[FPEnv] Use single enum to represent rounding mode

[sepavloff]

Summary

Now compiler defines 5 sets of constants to represent rounding mode.
These are:

1. llvm::APFloatBase::roundingMode. It specifies all 5 rounding modes

defined by IEEE-754 and is used in APFloat implementation.

2. clang::LangOptions::FPRoundingModeKind. It specifies 4 of 5 IEEE-754

rounding modes and a special value for dynamic rounding mode. It is used
in clang frontend.

3. llvm::fp::RoundingMode. Defines the same values as

clang::LangOptions::FPRoundingModeKind but in different order. It is
used to specify rounding mode in IR and functions that operate IR.

4. Rounding mode representation used by FLT_ROUNDS (C11, 5.2.4.2.2p7).

Besides constants for rounding mode it also defines a special value to
indicate errors. It is convenient to use in intrinsic functions, as it
is a platform-independent representation for rounding mode. In this
role it is used in some pending patches.

5. Values like FE_DOWNWARD and other, which specify rounding mode in

library calls fesetround and fegetround. Often they represent bits
of some control register, so they are target-dependent. The same names
(not values) and a special name FE_DYNAMIC are used in
#pragma STDC FENV_ROUND.

The first 4 sets of constants are target independent and could have the
same numerical representation. It would simplify conversion between the
representations. Also now clang::LangOptions::FPRoundingModeKind and
llvm::fp::RoundingMode do not contain the value for IEEE-754 rounding
direction roundTiesToAway, although it is supported natively on
some targets.

This change defines all the rounding mode types via one enumeration
llvm::RoundingMode, which also contains rounding mode for IEEE
rounding direction roundTiesToAway. This enumeration uses the encoding
specified by C standard for FLT_ROUNDS, as it is the only representation
in which numerical values of rounding modes is fixed.

[rjmccall]

llvm/include/llvm/ADT/FloatingPointMode.h

26

I agree that we should use one enum across LLVM and Clang. I'm not sure that using the FLT_ROUNDS values is worthwhile, especially since (1) FLT_ROUNDS doesn't specify a value for some of these (like NearestTiesToAway) and (2) some of the values it does use (e.g. for "indeterminable") make this actively more awkward to store. And the most useful thing we could do — matching the values of FE_TONEAREST and so on — isn't possible because those values are unportable. I'd rather we just pick arbitrary, non-ABI-stable values, like we normally would, and then make the places that rely on matching some external schema translate.

[sepavloff]

Updated patch

• enhancements of LangOptions.def are moved into separate patch,
• fixed some clang-format errors,
• added more comments to RoundingMode.

[sepavloff]

llvm/include/llvm/ADT/FloatingPointMode.h

26

(1) FLT_ROUNDS doesn't specify a value for some of these (like NearestTiesToAway)

In the recent C standard draft (http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2454.pdf), there is support of all 5 rounding modes including values returned by FLT_ROUNDS (5.2.4.2.2p11), values used by fegetround and fesetround (FE_TONEARESTFROMZERO in 7.6p13)

(2) some of the values it does use (e.g. for "indeterminable") make this actively more awkward to store.

This is not a rounding mode value, it is just error indicator returned by intrinsic functions, it does not need to be stored. Added comment about that.

And the most useful thing we could do — matching the values of FE_TONEAREST and so on — isn't possible because those values are unportable.

I am working on patch that implements fesetround as intrinsic function. It introduces two intrinsic functions, one is llvm.set_rounding (D74729 [FPEnv] Intrinsic for setting rounding mode) and the other is llvm.convert_rounding (unpublished yet). The latter translates platform-dependent values like FE_DOWNWARD into platform independent representation, which is the same as used by FLT_ROUNDS.

Actually the motivation for this patch was just the need to have platform-independent representation of rounding mode that could be used in IR, which is more or less platform-independent. The representation used by FLT_ROUNDS fits these purposes because:

• it is platform-neutral,
• it is defined by standard, and
• it encodes all IEEE rounding modes.

[rjmccall]

llvm/include/llvm/ADT/FloatingPointMode.h

26

Okay. I'm just worried that trying to match FLT_ROUNDS in our internal representation is ultimately going to cause unnecessary problems. If C has standardized a value for NearesetTiesToAway, that certainly helps avoid that. FLT_ROUNDS allows targets to add implementation-defined values; are there any targets that support other rounding modes that aren't currently described?

[sepavloff]

llvm/include/llvm/ADT/FloatingPointMode.h

26

There are 10 possible "arithmetic" rounding modes (https://upload.wikimedia.org/wikipedia/commons/8/8a/Comparison_rounding_graphs_SMIL.svg). Current implementation of glibc does not define anything beyond the standard. I know there are cores that use non-IEEE modes "stochastic rounding to nearest" and "away from zero", don't know if they expose these modes through standard interface like FLT_ROUNDS. It looks like there is no precedent of extending the value set returned by FLT_ROUNDS.

[rjmccall]

llvm/include/llvm/ADT/FloatingPointMode.h

26

Okay, thanks, that's the kind of thing I meant. I have no objection to matching FLT_ROUNDS for the basic IEEE set of rounding modes that we have today — hey, we might as well — but we should acknowledge that fundamentally this is not the same as FLT_ROUNDS, and that places that need to generate a FLT_ROUNDS value may need a conversion, even if that conversion is likely very simple. For example, if targets with non-IEEE rounding modes wanted to support them in LLVM, we will just assign them the next available enumerators in RoundingMode, and they can separately make a policy decision about what they want to return from FLT_ROUNDS (either -1 or some implementation-defined value).

[sepavloff]

Removed dependency

[sepavloff]

Updated patch

• Fixed comment,
• Mention "round.tonearestaway" in documentation.

[sepavloff]

llvm/include/llvm/ADT/FloatingPointMode.h

26

Yes, we can require identical rounding mode definitions only for standard modes, mentioned in the standard. Fixed comment accordingly.

[rjmccall]

Okay. I can accept this.

clang/include/clang/Basic/LangOptions.h

 //===- LangOptions.h - C Language Family Language Options -------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file
 /// Defines the clang::LangOptions interface.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_CLANG_BASIC_LANGOPTIONS_H
 #define LLVM_CLANG_BASIC_LANGOPTIONS_H
 
 #include "clang/Basic/CommentOptions.h"
 #include "clang/Basic/LLVM.h"
 #include "clang/Basic/ObjCRuntime.h"
 #include "clang/Basic/Sanitizers.h"
 #include "clang/Basic/Visibility.h"
+#include "llvm/ADT/FloatingPointMode.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/Triple.h"
 #include <string>
 #include <vector>
 
 namespace clang {
 
 /// Bitfields of LangOptions, split out from LangOptions in order to ensure that

 /// members used to implement virtual inheritance.
 enum class MSVtorDispMode { Never, ForVBaseOverride, ForVFTable };
 
 /// Keeps track of the various options that can be
 /// enabled, which controls the dialect of C or C++ that is accepted.
 class LangOptions : public LangOptionsBase {
 public:
   using Visibility = clang::Visibility;
+  using RoundingMode = llvm::RoundingMode;
 
   enum GCMode { NonGC, GCOnly, HybridGC };
   enum StackProtectorMode { SSPOff, SSPOn, SSPStrong, SSPReq };
 
   // Automatic variables live on the stack, and when trivial they're usually
   // uninitialized because it's undefined behavior to use them without
   // initializing them.
   enum class TrivialAutoVarInitKind { Uninitialized, Zero, Pattern };
public:
 
   // TODO: merge FEnvAccessModeKind and FPContractModeKind
   enum FEnvAccessModeKind {
     FEA_Off,
 
     FEA_On
   };
 
-  // Values of the following enumerations correspond to metadata arguments
-  // specified for constrained floating-point intrinsics:
-  // http://llvm.org/docs/LangRef.html#constrained-floating-point-intrinsics.
-
-  /// Possible rounding modes.
-  enum FPRoundingModeKind {
-    /// Rounding to nearest, corresponds to "round.tonearest".
-    FPR_ToNearest,
-    /// Rounding toward -Inf, corresponds to "round.downward".
-    FPR_Downward,
-    /// Rounding toward +Inf, corresponds to "round.upward".
-    FPR_Upward,
-    /// Rounding toward zero, corresponds to "round.towardzero".
-    FPR_TowardZero,
-    /// Is determined by runtime environment, corresponds to "round.dynamic".
-    FPR_Dynamic
-  };
+  /// Alias for RoundingMode::NearestTiesToEven.
+  static constexpr unsigned FPR_ToNearest =
+      static_cast<unsigned>(llvm::RoundingMode::NearestTiesToEven);
 
   /// Possible floating point exception behavior.
   enum FPExceptionModeKind {
     /// Assume that floating-point exceptions are masked.
     FPE_Ignore,
     /// Transformations do not cause new exceptions but may hide some.
     FPE_MayTrap,
     /// Strictly preserve the floating-point exception semantics.
#include "clang/Basic/LangOptions.def"
   /// Check if leaf functions are also signed.
   bool isSignReturnAddressScopeAll() const {
     return getSignReturnAddressScope() == SignReturnAddressScopeKind::All;
   }
 };
 
 /// Floating point control options
 class FPOptions {
+  using RoundingMode = llvm::RoundingMode;
+
 public:
   FPOptions() : fp_contract(LangOptions::FPC_Off),
                 fenv_access(LangOptions::FEA_Off),
                 rounding(LangOptions::FPR_ToNearest),
                 exceptions(LangOptions::FPE_Ignore)
         {}
 
   // Used for serializing.
   explicit FPOptions(unsigned I)
-      : fp_contract(static_cast<LangOptions::FPContractModeKind>(I & 3)),
-        fenv_access(static_cast<LangOptions::FEnvAccessModeKind>((I >> 2) & 1)),
-        rounding(static_cast<LangOptions::FPRoundingModeKind>((I >> 3) & 7)),
-        exceptions(static_cast<LangOptions::FPExceptionModeKind>((I >> 6) & 3))
+      : fp_contract(I & 3),
+        fenv_access((I >> 2) & 1),
+        rounding   ((I >> 3) & 7),
+        exceptions ((I >> 6) & 3)
         {}
 
   explicit FPOptions(const LangOptions &LangOpts)
       : fp_contract(LangOpts.getDefaultFPContractMode()),
         fenv_access(LangOptions::FEA_Off),
         rounding(LangOptions::FPR_ToNearest),
         exceptions(LangOptions::FPE_Ignore)
         {}
public:
   }
 
   void setAllowFEnvAccess() {
     fenv_access = LangOptions::FEA_On;
   }
 
   void setDisallowFEnvAccess() { fenv_access = LangOptions::FEA_Off; }
 
-  LangOptions::FPRoundingModeKind getRoundingMode() const {
-    return static_cast<LangOptions::FPRoundingModeKind>(rounding);
+  RoundingMode getRoundingMode() const {
+    return static_cast<RoundingMode>(rounding);
   }
 
-  void setRoundingMode(LangOptions::FPRoundingModeKind RM) {
-    rounding = RM;
+  void setRoundingMode(RoundingMode RM) {
+    rounding = static_cast<unsigned>(RM);
   }
 
   LangOptions::FPExceptionModeKind getExceptionMode() const {
     return static_cast<LangOptions::FPExceptionModeKind>(exceptions);
   }
 
   void setExceptionMode(LangOptions::FPExceptionModeKind EM) {
     exceptions = EM;
   }
 
   bool isFPConstrained() const {
-    return getRoundingMode() != LangOptions::FPR_ToNearest ||
+    return getRoundingMode() != RoundingMode::NearestTiesToEven ||
            getExceptionMode() != LangOptions::FPE_Ignore ||
            allowFEnvAccess();
   }
 
   /// Used to serialize this.
   unsigned getInt() const {
     return fp_contract | (fenv_access << 2) | (rounding << 3)
         | (exceptions << 6);
   }

clang/include/clang/Basic/LangOptions.def

 BENIGN_LANGOPT(DebuggerCastResultToId, 1, 0, "for 'po' in the debugger, cast the result to id if it is of unknown type")
 BENIGN_LANGOPT(DebuggerObjCLiteral , 1, 0, "debugger Objective-C literals and subscripting support")
 
 BENIGN_LANGOPT(SpellChecking , 1, 1, "spell-checking")
 LANGOPT(SinglePrecisionConstants , 1, 0, "treating double-precision floating point constants as single precision constants")
 LANGOPT(FastRelaxedMath , 1, 0, "OpenCL fast relaxed math")
 /// FP_CONTRACT mode (on/off/fast).
 ENUM_LANGOPT(DefaultFPContractMode, FPContractModeKind, 2, FPC_Off, "FP contraction type")
-ENUM_LANGOPT(FPRoundingMode, FPRoundingModeKind, 3, FPR_ToNearest, "FP Rounding Mode type")
+ENUM_LANGOPT(FPRoundingMode, RoundingMode, 3, RoundingMode::NearestTiesToEven, "FP Rounding Mode type")
 ENUM_LANGOPT(FPExceptionMode, FPExceptionModeKind, 2, FPE_Ignore, "FP Exception Behavior Mode type")
 LANGOPT(NoBitFieldTypeAlign , 1, 0, "bit-field type alignment")
 LANGOPT(HexagonQdsp6Compat , 1, 0, "hexagon-qdsp6 backward compatibility")
 LANGOPT(ObjCAutoRefCount , 1, 0, "Objective-C automated reference counting")
 LANGOPT(ObjCWeakRuntime     , 1, 0, "__weak support in the ARC runtime")
 LANGOPT(ObjCWeak            , 1, 0, "Objective-C __weak in ARC and MRC files")
 LANGOPT(ObjCSubscriptingLegacyRuntime         , 1, 0, "Subscripting support in legacy ObjectiveC runtime")
 LANGOPT(CFProtectionBranch , 1, 0, "Control-Flow Branch Protection enabled")

clang/include/clang/Sema/Sema.h

 //===--- Sema.h - Semantic Analysis & AST Building --------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //

   /// \#pragma clang fp contract
   void ActOnPragmaFPContract(LangOptions::FPContractModeKind FPC);
 
   /// ActOnPragmaFenvAccess - Called on well formed
   /// \#pragma STDC FENV_ACCESS
   void ActOnPragmaFEnvAccess(LangOptions::FEnvAccessModeKind FPC);
 
   /// Called to set rounding mode for floating point operations.
-  void setRoundingMode(LangOptions::FPRoundingModeKind);
+  void setRoundingMode(llvm::RoundingMode);
 
   /// Called to set exception behavior for floating point operations.
   void setExceptionMode(LangOptions::FPExceptionModeKind);
 
   /// AddAlignmentAttributesForRecord - Adds any needed alignment attributes to
   /// a the record decl, to handle '\#pragma pack' and '\#pragma options align'.
   void AddAlignmentAttributesForRecord(RecordDecl *RD);
 

clang/lib/CodeGen/CodeGenFunction.cpp

 //===--- CodeGenFunction.cpp - Emit LLVM Code from ASTs for a Function ----===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
CodeGenFunction::~CodeGenFunction() {
   // outlining etc) at some point. Doing it once the function codegen is done
   // seems to be a reasonable spot. We do it here, as opposed to the deletion
   // time of the CodeGenModule, because we have to ensure the IR has not yet
   // been "emitted" to the outside, thus, modifications are still sensible.
   if (llvm::OpenMPIRBuilder *OMPBuilder = CGM.getOpenMPIRBuilder())
     OMPBuilder->finalize();
 }
 
-// Map the LangOption for rounding mode into
-// the corresponding enum in the IR.
-static llvm::fp::RoundingMode ToConstrainedRoundingMD(
-  LangOptions::FPRoundingModeKind Kind) {
-
-  switch (Kind) {
-  case LangOptions::FPR_ToNearest:  return llvm::fp::rmToNearest;
-  case LangOptions::FPR_Downward:   return llvm::fp::rmDownward;
-  case LangOptions::FPR_Upward:     return llvm::fp::rmUpward;
-  case LangOptions::FPR_TowardZero: return llvm::fp::rmTowardZero;
-  case LangOptions::FPR_Dynamic:    return llvm::fp::rmDynamic;
-  }
-  llvm_unreachable("Unsupported FP RoundingMode");
-}
-
 // Map the LangOption for exception behavior into
 // the corresponding enum in the IR.
 static llvm::fp::ExceptionBehavior ToConstrainedExceptMD(
   LangOptions::FPExceptionModeKind Kind) {
 
   switch (Kind) {
   case LangOptions::FPE_Ignore:  return llvm::fp::ebIgnore;
   case LangOptions::FPE_MayTrap: return llvm::fp::ebMayTrap;
   case LangOptions::FPE_Strict:  return llvm::fp::ebStrict;
   }
   llvm_unreachable("Unsupported FP Exception Behavior");
 }
 
 void CodeGenFunction::SetFPModel() {
-  auto fpRoundingMode = ToConstrainedRoundingMD(
-                          getLangOpts().getFPRoundingMode());
+  llvm::RoundingMode RM = getLangOpts().getFPRoundingMode();
   auto fpExceptionBehavior = ToConstrainedExceptMD(
                                getLangOpts().getFPExceptionMode());
 
   if (fpExceptionBehavior == llvm::fp::ebIgnore &&
-      fpRoundingMode == llvm::fp::rmToNearest)
+      RM == llvm::RoundingMode::NearestTiesToEven)
     // Constrained intrinsics are not used.
     ;
   else {
     Builder.setIsFPConstrained(true);
-    Builder.setDefaultConstrainedRounding(fpRoundingMode);
+    Builder.setDefaultConstrainedRounding(RM);
     Builder.setDefaultConstrainedExcept(fpExceptionBehavior);
   }
 }
 
 CharUnits CodeGenFunction::getNaturalPointeeTypeAlignment(QualType T,
                                                     LValueBaseInfo *BaseInfo,
                                                     TBAAAccessInfo *TBAAInfo) {
   return getNaturalTypeAlignment(T->getPointeeType(), BaseInfo, TBAAInfo,

clang/lib/Frontend/CompilerInvocation.cpp

 //===- CompilerInvocation.cpp ---------------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 

 #include "clang/Lex/PreprocessorOptions.h"
 #include "clang/Sema/CodeCompleteOptions.h"
 #include "clang/Serialization/ASTBitCodes.h"
 #include "clang/Serialization/ModuleFileExtension.h"
 #include "clang/StaticAnalyzer/Core/AnalyzerOptions.h"
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/CachedHashString.h"
+#include "llvm/ADT/FloatingPointMode.h"
 #include "llvm/ADT/Hashing.h"
 #include "llvm/ADT/None.h"
 #include "llvm/ADT/Optional.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/ADT/Triple.h"
if (Arg *A = Args.getLastArg(OPT_ffp_contract)) {
     else if (Val == "on")
       Opts.setDefaultFPContractMode(LangOptions::FPC_On);
     else if (Val == "off")
       Opts.setDefaultFPContractMode(LangOptions::FPC_Off);
     else
       Diags.Report(diag::err_drv_invalid_value) << A->getAsString(Args) << Val;
   }
 
-  LangOptions::FPRoundingModeKind FPRM = LangOptions::FPR_ToNearest;
+  auto FPRM = llvm::RoundingMode::NearestTiesToEven;
   if (Args.hasArg(OPT_frounding_math)) {
-    FPRM = LangOptions::FPR_Dynamic;
+    FPRM = llvm::RoundingMode::Dynamic;
   }
   Opts.setFPRoundingMode(FPRM);
 
   if (Args.hasArg(OPT_ftrapping_math)) {
     Opts.setFPExceptionMode(LangOptions::FPE_Strict);
   }
 
   if (Args.hasArg(OPT_fno_trapping_math)) {

clang/lib/Sema/SemaAttr.cpp

 //===--- SemaAttr.cpp - Semantic Analysis for Attributes ------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
case LangOptions::FPC_Fast:
     FPFeatures.setAllowFPContractAcrossStatement();
     break;
   case LangOptions::FPC_Off:
     FPFeatures.setDisallowFPContract();
     break;
   }
 }
 
-void Sema::setRoundingMode(LangOptions::FPRoundingModeKind FPR) {
+void Sema::setRoundingMode(llvm::RoundingMode FPR) {
   FPFeatures.setRoundingMode(FPR);
 }
 
 void Sema::setExceptionMode(LangOptions::FPExceptionModeKind FPE) {
   FPFeatures.setExceptionMode(FPE);
 }
 
 void Sema::ActOnPragmaFEnvAccess(LangOptions::FEnvAccessModeKind FPC) {

clang/lib/Sema/SemaExpr.cpp

 //===--- SemaExpr.cpp - Semantic Analysis for Expressions -----------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //

 #include "clang/Sema/ScopeInfo.h"
 #include "clang/Sema/SemaFixItUtils.h"
 #include "clang/Sema/SemaInternal.h"
 #include "clang/Sema/Template.h"
 #include "llvm/Support/ConvertUTF.h"
 #include "llvm/Support/SaveAndRestore.h"
 using namespace clang;
 using namespace sema;
+using llvm::RoundingMode;
 
 /// Determine whether the use of this declaration is valid, without
 /// emitting diagnostics.
 bool Sema::CanUseDecl(NamedDecl *D, bool TreatUnavailableAsInvalid) {
   // See if this is an auto-typed variable whose initializer we are parsing.
   if (ParsingInitForAutoVars.count(D))
     return false;
 
if (getLangOpts().CPlusPlus && !RHS.isInvalid()) {
       OK = RHS.get()->getObjectKind();
     }
     break;
   }
   if (ResultTy.isNull() || LHS.isInvalid() || RHS.isInvalid())
     return ExprError();
 
   if (ResultTy->isRealFloatingType() &&
-      (getLangOpts().getFPRoundingMode() != LangOptions::FPR_ToNearest ||
+      (getLangOpts().getFPRoundingMode() != RoundingMode::NearestTiesToEven ||
        getLangOpts().getFPExceptionMode() != LangOptions::FPE_Ignore))
     // Mark the current function as usng floating point constrained intrinsics
     if (FunctionDecl *F = dyn_cast<FunctionDecl>(CurContext)) {
       F->setUsesFPIntrin(true);
     }
 
   // Some of the binary operations require promoting operands of half vector to
   // float vectors and truncating the result back to half vector. For now, we do

clang/lib/Serialization/ASTReader.cpp

 //===- ASTReader.cpp - AST File Reader ------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 //  This file defines the ASTReader class, which reads AST files.
 //
 //===----------------------------------------------------------------------===//
 
 #include "clang/Basic/OpenMPKinds.h"
 #include "clang/Serialization/ASTRecordReader.h"
 #include "ASTCommon.h"
 #include "ASTReaderInternals.h"
 #include "clang/AST/AbstractTypeReader.h"
 #include "clang/AST/ASTConsumer.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/ASTMutationListener.h"
 #include "clang/AST/ASTUnresolvedSet.h"
 #include "clang/AST/Decl.h"
 #include "clang/AST/DeclBase.h"
 #include "clang/AST/DeclCXX.h"
 #include "clang/AST/DeclFriend.h"
 #include "clang/AST/DeclGroup.h"
 #include "clang/AST/DeclObjC.h"
 #include "clang/AST/DeclTemplate.h"
 #include "clang/AST/DeclarationName.h"
 #include "clang/AST/Expr.h"
 #include "clang/AST/ExprCXX.h"
 #include "clang/AST/ExternalASTSource.h"
 #include "clang/AST/NestedNameSpecifier.h"
 #include "clang/AST/OpenMPClause.h"
 #include "clang/AST/ODRHash.h"
 #include "clang/AST/RawCommentList.h"
 #include "clang/AST/TemplateBase.h"
 #include "clang/AST/TemplateName.h"
 #include "clang/AST/Type.h"
 #include "clang/AST/TypeLoc.h"
 #include "clang/AST/TypeLocVisitor.h"
 #include "clang/AST/UnresolvedSet.h"
 #include "clang/Basic/CommentOptions.h"
 #include "clang/Basic/Diagnostic.h"
 #include "clang/Basic/DiagnosticOptions.h"
 #include "clang/Basic/ExceptionSpecificationType.h"
 #include "clang/Basic/FileManager.h"
 #include "clang/Basic/FileSystemOptions.h"
 #include "clang/Basic/IdentifierTable.h"
 #include "clang/Basic/LLVM.h"
 #include "clang/Basic/LangOptions.h"
 #include "clang/Basic/Module.h"
 #include "clang/Basic/ObjCRuntime.h"
 #include "clang/Basic/OperatorKinds.h"
 #include "clang/Basic/PragmaKinds.h"
 #include "clang/Basic/Sanitizers.h"
 #include "clang/Basic/SourceLocation.h"
 #include "clang/Basic/SourceManager.h"
 #include "clang/Basic/SourceManagerInternals.h"
 #include "clang/Basic/Specifiers.h"
 #include "clang/Basic/TargetInfo.h"
 #include "clang/Basic/TargetOptions.h"
 #include "clang/Basic/TokenKinds.h"
 #include "clang/Basic/Version.h"
 #include "clang/Lex/HeaderSearch.h"
 #include "clang/Lex/HeaderSearchOptions.h"
 #include "clang/Lex/MacroInfo.h"
 #include "clang/Lex/ModuleMap.h"
 #include "clang/Lex/PreprocessingRecord.h"
 #include "clang/Lex/Preprocessor.h"
 #include "clang/Lex/PreprocessorOptions.h"
 #include "clang/Lex/Token.h"
 #include "clang/Sema/ObjCMethodList.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/Sema.h"
 #include "clang/Sema/Weak.h"
 #include "clang/Serialization/ASTBitCodes.h"
 #include "clang/Serialization/ASTDeserializationListener.h"
 #include "clang/Serialization/ContinuousRangeMap.h"
 #include "clang/Serialization/GlobalModuleIndex.h"
 #include "clang/Serialization/InMemoryModuleCache.h"
 #include "clang/Serialization/ModuleFile.h"
 #include "clang/Serialization/ModuleFileExtension.h"
 #include "clang/Serialization/ModuleManager.h"
 #include "clang/Serialization/PCHContainerOperations.h"
 #include "clang/Serialization/SerializationDiagnostic.h"
 #include "llvm/ADT/APFloat.h"
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/APSInt.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/FloatingPointMode.h"
 #include "llvm/ADT/FoldingSet.h"
 #include "llvm/ADT/Hashing.h"
 #include "llvm/ADT/IntrusiveRefCntPtr.h"
 #include "llvm/ADT/None.h"
 #include "llvm/ADT/Optional.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/ScopeExit.h"
 #include "llvm/ADT/SmallPtrSet.h"

 #include <tuple>
 #include <utility>
 #include <vector>
 
 using namespace clang;
 using namespace clang::serialization;
 using namespace clang::serialization::reader;
 using llvm::BitstreamCursor;
+using llvm::RoundingMode;
 
 //===----------------------------------------------------------------------===//
 // ChainedASTReaderListener implementation
 //===----------------------------------------------------------------------===//
 
 bool
 ChainedASTReaderListener::ReadFullVersionInformation(StringRef FullVersion) {
   return First->ReadFullVersionInformation(FullVersion) ||

llvm/docs/LangRef.rst

 
 ::
 
       "round.dynamic"
       "round.tonearest"
       "round.downward"
       "round.upward"
       "round.towardzero"
+      "round.tonearestaway"
 
 If this argument is "round.dynamic" optimization passes must assume that the
 rounding mode is unknown and may change at runtime.  No transformations that
 depend on rounding mode may be performed in this case.
 
 The other possible values for the rounding mode argument correspond to the
 similarly named IEEE rounding modes.  If the argument is any of these values
 optimization passes may perform transformations as long as they are consistent

llvm/include/llvm/ADT/APFloat.h

 //===- llvm/ADT/APFloat.h - Arbitrary Precision Floating Point ---*- C++ -*-==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 ///
 /// \file
 /// \brief
 /// This file declares a class to represent arbitrary precision floating point
 /// values and provide a variety of arithmetic operations on them.
 ///
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_ADT_APFLOAT_H
 #define LLVM_ADT_APFLOAT_H
 
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/FloatingPointMode.h"
 #include "llvm/Support/ErrorHandling.h"
 #include <memory>
 
 #define APFLOAT_DISPATCH_ON_SEMANTICS(METHOD_CALL)                             \
   do {                                                                         \
     if (usesLayout<IEEEFloat>(getSemantics()))                                 \
       return U.IEEE.METHOD_CALL;                                               \
     if (usesLayout<DoubleAPFloat>(getSemantics()))                             \
struct APFloatBase {
   enum cmpResult {
     cmpLessThan,
     cmpEqual,
     cmpGreaterThan,
     cmpUnordered
   };
 
   /// IEEE-754R 4.3: Rounding-direction attributes.
-  enum roundingMode {
-    rmNearestTiesToEven,
-    rmTowardPositive,
-    rmTowardNegative,
-    rmTowardZero,
-    rmNearestTiesToAway
-  };
+  using roundingMode = llvm::RoundingMode;
+
+  static constexpr roundingMode rmNearestTiesToEven =
+                                                RoundingMode::NearestTiesToEven;
+  static constexpr roundingMode rmTowardPositive = RoundingMode::TowardPositive;
+  static constexpr roundingMode rmTowardNegative = RoundingMode::TowardNegative;
+  static constexpr roundingMode rmTowardZero     = RoundingMode::TowardZero;
+  static constexpr roundingMode rmNearestTiesToAway =
+                                                RoundingMode::NearestTiesToAway;
 
   /// IEEE-754R 7: Default exception handling.
   ///
   /// opUnderflow or opOverflow are always returned or-ed with opInexact.
   ///
   /// APFloat models this behavior specified by IEEE-754:
   ///   "For operations producing results in floating-point format, the default
   ///    result of an operation that signals the invalid operation exception

llvm/include/llvm/ADT/FloatingPointMode.h

 //===- llvm/Support/FloatingPointMode.h -------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // Utilities for dealing with flags related to floating point mode controls.
 //
 //===----------------------------------------------------------------------===/
 
 #ifndef LLVM_FLOATINGPOINTMODE_H
 #define LLVM_FLOATINGPOINTMODE_H
 
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/Support/raw_ostream.h"
 
 namespace llvm {
 
+/// Rounding mode.
+///
+/// Enumerates supported rounding modes, as well as some special values. The set
+/// of the modes must agree with IEEE-754, 4.3.1 and 4.3.2. The constants
+/// assigned to the IEEE rounding modes must agree with the values used by
+/// FLT_ROUNDS (C11, 5.2.4.2.2p8).
+///
+/// This value is packed into bitfield in some cases, including \c FPOptions, so
+/// the rounding mode values and the special value \c Dynamic must fit into the
+/// the bit field (now - 3 bits). The value \c Invalid is used only in values
+/// returned by intrinsics to indicate errors, it should never be stored as
+/// rounding mode value, so it does not need to fit the bit fields.
+///
+enum class RoundingMode : int8_t {
+  // Rounding mode defined in IEEE-754.
+  TowardZero        = 0,    ///< roundTowardZero.
+  NearestTiesToEven = 1,    ///< roundTiesToEven.
+  TowardPositive    = 2,    ///< roundTowardPositive.
+  TowardNegative    = 3,    ///< roundTowardNegative.
+  NearestTiesToAway = 4,    ///< roundTiesToAway.
+
+  // Special values.
+  Dynamic = 7,    ///< Denotes mode unknown at compile time.
+  Invalid = -1    ///< Denotes invalid value.
+};
+
 /// Represent ssubnormal handling kind for floating point instruction inputs and
 /// outputs.
 struct DenormalMode {
   /// Represent handled modes for denormal (aka subnormal) modes in the floating
   /// point environment.
   enum DenormalModeKind : int8_t {
     Invalid = -1,
 

llvm/include/llvm/IR/FPEnv.h

 //===- FPEnv.h ---- FP Environment ------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //

 #include "llvm/ADT/Optional.h"
 #include "llvm/ADT/StringRef.h"
 #include <stdint.h>
 
 namespace llvm {
 
 namespace fp {
 
-/// Rounding mode used for floating point operations.
-///
-/// Each of these values correspond to some metadata argument value of a
-/// constrained floating point intrinsic. See the LLVM Language Reference Manual
-/// for details.
-enum RoundingMode : uint8_t {
-  rmDynamic,   ///< This corresponds to "fpround.dynamic".
-  rmToNearest, ///< This corresponds to "fpround.tonearest".
-  rmDownward,  ///< This corresponds to "fpround.downward".
-  rmUpward,    ///< This corresponds to "fpround.upward".
-  rmTowardZero ///< This corresponds to "fpround.tozero".
-};
-
 /// Exception behavior used for floating point operations.
 ///
 /// Each of these values correspond to some metadata argument value of a
 /// constrained floating point intrinsic. See the LLVM Language Reference Manual
 /// for details.
 enum ExceptionBehavior : uint8_t {
   ebIgnore,  ///< This corresponds to "fpexcept.ignore".
   ebMayTrap, ///< This corresponds to "fpexcept.maytrap".
   ebStrict   ///< This corresponds to "fpexcept.strict".
 };
 
 }
 
 /// Returns a valid RoundingMode enumerator when given a string
 /// that is valid as input in constrained intrinsic rounding mode
 /// metadata.
-Optional<fp::RoundingMode> StrToRoundingMode(StringRef);
+Optional<RoundingMode> StrToRoundingMode(StringRef);
 
 /// For any RoundingMode enumerator, returns a string valid as input in
 /// constrained intrinsic rounding mode metadata.
-Optional<StringRef> RoundingModeToStr(fp::RoundingMode);
+Optional<StringRef> RoundingModeToStr(RoundingMode);
 
 /// Returns a valid ExceptionBehavior enumerator when given a string
 /// valid as input in constrained intrinsic exception behavior metadata.
 Optional<fp::ExceptionBehavior> StrToExceptionBehavior(StringRef);
 
 /// For any ExceptionBehavior enumerator, returns a string valid as
 /// input in constrained intrinsic exception behavior metadata.
 Optional<StringRef> ExceptionBehaviorToStr(fp::ExceptionBehavior);
-
-/// Converts rounding mode represented by fp::RoundingMode to the rounding mode
-/// index used by APFloat. For fp::rmDynamic it returns None.
-Optional<APFloatBase::roundingMode> getAPFloatRoundingMode(fp::RoundingMode);
 }
 #endif

llvm/include/llvm/IR/IRBuilder.h

 //===- llvm/IRBuilder.h - Builder for LLVM Instructions ---------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
protected:
   const IRBuilderFolder &Folder;
   const IRBuilderDefaultInserter &Inserter;
 
   MDNode *DefaultFPMathTag;
   FastMathFlags FMF;
 
   bool IsFPConstrained;
   fp::ExceptionBehavior DefaultConstrainedExcept;
-  fp::RoundingMode DefaultConstrainedRounding;
+  RoundingMode DefaultConstrainedRounding;
 
   ArrayRef<OperandBundleDef> DefaultOperandBundles;
 
 public:
   IRBuilderBase(LLVMContext &context, const IRBuilderFolder &Folder,
                 const IRBuilderDefaultInserter &Inserter,
                 MDNode *FPMathTag, ArrayRef<OperandBundleDef> OpBundles)
       : Context(context), Folder(Folder), Inserter(Inserter),
         DefaultFPMathTag(FPMathTag), IsFPConstrained(false),
         DefaultConstrainedExcept(fp::ebStrict),
-        DefaultConstrainedRounding(fp::rmDynamic),
+        DefaultConstrainedRounding(RoundingMode::Dynamic),
         DefaultOperandBundles(OpBundles) {
     ClearInsertionPoint();
   }
 
   /// Insert and return the specified instruction.
   template<typename InstTy>
   InstTy *Insert(InstTy *I, const Twine &Name = "") const {
     Inserter.InsertHelper(I, Name, BB, InsertPt);
public:
   bool getIsFPConstrained() { return IsFPConstrained; }
 
   /// Set the exception handling to be used with constrained floating point
   void setDefaultConstrainedExcept(fp::ExceptionBehavior NewExcept) {
     DefaultConstrainedExcept = NewExcept;
   }
 
   /// Set the rounding mode handling to be used with constrained floating point
-  void setDefaultConstrainedRounding(fp::RoundingMode NewRounding) {
+  void setDefaultConstrainedRounding(RoundingMode NewRounding) {
     DefaultConstrainedRounding = NewRounding;
   }
 
   /// Get the exception handling used with constrained floating point
   fp::ExceptionBehavior getDefaultConstrainedExcept() {
     return DefaultConstrainedExcept;
   }
 
   /// Get the rounding mode handling used with constrained floating point
-  fp::RoundingMode getDefaultConstrainedRounding() {
+  RoundingMode getDefaultConstrainedRounding() {
     return DefaultConstrainedRounding;
   }
 
   void setConstrainedFPFunctionAttr() {
     assert(BB && "Must have a basic block to set any function attributes!");
 
     Function *F = BB->getParent();
     if (!F->hasFnAttribute(Attribute::StrictFP)) {
private:
 
   Value *foldConstant(Instruction::BinaryOps Opc, Value *L,
                       Value *R, const Twine &Name) const {
     auto *LC = dyn_cast<Constant>(L);
     auto *RC = dyn_cast<Constant>(R);
     return (LC && RC) ? Insert(Folder.CreateBinOp(Opc, LC, RC), Name) : nullptr;
   }
 
-  Value *getConstrainedFPRounding(Optional<fp::RoundingMode> Rounding) {
-    fp::RoundingMode UseRounding = DefaultConstrainedRounding;
+  Value *getConstrainedFPRounding(Optional<RoundingMode> Rounding) {
+    RoundingMode UseRounding = DefaultConstrainedRounding;
 
     if (Rounding.hasValue())
       UseRounding = Rounding.getValue();
 
     Optional<StringRef> RoundingStr = RoundingModeToStr(UseRounding);
     assert(RoundingStr.hasValue() && "Garbage strict rounding mode!");
     auto *RoundingMDS = MDString::get(Context, RoundingStr.getValue());
 
Value *CreateFMulFMF(Value *L, Value *R, Instruction *FMFSource,
     if (IsFPConstrained)
       return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fmul,
                                       L, R, FMFSource, Name);
 
     if (Value *V = foldConstant(Instruction::FMul, L, R, Name)) return V;
     Instruction *I = setFPAttrs(BinaryOperator::CreateFMul(L, R), nullptr,
                                 FMFSource->getFastMathFlags());
     return Insert(I, Name);
   }
 
   Value *CreateFDiv(Value *L, Value *R, const Twine &Name = "",
                     MDNode *FPMD = nullptr) {
     if (IsFPConstrained)
       return CreateConstrainedFPBinOp(Intrinsic::experimental_constrained_fdiv,
                                       L, R, nullptr, Name, FPMD);
 
     if (Value *V = foldConstant(Instruction::FDiv, L, R, Name)) return V;
Value *CreateBinOp(Instruction::BinaryOps Opc,
     if (isa<FPMathOperator>(BinOp))
       setFPAttrs(BinOp, FPMathTag, FMF);
     return Insert(BinOp, Name);
   }
 
   CallInst *CreateConstrainedFPBinOp(
       Intrinsic::ID ID, Value *L, Value *R, Instruction *FMFSource = nullptr,
       const Twine &Name = "", MDNode *FPMathTag = nullptr,
-      Optional<fp::RoundingMode> Rounding = None,
+      Optional<RoundingMode> Rounding = None,
       Optional<fp::ExceptionBehavior> Except = None);
 
   Value *CreateNeg(Value *V, const Twine &Name = "",
                    bool HasNUW = false, bool HasNSW = false) {
     if (auto *VC = dyn_cast<Constant>(V))
       return Insert(Folder.CreateNeg(VC, HasNUW, HasNSW), Name);
     BinaryOperator *BO = Insert(BinaryOperator::CreateNeg(V), Name);
     if (HasNUW) BO->setHasNoUnsignedWrap();
public:
   Value *CreateTruncOrBitCast(Value *V, Type *DestTy,
                               const Twine &Name = "") {
     if (V->getType() == DestTy)
       return V;
     if (auto *VC = dyn_cast<Constant>(V))
       return Insert(Folder.CreateTruncOrBitCast(VC, DestTy), Name);
     return Insert(CastInst::CreateTruncOrBitCast(V, DestTy), Name);
   }
 
   Value *CreateCast(Instruction::CastOps Op, Value *V, Type *DestTy,
                     const Twine &Name = "") {
     if (V->getType() == DestTy)
       return V;
     if (auto *VC = dyn_cast<Constant>(V))
       return Insert(Folder.CreateCast(Op, VC, DestTy), Name);
     return Insert(CastInst::Create(Op, V, DestTy), Name);
   }
if (auto *VC = dyn_cast<Constant>(V))
       return Insert(Folder.CreateFPCast(VC, DestTy), Name);
     return Insert(CastInst::CreateFPCast(V, DestTy), Name);
   }
 
   CallInst *CreateConstrainedFPCast(
       Intrinsic::ID ID, Value *V, Type *DestTy,
       Instruction *FMFSource = nullptr, const Twine &Name = "",
       MDNode *FPMathTag = nullptr,
-      Optional<fp::RoundingMode> Rounding = None,
+      Optional<RoundingMode> Rounding = None,
       Optional<fp::ExceptionBehavior> Except = None);
 
   // Provided to resolve 'CreateIntCast(Ptr, Ptr, "...")', giving a
   // compile time error, instead of converting the string to bool for the
   // isSigned parameter.
   Value *CreateIntCast(Value *, Type *, const char *) = delete;
 
   //===--------------------------------------------------------------------===//

 
 public:
   CallInst *CreateConstrainedFPCmp(
       Intrinsic::ID ID, CmpInst::Predicate P, Value *L, Value *R,
       const Twine &Name = "", Optional<fp::ExceptionBehavior> Except = None);
 
   //===--------------------------------------------------------------------===//
   // Instruction creation methods: Other Instructions
   //===--------------------------------------------------------------------===//
 
   PHINode *CreatePHI(Type *Ty, unsigned NumReservedValues,
                      const Twine &Name = "") {
     PHINode *Phi = PHINode::Create(Ty, NumReservedValues);
     if (isa<FPMathOperator>(Phi))
       setFPAttrs(Phi, nullptr /* MDNode* */, FMF);
     return Insert(Phi, Name);
   }
CallInst *CreateCall(Value *Callee, ArrayRef<Value *> Args,
                        const Twine &Name = "", MDNode *FPMathTag = nullptr) {
     return CreateCall(
         cast<FunctionType>(Callee->getType()->getPointerElementType()), Callee,
         Args, OpBundles, Name, FPMathTag);
   }
 
   CallInst *CreateConstrainedFPCall(
       Function *Callee, ArrayRef<Value *> Args, const Twine &Name = "",
-      Optional<fp::RoundingMode> Rounding = None,
+      Optional<RoundingMode> Rounding = None,
       Optional<fp::ExceptionBehavior> Except = None);
 
   Value *CreateSelect(Value *C, Value *True, Value *False,
                       const Twine &Name = "", Instruction *MDFrom = nullptr);
 
   VAArgInst *CreateVAArg(Value *List, Type *Ty, const Twine &Name = "") {
     return Insert(new VAArgInst(List, Ty), Name);
   }

llvm/include/llvm/IR/IntrinsicInst.h

 //===-- llvm/IntrinsicInst.h - Intrinsic Instruction Wrappers ---*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
public:
     static unsigned GetFunctionalOpcodeForVP(Intrinsic::ID ID);
   };
 
   /// This is the common base class for constrained floating point intrinsics.
   class ConstrainedFPIntrinsic : public IntrinsicInst {
   public:
     bool isUnaryOp() const;
     bool isTernaryOp() const;
-    Optional<fp::RoundingMode> getRoundingMode() const;
+    Optional<RoundingMode> getRoundingMode() const;
     Optional<fp::ExceptionBehavior> getExceptionBehavior() const;
 
     // Methods for support type inquiry through isa, cast, and dyn_cast:
     static bool classof(const IntrinsicInst *I);
     static bool classof(const Value *V) {
       return isa<IntrinsicInst>(V) && classof(cast<IntrinsicInst>(V));
     }
   };

llvm/lib/Analysis/ConstantFolding.cpp

 //===-- ConstantFolding.cpp - Fold instructions into constants ------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
if (auto *Op = dyn_cast<ConstantFP>(Operands[0])) {
 
     Optional<APFloat::roundingMode> RM;
     switch (IntrinsicID) {
     default:
       break;
     case Intrinsic::experimental_constrained_nearbyint:
     case Intrinsic::experimental_constrained_rint: {
       auto CI = cast<ConstrainedFPIntrinsic>(Call);
-      Optional<fp::RoundingMode> RMOp = CI->getRoundingMode();
-      if (RMOp)
-        RM = getAPFloatRoundingMode(*RMOp);
-      if (!RM)
+      RM = CI->getRoundingMode();
+      if (!RM || RM.getValue() == RoundingMode::Dynamic)
         return nullptr;
       break;
     }
     case Intrinsic::experimental_constrained_round:
       RM = APFloat::rmNearestTiesToAway;
       break;
     case Intrinsic::experimental_constrained_ceil:
       RM = APFloat::rmTowardPositive;

llvm/lib/IR/FPEnv.cpp

 //===-- FPEnv.cpp ---- FP Environment -------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// @file
 /// This file contains the implementations of entities that describe floating
 /// point environment.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/IR/FPEnv.h"
 
 namespace llvm {
 
-Optional<fp::RoundingMode> StrToRoundingMode(StringRef RoundingArg) {
+Optional<RoundingMode> StrToRoundingMode(StringRef RoundingArg) {
   // For dynamic rounding mode, we use round to nearest but we will set the
   // 'exact' SDNodeFlag so that the value will not be rounded.
-  return StringSwitch<Optional<fp::RoundingMode>>(RoundingArg)
-      .Case("round.dynamic", fp::rmDynamic)
-      .Case("round.tonearest", fp::rmToNearest)
-      .Case("round.downward", fp::rmDownward)
-      .Case("round.upward", fp::rmUpward)
-      .Case("round.towardzero", fp::rmTowardZero)
+  return StringSwitch<Optional<RoundingMode>>(RoundingArg)
+      .Case("round.dynamic", RoundingMode::Dynamic)
+      .Case("round.tonearest", RoundingMode::NearestTiesToEven)
+      .Case("round.tonearestaway", RoundingMode::NearestTiesToAway)
+      .Case("round.downward", RoundingMode::TowardNegative)
+      .Case("round.upward", RoundingMode::TowardPositive)
+      .Case("round.towardzero", RoundingMode::TowardZero)
       .Default(None);
 }
 
-Optional<StringRef> RoundingModeToStr(fp::RoundingMode UseRounding) {
+Optional<StringRef> RoundingModeToStr(RoundingMode UseRounding) {
   Optional<StringRef> RoundingStr = None;
   switch (UseRounding) {
-  case fp::rmDynamic:
+  case RoundingMode::Dynamic:
     RoundingStr = "round.dynamic";
     break;
-  case fp::rmToNearest:
+  case RoundingMode::NearestTiesToEven:
     RoundingStr = "round.tonearest";
     break;
-  case fp::rmDownward:
+  case RoundingMode::NearestTiesToAway:
+    RoundingStr = "round.tonearestaway";
+    break;
+  case RoundingMode::TowardNegative:
     RoundingStr = "round.downward";
     break;
-  case fp::rmUpward:
+  case RoundingMode::TowardPositive:
     RoundingStr = "round.upward";
     break;
-  case fp::rmTowardZero:
+  case RoundingMode::TowardZero:
     RoundingStr = "round.towardzero";
     break;
+  default:
+    break;
   }
   return RoundingStr;
 }
 
 Optional<fp::ExceptionBehavior> StrToExceptionBehavior(StringRef ExceptionArg) {
   return StringSwitch<Optional<fp::ExceptionBehavior>>(ExceptionArg)
       .Case("fpexcept.ignore", fp::ebIgnore)
       .Case("fpexcept.maytrap", fp::ebMayTrap)
case fp::ebIgnore:
     ExceptStr = "fpexcept.ignore";
     break;
   case fp::ebMayTrap:
     ExceptStr = "fpexcept.maytrap";
     break;
   }
   return ExceptStr;
 }
-
-Optional<APFloatBase::roundingMode>
-getAPFloatRoundingMode(fp::RoundingMode RM) {
-  switch (RM) {
-  case fp::rmDynamic:
-    return None;
-  case fp::rmToNearest:
-    return APFloat::rmNearestTiesToEven;
-  case fp::rmDownward:
-    return APFloat::rmTowardNegative;
-  case fp::rmUpward:
-    return APFloat::rmTowardPositive;
-  case fp::rmTowardZero:
-    return APFloat::rmTowardZero;
-  }
-  llvm_unreachable("Unexpected rounding mode");
-}
 }

llvm/lib/IR/IRBuilder.cpp

 //===- IRBuilder.cpp - Builder for LLVM Instrs ----------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
CallInst *IRBuilderBase::CreateBinaryIntrinsic(Intrinsic::ID ID, Value *LHS,
                                                const Twine &Name) {
   Module *M = BB->getModule();
   Function *Fn = Intrinsic::getDeclaration(M, ID, { LHS->getType() });
   return createCallHelper(Fn, {LHS, RHS}, this, Name, FMFSource);
 }
 
 CallInst *IRBuilderBase::CreateIntrinsic(Intrinsic::ID ID,
                                          ArrayRef<Type *> Types,
                                          ArrayRef<Value *> Args,
                                          Instruction *FMFSource,
                                          const Twine &Name) {
   Module *M = BB->getModule();
   Function *Fn = Intrinsic::getDeclaration(M, ID, Types);
   return createCallHelper(Fn, Args, this, Name, FMFSource);
 }
 
 CallInst *IRBuilderBase::CreateConstrainedFPBinOp(
     Intrinsic::ID ID, Value *L, Value *R, Instruction *FMFSource,
     const Twine &Name, MDNode *FPMathTag,
-    Optional<fp::RoundingMode> Rounding,
+    Optional<RoundingMode> Rounding,
     Optional<fp::ExceptionBehavior> Except) {
   Value *RoundingV = getConstrainedFPRounding(Rounding);
   Value *ExceptV = getConstrainedFPExcept(Except);
 
   FastMathFlags UseFMF = FMF;
   if (FMFSource)
     UseFMF = FMFSource->getFastMathFlags();
 
   CallInst *C = CreateIntrinsic(ID, {L->getType()},
                                 {L, R, RoundingV, ExceptV}, nullptr, Name);
   setConstrainedFPCallAttr(C);
   setFPAttrs(C, FPMathTag, UseFMF);
   return C;
 }
 
 Value *IRBuilderBase::CreateNAryOp(unsigned Opc, ArrayRef<Value *> Ops,
                                    const Twine &Name, MDNode *FPMathTag) {
   if (Instruction::isBinaryOp(Opc)) {
     assert(Ops.size() == 2 && "Invalid number of operands!");
     return CreateBinOp(static_cast<Instruction::BinaryOps>(Opc),
                        Ops[0], Ops[1], Name, FPMathTag);
   }
   if (Instruction::isUnaryOp(Opc)) {
     assert(Ops.size() == 1 && "Invalid number of operands!");
     return CreateUnOp(static_cast<Instruction::UnaryOps>(Opc),
                       Ops[0], Name, FPMathTag);
   }
   llvm_unreachable("Unexpected opcode!");
 }
 
 CallInst *IRBuilderBase::CreateConstrainedFPCast(
     Intrinsic::ID ID, Value *V, Type *DestTy,
     Instruction *FMFSource, const Twine &Name, MDNode *FPMathTag,
-    Optional<fp::RoundingMode> Rounding,
+    Optional<RoundingMode> Rounding,
     Optional<fp::ExceptionBehavior> Except) {
   Value *ExceptV = getConstrainedFPExcept(Except);
 
   FastMathFlags UseFMF = FMF;
   if (FMFSource)
     UseFMF = FMFSource->getFastMathFlags();
 
   CallInst *C;
if (auto *RC = dyn_cast<Constant>(RHS))
       return Insert(Folder.CreateFCmp(P, LC, RC), Name);
   return Insert(setFPAttrs(new FCmpInst(P, LHS, RHS), FPMathTag, FMF), Name);
 }
 
 CallInst *IRBuilderBase::CreateConstrainedFPCmp(
     Intrinsic::ID ID, CmpInst::Predicate P, Value *L, Value *R,
     const Twine &Name, Optional<fp::ExceptionBehavior> Except) {
   Value *PredicateV = getConstrainedFPPredicate(P);
   Value *ExceptV = getConstrainedFPExcept(Except);
 
   CallInst *C = CreateIntrinsic(ID, {L->getType()},
                                 {L, R, PredicateV, ExceptV}, nullptr, Name);
   setConstrainedFPCallAttr(C);
   return C;
 }
 
 CallInst *IRBuilderBase::CreateConstrainedFPCall(
     Function *Callee, ArrayRef<Value *> Args, const Twine &Name,
-    Optional<fp::RoundingMode> Rounding,
+    Optional<RoundingMode> Rounding,
     Optional<fp::ExceptionBehavior> Except) {
   llvm::SmallVector<Value *, 6> UseArgs;
 
   for (auto *OneArg : Args)
     UseArgs.push_back(OneArg);
   bool HasRoundingMD = false;
   switch (Callee->getIntrinsicID()) {
   default:

llvm/lib/IR/IntrinsicInst.cpp

 //===-- InstrinsicInst.cpp - Intrinsic Instruction Wrappers ---------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
Value *InstrProfIncrementInst::getStep() const {
   if (InstrProfIncrementInstStep::classof(this)) {
     return const_cast<Value *>(getArgOperand(4));
   }
   const Module *M = getModule();
   LLVMContext &Context = M->getContext();
   return ConstantInt::get(Type::getInt64Ty(Context), 1);
 }
 
-Optional<fp::RoundingMode> ConstrainedFPIntrinsic::getRoundingMode() const {
+Optional<RoundingMode> ConstrainedFPIntrinsic::getRoundingMode() const {
   unsigned NumOperands = getNumArgOperands();
   Metadata *MD =
       cast<MetadataAsValue>(getArgOperand(NumOperands - 2))->getMetadata();
   if (!MD || !isa<MDString>(MD))
     return None;
   return StrToRoundingMode(cast<MDString>(MD)->getString());
 }
 

llvm/lib/Support/APFloat.cpp

 //===-- APFloat.cpp - Implement APFloat class -----------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
namespace llvm {
   }
   const fltSemantics &APFloatBase::Bogus() {
     return semBogus;
   }
   const fltSemantics &APFloatBase::PPCDoubleDouble() {
     return semPPCDoubleDouble;
   }
 
+  constexpr RoundingMode APFloatBase::rmNearestTiesToEven;
+  constexpr RoundingMode APFloatBase::rmTowardPositive;
+  constexpr RoundingMode APFloatBase::rmTowardNegative;
+  constexpr RoundingMode APFloatBase::rmTowardZero;
+  constexpr RoundingMode APFloatBase::rmNearestTiesToAway;
+
   /* A tight upper bound on number of parts required to hold the value
      pow(5, power) is
 
        power * 815 / (351 * integerPartWidth) + 1
 
      However, whilst the result may require only this many parts,
      because we are multiplying two values to get it, the
      multiplication may require an extra part with the excess part
bool IEEEFloat::roundAwayFromZero(roundingMode rounding_mode,
   case rmTowardZero:
     return false;
 
   case rmTowardPositive:
     return !sign;
 
   case rmTowardNegative:
     return sign;
+
+  default:
+    break;
   }
   llvm_unreachable("Invalid rounding mode found");
 }
 
 IEEEFloat::opStatus IEEEFloat::normalize(roundingMode rounding_mode,
                                          lostFraction lost_fraction) {
   unsigned int omsb;                /* One, not zero, based MSB.  */
   int exponentChange;

llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp

 //===- InstCombineAddSub.cpp ------------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //

 void FAddendCoef::operator=(const FAddendCoef &That) {
   if (That.isInt())
     set(That.IntVal);
   else
     set(That.getFpVal());
 }
 
 void FAddendCoef::operator+=(const FAddendCoef &That) {
-  enum APFloat::roundingMode RndMode = APFloat::rmNearestTiesToEven;
+  RoundingMode RndMode = RoundingMode::NearestTiesToEven;
   if (isInt() == That.isInt()) {
     if (isInt())
       IntVal += That.IntVal;
     else
       getFpVal().add(That.getFpVal(), RndMode);
     return;
   }
 

llvm/unittests/IR/IRBuilderTest.cpp

 //===- llvm/unittest/IR/IRBuilderTest.cpp - IRBuilder tests ---------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
TEST_F(IRBuilderTest, ConstrainedFP) {
   AttributeSet FnAttrs = Attrs.getFnAttributes();
   EXPECT_EQ(FnAttrs.hasAttribute(Attribute::StrictFP), true);
 
   // Verify the codepaths for setting and overriding the default metadata.
   V = Builder.CreateFAdd(V, V);
   ASSERT_TRUE(isa<ConstrainedFPIntrinsic>(V));
   auto *CII = cast<ConstrainedFPIntrinsic>(V);
   EXPECT_EQ(fp::ebStrict, CII->getExceptionBehavior());
-  EXPECT_EQ(fp::rmDynamic, CII->getRoundingMode());
+  EXPECT_EQ(RoundingMode::Dynamic, CII->getRoundingMode());
 
   Builder.setDefaultConstrainedExcept(fp::ebIgnore);
-  Builder.setDefaultConstrainedRounding(fp::rmUpward);
+  Builder.setDefaultConstrainedRounding(RoundingMode::TowardPositive);
   V = Builder.CreateFAdd(V, V);
   CII = cast<ConstrainedFPIntrinsic>(V);
   EXPECT_EQ(fp::ebIgnore, CII->getExceptionBehavior());
-  EXPECT_EQ(CII->getRoundingMode(), fp::rmUpward);
+  EXPECT_EQ(CII->getRoundingMode(), RoundingMode::TowardPositive);
 
   Builder.setDefaultConstrainedExcept(fp::ebIgnore);
-  Builder.setDefaultConstrainedRounding(fp::rmToNearest);
+  Builder.setDefaultConstrainedRounding(RoundingMode::NearestTiesToEven);
   V = Builder.CreateFAdd(V, V);
   CII = cast<ConstrainedFPIntrinsic>(V);
   EXPECT_EQ(fp::ebIgnore, CII->getExceptionBehavior());
-  EXPECT_EQ(fp::rmToNearest, CII->getRoundingMode());
+  EXPECT_EQ(RoundingMode::NearestTiesToEven, CII->getRoundingMode());
 
   Builder.setDefaultConstrainedExcept(fp::ebMayTrap);
-  Builder.setDefaultConstrainedRounding(fp::rmDownward);
+  Builder.setDefaultConstrainedRounding(RoundingMode::TowardNegative);
   V = Builder.CreateFAdd(V, V);
   CII = cast<ConstrainedFPIntrinsic>(V);
   EXPECT_EQ(fp::ebMayTrap, CII->getExceptionBehavior());
-  EXPECT_EQ(fp::rmDownward, CII->getRoundingMode());
+  EXPECT_EQ(RoundingMode::TowardNegative, CII->getRoundingMode());
 
   Builder.setDefaultConstrainedExcept(fp::ebStrict);
-  Builder.setDefaultConstrainedRounding(fp::rmTowardZero);
+  Builder.setDefaultConstrainedRounding(RoundingMode::TowardZero);
   V = Builder.CreateFAdd(V, V);
   CII = cast<ConstrainedFPIntrinsic>(V);
   EXPECT_EQ(fp::ebStrict, CII->getExceptionBehavior());
-  EXPECT_EQ(fp::rmTowardZero, CII->getRoundingMode());
+  EXPECT_EQ(RoundingMode::TowardZero, CII->getRoundingMode());
 
   Builder.setDefaultConstrainedExcept(fp::ebIgnore);
-  Builder.setDefaultConstrainedRounding(fp::rmDynamic);
+  Builder.setDefaultConstrainedRounding(RoundingMode::Dynamic);
   V = Builder.CreateFAdd(V, V);
   CII = cast<ConstrainedFPIntrinsic>(V);
   EXPECT_EQ(fp::ebIgnore, CII->getExceptionBehavior());
-  EXPECT_EQ(fp::rmDynamic, CII->getRoundingMode());
+  EXPECT_EQ(RoundingMode::Dynamic, CII->getRoundingMode());
 
   // Now override the defaults.
   Call = Builder.CreateConstrainedFPBinOp(
         Intrinsic::experimental_constrained_fadd, V, V, nullptr, "", nullptr,
-        fp::rmDownward, fp::ebMayTrap);
+        RoundingMode::TowardNegative, fp::ebMayTrap);
   CII = cast<ConstrainedFPIntrinsic>(Call);
   EXPECT_EQ(CII->getIntrinsicID(), Intrinsic::experimental_constrained_fadd);
   EXPECT_EQ(fp::ebMayTrap, CII->getExceptionBehavior());
-  EXPECT_EQ(fp::rmDownward, CII->getRoundingMode());
+  EXPECT_EQ(RoundingMode::TowardNegative, CII->getRoundingMode());
 
   Builder.CreateRetVoid();
   EXPECT_FALSE(verifyModule(*M));
 }
 
 TEST_F(IRBuilderTest, Lifetime) {
   IRBuilder<> Builder(BB);
   AllocaInst *Var1 = Builder.CreateAlloca(Builder.getInt8Ty());