[clang][Interp] Support arbitrary precision constants

[tbaederr]

Add (de)serialization support for them, like we do for Floating values.

Unfortunately I wasn't able to eliminate the code duplication coming from the signed/unsigned split here. I was trying to avoid adding a third function and calling that from the other two, but maybe that's the best thing to do.

[llvmbot]

@llvm/pr-subscribers-platform-windows
@llvm/pr-subscribers-llvm-support

@llvm/pr-subscribers-clang

Author: Timm Baeder (tbaederr)

Changes

Add (de)serialization support for them, like we do for Floating values.

Unfortunately I wasn't able to eliminate the code duplication coming from the signed/unsigned split here. I was trying to avoid adding a third function and calling that from the other two, but maybe that's the best thing to do.

---

Full diff: https://github.com/llvm/llvm-project/pull/79747.diff

7 Files Affected:

• (modified) clang/lib/AST/Interp/ByteCodeEmitter.cpp (+37)
• (modified) clang/lib/AST/Interp/ByteCodeExprGen.cpp (+7-4)
• (modified) clang/lib/AST/Interp/Disasm.cpp (+15)
• (modified) clang/lib/AST/Interp/IntegralAP.h (+29)
• (modified) clang/lib/AST/Interp/Interp.h (+16)
• (modified) clang/lib/AST/Interp/Opcodes.td (+4)
• (modified) clang/test/AST/Interp/intap.cpp (+14)

diff --git a/clang/lib/AST/Interp/ByteCodeEmitter.cpp b/clang/lib/AST/Interp/ByteCodeEmitter.cpp
index fd2a92d9d3f91e0..4f7ad51773a6546 100644
--- a/clang/lib/AST/Interp/ByteCodeEmitter.cpp
+++ b/clang/lib/AST/Interp/ByteCodeEmitter.cpp
@@ -10,6 +10,7 @@
 #include "ByteCodeGenError.h"
 #include "Context.h"
 #include "Floating.h"
+#include "IntegralAP.h"
 #include "Opcode.h"
 #include "Program.h"
 #include "clang/AST/ASTLambda.h"
@@ -228,6 +229,42 @@ void emit(Program &P, std::vector<std::byte> &Code, const Floating &Val,
   Val.serialize(Code.data() + ValPos);
 }
 
+template <>
+void emit(Program &P, std::vector<std::byte> &Code,
+          const IntegralAP<false> &Val, bool &Success) {
+  size_t Size = Val.bytesToSerialize();
+
+  if (Code.size() + Size > std::numeric_limits<unsigned>::max()) {
+    Success = false;
+    return;
+  }
+
+  // Access must be aligned!
+  size_t ValPos = align(Code.size());
+  Size = align(Size);
+  assert(aligned(ValPos + Size));
+  Code.resize(ValPos + Size);
+  Val.serialize(Code.data() + ValPos);
+}
+
+template <>
+void emit(Program &P, std::vector<std::byte> &Code, const IntegralAP<true> &Val,
+          bool &Success) {
+  size_t Size = Val.bytesToSerialize();
+
+  if (Code.size() + Size > std::numeric_limits<unsigned>::max()) {
+    Success = false;
+    return;
+  }
+
+  // Access must be aligned!
+  size_t ValPos = align(Code.size());
+  Size = align(Size);
+  assert(aligned(ValPos + Size));
+  Code.resize(ValPos + Size);
+  Val.serialize(Code.data() + ValPos);
+}
+
 template <typename... Tys>
 bool ByteCodeEmitter::emitOp(Opcode Op, const Tys &... Args, const SourceInfo &SI) {
   bool Success = true;
diff --git a/clang/lib/AST/Interp/ByteCodeExprGen.cpp b/clang/lib/AST/Interp/ByteCodeExprGen.cpp
index d4501cefb2131da..c2110834f64c045 100644
--- a/clang/lib/AST/Interp/ByteCodeExprGen.cpp
+++ b/clang/lib/AST/Interp/ByteCodeExprGen.cpp
@@ -2191,15 +2191,13 @@ bool ByteCodeExprGen<Emitter>::emitConst(T Value, PrimType Ty, const Expr *E) {
     return this->emitConstSint64(Value, E);
   case PT_Uint64:
     return this->emitConstUint64(Value, E);
-  case PT_IntAP:
-  case PT_IntAPS:
-    assert(false);
-    return false;
   case PT_Bool:
     return this->emitConstBool(Value, E);
   case PT_Ptr:
   case PT_FnPtr:
   case PT_Float:
+  case PT_IntAP:
+  case PT_IntAPS:
     llvm_unreachable("Invalid integral type");
     break;
   }
@@ -2215,6 +2213,11 @@ bool ByteCodeExprGen<Emitter>::emitConst(T Value, const Expr *E) {
 template <class Emitter>
 bool ByteCodeExprGen<Emitter>::emitConst(const APSInt &Value, PrimType Ty,
                                          const Expr *E) {
+  if (Ty == PT_IntAPS)
+    return this->emitConstIntAPS(Value, E);
+  else if (Ty == PT_IntAP)
+    return this->emitConstIntAP(Value, E);
+
   if (Value.isSigned())
     return this->emitConst(Value.getSExtValue(), Ty, E);
   return this->emitConst(Value.getZExtValue(), Ty, E);
diff --git a/clang/lib/AST/Interp/Disasm.cpp b/clang/lib/AST/Interp/Disasm.cpp
index d276df8f2926229..eba437e05f59d47 100644
--- a/clang/lib/AST/Interp/Disasm.cpp
+++ b/clang/lib/AST/Interp/Disasm.cpp
@@ -12,6 +12,7 @@
 
 #include "Floating.h"
 #include "Function.h"
+#include "IntegralAP.h"
 #include "Opcode.h"
 #include "PrimType.h"
 #include "Program.h"
@@ -37,6 +38,20 @@ template <> inline Floating ReadArg<Floating>(Program &P, CodePtr &OpPC) {
   return F;
 }
 
+template <>
+inline IntegralAP<false> ReadArg<IntegralAP<false>>(Program &P, CodePtr &OpPC) {
+  IntegralAP<false> I = IntegralAP<false>::deserialize(*OpPC);
+  OpPC += align(I.bytesToSerialize());
+  return I;
+}
+
+template <>
+inline IntegralAP<true> ReadArg<IntegralAP<true>>(Program &P, CodePtr &OpPC) {
+  IntegralAP<true> I = IntegralAP<true>::deserialize(*OpPC);
+  OpPC += align(I.bytesToSerialize());
+  return I;
+}
+
 LLVM_DUMP_METHOD void Function::dump() const { dump(llvm::errs()); }
 
 LLVM_DUMP_METHOD void Function::dump(llvm::raw_ostream &OS) const {
diff --git a/clang/lib/AST/Interp/IntegralAP.h b/clang/lib/AST/Interp/IntegralAP.h
index 55e29caa1cd7471..cf040c5d818d23d 100644
--- a/clang/lib/AST/Interp/IntegralAP.h
+++ b/clang/lib/AST/Interp/IntegralAP.h
@@ -263,6 +263,30 @@ template <bool Signed> class IntegralAP final {
       *R = IntegralAP(A.V.lshr(ShiftAmount));
   }
 
+  // === Serialization support ===
+  size_t bytesToSerialize() const {
+    // 1 byte for the BitWidth followed by N bytes for the actual APInt.
+    // FIXME: Is 1 byte enough? APInt::getBitWidth() returns an unsigned.
+    return 1 + (V.getBitWidth() / CHAR_BIT);
+  }
+
+  void serialize(std::byte *Buff) const {
+    assert(V.getBitWidth() < std::numeric_limits<uint8_t>::max());
+    uint8_t BitWidth = V.getBitWidth();
+
+    *Buff = static_cast<std::byte>(BitWidth);
+    llvm::StoreIntToMemory(V, (uint8_t *)(Buff + 1), BitWidth / CHAR_BIT);
+  }
+
+  static IntegralAP<Signed> deserialize(const std::byte *Buff) {
+    uint8_t BitWidth = static_cast<uint8_t>(*Buff);
+    IntegralAP<Signed> Val(APInt(BitWidth, 0ull, !Signed));
+
+    llvm::LoadIntFromMemory(Val.V, (const uint8_t *)Buff + 1,
+                            BitWidth / CHAR_BIT);
+    return Val;
+  }
+
 private:
   template <template <typename T> class Op>
   static bool CheckAddSubMulUB(const IntegralAP &A, const IntegralAP &B,
@@ -289,6 +313,11 @@ inline llvm::raw_ostream &operator<<(llvm::raw_ostream &OS,
   return OS;
 }
 
+template <bool Signed>
+IntegralAP<Signed> getSwappedBytes(IntegralAP<Signed> F) {
+  return F;
+}
+
 } // namespace interp
 } // namespace clang
 
diff --git a/clang/lib/AST/Interp/Interp.h b/clang/lib/AST/Interp/Interp.h
index 65c54ed9c89b61c..ec43481ebe6cfc7 100644
--- a/clang/lib/AST/Interp/Interp.h
+++ b/clang/lib/AST/Interp/Interp.h
@@ -2041,6 +2041,22 @@ template <> inline Floating ReadArg<Floating>(InterpState &S, CodePtr &OpPC) {
   return F;
 }
 
+template <>
+inline IntegralAP<false> ReadArg<IntegralAP<false>>(InterpState &S,
+                                                    CodePtr &OpPC) {
+  IntegralAP<false> I = IntegralAP<false>::deserialize(*OpPC);
+  OpPC += align(I.bytesToSerialize());
+  return I;
+}
+
+template <>
+inline IntegralAP<true> ReadArg<IntegralAP<true>>(InterpState &S,
+                                                  CodePtr &OpPC) {
+  IntegralAP<true> I = IntegralAP<true>::deserialize(*OpPC);
+  OpPC += align(I.bytesToSerialize());
+  return I;
+}
+
 } // namespace interp
 } // namespace clang
 
diff --git a/clang/lib/AST/Interp/Opcodes.td b/clang/lib/AST/Interp/Opcodes.td
index 24747b6b98c163e..b0fafb973dff45c 100644
--- a/clang/lib/AST/Interp/Opcodes.td
+++ b/clang/lib/AST/Interp/Opcodes.td
@@ -45,6 +45,8 @@ def ArgUint32 : ArgType { let Name = "uint32_t"; }
 def ArgSint64 : ArgType { let Name = "int64_t"; }
 def ArgUint64 : ArgType { let Name = "uint64_t"; }
 def ArgFloat : ArgType { let Name = "Floating"; }
+def ArgIntAP : ArgType { let Name = "IntegralAP<false>"; }
+def ArgIntAPS : ArgType { let Name = "IntegralAP<true>"; }
 def ArgBool : ArgType { let Name = "bool"; }
 
 def ArgFunction : ArgType { let Name = "const Function *"; }
@@ -244,6 +246,8 @@ def ConstUint32 : ConstOpcode<Uint32, ArgUint32>;
 def ConstSint64 : ConstOpcode<Sint64, ArgSint64>;
 def ConstUint64 : ConstOpcode<Uint64, ArgUint64>;
 def ConstFloat : ConstOpcode<Float, ArgFloat>;
+def constIntAP : ConstOpcode<IntAP, ArgIntAP>;
+def constIntAPS : ConstOpcode<IntAPS, ArgIntAPS>;
 def ConstBool : ConstOpcode<Bool, ArgBool>;
 
 // [] -> [Integer]
diff --git a/clang/test/AST/Interp/intap.cpp b/clang/test/AST/Interp/intap.cpp
index 118dc21b67e875f..d4440124856915d 100644
--- a/clang/test/AST/Interp/intap.cpp
+++ b/clang/test/AST/Interp/intap.cpp
@@ -154,6 +154,20 @@ namespace i128 {
   constexpr uint128_t ui128Zero{};
   static_assert(ui128Zero == 0, "");
 
+
+  enum LargeEnum : signed __int128 {
+    LV = (signed __int128)1 << 127,
+  };
+
+  constexpr LargeEnum F = LV;
+  static_assert(F ==  (signed __int128)1 << 127, "");
+  constexpr LargeEnum getLargeEnum() {
+    return LV;
+  }
+  static_assert(getLargeEnum() ==  (signed __int128)1 << 127, "");
+
+
+
 #if __cplusplus >= 201402L
   template <typename T>
   constexpr T CastFrom(__int128_t A) {

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[AaronBallman]

LGTM aside from a minor formatting nit.