Add the 'initializes' attribute langref and support

llvm/docs/LangRef.rst

@@ -1625,6 +1625,27 @@ Currently, only the following parameter attributes are defined:
     ``readonly`` or a ``memory`` attribute that does not contain
     ``argmem: write``.
 
+``initializes((Lo1, Hi1), ...)``
+    This attribute indicates that the function initializes the ranges of the
+    pointer parameter's memory, ``[%p+LoN, %p+HiN)``. Initialization of memory
+    means the first memory access is a non-volatile, non-atomic write. The
+    write must happen before the function returns. If the function unwinds,
+    the write may not happen.
+
+    This attribute only holds for the memory accessed via this pointer
+    parameter. Other arbitrary accesses to the same memory via other pointers
+    are allowed.
+
+    The ``writable`` or ``dereferenceable`` attribute do not imply the
+    ``initializes`` attribute. The ``initializes`` attribute does not imply
+    ``writeonly`` since ``initializes`` allows reading from the pointer
+    after writing.
+
+    This attribute is a list of constant ranges in ascending order with no
+    overlapping or consecutive list elements. ``LoN/HiN`` are 64-bit integers,
+    and negative values are allowed in case the argument points partway into
+    an allocation. An empty list is not allowed.
+
 ``dead_on_unwind``
     At a high level, this attribute indicates that the pointer argument is dead
     if the call unwinds, in the sense that the caller will not depend on the

llvm/include/llvm/AsmParser/LLParser.h

@@ -372,6 +372,7 @@ namespace llvm {
                                     std::vector<unsigned> &FwdRefAttrGrps,
                                     bool inAttrGrp, LocTy &BuiltinLoc);
     bool parseRangeAttr(AttrBuilder &B);
+    bool parseInitializesAttr(AttrBuilder &B);
     bool parseRequiredTypeAttr(AttrBuilder &B, lltok::Kind AttrToken,
                                Attribute::AttrKind AttrKind);
 

llvm/include/llvm/Bitcode/LLVMBitCodes.h

@@ -755,6 +755,7 @@ enum AttributeKindCodes {
   ATTR_KIND_DEAD_ON_UNWIND = 91,
   ATTR_KIND_RANGE = 92,
   ATTR_KIND_SANITIZE_NUMERICAL_STABILITY = 93,
+  ATTR_KIND_INITIALIZES = 94,
 };
 
 enum ComdatSelectionKindCodes {

llvm/include/llvm/IR/Attributes.h

@@ -38,6 +38,7 @@ class AttributeImpl;
 class AttributeListImpl;
 class AttributeSetNode;
 class ConstantRange;
+class ConstantRangeList;
 class FoldingSetNodeID;
 class Function;
 class LLVMContext;
@@ -107,6 +108,10 @@ class Attribute {
   static bool isConstantRangeAttrKind(AttrKind Kind) {
     return Kind >= FirstConstantRangeAttr && Kind <= LastConstantRangeAttr;
   }
+  static bool isConstantRangeListAttrKind(AttrKind Kind) {
+    return Kind >= FirstConstantRangeListAttr &&
+           Kind <= LastConstantRangeListAttr;
+  }
 
   static bool canUseAsFnAttr(AttrKind Kind);
   static bool canUseAsParamAttr(AttrKind Kind);
@@ -131,6 +136,8 @@ class Attribute {
   static Attribute get(LLVMContext &Context, AttrKind Kind, Type *Ty);
   static Attribute get(LLVMContext &Context, AttrKind Kind,
                        const ConstantRange &CR);
+  static Attribute get(LLVMContext &Context, AttrKind Kind,
+                       const ConstantRangeList &CRL);
 
   /// Return a uniquified Attribute object that has the specific
   /// alignment set.
@@ -189,6 +196,9 @@ class Attribute {
   /// Return true if the attribute is a ConstantRange attribute.
   bool isConstantRangeAttribute() const;
 
+  /// Return true if the attribute is a ConstantRangeList attribute.
+  bool isConstantRangeListAttribute() const;
+
   /// Return true if the attribute is any kind of attribute.
   bool isValid() const { return pImpl; }
 
@@ -226,6 +236,10 @@ class Attribute {
   /// attribute to be a ConstantRange attribute.
   const ConstantRange &getValueAsConstantRange() const;
 
+  /// Return the attribute's value as a ConstantRangeList. This requires the
+  /// attribute to be a ConstantRangeList attribute.
+  ConstantRangeList getValueAsConstantRangeList() const;
+
   /// Returns the alignment field of an attribute as a byte alignment
   /// value.
   MaybeAlign getAlignment() const;
@@ -267,6 +281,9 @@ class Attribute {
   /// Returns the value of the range attribute.
   const ConstantRange &getRange() const;
 
+  /// Returns the value of the initializes attribute.
+  ConstantRangeList getInitializes() 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;
@@ -1222,6 +1239,13 @@ class AttrBuilder {
   /// Add range attribute.
   AttrBuilder &addRangeAttr(const ConstantRange &CR);
 
+  /// Add a ConstantRangeList attribute with the given ranges.
+  AttrBuilder &addConstantRangeListAttr(Attribute::AttrKind Kind,
+                                        const ConstantRangeList &CRL);
+
+  /// Add initializes attribute.
+  AttrBuilder &addInitializesAttr(const ConstantRangeList &CRL);
+
   ArrayRef<Attribute> attrs() const { return Attrs; }
 
   bool operator==(const AttrBuilder &B) const;

llvm/include/llvm/IR/Attributes.td

@@ -47,6 +47,9 @@ class ComplexStrAttr<string S, list<AttrProperty> P> : Attr<S, P>;
 /// ConstantRange attribute.
 class ConstantRangeAttr<string S, list<AttrProperty> P> : Attr<S, P>;
 
+/// ConstantRangeList attribute.
+class ConstantRangeListAttr<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.
@@ -112,6 +115,9 @@ def FnRetThunkExtern : EnumAttr<"fn_ret_thunk_extern", [FnAttr]>;
 /// Pass structure in an alloca.
 def InAlloca : TypeAttr<"inalloca", [ParamAttr]>;
 
+/// Pointer argument memory is initialized.
+def Initializes : ConstantRangeListAttr<"initializes", [ParamAttr]>;
+
 /// Source said inlining was desirable.
 def InlineHint : EnumAttr<"inlinehint", [FnAttr]>;
 

llvm/include/llvm/IR/ConstantRangeList.h

@@ -0,0 +1,92 @@
+//===- ConstantRangeList.h - A list of constant ranges ----------*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Represent a list of signed ConstantRange and do NOT support wrap around the
+// end of the numeric range. Ranges in the list are ordered and not overlapping.
+// Ranges should have the same bitwidth. Each range's lower should be less than
+// its upper.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_IR_CONSTANTRANGELIST_H
+#define LLVM_IR_CONSTANTRANGELIST_H
+
+#include "llvm/ADT/APInt.h"
+#include "llvm/IR/ConstantRange.h"
+#include "llvm/Support/Debug.h"
+#include <cstddef>
+#include <cstdint>
+
+namespace llvm {
+
+class raw_ostream;
+
+/// This class represents a list of constant ranges.
+class [[nodiscard]] ConstantRangeList {
+  SmallVector<ConstantRange, 2> Ranges;
+
+public:
+  ConstantRangeList() = default;
+  ConstantRangeList(ArrayRef<ConstantRange> RangesRef) {
+    assert(isOrderedRanges(RangesRef));
+    for (const ConstantRange &R : RangesRef) {
+      assert(R.getBitWidth() == getBitWidth());
+      Ranges.push_back(R);
+    }
+  }
+
+  // Return true if the ranges are non-overlapping and increasing.
+  static bool isOrderedRanges(ArrayRef<ConstantRange> RangesRef);
+  static std::optional<ConstantRangeList>
+  getConstantRangeList(ArrayRef<ConstantRange> RangesRef);
+
+  SmallVectorImpl<ConstantRange>::iterator begin() { return Ranges.begin(); }
+  SmallVectorImpl<ConstantRange>::iterator end() { return Ranges.end(); }
+  SmallVectorImpl<ConstantRange>::const_iterator begin() const {
+    return Ranges.begin();
+  }
+  SmallVectorImpl<ConstantRange>::const_iterator end() const {
+    return Ranges.end();
+  }
+  ConstantRange getRange(unsigned i) const { return Ranges[i]; }
+
+  /// Return true if this list contains no members.
+  bool empty() const { return Ranges.empty(); }
+
+  /// Get the bit width of this ConstantRangeList.
+  uint32_t getBitWidth() const { return 64; }
+
+  /// Return the number of ranges in this ConstantRangeList.
+  size_t size() const { return Ranges.size(); }
+
+  /// Insert a new range to Ranges and keep the list ordered.
+  void insert(const ConstantRange &NewRange);
+  void insert(int64_t Lower, int64_t Upper) {
+    insert(ConstantRange(APInt(64, Lower, /*isSigned=*/true),
+                         APInt(64, Upper, /*isSigned=*/true)));
+  }
+
+  /// Return true if this range list is equal to another range list.
+  bool operator==(const ConstantRangeList &CRL) const {
+    return Ranges == CRL.Ranges;
+  }
+  bool operator!=(const ConstantRangeList &CRL) const {
+    return !operator==(CRL);
+  }
+
+  /// Print out the ranges to a stream.
+  void print(raw_ostream &OS) const;
+
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
+  void dump() const;
+#endif
+};
+
+} // end namespace llvm
+
+#endif // LLVM_IR_CONSTANTRANGELIST_H

llvm/lib/AsmParser/LLParser.cpp

@@ -25,6 +25,7 @@
 #include "llvm/IR/CallingConv.h"
 #include "llvm/IR/Comdat.h"
 #include "llvm/IR/ConstantRange.h"
+#include "llvm/IR/ConstantRangeList.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DebugInfoMetadata.h"
 #include "llvm/IR/DerivedTypes.h"
@@ -1626,6 +1627,8 @@ bool LLParser::parseEnumAttribute(Attribute::AttrKind Attr, AttrBuilder &B,
   }
   case Attribute::Range:
     return parseRangeAttr(B);
+  case Attribute::Initializes:
+    return parseInitializesAttr(B);
   default:
     B.addAttribute(Attr);
     Lex.Lex();
@@ -3101,6 +3104,52 @@ bool LLParser::parseRangeAttr(AttrBuilder &B) {
   return false;
 }
 
+/// parseInitializesAttr
+///   ::= initializes((Lo1,Hi1),(Lo2,Hi2),...)
+bool LLParser::parseInitializesAttr(AttrBuilder &B) {
+  Lex.Lex();
+
+  auto ParseAPSInt = [&](APInt &Val) {
+    if (Lex.getKind() != lltok::APSInt)
+      return tokError("expected integer");
+    Val = Lex.getAPSIntVal().extend(64);
+    Lex.Lex();
+    return false;
+  };
+
+  if (parseToken(lltok::lparen, "expected '('"))
+    return true;
+
+  SmallVector<ConstantRange, 2> RangeList;
+  // Parse each constant range.
+  do {
+    APInt Lower, Upper;
+    if (parseToken(lltok::lparen, "expected '('"))
+      return true;
+
+    if (ParseAPSInt(Lower) || parseToken(lltok::comma, "expected ','") ||
+        ParseAPSInt(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;
+
+    RangeList.push_back(ConstantRange(Lower, Upper));
+  } while (EatIfPresent(lltok::comma));
+
+  if (parseToken(lltok::rparen, "expected ')'"))
+    return true;
+
+  auto CRLOrNull = ConstantRangeList::getConstantRangeList(RangeList);
+  if (!CRLOrNull.has_value())
+    return tokError("Invalid (unordered or overlapping) range list");
+  B.addInitializesAttr(*CRLOrNull);
+  return false;
+}
+
 /// parseOptionalOperandBundles
 ///    ::= /*empty*/
 ///    ::= '[' OperandBundle [, OperandBundle ]* ']'

llvm/lib/Bitcode/Reader/BitcodeReader.cpp

@@ -30,6 +30,7 @@
 #include "llvm/IR/CallingConv.h"
 #include "llvm/IR/Comdat.h"
 #include "llvm/IR/Constant.h"
+#include "llvm/IR/ConstantRangeList.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/DebugInfo.h"
@@ -836,10 +837,10 @@ class BitcodeReader : public BitcodeReaderBase, public GVMaterializer {
   }
 
   Expected<ConstantRange> readConstantRange(ArrayRef<uint64_t> Record,
-                                            unsigned &OpNum) {
-    if (Record.size() - OpNum < 3)
+                                            unsigned &OpNum,
+                                            unsigned BitWidth) {
+    if (Record.size() - OpNum < 2)
       return error("Too few records for range");
-    unsigned BitWidth = Record[OpNum++];
     if (BitWidth > 64) {
       unsigned LowerActiveWords = Record[OpNum];
       unsigned UpperActiveWords = Record[OpNum++] >> 32;
@@ -859,6 +860,14 @@ class BitcodeReader : public BitcodeReaderBase, public GVMaterializer {
     }
   }
 
+  Expected<ConstantRange>
+  readBitWidthAndConstantRange(ArrayRef<uint64_t> Record, unsigned &OpNum) {
+    if (Record.size() - OpNum < 1)
+      return error("Too few records for range");
+    unsigned BitWidth = Record[OpNum++];
+    return readConstantRange(Record, OpNum, BitWidth);
+  }
+
   /// 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.
@@ -2172,6 +2181,8 @@ static Attribute::AttrKind getAttrFromCode(uint64_t Code) {
     return Attribute::DeadOnUnwind;
   case bitc::ATTR_KIND_RANGE:
     return Attribute::Range;
+  case bitc::ATTR_KIND_INITIALIZES:
+    return Attribute::Initializes;
   }
 }
 
@@ -2350,12 +2361,42 @@ Error BitcodeReader::parseAttributeGroupBlock() {
           if (!Attribute::isConstantRangeAttrKind(Kind))
             return error("Not a ConstantRange attribute");
 
-          Expected<ConstantRange> MaybeCR = readConstantRange(Record, i);
+          Expected<ConstantRange> MaybeCR =
+              readBitWidthAndConstantRange(Record, i);
           if (!MaybeCR)
             return MaybeCR.takeError();
           i--;
 
           B.addConstantRangeAttr(Kind, MaybeCR.get());
+        } else if (Record[i] == 8) {
+          Attribute::AttrKind Kind;
+
+          i++;
+          if (Error Err = parseAttrKind(Record[i++], &Kind))
+            return Err;
+          if (!Attribute::isConstantRangeListAttrKind(Kind))
+            return error("Not a constant range list attribute");
+
+          SmallVector<ConstantRange, 2> Val;
+          if (i + 2 > e)
+            return error("Too few records for constant range list");
+          unsigned RangeSize = Record[i++];
+          unsigned BitWidth = Record[i++];
+          if (i + 2 * RangeSize > e)
+            return error("Incomplete constant range list");
+          for (unsigned Idx = 0; Idx < RangeSize; ++Idx) {
+            Expected<ConstantRange> MaybeCR =
+                readConstantRange(Record, i, BitWidth);
+            if (!MaybeCR)
+              return MaybeCR.takeError();
+            Val.push_back(MaybeCR.get());
+          }
+          i--;
+
+          auto CRLOrNull = ConstantRangeList::getConstantRangeList(Val);
+          if (!CRLOrNull.has_value())
+            return error("Invalid (unordered or overlapping) range list");
+          B.addConstantRangeListAttr(Kind, *CRLOrNull);
         } else {
           return error("Invalid attribute group entry");
         }
@@ -3370,7 +3411,8 @@ Error BitcodeReader::parseConstants() {
         (void)InRangeIndex;
       } else if (BitCode == bitc::CST_CODE_CE_GEP_WITH_INRANGE) {
         Flags = Record[OpNum++];
-        Expected<ConstantRange> MaybeInRange = readConstantRange(Record, OpNum);
+        Expected<ConstantRange> MaybeInRange =
+            readBitWidthAndConstantRange(Record, OpNum);
         if (!MaybeInRange)
           return MaybeInRange.takeError();
         InRange = MaybeInRange.get();