Reapply [IR] Add new Range attribute using new ConstantRange Attribute type (#84617)

The only change from #83171 is the
change of the allocator so the destructor is called for
ConstantRangeAttributeImpl.

reverts #84549

Author: andjo403

llvm/docs/LangRef.rst

@@ -1635,6 +1635,22 @@ Currently, only the following parameter attributes are defined:
 
     This attribute cannot be applied to return values.
 
+``range(<ty> <a>, <b>)``
+    This attribute expresses the possible range of the parameter or return value.
+    If the value is not in the specified range, it is converted to poison.
+    The arguments passed to ``range`` have the following properties:
+
+    -  The type must match the scalar type of the parameter or return value.
+    -  The pair ``a,b`` represents the range ``[a,b)``.
+    -  Both ``a`` and ``b`` are constants.
+    -  The range is allowed to wrap.
+    -  The range should not represent the full or empty set. That is, ``a!=b``.
+    
+    This attribute may only be applied to parameters or return values with integer 
+    or vector of integer types.
+    
+    For vector-typed parameters, the range is applied element-wise.
+
 .. _gc:
 
 Garbage Collector Strategy Names

llvm/include/llvm/ADT/FoldingSet.h

@@ -16,6 +16,7 @@
 #ifndef LLVM_ADT_FOLDINGSET_H
 #define LLVM_ADT_FOLDINGSET_H
 
+#include "llvm/ADT/APInt.h"
 #include "llvm/ADT/Hashing.h"
 #include "llvm/ADT/STLForwardCompat.h"
 #include "llvm/ADT/SmallVector.h"
@@ -354,6 +355,12 @@ class FoldingSetNodeID {
     AddInteger(unsigned(I));
     AddInteger(unsigned(I >> 32));
   }
+  void AddInteger(const APInt &Int) {
+    const auto *Parts = Int.getRawData();
+    for (int i = 0, N = Int.getNumWords(); i < N; ++i) {
+      AddInteger(Parts[i]);
+    }
+  }
 
   void AddBoolean(bool B) { AddInteger(B ? 1U : 0U); }
   void AddString(StringRef String);

llvm/include/llvm/AsmParser/LLParser.h

@@ -369,6 +369,7 @@ namespace llvm {
     bool parseFnAttributeValuePairs(AttrBuilder &B,
                                     std::vector<unsigned> &FwdRefAttrGrps,
                                     bool inAttrGrp, LocTy &BuiltinLoc);
+    bool parseRangeAttr(AttrBuilder &B);
     bool parseRequiredTypeAttr(AttrBuilder &B, lltok::Kind AttrToken,
                                Attribute::AttrKind AttrKind);
 

llvm/include/llvm/Bitcode/LLVMBitCodes.h

@@ -724,6 +724,7 @@ enum AttributeKindCodes {
   ATTR_KIND_WRITABLE = 89,
   ATTR_KIND_CORO_ONLY_DESTROY_WHEN_COMPLETE = 90,
   ATTR_KIND_DEAD_ON_UNWIND = 91,
+  ATTR_KIND_RANGE = 92,
 };
 
 enum ComdatSelectionKindCodes {

llvm/include/llvm/IR/Attributes.h

@@ -37,6 +37,7 @@ class AttributeMask;
 class AttributeImpl;
 class AttributeListImpl;
 class AttributeSetNode;
+class ConstantRange;
 class FoldingSetNodeID;
 class Function;
 class LLVMContext;
@@ -103,6 +104,9 @@ class Attribute {
   static bool isTypeAttrKind(AttrKind Kind) {
     return Kind >= FirstTypeAttr && Kind <= LastTypeAttr;
   }
+  static bool isConstantRangeAttrKind(AttrKind Kind) {
+    return Kind >= FirstConstantRangeAttr && Kind <= LastConstantRangeAttr;
+  }
 
   static bool canUseAsFnAttr(AttrKind Kind);
   static bool canUseAsParamAttr(AttrKind Kind);
@@ -125,6 +129,8 @@ class Attribute {
   static Attribute get(LLVMContext &Context, StringRef Kind,
                        StringRef Val = StringRef());
   static Attribute get(LLVMContext &Context, AttrKind Kind, Type *Ty);
+  static Attribute get(LLVMContext &Context, AttrKind Kind,
+                       const ConstantRange &CR);
 
   /// Return a uniquified Attribute object that has the specific
   /// alignment set.
@@ -180,6 +186,9 @@ class Attribute {
   /// Return true if the attribute is a type attribute.
   bool isTypeAttribute() const;
 
+  /// Return true if the attribute is a ConstantRange attribute.
+  bool isConstantRangeAttribute() const;
+
   /// Return true if the attribute is any kind of attribute.
   bool isValid() const { return pImpl; }
 
@@ -213,6 +222,10 @@ class Attribute {
   /// a type attribute.
   Type *getValueAsType() const;
 
+  /// Return the attribute's value as a ConstantRange. This requires the
+  /// attribute to be a ConstantRange attribute.
+  ConstantRange getValueAsConstantRange() const;
+
   /// Returns the alignment field of an attribute as a byte alignment
   /// value.
   MaybeAlign getAlignment() const;
@@ -251,6 +264,9 @@ class Attribute {
   /// Return the FPClassTest for nofpclass
   FPClassTest getNoFPClass() const;
 
+  /// Returns the value of the range attribute.
+  ConstantRange getRange() const;
+
   /// The Attribute is converted to a string of equivalent mnemonic. This
   /// is, presumably, for writing out the mnemonics for the assembly writer.
   std::string getAsString(bool InAttrGrp = false) const;
@@ -1189,6 +1205,13 @@ class AttrBuilder {
   // Add nofpclass attribute
   AttrBuilder &addNoFPClassAttr(FPClassTest NoFPClassMask);
 
+  /// Add a ConstantRange attribute with the given range.
+  AttrBuilder &addConstantRangeAttr(Attribute::AttrKind Kind,
+                                    const ConstantRange &CR);
+
+  /// Add range attribute.
+  AttrBuilder &addRangeAttr(const ConstantRange &CR);
+
   ArrayRef<Attribute> attrs() const { return Attrs; }
 
   bool operator==(const AttrBuilder &B) const;

llvm/include/llvm/IR/Attributes.td

@@ -44,6 +44,9 @@ class StrBoolAttr<string S> : Attr<S, []>;
 /// Arbitrary string attribute.
 class ComplexStrAttr<string S, list<AttrProperty> P> : Attr<S, P>;
 
+/// ConstantRange attribute.
+class ConstantRangeAttr<string S, list<AttrProperty> P> : Attr<S, P>;
+
 /// Target-independent enum attributes.
 
 /// Alignment of parameter (5 bits) stored as log2 of alignment with +1 bias.
@@ -218,6 +221,9 @@ def OptimizeNone : EnumAttr<"optnone", [FnAttr]>;
 /// Similar to byval but without a copy.
 def Preallocated : TypeAttr<"preallocated", [FnAttr, ParamAttr]>;
 
+/// Parameter or return value is within the specified range.
+def Range : ConstantRangeAttr<"range", [ParamAttr, RetAttr]>;
+
 /// Function does not access memory.
 def ReadNone : EnumAttr<"readnone", [ParamAttr]>;
 

llvm/lib/AsmParser/LLParser.cpp

@@ -1596,6 +1596,8 @@ bool LLParser::parseEnumAttribute(Attribute::AttrKind Attr, AttrBuilder &B,
 
     return true;
   }
+  case Attribute::Range:
+    return parseRangeAttr(B);
   default:
     B.addAttribute(Attr);
     Lex.Lex();
@@ -3008,6 +3010,47 @@ bool LLParser::parseRequiredTypeAttr(AttrBuilder &B, lltok::Kind AttrToken,
   return false;
 }
 
+/// parseRangeAttr
+///   ::= range(<ty> <n>,<n>)
+bool LLParser::parseRangeAttr(AttrBuilder &B) {
+  Lex.Lex();
+
+  APInt Lower;
+  APInt Upper;
+  Type *Ty = nullptr;
+  LocTy TyLoc;
+
+  auto ParseAPSInt = [&](unsigned BitWidth, APInt &Val) {
+    if (Lex.getKind() != lltok::APSInt)
+      return tokError("expected integer");
+    if (Lex.getAPSIntVal().getBitWidth() > BitWidth)
+      return tokError(
+          "integer is too large for the bit width of specified type");
+    Val = Lex.getAPSIntVal().extend(BitWidth);
+    Lex.Lex();
+    return false;
+  };
+
+  if (parseToken(lltok::lparen, "expected '('") || parseType(Ty, TyLoc))
+    return true;
+  if (!Ty->isIntegerTy())
+    return error(TyLoc, "the range must have integer type!");
+
+  unsigned BitWidth = Ty->getPrimitiveSizeInBits();
+
+  if (ParseAPSInt(BitWidth, Lower) ||
+      parseToken(lltok::comma, "expected ','") || ParseAPSInt(BitWidth, Upper))
+    return true;
+  if (Lower == Upper)
+    return tokError("the range should not represent the full or empty set!");
+
+  if (parseToken(lltok::rparen, "expected ')'"))
+    return true;
+
+  B.addRangeAttr(ConstantRange(Lower, Upper));
+  return false;
+}
+
 /// parseOptionalOperandBundles
 ///    ::= /*empty*/
 ///    ::= '[' OperandBundle [, OperandBundle ]* ']'

llvm/lib/Bitcode/Reader/BitcodeReader.cpp

@@ -815,6 +815,30 @@ class BitcodeReader : public BitcodeReaderBase, public GVMaterializer {
     return getFnValueByID(ValNo, Ty, TyID, ConstExprInsertBB);
   }
 
+  Expected<ConstantRange> readConstantRange(ArrayRef<uint64_t> Record,
+                                            unsigned &OpNum) {
+    if (Record.size() - OpNum < 3)
+      return error("Too few records for range");
+    unsigned BitWidth = Record[OpNum++];
+    if (BitWidth > 64) {
+      unsigned LowerActiveWords = Record[OpNum];
+      unsigned UpperActiveWords = Record[OpNum++] >> 32;
+      if (Record.size() - OpNum < LowerActiveWords + UpperActiveWords)
+        return error("Too few records for range");
+      APInt Lower =
+          readWideAPInt(ArrayRef(&Record[OpNum], LowerActiveWords), BitWidth);
+      OpNum += LowerActiveWords;
+      APInt Upper =
+          readWideAPInt(ArrayRef(&Record[OpNum], UpperActiveWords), BitWidth);
+      OpNum += UpperActiveWords;
+      return ConstantRange(Lower, Upper);
+    } else {
+      int64_t Start = BitcodeReader::decodeSignRotatedValue(Record[OpNum++]);
+      int64_t End = BitcodeReader::decodeSignRotatedValue(Record[OpNum++]);
+      return ConstantRange(APInt(BitWidth, Start), APInt(BitWidth, End));
+    }
+  }
+
   /// Upgrades old-style typeless byval/sret/inalloca attributes by adding the
   /// corresponding argument's pointee type. Also upgrades intrinsics that now
   /// require an elementtype attribute.
@@ -2103,6 +2127,8 @@ static Attribute::AttrKind getAttrFromCode(uint64_t Code) {
     return Attribute::CoroDestroyOnlyWhenComplete;
   case bitc::ATTR_KIND_DEAD_ON_UNWIND:
     return Attribute::DeadOnUnwind;
+  case bitc::ATTR_KIND_RANGE:
+    return Attribute::Range;
   }
 }
 
@@ -2272,6 +2298,21 @@ Error BitcodeReader::parseAttributeGroupBlock() {
             return error("Not a type attribute");
 
           B.addTypeAttr(Kind, HasType ? getTypeByID(Record[++i]) : nullptr);
+        } else if (Record[i] == 7) {
+          Attribute::AttrKind Kind;
+
+          i++;
+          if (Error Err = parseAttrKind(Record[i++], &Kind))
+            return Err;
+          if (!Attribute::isConstantRangeAttrKind(Kind))
+            return error("Not a ConstantRange attribute");
+
+          Expected<ConstantRange> MaybeCR = readConstantRange(Record, i);
+          if (!MaybeCR)
+            return MaybeCR.takeError();
+          i--;
+
+          B.addConstantRangeAttr(Kind, MaybeCR.get());
         } else {
           return error("Invalid attribute group entry");
         }

llvm/lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -844,6 +844,8 @@ static uint64_t getAttrKindEncoding(Attribute::AttrKind Kind) {
     return bitc::ATTR_KIND_CORO_ONLY_DESTROY_WHEN_COMPLETE;
   case Attribute::DeadOnUnwind:
     return bitc::ATTR_KIND_DEAD_ON_UNWIND;
+  case Attribute::Range:
+    return bitc::ATTR_KIND_RANGE;
   case Attribute::EndAttrKinds:
     llvm_unreachable("Can not encode end-attribute kinds marker.");
   case Attribute::None:
@@ -856,6 +858,39 @@ static uint64_t getAttrKindEncoding(Attribute::AttrKind Kind) {
   llvm_unreachable("Trying to encode unknown attribute");
 }
 
+static void emitSignedInt64(SmallVectorImpl<uint64_t> &Vals, uint64_t V) {
+  if ((int64_t)V >= 0)
+    Vals.push_back(V << 1);
+  else
+    Vals.push_back((-V << 1) | 1);
+}
+
+static void emitWideAPInt(SmallVectorImpl<uint64_t> &Vals, const APInt &A) {
+  // We have an arbitrary precision integer value to write whose
+  // bit width is > 64. However, in canonical unsigned integer
+  // format it is likely that the high bits are going to be zero.
+  // So, we only write the number of active words.
+  unsigned NumWords = A.getActiveWords();
+  const uint64_t *RawData = A.getRawData();
+  for (unsigned i = 0; i < NumWords; i++)
+    emitSignedInt64(Vals, RawData[i]);
+}
+
+static void emitConstantRange(SmallVectorImpl<uint64_t> &Record,
+                              const ConstantRange &CR) {
+  unsigned BitWidth = CR.getBitWidth();
+  Record.push_back(BitWidth);
+  if (BitWidth > 64) {
+    Record.push_back(CR.getLower().getActiveWords() |
+                     (uint64_t(CR.getUpper().getActiveWords()) << 32));
+    emitWideAPInt(Record, CR.getLower());
+    emitWideAPInt(Record, CR.getUpper());
+  } else {
+    emitSignedInt64(Record, CR.getLower().getSExtValue());
+    emitSignedInt64(Record, CR.getUpper().getSExtValue());
+  }
+}
+
 void ModuleBitcodeWriter::writeAttributeGroupTable() {
   const std::vector<ValueEnumerator::IndexAndAttrSet> &AttrGrps =
       VE.getAttributeGroups();
@@ -889,13 +924,17 @@ void ModuleBitcodeWriter::writeAttributeGroupTable() {
           Record.append(Val.begin(), Val.end());
           Record.push_back(0);
         }
-      } else {
-        assert(Attr.isTypeAttribute());
+      } else if (Attr.isTypeAttribute()) {
         Type *Ty = Attr.getValueAsType();
         Record.push_back(Ty ? 6 : 5);
         Record.push_back(getAttrKindEncoding(Attr.getKindAsEnum()));
         if (Ty)
           Record.push_back(VE.getTypeID(Attr.getValueAsType()));
+      } else {
+        assert(Attr.isConstantRangeAttribute());
+        Record.push_back(7);
+        Record.push_back(getAttrKindEncoding(Attr.getKindAsEnum()));
+        emitConstantRange(Record, Attr.getValueAsConstantRange());
       }
     }
 
@@ -1716,24 +1755,6 @@ void ModuleBitcodeWriter::writeDIGenericSubrange(
   Record.clear();
 }
 
-static void emitSignedInt64(SmallVectorImpl<uint64_t> &Vals, uint64_t V) {
-  if ((int64_t)V >= 0)
-    Vals.push_back(V << 1);
-  else
-    Vals.push_back((-V << 1) | 1);
-}
-
-static void emitWideAPInt(SmallVectorImpl<uint64_t> &Vals, const APInt &A) {
-  // We have an arbitrary precision integer value to write whose
-  // bit width is > 64. However, in canonical unsigned integer
-  // format it is likely that the high bits are going to be zero.
-  // So, we only write the number of active words.
-  unsigned NumWords = A.getActiveWords();
-  const uint64_t *RawData = A.getRawData();
-  for (unsigned i = 0; i < NumWords; i++)
-    emitSignedInt64(Vals, RawData[i]);
-}
-
 void ModuleBitcodeWriter::writeDIEnumerator(const DIEnumerator *N,
                                             SmallVectorImpl<uint64_t> &Record,
                                             unsigned Abbrev) {