 // COMMONIR:      [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8>
 // COMMONIR:      [[TMP1:%.*]] = bitcast <4 x half> [[SUB]] to <8 x i8>
 // COMMONIR:      [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8>
 // COMMONIR:      [[TMP3:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
 // COMMONIR:      [[TMP4:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half>
 // COMMONIR:      [[TMP5:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half>
 // COMMONIR:      [[LANE:%.*]] = shufflevector <8 x half> [[TMP5]], <8 x half> [[TMP5]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
 // UNCONSTRAINED: [[FMLA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[LANE]], <4 x half> [[TMP4]], <4 x half> [[TMP3]])
-// CONSTRAINED:   [[FMLA:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[LANE]], <4 x half> [[TMP4]], <4 x half> [[TMP3]], metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMLA:%.*]] = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[LANE]], <4 x half> [[TMP4]], <4 x half> [[TMP3]], metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmls v{{[0-9]+}}.4h, v{{[0-9]+}}.4h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <4 x half> [[FMLA]]
 float16x4_t test_vfms_laneq_f16(float16x4_t a, float16x4_t b, float16x8_t c) {
   return vfms_laneq_f16(a, b, c, 7);
 }
 
 // COMMON-LABEL: test_vfmsq_laneq_f16
 // COMMONIR:      [[SUB:%.*]]  = fneg <8 x half> %b
 // CHECK-ASM-NOT: fneg
 // COMMONIR:      [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8>
 // COMMONIR:      [[TMP1:%.*]] = bitcast <8 x half> [[SUB]] to <16 x i8>
 // COMMONIR:      [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8>
 // COMMONIR:      [[TMP3:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half>
 // COMMONIR:      [[TMP4:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half>
 // COMMONIR:      [[TMP5:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half>
 // COMMONIR:      [[LANE:%.*]] = shufflevector <8 x half> [[TMP5]], <8 x half> [[TMP5]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
 // UNCONSTRAINED: [[FMLA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[LANE]], <8 x half> [[TMP4]], <8 x half> [[TMP3]])
-// CONSTRAINED:   [[FMLA:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[LANE]], <8 x half> [[TMP4]], <8 x half> [[TMP3]], metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMLA:%.*]] = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[LANE]], <8 x half> [[TMP4]], <8 x half> [[TMP3]], metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmls v{{[0-9]+}}.8h, v{{[0-9]+}}.8h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <8 x half> [[FMLA]]
 float16x8_t test_vfmsq_laneq_f16(float16x8_t a, float16x8_t b, float16x8_t c) {
   return vfmsq_laneq_f16(a, b, c, 7);
 }
 
 // COMMON-LABEL: test_vfms_n_f16
 // COMMONIR:      [[SUB:%.*]]  = fneg <4 x half> %b
 // COMMONIR:      [[TMP0:%.*]] = insertelement <4 x half> undef, half %c, i32 0
 // COMMONIR:      [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %c, i32 1
 // COMMONIR:      [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %c, i32 2
 // COMMONIR:      [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half %c, i32 3
 // UNCONSTRAINED: [[FMA:%.*]]  = call <4 x half> @llvm.fma.v4f16(<4 x half> [[SUB]], <4 x half> [[TMP3]], <4 x half> %a)
-// CONSTRAINED:   [[FMA:%.*]]  = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[SUB]], <4 x half> [[TMP3]], <4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]]  = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> [[SUB]], <4 x half> [[TMP3]], <4 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmls v{{[0-9]+}}.4h, v{{[0-9]+}}.4h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <4 x half> [[FMA]]
 float16x4_t test_vfms_n_f16(float16x4_t a, float16x4_t b, float16_t c) {
   return vfms_n_f16(a, b, c);
 }
 
 // COMMON-LABEL: test_vfmsq_n_f16
 // COMMONIR:      [[SUB:%.*]]  = fneg <8 x half> %b
 // COMMONIR:      [[TMP0:%.*]] = insertelement <8 x half> undef, half %c, i32 0
 // COMMONIR:      [[TMP1:%.*]] = insertelement <8 x half> [[TMP0]], half %c, i32 1
 // COMMONIR:      [[TMP2:%.*]] = insertelement <8 x half> [[TMP1]], half %c, i32 2
 // COMMONIR:      [[TMP3:%.*]] = insertelement <8 x half> [[TMP2]], half %c, i32 3
 // COMMONIR:      [[TMP4:%.*]] = insertelement <8 x half> [[TMP3]], half %c, i32 4
 // COMMONIR:      [[TMP5:%.*]] = insertelement <8 x half> [[TMP4]], half %c, i32 5
 // COMMONIR:      [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %c, i32 6
 // COMMONIR:      [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %c, i32 7
 // UNCONSTRAINED: [[FMA:%.*]]  = call <8 x half> @llvm.fma.v8f16(<8 x half> [[SUB]], <8 x half> [[TMP7]], <8 x half> %a)
-// CONSTRAINED:   [[FMA:%.*]]  = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[SUB]], <8 x half> [[TMP7]], <8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]]  = call <8 x half> @llvm.experimental.constrained.fma.v8f16(<8 x half> [[SUB]], <8 x half> [[TMP7]], <8 x half> %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmls v{{[0-9]+}}.8h, v{{[0-9]+}}.8h, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret <8 x half> [[FMA]]
 float16x8_t test_vfmsq_n_f16(float16x8_t a, float16x8_t b, float16_t c) {
   return vfmsq_n_f16(a, b, c);
 }
 
 // COMMON-LABEL: test_vfmsh_lane_f16
 // UNCONSTRAINED: [[TMP0:%.*]] = fpext half %b to float
 // CONSTRAINED:   [[TMP0:%.*]] = call float @llvm.experimental.constrained.fpext.f32.f16(half %b, metadata !"fpexcept.strict")
 // CHECK-ASM:     fcvt s{{[0-9]+}}, h{{[0-9]+}}
 // COMMONIR:      [[TMP1:%.*]] = fneg float [[TMP0]]
 // CHECK-ASM:     fneg s{{[0-9]+}}, s{{[0-9]+}}
 // UNCONSTRAINED: [[SUB:%.*]]  = fptrunc float [[TMP1]] to half
 // CONSTRAINED:   [[SUB:%.*]]  = call half @llvm.experimental.constrained.fptrunc.f16.f32(float [[TMP1]], metadata !"round.tonearest", metadata !"fpexcept.strict")
 // CHECK-ASM:     fcvt h{{[0-9]+}}, s{{[0-9]+}}
 // COMMONIR:      [[EXTR:%.*]] = extractelement <4 x half> %c, i32 3
 // UNCONSTRAINED: [[FMA:%.*]]  = call half @llvm.fma.f16(half [[SUB]], half [[EXTR]], half %a)
-// CONSTRAINED:   [[FMA:%.*]]  = call half @llvm.experimental.constrained.fma.f16(half [[SUB]], half [[EXTR]], half %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]]  = call half @llvm.experimental.constrained.fma.f16(half [[SUB]], half [[EXTR]], half %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla h{{[0-9]+}}, h{{[0-9]+}}, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret half [[FMA]]
 float16_t test_vfmsh_lane_f16(float16_t a, float16_t b, float16x4_t c) {
   return vfmsh_lane_f16(a, b, c, 3);
 }
 
 // COMMON-LABEL: test_vfmsh_laneq_f16
 // UNCONSTRAINED: [[TMP0:%.*]] = fpext half %b to float
 // CONSTRAINED:   [[TMP0:%.*]] = call float @llvm.experimental.constrained.fpext.f32.f16(half %b, metadata !"fpexcept.strict")
 // CHECK-ASM:     fcvt s{{[0-9]+}}, h{{[0-9]+}}
 // COMMONIR:      [[TMP1:%.*]] = fneg float [[TMP0]]
 // CHECK-ASM:     fneg s{{[0-9]+}}, s{{[0-9]+}}
 // UNCONSTRAINED: [[SUB:%.*]]  = fptrunc float [[TMP1]] to half
 // CONSTRAINED:   [[SUB:%.*]]  = call half @llvm.experimental.constrained.fptrunc.f16.f32(float [[TMP1]], metadata !"round.tonearest", metadata !"fpexcept.strict")
 // CHECK-ASM:     fcvt h{{[0-9]+}}, s{{[0-9]+}}
 // COMMONIR:      [[EXTR:%.*]] = extractelement <8 x half> %c, i32 7
 // UNCONSTRAINED: [[FMA:%.*]]  = call half @llvm.fma.f16(half [[SUB]], half [[EXTR]], half %a)
-// CONSTRAINED:   [[FMA:%.*]]  = call half @llvm.experimental.constrained.fma.f16(half [[SUB]], half [[EXTR]], half %a, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CONSTRAINED:   [[FMA:%.*]]  = call half @llvm.experimental.constrained.fma.f16(half [[SUB]], half [[EXTR]], half %a, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
 // CHECK-ASM:     fmla h{{[0-9]+}}, h{{[0-9]+}}, v{{[0-9]+}}.h[{{[0-9]+}}]
 // COMMONIR:      ret half [[FMA]]
 float16_t test_vfmsh_laneq_f16(float16_t a, float16_t b, float16x8_t c) {
   return vfmsh_laneq_f16(a, b, c, 7);
 }
-

clang/test/CodeGen/builtin_float_strictfp.c

+// RUN: %clang_cc1 -emit-llvm -triple x86_64-windows-pc -ffp-exception-behavior=maytrap -o - %s | FileCheck %s --check-prefixes=CHECK,FP16
+// RUN: %clang_cc1 -emit-llvm -triple ppc64-be -ffp-exception-behavior=maytrap -o - %s | FileCheck %s --check-prefixes=CHECK,NOFP16
+
+// test to ensure that these builtins don't do the variadic promotion of float->double.
+
+// Test that the constrained intrinsics are picking up the exception
+// metadata from the AST instead of the global default from the command line.
+// FIXME: All cases of "fpexcept.maytrap" in this test are wrong.
+
+#pragma float_control(except, on)
+
+// CHECK-LABEL: @test_half
+void test_half(__fp16 *H, __fp16 *H2) {
+  (void)__builtin_isgreater(*H, *H2);
+  // FP16: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // FP16: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"ogt", metadata !"fpexcept.maytrap")
+  // CHECK-NEXT: zext i1
+  (void)__builtin_isinf(*H);
+  // NOFP16: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float 0x7FF0000000000000, metadata !"oeq", metadata !"fpexcept.maytrap")
+  // FP16: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half 0xH7C00, metadata !"oeq", metadata !"fpexcept.maytrap")
+}
+
+// CHECK-LABEL: @test_mixed
+void test_mixed(double d1, float f2) {
+  (void)__builtin_isgreater(d1, f2);
+  // CHECK: [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK-NEXT: call i1 @llvm.experimental.constrained.fcmp.f64(double %{{.*}}, double [[CONV]], metadata !"ogt", metadata !"fpexcept.maytrap")
+  // CHECK-NEXT: zext i1
+  (void)__builtin_isgreaterequal(d1, f2);
+  // CHECK: [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK-NEXT: call i1 @llvm.experimental.constrained.fcmp.f64(double %{{.*}}, double [[CONV]], metadata !"oge", metadata !"fpexcept.maytrap")
+  // CHECK-NEXT: zext i1
+  (void)__builtin_isless(d1, f2);
+  // CHECK: [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK-NEXT: call i1 @llvm.experimental.constrained.fcmp.f64(double %{{.*}}, double [[CONV]], metadata !"olt", metadata !"fpexcept.maytrap")
+  // CHECK-NEXT: zext i1
+  (void)__builtin_islessequal(d1, f2);
+  // CHECK: [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK-NEXT: call i1 @llvm.experimental.constrained.fcmp.f64(double %{{.*}}, double [[CONV]], metadata !"ole", metadata !"fpexcept.maytrap")
+  // CHECK-NEXT: zext i1
+  (void)__builtin_islessgreater(d1, f2);
+  // CHECK: [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK-NEXT: call i1 @llvm.experimental.constrained.fcmp.f64(double %{{.*}}, double [[CONV]], metadata !"one", metadata !"fpexcept.maytrap")
+  // CHECK-NEXT: zext i1
+  (void)__builtin_isunordered(d1, f2);
+  // CHECK: [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK-NEXT: call i1 @llvm.experimental.constrained.fcmp.f64(double %{{.*}}, double [[CONV]], metadata !"uno", metadata !"fpexcept.maytrap")
+  // CHECK-NEXT: zext i1
+}

clang/test/CodeGen/complex-strictfp.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+// RUN: %clang_cc1 -triple x86_64-unknown-unknown -ffp-exception-behavior=maytrap -emit-llvm -o - %s | FileCheck %s
+
+
+// Test that the constrained intrinsics are picking up the exception
+// metadata from the AST instead of the global default from the command line.
+// Include rounding metadata in the testing.
+// FIXME: All cases of "fpexcept.maytrap" in this test are wrong.
+// FIXME: All cases of "round.tonearest" in this test are wrong.
+
+#pragma float_control(except, on)
+#pragma STDC FENV_ROUND FE_UPWARD
+
+_Complex double g1, g2;
+_Complex float cf;
+double D;
+
+// CHECK-LABEL: @test3a(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = load double, double* @D, align 8
+// CHECK-NEXT:    [[CF_REAL:%.*]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 0), align 4
+// CHECK-NEXT:    [[CF_IMAG:%.*]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 1), align 4
+// CHECK-NEXT:    [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[CF_REAL]], metadata !"fpexcept.strict") [[ATTR2:#.*]]
+// CHECK-NEXT:    [[CONV1:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[CF_IMAG]], metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    [[ADD_R:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[CONV]], double [[TMP0]], metadata !"round.upward", metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    [[CONV2:%.*]] = call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[ADD_R]], metadata !"round.upward", metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    [[CONV3:%.*]] = call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[CONV1]], metadata !"round.upward", metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    store float [[CONV2]], float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 0), align 4
+// CHECK-NEXT:    store float [[CONV3]], float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 1), align 4
+// CHECK-NEXT:    ret void
+//
+void test3a() {
+  cf += D;
+}
+
+// CHECK-LABEL: @test3b(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[CF_REAL:%.*]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 0), align 4
+// CHECK-NEXT:    [[CF_IMAG:%.*]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 1), align 4
+// CHECK-NEXT:    [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[CF_REAL]], metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    [[CONV1:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[CF_IMAG]], metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = load double, double* @D, align 8
+// CHECK-NEXT:    [[ADD_R:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[TMP0]], double [[CONV]], metadata !"round.upward", metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    store double [[ADD_R]], double* @D, align 8
+// CHECK-NEXT:    ret void
+//
+void test3b() {
+  D += cf;
+}
+
+// CHECK-LABEL: @test3c(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[G1_REAL:%.*]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 0), align 8
+// CHECK-NEXT:    [[G1_IMAG:%.*]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 1), align 8
+// CHECK-NEXT:    [[CF_REAL:%.*]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 0), align 4
+// CHECK-NEXT:    [[CF_IMAG:%.*]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 1), align 4
+// CHECK-NEXT:    [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[CF_REAL]], metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    [[CONV1:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[CF_IMAG]], metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    [[CALL:%.*]] = call { double, double } @__divdc3(double [[CONV]], double [[CONV1]], double [[G1_REAL]], double [[G1_IMAG]]) [[ATTR3:#.*]]
+// CHECK-NEXT:    [[TMP0:%.*]] = extractvalue { double, double } [[CALL]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { double, double } [[CALL]], 1
+// CHECK-NEXT:    [[CONV2:%.*]] = call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[TMP0]], metadata !"round.upward", metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    [[CONV3:%.*]] = call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[TMP1]], metadata !"round.upward", metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    store float [[CONV2]], float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 0), align 4
+// CHECK-NEXT:    store float [[CONV3]], float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 1), align 4
+// CHECK-NEXT:    ret void
+//
+void test3c() {
+  cf /= g1;
+}
+
+// CHECK-LABEL: @test3d(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[G1_REAL:%.*]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 0), align 8
+// CHECK-NEXT:    [[G1_IMAG:%.*]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 1), align 8
+// CHECK-NEXT:    [[TMP0:%.*]] = load double, double* @D, align 8
+// CHECK-NEXT:    [[ADD_R:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[G1_REAL]], double [[TMP0]], metadata !"round.tonearest", metadata !"fpexcept.maytrap") [[ATTR2]]
+// CHECK-NEXT:    store double [[ADD_R]], double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 0), align 8
+// CHECK-NEXT:    store double [[G1_IMAG]], double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 1), align 8
+// CHECK-NEXT:    ret void
+//
+void test3d() {
+  g1 = g1 + D;
+}
+
+// CHECK-LABEL: @test3e(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = load double, double* @D, align 8
+// CHECK-NEXT:    [[G1_REAL:%.*]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 0), align 8
+// CHECK-NEXT:    [[G1_IMAG:%.*]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 1), align 8
+// CHECK-NEXT:    [[ADD_R:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[TMP0]], double [[G1_REAL]], metadata !"round.tonearest", metadata !"fpexcept.maytrap") [[ATTR2]]
+// CHECK-NEXT:    store double [[ADD_R]], double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 0), align 8
+// CHECK-NEXT:    store double [[G1_IMAG]], double* getelementptr inbounds ({ double, double }, { double, double }* @g1, i32 0, i32 1), align 8
+// CHECK-NEXT:    ret void
+//
+void test3e() {
+  g1 = D + g1;
+}
+
+// CHECK-LABEL: @t1(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[CONV:%.*]] = call float @llvm.experimental.constrained.fptrunc.f32.f64(double 4.000000e+00, metadata !"round.upward", metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    store float [[CONV]], float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 0), align 4
+// CHECK-NEXT:    ret void
+//
+void t1() {
+  (__real__ cf) = 4.0;
+}
+
+// CHECK-LABEL: @t2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[CONV:%.*]] = call float @llvm.experimental.constrained.fptrunc.f32.f64(double 4.000000e+00, metadata !"round.upward", metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    store float [[CONV]], float* getelementptr inbounds ({ float, float }, { float, float }* @cf, i32 0, i32 1), align 4
+// CHECK-NEXT:    ret void
+//
+void t2() {
+  (__imag__ cf) = 4.0;
+}
+
+// CHECK-LABEL: @t91(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[C:%.*]] = alloca [0 x i8], align 1
+// CHECK-NEXT:    br i1 false, label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK:       cond.true:
+// CHECK-NEXT:    [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float 2.000000e+00, metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    br label [[COND_END:%.*]]
+// CHECK:       cond.false:
+// CHECK-NEXT:    [[CONV1:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float 2.000000e+00, metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    br label [[COND_END]]
+// CHECK:       cond.end:
+// CHECK-NEXT:    [[COND_R:%.*]] = phi double [ [[CONV]], [[COND_TRUE]] ], [ [[CONV1]], [[COND_FALSE]] ]
+// CHECK-NEXT:    [[COND_I:%.*]] = phi double [ 0.000000e+00, [[COND_TRUE]] ], [ 0.000000e+00, [[COND_FALSE]] ]
+// CHECK-NEXT:    ret void
+//
+void t91() {
+  // Check for proper type promotion of conditional expression
+  char c[(int)(sizeof(typeof((0 ? 2.0f : (_Complex double) 2.0f))) - sizeof(_Complex double))];
+  // Check for proper codegen
+  (0 ? 2.0f : (_Complex double) 2.0f);
+}
+
+// CHECK-LABEL: @t92(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[C:%.*]] = alloca [0 x i8], align 1
+// CHECK-NEXT:    br i1 false, label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK:       cond.true:
+// CHECK-NEXT:    [[CONV:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float 2.000000e+00, metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    br label [[COND_END:%.*]]
+// CHECK:       cond.false:
+// CHECK-NEXT:    [[CONV1:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float 2.000000e+00, metadata !"fpexcept.strict") [[ATTR2]]
+// CHECK-NEXT:    br label [[COND_END]]
+// CHECK:       cond.end:
+// CHECK-NEXT:    [[COND_R:%.*]] = phi double [ [[CONV]], [[COND_TRUE]] ], [ [[CONV1]], [[COND_FALSE]] ]
+// CHECK-NEXT:    [[COND_I:%.*]] = phi double [ 0.000000e+00, [[COND_TRUE]] ], [ 0.000000e+00, [[COND_FALSE]] ]
+// CHECK-NEXT:    ret void
+//
+void t92() {
+  // Check for proper type promotion of conditional expression
+  char c[(int)(sizeof(typeof((0 ? (_Complex double) 2.0f : 2.0f))) - sizeof(_Complex double))];
+  // Check for proper codegen
+  (0 ? (_Complex double) 2.0f : 2.0f);
+}
+

clang/test/CodeGen/constrained-math-builtins.c

-// RUN: %clang_cc1 -triple x86_64-linux -ffp-exception-behavior=strict -w -S -o - -emit-llvm %s | FileCheck %s
+// RUN: %clang_cc1 -triple x86_64-linux -ffp-exception-behavior=maytrap -w -S -o - -emit-llvm %s | FileCheck %s
 
 // Test codegen of constrained math builtins.
+//
+// Test that the constrained intrinsics are picking up the exception
+// metadata from the AST instead of the global default from the command line.
+
+#pragma float_control(except, on)
 
 void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
   f = __builtin_fmod(f,f);    f = __builtin_fmodf(f,f);   f =  __builtin_fmodl(f,f); f = __builtin_fmodf128(f,f);
 
 // CHECK: declare double @llvm.experimental.constrained.frem.f64(double, double, metadata, metadata)
 // CHECK: declare float @llvm.experimental.constrained.frem.f32(float, float, metadata, metadata)
 // CHECK: declare x86_fp80 @llvm.experimental.constrained.frem.f80(x86_fp80, x86_fp80, metadata, metadata)
 // CHECK: declare fp128 @llvm.experimental.constrained.frem.f128(fp128, fp128, metadata, metadata)

 };
 
 #pragma STDC FP_CONTRACT ON
 void bar(float f) {
   f * f + f;
   (double)f * f - f;
   (long double)-f * f + f;
 
-  // CHECK: call float @llvm.experimental.constrained.fmuladd.f32
+  // CHECK: call float @llvm.experimental.constrained.fmuladd.f32(float %{{.*}}, float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
   // CHECK: fneg
-  // CHECK: call double @llvm.experimental.constrained.fmuladd.f64
+  // CHECK: call double @llvm.experimental.constrained.fmuladd.f64(double %{{.*}}, double %{{.*}}, double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
   // CHECK: fneg
-  // CHECK: call x86_fp80 @llvm.experimental.constrained.fmuladd.f80
+  // CHECK: call x86_fp80 @llvm.experimental.constrained.fmuladd.f80(x86_fp80 %{{.*}}, x86_fp80 %{{.*}}, x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 };

clang/test/CodeGen/exprs-strictfp.c

+// RUN: %clang_cc1 -triple x86_64-unknown-unknown %s -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 -triple x86_64-unknown-unknown %s -ffp-exception-behavior=maytrap -emit-llvm -o - | FileCheck %s
+
+// Test codegen of constrained floating point to bool conversion
+//
+// Test that the constrained intrinsics are picking up the exception
+// metadata from the AST instead of the global default from the command line.
+
+#pragma float_control(except, on)
+
+void eMaisUma() {
+  double t[1];
+  if (*t)
+    return;
+// CHECK: call i1 @llvm.experimental.constrained.fcmp.f64(double %{{.*}}, double 0.000000e+00, metadata !"une", metadata !"fpexcept.strict")
+}
+

clang/test/CodeGen/fp16-ops-strictfp.c

+// REQUIRES: arm-registered-target
+// RUN: %clang_cc1 -ffp-exception-behavior=maytrap -fexperimental-strict-floating-point -emit-llvm -o - -triple arm-none-linux-gnueabi %s | FileCheck %s --check-prefix=NOTNATIVE --check-prefix=CHECK -vv -dump-input=fail
+// RUN: %clang_cc1 -ffp-exception-behavior=maytrap -fexperimental-strict-floating-point -emit-llvm -o - -triple aarch64-none-linux-gnueabi %s | FileCheck %s --check-prefix=NOTNATIVE --check-prefix=CHECK
+// RUN: %clang_cc1 -ffp-exception-behavior=maytrap -fexperimental-strict-floating-point -emit-llvm -o - -triple x86_64-linux-gnu %s | FileCheck %s --check-prefix=NOTNATIVE --check-prefix=CHECK
+// RUN: %clang_cc1 -ffp-exception-behavior=maytrap -fexperimental-strict-floating-point -emit-llvm -o - -triple arm-none-linux-gnueabi -fallow-half-arguments-and-returns %s | FileCheck %s --check-prefix=NOTNATIVE --check-prefix=CHECK
+// RUN: %clang_cc1 -ffp-exception-behavior=maytrap -fexperimental-strict-floating-point -emit-llvm -o - -triple aarch64-none-linux-gnueabi -fallow-half-arguments-and-returns %s | FileCheck %s --check-prefix=NOTNATIVE --check-prefix=CHECK
+// RUN: %clang_cc1 -ffp-exception-behavior=maytrap -fexperimental-strict-floating-point -emit-llvm -o - -triple arm-none-linux-gnueabi -fnative-half-type %s \
+// RUN:   | FileCheck %s --check-prefix=NATIVE-HALF --check-prefix=CHECK
+// RUN: %clang_cc1 -ffp-exception-behavior=maytrap -fexperimental-strict-floating-point -emit-llvm -o - -triple aarch64-none-linux-gnueabi -fnative-half-type %s \
+// RUN:   | FileCheck %s --check-prefix=NATIVE-HALF --check-prefix=CHECK
+//
+// Test that the constrained intrinsics are picking up the exception
+// metadata from the AST instead of the global default from the command line.
+// FIXME: All cases of "fpexcept.maytrap" in this test are wrong.
+
+#pragma float_control(except, on)
+
+typedef unsigned cond_t;
+typedef __fp16 float16_t;
+
+volatile cond_t test;
+volatile int i0;
+volatile __fp16 h0 = 0.0, h1 = 1.0, h2;
+volatile float f0, f1, f2;
+volatile double d0;
+short s0;
+
+void foo(void) {
+  // CHECK-LABEL: define void @foo()
+
+  // Check unary ops
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i32 @llvm.experimental.constrained.fptoui.i32.f32(float %{{.*}}, metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i32 @llvm.experimental.constrained.fptoui.i32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.uitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.uitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 = (test);
+
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half 0xH0000, metadata !"une", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float 0.000000e+00, metadata !"une", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (!h1);
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: fneg float
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: fneg half
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = -h1;
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: load volatile half
+  // NATIVE-HALF-NEXT: store volatile half
+  // NOTNATIVE: store {{.*}} half {{.*}}, half*
+  h1 = +h1;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half 0xH3C00, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float {{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1++;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half 0xH3C00, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float {{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  ++h1;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half 0xHBC00, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float {{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  --h1;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half 0xHBC00, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float {{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1--;
+
+  // Check binary ops with various operands
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fmul.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = h0 * h2;
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float -2.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fmul.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = h0 * (__fp16) -2.0f;
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = h0 * f2;
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = f0 * h2;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fmul.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = h0 * i0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fdiv.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h0 / h2);
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float -2.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fdiv.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h0 / (__fp16) -2.0f);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h0 / f2);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (f0 / h2);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fdiv.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h0 / i0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h2 + h0);
+
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f64(double -2.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = ((__fp16)-2.0 + h0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h2 + f0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (f2 + h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h0 + i0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fsub.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h2 - h0);
+
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float -2.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fsub.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = ((__fp16)-2.0f - h0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h2 - f0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (f2 - h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fsub.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h0 - i0);
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h2 < h0);
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 4.200000e+01, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 4.200000e+01, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h2 < (__fp16)42.0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h2 < f0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (f2 < h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (i0 < h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"olt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 < i0);
+
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 > h2);
+
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 4.200000e+01, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 4.200000e+01, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = ((__fp16)42.0 > h2);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 > f2);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (f0 > h2);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (i0 > h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ogt", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 > i0);
+
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h2 <= h0);
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 4.200000e+01, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 4.200000e+01, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %98, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h2 <= (__fp16)42.0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h2 <= f0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (f2 <= h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (i0 <= h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"ole", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 <= i0);
+
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 >= h2);
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f64(double -2.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 >= (__fp16)-2.0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 >= f2);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (f0 >= h2);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (i0 >= h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmps.f16(half %{{.*}}, half %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmps.f32(float %{{.*}}, float %{{.*}}, metadata !"oge", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 >= i0);
+
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h1 == h2);
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %122, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h1 == (__fp16)1.0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h1 == f1);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (f1 == h1);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (i0 == h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"oeq", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 == i0);
+
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h1 != h2);
+
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h1 != (__fp16)1.0);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h1 != f1);
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (f1 != h1);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (i0 != h0);
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float %{{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  test = (h0 != i0);
+
+  // NATIVE-HALF: call i1 @llvm.experimental.constrained.fcmp.f16(half %{{.*}}, half 0xH0000, metadata !"une", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i1 @llvm.experimental.constrained.fcmp.f32(float %{{.*}}, float {{.*}}, metadata !"une", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h1 = (h1 ? h2 : h0);
+
+  // Check assignments (inc. compound)
+  // CHECK: store {{.*}} half {{.*}}, half*
+  // xATIVE-HALF: store {{.*}} half 0xHC000 // FIXME: We should be folding here.
+  h0 = h1;
+
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float -2.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 = (__fp16)-2.0f;
+
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 = f0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 = i0;
+
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %{{.*}}, metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  i0 = h0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 += h1;
+
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fptrunc.f16.f32(float 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 += (__fp16)1.0f;
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // CHECK: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 += f2;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %{{.*}}, metadata !"fpexcept.maytrap") 
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %{{.*}}, metadata !"fpexcept.maytrap") 
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  i0 += h0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fadd.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fadd.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 += i0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fsub.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 -= h1;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fsub.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 -= (__fp16)1.0;
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // CHECK: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 -= f2;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fsub.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %{{.*}}, metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  i0 -= h0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fsub.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fsub.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 -= i0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fmul.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 *= h1;
+
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fmul.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 *= (__fp16)1.0;
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // CHECK: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 *= f2;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fmul.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %{{.*}}, metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  i0 *= h0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fmul.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fmul.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 *= i0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fdiv.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 /= h1;
+
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fptrunc.f16.f64(double 1.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fdiv.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 /= (__fp16)1.0;
+
+  // CHECK: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // CHECK: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 /= f2;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fdiv.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %{{.*}}, metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} i32 {{.*}}, i32*
+  i0 /= h0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NATIVE-HALF: call half @llvm.experimental.constrained.fdiv.f16(half %{{.*}}, half %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.maytrap")
+  // NOTNATIVE: call float @llvm.experimental.constrained.fdiv.f32(float %{{.*}}, float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 /= i0;
+
+  // Check conversions to/from double
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 = d0;
+
+  // CHECK: call float @llvm.experimental.constrained.fptrunc.f32.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 = (float)d0;
+
+  // CHECK: call double @llvm.experimental.constrained.fpext.f64.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} double {{.*}}, double*
+  d0 = h0;
+
+  // CHECK: [[MID:%.*]] = call float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
+  // CHECK: call double @llvm.experimental.constrained.fpext.f64.f32(float [[MID]], metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} double {{.*}}, double*
+  d0 = (float)h0;
+
+  // NATIVE-HALF: call half @llvm.experimental.constrained.sitofp.f16.i16(i16 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call float @llvm.experimental.constrained.sitofp.f32.i16(i16 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // NOTNATIVE: call half @llvm.experimental.constrained.fptrunc.f16.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK: store {{.*}} half {{.*}}, half*
+  h0 = s0;
+}
+
+// CHECK-LABEL: define void @testTypeDef(
+// NATIVE-HALF: call <4 x half> @llvm.experimental.constrained.fadd.v4f16(<4 x half> %{{.*}}, <4 x half> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// NOTNATIVE: %[[CONV:.*]] = call <4 x float> @llvm.experimental.constrained.fpext.v4f32.v4f16(<4 x half> %{{.*}}, metadata !"fpexcept.strict")
+// NOTNATIVE: %[[CONV1:.*]] = call <4 x float> @llvm.experimental.constrained.fpext.v4f32.v4f16(<4 x half> %{{.*}}, metadata !"fpexcept.strict")
+// NOTNATIVE: %[[ADD:.*]] = call <4 x float> @llvm.experimental.constrained.fadd.v4f32(<4 x float> %conv, <4 x float> %conv1, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// NOTNATIVE: call <4 x half> @llvm.experimental.constrained.fptrunc.v4f16.v4f32(<4 x float> %add, metadata !"round.tonearest", metadata !"fpexcept.strict")
+
+void testTypeDef() {
+  __fp16 t0 __attribute__((vector_size(8)));
+  float16_t t1 __attribute__((vector_size(8)));
+  t1 = t0 + t1;
+}
+

llvm/include/llvm/IR/IRBuilder.h

public:
   /// by this setting.
   void setIsFPConstrained(bool IsCon) { IsFPConstrained = IsCon; }
 
   /// Query for the use of constrained floating point math
   bool getIsFPConstrained() { return IsFPConstrained; }
 
   /// Set the exception handling to be used with constrained floating point
   void setDefaultConstrainedExcept(fp::ExceptionBehavior NewExcept) {
+#ifndef NDEBUG
+    Optional<StringRef> ExceptStr = ExceptionBehaviorToStr(NewExcept);
+    assert(ExceptStr.hasValue() && "Garbage strict exception behavior!");
+#endif
     DefaultConstrainedExcept = NewExcept;
   }
 
   /// Set the rounding mode handling to be used with constrained floating point
   void setDefaultConstrainedRounding(RoundingMode NewRounding) {
+#ifndef NDEBUG
+    Optional<StringRef> RoundingStr = RoundingModeToStr(NewRounding);
+    assert(RoundingStr.hasValue() && "Garbage strict rounding mode!");
+#endif
     DefaultConstrainedRounding = NewRounding;
   }
 
   /// Get the exception handling used with constrained floating point
   fp::ExceptionBehavior getDefaultConstrainedExcept() {
     return DefaultConstrainedExcept;
   }