[LLVM][IR] Add native vector support to ConstantInt & ConstantFP.

[paulwalker-arm]

NOTE: For brevity the following talks about ConstantInt but
everything extends to cover ConstantFP as well.

Whilst ConstantInt::get() supports the creation of vectors whereby
each lane has the same value, it achieves this via other constants:

• ConstantVector for fixed-length vectors
• ConstantExprs for scalable vectors

However, ConstantExprs are being deprecated and ConstantVector is
not space efficient for larger vector types. By extending ConstantInt
we can represent vector splats by only storing the underlying scalar
value.

More specifically:

• ConstantInt gains an ElementCount variant of get().
• LLVMContext is extended to map <EC,APInt>->ConstantInt.
• BitcodeReader/Writer support is extended to allow vector types.

Whilst this patch adds the base support, more work is required
before it's production ready. For example, there's likely to be
many places where isa assumes a scalar type. Accordingly
the default behaviour of ConstantInt::get() remains unchanged but a
set of flags are added to allow wider testing and thus help with the
migration:

--use-constant-int-for-fixed-length-splat
--use-constant-fp-for-fixed-length-splat
--use-constant-int-for-scalable-splat
--use-constant-fp-for-scalable-splat

NOTE: No change is required to the bitcode format because types and
values are handled separately.

NOTE: For similar reasons as above, code generation doesn't work
out-the-box.

[llvmbot]

@llvm/pr-subscribers-llvm-ir
@llvm/pr-subscribers-clang
@llvm/pr-subscribers-llvm-analysis

@llvm/pr-subscribers-llvm-transforms

Author: Paul Walker (paulwalker-arm)

Changes

[LLVM][IR] Add native vector support to ConstantInt & ConstantFP.

NOTE: For brevity the following takes about ConstantInt but
everything extends to cover ConstantFP as well.

Whilst ConstantInt::get() supports the creation of vectors whereby
each lane has the same value, it achieves this via other constants:

• ConstantVector for fixed-length vectors
• ConstantExprs for scalable vectors

ConstantExprs are being deprecated and ConstantVector is not space
efficient for larger vector types. This patch introduces an
alternative by allowing ConstantInt to natively support vector
splats via the IR syntax:

<N x ty> splat(ty <imm>)

More specifically:

• IR parsing is extended to support the new syntax.
• ConstantInt gains the interface getSplat().
• LLVMContext is extended to map <EC,APInt>->ConstantInt.
• BitCodeReader/Writer is extended to support vector types.

Whilst this patch adds the base support, more work is required
before it's production ready. For example, there's likely to be
many places where isa<ConstantInt> assumes a scalar type. Accordingly
the default behaviour of ConstantInt::get() remains unchanged but a
set of flag are added to allow wider testing and thus help with the
migration:

--use-constant-int-for-fixed-length-splat
--use-constant-fp-for-fixed-length-splat
--use-constant-int-for-scalable-splat
--use-constant-fp-for-scalable-splat

NOTE: No change is required to the bitcode format because types and
values are handled separately.

NOTE: Code generation doesn't work out-the-box but the issues look
limited to calls to ConstantInt::getBitWidth() that will need to be
ported.

---

Patch is 41.22 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/74502.diff

22 Files Affected:

• (modified) clang/lib/CodeGen/CGBuiltin.cpp (+4-3)
• (modified) llvm/include/llvm/AsmParser/LLParser.h (+3-1)
• (modified) llvm/include/llvm/AsmParser/LLToken.h (+1)
• (modified) llvm/include/llvm/IR/Constants.h (+23-5)
• (modified) llvm/lib/Analysis/InstructionSimplify.cpp (+1-1)
• (modified) llvm/lib/AsmParser/LLLexer.cpp (+1)
• (modified) llvm/lib/AsmParser/LLParser.cpp (+50-6)
• (modified) llvm/lib/Bitcode/Reader/BitcodeReader.cpp (+41-29)
• (modified) llvm/lib/Bitcode/Writer/BitcodeWriter.cpp (+3-3)
• (modified) llvm/lib/IR/AsmWriter.cpp (+32-5)
• (modified) llvm/lib/IR/ConstantFold.cpp (+1-1)
• (modified) llvm/lib/IR/Constants.cpp (+92-1)
• (modified) llvm/lib/IR/LLVMContextImpl.cpp (+2)
• (modified) llvm/lib/IR/LLVMContextImpl.h (+4)
• (modified) llvm/lib/IR/Verifier.cpp (+2-2)
• (modified) llvm/lib/Target/Hexagon/HexagonLoopIdiomRecognition.cpp (+3-3)
• (modified) llvm/lib/Transforms/IPO/OpenMPOpt.cpp (+8-7)
• (modified) llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp (+2-2)
• (modified) llvm/lib/Transforms/Scalar/ConstantHoisting.cpp (+3-3)
• (modified) llvm/lib/Transforms/Scalar/LoopFlatten.cpp (+1-1)
• (modified) llvm/lib/Transforms/Utils/SimplifyCFG.cpp (+4-4)
• (added) llvm/test/Bitcode/constant-splat.ll (+53)

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 65d9862621061..8dc828abf8aec 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -3218,7 +3218,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
     Value *AlignmentValue = EmitScalarExpr(E->getArg(1));
     ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue);
     if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment))
-      AlignmentCI = ConstantInt::get(AlignmentCI->getType(),
+      AlignmentCI = ConstantInt::get(AlignmentCI->getIntegerType(),
                                      llvm::Value::MaximumAlignment);
 
     emitAlignmentAssumption(PtrValue, Ptr,
@@ -17010,7 +17010,7 @@ Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID,
     Value *Op1 = EmitScalarExpr(E->getArg(1));
     ConstantInt *AlignmentCI = cast<ConstantInt>(Op0);
     if (AlignmentCI->getValue().ugt(llvm::Value::MaximumAlignment))
-      AlignmentCI = ConstantInt::get(AlignmentCI->getType(),
+      AlignmentCI = ConstantInt::get(AlignmentCI->getIntegerType(),
                                      llvm::Value::MaximumAlignment);
 
     emitAlignmentAssumption(Op1, E->getArg(1),
@@ -17248,7 +17248,8 @@ Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID,
         Op0, llvm::FixedVectorType::get(ConvertType(E->getType()), 2));
 
     if (getTarget().isLittleEndian())
-      Index = ConstantInt::get(Index->getType(), 1 - Index->getZExtValue());
+      Index =
+          ConstantInt::get(Index->getIntegerType(), 1 - Index->getZExtValue());
 
     return Builder.CreateExtractElement(Unpacked, Index);
   }
diff --git a/llvm/include/llvm/AsmParser/LLParser.h b/llvm/include/llvm/AsmParser/LLParser.h
index 810f3668d05d4..38f6f08b8f3a1 100644
--- a/llvm/include/llvm/AsmParser/LLParser.h
+++ b/llvm/include/llvm/AsmParser/LLParser.h
@@ -59,7 +59,9 @@ namespace llvm {
       t_Constant,                      // Value in ConstantVal.
       t_InlineAsm,                     // Value in FTy/StrVal/StrVal2/UIntVal.
       t_ConstantStruct,                // Value in ConstantStructElts.
-      t_PackedConstantStruct           // Value in ConstantStructElts.
+      t_PackedConstantStruct,          // Value in ConstantStructElts.
+      t_APSIntSplat,                   // Value in APSIntVal.
+      t_APFloatSplat                   // Value in APFloatVal.
     } Kind = t_LocalID;
 
     LLLexer::LocTy Loc;
diff --git a/llvm/include/llvm/AsmParser/LLToken.h b/llvm/include/llvm/AsmParser/LLToken.h
index 0683291faae72..dd55afee21033 100644
--- a/llvm/include/llvm/AsmParser/LLToken.h
+++ b/llvm/include/llvm/AsmParser/LLToken.h
@@ -335,6 +335,7 @@ enum Kind {
   kw_extractelement,
   kw_insertelement,
   kw_shufflevector,
+  kw_splat,
   kw_extractvalue,
   kw_insertvalue,
   kw_blockaddress,
diff --git a/llvm/include/llvm/IR/Constants.h b/llvm/include/llvm/IR/Constants.h
index 2f7fc5652c2cd..b76cb1beecf3c 100644
--- a/llvm/include/llvm/IR/Constants.h
+++ b/llvm/include/llvm/IR/Constants.h
@@ -81,6 +81,7 @@ class ConstantInt final : public ConstantData {
   APInt Val;
 
   ConstantInt(IntegerType *Ty, const APInt &V);
+  ConstantInt(VectorType *Ty, const APInt &V);
 
   void destroyConstantImpl();
 
@@ -98,6 +99,13 @@ class ConstantInt final : public ConstantData {
   /// value. Otherwise return a ConstantInt for the given value.
   static Constant *get(Type *Ty, uint64_t V, bool IsSigned = false);
 
+  /// WARNING: Incomplete support, do not use. These methods exist for early
+  /// prototyping, for most use cases ConstantInt::get() should be used.
+  /// Return a ConstantInt with a splat of the given value.
+  static ConstantInt *getSplat(LLVMContext &Context, ElementCount EC,
+                               const APInt &V);
+  static ConstantInt *getSplat(const VectorType *Ty, const APInt &V);
+
   /// Return a ConstantInt with the specified integer value for the specified
   /// type. If the type is wider than 64 bits, the value will be zero-extended
   /// to fit the type, unless IsSigned is true, in which case the value will
@@ -136,7 +144,11 @@ class ConstantInt final : public ConstantData {
   inline const APInt &getValue() const { return Val; }
 
   /// getBitWidth - Return the bitwidth of this constant.
-  unsigned getBitWidth() const { return Val.getBitWidth(); }
+  unsigned getBitWidth() const {
+    assert(getType()->isIntegerTy() &&
+           "Returning the bitwidth of a vector constant is not support!");
+    return Val.getBitWidth();
+  }
 
   /// Return the constant as a 64-bit unsigned integer value after it
   /// has been zero extended as appropriate for the type of this constant. Note
@@ -170,10 +182,9 @@ class ConstantInt final : public ConstantData {
   /// Determine if this constant's value is same as an unsigned char.
   bool equalsInt(uint64_t V) const { return Val == V; }
 
-  /// getType - Specialize the getType() method to always return an IntegerType,
-  /// which reduces the amount of casting needed in parts of the compiler.
-  ///
-  inline IntegerType *getType() const {
+  /// Variant of the getType() method to always return an IntegerType, which
+  /// reduces the amount of casting needed in parts of the compiler.
+  inline IntegerType *getIntegerType() const {
     return cast<IntegerType>(Value::getType());
   }
 
@@ -279,6 +290,13 @@ class ConstantFP final : public ConstantData {
   /// value. Otherwise return a ConstantFP for the given value.
   static Constant *get(Type *Ty, const APFloat &V);
 
+  /// WARNING: Incomplete support, do not use. These methods exist for early
+  /// prototyping, for most use cases ConstantFP::get() should be used.
+  /// Return a ConstantFP with a splat of the given value.
+  static ConstantFP *getSplat(LLVMContext &Context, ElementCount EC,
+                              const APFloat &V);
+  static ConstantFP *getSplat(const VectorType *Ty, const APFloat &V);
+
   static Constant *get(Type *Ty, StringRef Str);
   static ConstantFP *get(LLVMContext &Context, const APFloat &V);
   static Constant *getNaN(Type *Ty, bool Negative = false,
diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index cef9f6ec179ba..c24bb1bb2cf9f 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -6081,7 +6081,7 @@ static Value *simplifyRelativeLoad(Constant *Ptr, Constant *Offset,
   Type *Int32Ty = Type::getInt32Ty(Ptr->getContext());
 
   auto *OffsetConstInt = dyn_cast<ConstantInt>(Offset);
-  if (!OffsetConstInt || OffsetConstInt->getType()->getBitWidth() > 64)
+  if (!OffsetConstInt || OffsetConstInt->getIntegerType()->getBitWidth() > 64)
     return nullptr;
 
   APInt OffsetInt = OffsetConstInt->getValue().sextOrTrunc(
diff --git a/llvm/lib/AsmParser/LLLexer.cpp b/llvm/lib/AsmParser/LLLexer.cpp
index 09a205c445dbe..eb47284feb218 100644
--- a/llvm/lib/AsmParser/LLLexer.cpp
+++ b/llvm/lib/AsmParser/LLLexer.cpp
@@ -697,6 +697,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(uinc_wrap);
   KEYWORD(udec_wrap);
 
+  KEYWORD(splat);
   KEYWORD(vscale);
   KEYWORD(x);
   KEYWORD(blockaddress);
diff --git a/llvm/lib/AsmParser/LLParser.cpp b/llvm/lib/AsmParser/LLParser.cpp
index d236b6cfa9000..94e1a51aa2e75 100644
--- a/llvm/lib/AsmParser/LLParser.cpp
+++ b/llvm/lib/AsmParser/LLParser.cpp
@@ -3952,6 +3952,31 @@ bool LLParser::parseValID(ValID &ID, PerFunctionState *PFS, Type *ExpectedTy) {
     return false;
   }
 
+  case lltok::kw_splat: {
+    Lex.Lex();
+    if (parseToken(lltok::lparen, "expected '(' after vector splat"))
+      return true;
+    Constant *C;
+    if (parseGlobalTypeAndValue(C))
+      return true;
+    if (parseToken(lltok::rparen, "expected ')' at end of vector splat"))
+      return true;
+
+    if (auto *CI = dyn_cast<ConstantInt>(C)) {
+      ID.APSIntVal = CI->getValue();
+      ID.Kind = ValID::t_APSIntSplat;
+      return false;
+    }
+
+    if (auto *CFP = dyn_cast<ConstantFP>(C)) {
+      ID.APFloatVal = CFP->getValue();
+      ID.Kind = ValID::t_APFloatSplat;
+      return false;
+    }
+
+    return tokError("invalid splat operand");
+  }
+
   case lltok::kw_getelementptr:
   case lltok::kw_shufflevector:
   case lltok::kw_insertelement:
@@ -5716,9 +5741,23 @@ bool LLParser::convertValIDToValue(Type *Ty, ValID &ID, Value *&V,
     ID.APSIntVal = ID.APSIntVal.extOrTrunc(Ty->getPrimitiveSizeInBits());
     V = ConstantInt::get(Context, ID.APSIntVal);
     return false;
+  case ValID::t_APSIntSplat:
+    if (!Ty->isVectorTy() || !Ty->getScalarType()->isIntegerTy())
+      return error(ID.Loc, "expected an integer vector result");
+    if (ID.APSIntVal.getBitWidth() !=
+        cast<IntegerType>(Ty->getScalarType())->getBitWidth())
+      return error(ID.Loc, "operand type must match result element type");
+    V = ConstantInt::getSplat(cast<VectorType>(Ty), ID.APSIntVal);
+    return false;
   case ValID::t_APFloat:
-    if (!Ty->isFloatingPointTy() ||
-        !ConstantFP::isValueValidForType(Ty, ID.APFloatVal))
+  case ValID::t_APFloatSplat: {
+    if ((ID.Kind == ValID::t_APFloat && !Ty->isFloatingPointTy()) ||
+        (ID.Kind == ValID::t_APFloatSplat && !Ty->isVectorTy()))
+      return error(ID.Loc, "floating point constant invalid for type");
+
+    Type *ScalarTy = Ty->getScalarType();
+    if (!ScalarTy->isFloatingPointTy() ||
+        !ConstantFP::isValueValidForType(ScalarTy, ID.APFloatVal))
       return error(ID.Loc, "floating point constant invalid for type");
 
     // The lexer has no type info, so builds all half, bfloat, float, and double
@@ -5727,13 +5766,13 @@ bool LLParser::convertValIDToValue(Type *Ty, ValID &ID, Value *&V,
       // Check for signaling before potentially converting and losing that info.
       bool IsSNAN = ID.APFloatVal.isSignaling();
       bool Ignored;
-      if (Ty->isHalfTy())
+      if (ScalarTy->isHalfTy())
         ID.APFloatVal.convert(APFloat::IEEEhalf(), APFloat::rmNearestTiesToEven,
                               &Ignored);
-      else if (Ty->isBFloatTy())
+      else if (ScalarTy->isBFloatTy())
         ID.APFloatVal.convert(APFloat::BFloat(), APFloat::rmNearestTiesToEven,
                               &Ignored);
-      else if (Ty->isFloatTy())
+      else if (ScalarTy->isFloatTy())
         ID.APFloatVal.convert(APFloat::IEEEsingle(), APFloat::rmNearestTiesToEven,
                               &Ignored);
       if (IsSNAN) {
@@ -5745,13 +5784,18 @@ bool LLParser::convertValIDToValue(Type *Ty, ValID &ID, Value *&V,
                                          ID.APFloatVal.isNegative(), &Payload);
       }
     }
-    V = ConstantFP::get(Context, ID.APFloatVal);
+
+    if (auto *VTy = dyn_cast<VectorType>(Ty))
+      V = ConstantFP::getSplat(VTy, ID.APFloatVal);
+    else
+      V = ConstantFP::get(Context, ID.APFloatVal);
 
     if (V->getType() != Ty)
       return error(ID.Loc, "floating point constant does not have type '" +
                                getTypeString(Ty) + "'");
 
     return false;
+  }
   case ValID::t_Null:
     if (!Ty->isPointerTy())
       return error(ID.Loc, "null must be a pointer type");
diff --git a/llvm/lib/Bitcode/Reader/BitcodeReader.cpp b/llvm/lib/Bitcode/Reader/BitcodeReader.cpp
index e4c3770946b3a..b661d36fb6854 100644
--- a/llvm/lib/Bitcode/Reader/BitcodeReader.cpp
+++ b/llvm/lib/Bitcode/Reader/BitcodeReader.cpp
@@ -3022,50 +3022,62 @@ Error BitcodeReader::parseConstants() {
       V = Constant::getNullValue(CurTy);
       break;
     case bitc::CST_CODE_INTEGER:   // INTEGER: [intval]
-      if (!CurTy->isIntegerTy() || Record.empty())
+      if (!CurTy->isIntOrIntVectorTy() || Record.empty())
         return error("Invalid integer const record");
-      V = ConstantInt::get(CurTy, decodeSignRotatedValue(Record[0]));
+
+      if (auto *VTy = dyn_cast<VectorType>(CurTy)) {
+        auto *ScalarTy = cast<IntegerType>(VTy->getScalarType());
+        unsigned BitWidth = ScalarTy->getBitWidth();
+        APInt VInt(BitWidth, decodeSignRotatedValue(Record[0]));
+        V = ConstantInt::getSplat(VTy, VInt);
+      } else
+        V = ConstantInt::get(CurTy, decodeSignRotatedValue(Record[0]));
       break;
     case bitc::CST_CODE_WIDE_INTEGER: {// WIDE_INTEGER: [n x intval]
-      if (!CurTy->isIntegerTy() || Record.empty())
+      if (!CurTy->isIntOrIntVectorTy() || Record.empty())
         return error("Invalid wide integer const record");
 
-      APInt VInt =
-          readWideAPInt(Record, cast<IntegerType>(CurTy)->getBitWidth());
-      V = ConstantInt::get(Context, VInt);
-
+      auto *ScalarTy = cast<IntegerType>(CurTy->getScalarType());
+      APInt VInt = readWideAPInt(Record, ScalarTy->getBitWidth());
+      if (auto *VTy = dyn_cast<VectorType>(CurTy))
+        V = ConstantInt::getSplat(VTy, VInt);
+      else
+        V = ConstantInt::get(Context, VInt);
       break;
     }
     case bitc::CST_CODE_FLOAT: {    // FLOAT: [fpval]
       if (Record.empty())
         return error("Invalid float const record");
-      if (CurTy->isHalfTy())
-        V = ConstantFP::get(Context, APFloat(APFloat::IEEEhalf(),
-                                             APInt(16, (uint16_t)Record[0])));
-      else if (CurTy->isBFloatTy())
-        V = ConstantFP::get(Context, APFloat(APFloat::BFloat(),
-                                             APInt(16, (uint32_t)Record[0])));
-      else if (CurTy->isFloatTy())
-        V = ConstantFP::get(Context, APFloat(APFloat::IEEEsingle(),
-                                             APInt(32, (uint32_t)Record[0])));
-      else if (CurTy->isDoubleTy())
-        V = ConstantFP::get(Context, APFloat(APFloat::IEEEdouble(),
-                                             APInt(64, Record[0])));
-      else if (CurTy->isX86_FP80Ty()) {
+
+      APFloat Val(APFloat::Bogus());
+      auto *ScalarTy = CurTy->getScalarType();
+      if (ScalarTy->isHalfTy())
+        Val = APFloat(APFloat::IEEEhalf(), APInt(16, (uint16_t)Record[0]));
+      else if (ScalarTy->isBFloatTy())
+        Val = APFloat(APFloat::BFloat(), APInt(16, (uint32_t)Record[0]));
+      else if (ScalarTy->isFloatTy())
+        Val = APFloat(APFloat::IEEEsingle(), APInt(32, (uint32_t)Record[0]));
+      else if (ScalarTy->isDoubleTy())
+        Val = APFloat(APFloat::IEEEdouble(), APInt(64, Record[0]));
+      else if (ScalarTy->isX86_FP80Ty()) {
         // Bits are not stored the same way as a normal i80 APInt, compensate.
         uint64_t Rearrange[2];
         Rearrange[0] = (Record[1] & 0xffffLL) | (Record[0] << 16);
         Rearrange[1] = Record[0] >> 48;
-        V = ConstantFP::get(Context, APFloat(APFloat::x87DoubleExtended(),
-                                             APInt(80, Rearrange)));
-      } else if (CurTy->isFP128Ty())
-        V = ConstantFP::get(Context, APFloat(APFloat::IEEEquad(),
-                                             APInt(128, Record)));
-      else if (CurTy->isPPC_FP128Ty())
-        V = ConstantFP::get(Context, APFloat(APFloat::PPCDoubleDouble(),
-                                             APInt(128, Record)));
-      else
+        Val = APFloat(APFloat::x87DoubleExtended(), APInt(80, Rearrange));
+      } else if (ScalarTy->isFP128Ty())
+        Val = APFloat(APFloat::IEEEquad(), APInt(128, Record));
+      else if (ScalarTy->isPPC_FP128Ty())
+        Val = APFloat(APFloat::PPCDoubleDouble(), APInt(128, Record));
+      else {
         V = UndefValue::get(CurTy);
+        break;
+      }
+
+      if (auto *VTy = dyn_cast<VectorType>(CurTy))
+        V = ConstantFP::getSplat(VTy, Val);
+      else
+        V = ConstantFP::get(Context, Val);
       break;
     }
 
diff --git a/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp b/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp
index 8239775d04865..0f5b9ff9ebd72 100644
--- a/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp
+++ b/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp
@@ -2577,18 +2577,18 @@ void ModuleBitcodeWriter::writeConstants(unsigned FirstVal, unsigned LastVal,
     } else if (isa<UndefValue>(C)) {
       Code = bitc::CST_CODE_UNDEF;
     } else if (const ConstantInt *IV = dyn_cast<ConstantInt>(C)) {
-      if (IV->getBitWidth() <= 64) {
+      if (IV->getValue().getBitWidth() <= 64) {
         uint64_t V = IV->getSExtValue();
         emitSignedInt64(Record, V);
         Code = bitc::CST_CODE_INTEGER;
         AbbrevToUse = CONSTANTS_INTEGER_ABBREV;
-      } else {                             // Wide integers, > 64 bits in size.
+      } else { // Wide integers, > 64 bits in size.
         emitWideAPInt(Record, IV->getValue());
         Code = bitc::CST_CODE_WIDE_INTEGER;
       }
     } else if (const ConstantFP *CFP = dyn_cast<ConstantFP>(C)) {
       Code = bitc::CST_CODE_FLOAT;
-      Type *Ty = CFP->getType();
+      Type *Ty = CFP->getType()->getScalarType();
       if (Ty->isHalfTy() || Ty->isBFloatTy() || Ty->isFloatTy() ||
           Ty->isDoubleTy()) {
         Record.push_back(CFP->getValueAPF().bitcastToAPInt().getZExtValue());
diff --git a/llvm/lib/IR/AsmWriter.cpp b/llvm/lib/IR/AsmWriter.cpp
index fabc79adbd33d..e37da7c460f26 100644
--- a/llvm/lib/IR/AsmWriter.cpp
+++ b/llvm/lib/IR/AsmWriter.cpp
@@ -1394,16 +1394,32 @@ static void WriteOptimizationInfo(raw_ostream &Out, const User *U) {
 static void WriteConstantInternal(raw_ostream &Out, const Constant *CV,
                                   AsmWriterContext &WriterCtx) {
   if (const ConstantInt *CI = dyn_cast<ConstantInt>(CV)) {
-    if (CI->getType()->isIntegerTy(1)) {
-      Out << (CI->getZExtValue() ? "true" : "false");
-      return;
+    if (CI->getType()->isVectorTy()) {
+      Out << "splat (";
+      WriterCtx.TypePrinter->print(CI->getType()->getScalarType(), Out);
+      Out << " ";
     }
-    Out << CI->getValue();
+
+    if (CI->getType()->getScalarType()->isIntegerTy(1))
+      Out << (CI->getZExtValue() ? "true" : "false");
+    else
+      Out << CI->getValue();
+
+    if (CI->getType()->isVectorTy())
+      Out << ")";
+
     return;
   }
 
   if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CV)) {
     const APFloat &APF = CFP->getValueAPF();
+
+    if (CFP->getType()->isVectorTy()) {
+      Out << "splat (";
+      WriterCtx.TypePrinter->print(CFP->getType()->getScalarType(), Out);
+      Out << " ";
+    }
+
     if (&APF.getSemantics() == &APFloat::IEEEsingle() ||
         &APF.getSemantics() == &APFloat::IEEEdouble()) {
       // We would like to output the FP constant value in exponential notation,
@@ -1429,6 +1445,10 @@ static void WriteConstantInternal(raw_ostream &Out, const Constant *CV,
         // Reparse stringized version!
         if (APFloat(APFloat::IEEEdouble(), StrVal).convertToDouble() == Val) {
           Out << StrVal;
+
+          if (CFP->getType()->isVectorTy())
+            Out << ")";
+
           return;
         }
       }
@@ -1454,6 +1474,10 @@ static void WriteConstantInternal(raw_ostream &Out, const Constant *CV,
         }
       }
       Out << format_hex(apf.bitcastToAPInt().getZExtValue(), 0, /*Upper=*/true);
+
+      if (CFP->getType()->isVectorTy())
+        Out << ")";
+
       return;
     }
 
@@ -1468,7 +1492,6 @@ static void WriteConstantInternal(raw_ostream &Out, const Constant *CV,
                                   /*Upper=*/true);
       Out << format_hex_no_prefix(API.getLoBits(64).getZExtValue(), 16,
                                   /*Upper=*/true);
-      return;
     } else if (&APF.getSemantics() == &APFloat::IEEEquad()) {
       Out << 'L';
       Out << format_hex_no_prefix(API.getLoBits(64).getZExtValue(), 16,
@@ -1491,6 +1514,10 @@ static void WriteConstantInternal(raw_ostream &Out, const Constant *CV,
                                   /*Upper=*/true);
     } else
       llvm_unreachable("Unsupported floating point type");
+
+    if (CFP->getType()->isVectorTy())
+      Out << ")";
+
     return;
   }
 
diff --git a/llvm/lib/IR/ConstantFold.cpp b/llvm/lib/IR/ConstantFold.cpp
index d499d74f7ba01..c478040234078 100644
--- a/llvm/lib/IR/ConstantFold.cpp
+++ b/llvm/lib/IR/ConstantFold.cpp
@@ -868,7 +868,7 @@ Constant *llvm::ConstantFoldBinaryInstruction(unsigned Opcode, Constant *C1,
           }
 
           if (GVAlign > 1) {
-            unsigned DstWidth = CI2->getType()->getBitWidth();
+            unsigned DstWidth = CI2->getIntegerType()->getBitWidth();
             unsigned SrcWidth = std::min(DstWidth, Log2(GVAlign));
             APInt BitsNotSet(APInt::...
[truncated]

[github-actions]

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

[paulwalker-arm]

The PR contains a couple of commits that unless there's disagreement I'm tempted to land directly but have held off just in case there's any buyer remorse about extending ConstantInt/ConstantFP to cover vector types.

For similar reasons I've not updated the LangRef as I don't really want people using the support until at least code generation works.

[nikic]

I don't think these APIs should exist. ConstantInt::get() already supports creation of splats, they just aren't represented as ConstantInts.

[paulwalker-arm]

We're in a transition period and thus I need an absolute way to create a vector ConstantInt (e.g. when parsing ll files and bitcode). Today ConstantInt::get() returns other Constant types to represent splats and that must be maintained for correctness because there are many code paths for which a vector ConstantInt will break.

[nikic]

I think this isn't the right way to phase in the change. I think the splat syntax should just return whatever ConstantVector::getSplat() produces, and what that produces can be controlled by the opt flags you have introduced.

That means that the splat syntax becomes usable right away as a short-hand for producing the representations we currently use, and will switch to producing plain ConstantInt/ConstantFP once the flag is flipped (or in tests that explicitly flip it).

[paulwalker-arm]

OK, thanks. It seems I've incorrectly assumed from an IR parsing and printing point of view there is a requirement for IR_out == IR_in. Your suggestion certainly means I can break the work up some more so I get that sorted.

[paulwalker-arm]

I've moved the textual IR side of things to #74620 following the suggestion to have splat(x) be synonymous with ConstantInt/FP::get().

[nikic]

Why? I think this API should work with vectors as well (with the current implementation, i.e. returning the bitwidth of the scalar value). You can just adjust the comment to clarify.

[paulwalker-arm]

Ultimately I think this should be more explicit, for example getScalarBitWidth(). For this patch though the need was tiny so I made this change purely to trigger asserts at this level when failure cases are hit once I start expanding the testing.

[nikic]

IMHO this is not necessary. If the name were getSizeInBits() I would agree, but the term "bit width" implies that we're talking about scalar. We don't use the term "bit width" to refer to full size of a vector.

[paulwalker-arm]

Fair enough.

[nikic]

Why are you changing usages like these? This code should work fine with getType().

[paulwalker-arm]

Much the same reason as with getBitWidth() but the need for immediate change is greater so I renamed this method so changes to existing code paths are minimal whilst still providing a route to trigger asserts once testing is expanded.

I did consider just removing the method and adding the necessary casts but figured somebody went to the trouble of adding the override in the first place so I maintained this but under a modified name. Do you think the shorthand is not worth it?

[nikic]

I don't really follow here. Even if you rename the overload on ConstantInt, there will still be the method inherited from Value::getType(), and using that method here should work (and be forward-compatible with vector ConstantInt), because ConstantInt::get doesn't actually require that the argument is an IntegerType.

[paulwalker-arm]

Could it be static ConstantInt *get(IntegerType *Ty, uint64_t V, bool IsSigned = false);?

I don't think I made this change up. I wanted a mechanical change so just removed the overload and the compiler told me all the places that relied on it.

[nikic]

Ooooh, I think the issue is that it's stored in a ConstantInt * variable. Changing it to Constant * should make it pick the other overload, I think.

[paulwalker-arm]

Hi @nikic, I had to put this on the back burner but I'm back now. I believe the rebase last year was the completion point for this stage of the work so the patch should be good to review. Given the time though I'll rebase and retest tomorrow but wanted to check if your assessment matches mine or whether you feel more is needed?

Separately to this I will create some NFC patches to port tests to the newer splat syntax so I can more easily find the failure paths when enabling the new functionality.

[nikic]

Looks conceptually fine to me.

[nikic]

LGTM

[nikic]

Just wondering whether this is something you want to keep long term or just initially?

[paulwalker-arm]

I'm happy to take offers but see it as temporary.

[nikic]

Something important I forgot: Can you please test what happens with <2 x i8> <i8 42, i8 42> in use-constant-int-for-fixed-length-splat mode? This needs to create the same representation as <2 x i8> splat (i8 42), to maintain a canonical form of constants.

(This applies to a lesser degree to the scalable case -- in that case it's "just" a missed constant folding opportunity.)

[nikic]

Just wanted to check back whether you're still pursuing this?

[paulwalker-arm]

Just wanted to check back whether you're still pursuing this?

Yes although I've not had chance to upload any related patches for a while. I started digging into the code generation side of things and then hit issues with generating slightly worse code due to code paths that relied on bogus behaviour. I was hoping to get it finished in time for LLVM 19 but I don't think that's realistic now but I'll create PRs for what I have.

[Dinistro]

@paulwalker-arm @nikic Do you think it makes sense to make these available in a header file? I tried to set them directly from within a custom tool, but failed to get access to these. For us, setting them via. the command line is impractical.
If so, I can provide a PR that moves their declaration to a header file.

[nikic]

I'd say tentatively "no". We only really export options if we need to read them from multiple files. If we wanted to make these settable via API, I think the way we'd typically do that is by moving them into LLVMContext and adding setters.

FWIW, I don't think these options are mature enough yet for practical usage.

I tried to set them directly from within a custom tool, but failed to get access to these. For us, setting them via. the command line is impractical.

Can you provide more details on this? The usual way to set the options is via cl::ParseCommandLineOptions, which doesn't mean that they actually have to come from the command line. It's pretty typical to just hardcode some options in the invocation.

[Dinistro]

Thanks for the fast response.

Can you provide more details on this?

I attempted to directly access the underlying Option structures, but that did not succeed.

The usual way to set the options is via cl::ParseCommandLineOptions, which doesn't mean that they actually have to come from the command line. It's pretty typical to just hardcode some options in the invocation.

I was not aware that one can executed this function twice to pass custom options to it, thanks for the hint. This should work, once some other, only tangentially related, issues have been addressed.