================ @@ -0,0 +1,1408 @@ +//===- X86.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 "llvm/ABI/ABIFunctionInfo.h" +#include "llvm/ABI/ABIInfo.h" +#include "llvm/ABI/ABITypeMapper.h" +#include "llvm/ABI/TargetCodegenInfo.h" +#include "llvm/ABI/Types.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Type.h" +#include "llvm/Support/Alignment.h" +#include "llvm/Support/Casting.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/MathExtras.h" +#include "llvm/Support/TypeSize.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/TargetParser/Triple.h" +#include <algorithm> +#include <cassert> +#include <cmath> +#include <cstdint> + +namespace llvm { +namespace abi { + +static unsigned getNativeVectorSizeForAVXABI(X86AVXABILevel AVXLevel) { + switch (AVXLevel) { + case X86AVXABILevel::AVX512: + return 512; + case X86AVXABILevel::AVX: + return 256; + case X86AVXABILevel::None: + return 128; + } + llvm_unreachable("Unknown AVXLevel"); +} + +class X86_64ABIInfo : public ABIInfo { +public: + enum Class { + Integer = 0, + SSE, + SSEUp, + X87, + X87UP, + Complex_X87, + NoClass, + Memory + }; + +private: + TypeBuilder &TB; + X86AVXABILevel AVXLevel; + bool Has64BitPointers; + const llvm::Triple &TargetTriple; + + static Class merge(Class Accum, Class Field); + + void postMerge(unsigned AggregateSize, Class &Lo, Class &Hi) const; + + void classify(const Type *T, uint64_t OffsetBase, Class &Lo, Class &Hi, + bool IsNamedArg, bool IsRegCall = false) const; + + const Type *getIntegerTypeAtOffset(const Type *IRType, unsigned IROffset, + const Type *SourceTy, + unsigned SourceOffset) const; + + const Type *getSSETypeAtOffset(const Type *ABIType, unsigned ABIOffset, + const Type *SourceTy, + unsigned SourceOffset) const; + bool isIllegalVectorType(const Type *Ty) const; + bool containsMatrixField(const StructType *ST) const; + + void computeInfo(ABIFunctionInfo &FI) const override; + ABIArgInfo getIndirectReturnResult(const Type *Ty) const; + const Type *getFPTypeAtOffset(const Type *Ty, unsigned Offset) const; + + const Type *isSingleElementStruct(const Type *Ty) const; + const Type *getByteVectorType(const Type *Ty) const; + + const Type *createPairType(const Type *Lo, const Type *Hi) const; + ABIArgInfo getIndirectResult(const Type *Ty, unsigned FreeIntRegs) const; + + ABIArgInfo classifyReturnType(const Type *RetTy) const; + const char *getClassName(Class C) const; + + ABIArgInfo classifyArgumentType(const Type *Ty, unsigned FreeIntRegs, + unsigned &NeededInt, unsigned &NeededSse, + bool IsNamedArg, + bool IsRegCall = false) const; + const Type *useFirstFieldIfTransparentUnion(const Type *Ty) const; + +public: + X86_64ABIInfo(TypeBuilder &TypeBuilder, const Triple &Triple, + X86AVXABILevel AVXABILevel, bool Has64BitPtrs, + const ABICompatInfo &Compat) + : ABIInfo(Compat), TB(TypeBuilder), AVXLevel(AVXABILevel), + Has64BitPointers(Has64BitPtrs), TargetTriple(Triple) {} + + bool has64BitPointers() const { return Has64BitPointers; } +}; + +void X86_64ABIInfo::postMerge(unsigned AggregateSize, Class &Lo, + Class &Hi) const { + // AMD64-ABI 3.2.3p2: Rule 5. Then a post merger cleanup is done: + // + // (a) If one of the classes is Memory, the whole argument is passed in + // memory. + // + // (b) If X87UP is not preceded by X87, the whole argument is passed in + // memory. + // + // (c) If the size of the aggregate exceeds two eightbytes and the first + // eightbyte isn't SSE or any other eightbyte isn't SSEUP, the whole + // argument is passed in memory. NOTE: This is necessary to keep the + // ABI working for processors that don't support the __m256 type. + // + // (d) If SSEUP is not preceded by SSE or SSEUP, it is converted to SSE. + // + // Some of these are enforced by the merging logic. Others can arise + // only with unions; for example: + // union { _Complex double; unsigned; } + // + // Note that clauses (b) and (c) were added in 0.98. + + if (Hi == Memory) + Lo = Memory; + if (Hi == X87UP && Lo != X87 && getABICompatInfo().Flags.HonorsRevision98) + Lo = Memory; + if (AggregateSize > 128 && (Lo != SSE || Hi != SSEUp)) + Lo = Memory; + if (Hi == SSEUp && Lo != SSE) + Hi = SSE; +} +X86_64ABIInfo::Class X86_64ABIInfo::merge(Class Accum, Class Field) { + // AMD64-ABI 3.2.3p2: Rule 4. Each field of an object is + // classified recursively so that always two fields are + // considered. The resulting class is calculated according to + // the classes of the fields in the eightbyte: + // + // (a) If both classes are equal, this is the resulting class. + // + // (b) If one of the classes is NO_CLASS, the resulting class is + // the other class. + // + // (c) If one of the classes is MEMORY, the result is the MEMORY + // class. + // + // (d) If one of the classes is INTEGER, the result is the + // INTEGER. + // + // (e) If one of the classes is X87, X87UP, COMPLEX_X87 class, + // MEMORY is used as class. + // + // (f) Otherwise class SSE is used. + + // Accum should never be memory (we should have returned) or + // ComplexX87 (because this cannot be passed in a structure). + assert((Accum != Memory && Accum != Complex_X87) && + "Invalid accumulated classification during merge."); + + if (Accum == Field || Field == NoClass) + return Accum; + if (Accum == NoClass) + return Field; + if (Field == Memory) + return Memory; + if (Accum == Integer || Field == Integer) + return Integer; + if (Field == X87 || Field == X87UP || Field == Complex_X87 || Accum == X87 || + Accum == X87UP) + return Memory; + + return SSE; +} + +bool X86_64ABIInfo::containsMatrixField(const StructType *ST) const { + for (const auto &Field : ST->getFields()) { + const Type *FieldType = Field.FieldType; + + if (const auto *AT = dyn_cast<ArrayType>(FieldType)) + return AT->isMatrixType(); + + if (const auto *NestedST = dyn_cast<StructType>(FieldType)) + return containsMatrixField(NestedST); + } + return false; +} + +void X86_64ABIInfo::classify(const Type *T, uint64_t OffsetBase, Class &Lo, + Class &Hi, bool IsNamedArg, bool IsRegCall) const { + Lo = Hi = NoClass; + Class &Current = OffsetBase < 64 ? Lo : Hi; + Current = Memory; + + if (T->isVoid()) { + Current = NoClass; + return; + } + + if (const auto *IT = dyn_cast<IntegerType>(T)) { + auto BitWidth = IT->getSizeInBits().getFixedValue(); + + if (BitWidth == 128 || + (IT->isBitInt() && BitWidth > 64 && BitWidth <= 128)) { + Lo = Integer; + Hi = Integer; + } else if (BitWidth <= 64) + Current = Integer; + + return; + } + + if (const auto *FT = dyn_cast<FloatType>(T)) { + const auto *FltSem = FT->getSemantics(); + + if (FltSem == &llvm::APFloat::IEEEsingle() || + FltSem == &llvm::APFloat::IEEEdouble() || + FltSem == &llvm::APFloat::IEEEhalf() || + FltSem == &llvm::APFloat::BFloat()) { + Current = SSE; + } else if (FltSem == &llvm::APFloat::IEEEquad()) { + Lo = SSE; + Hi = SSEUp; + } else if (FltSem == &llvm::APFloat::x87DoubleExtended()) { + Lo = X87; + Hi = X87UP; + } else + Current = SSE; + return; + } + if (T->isPointer()) { + Current = Integer; + return; + } + + if (const auto *MPT = dyn_cast<MemberPointerType>(T)) { + if (MPT->isFunctionPointer()) { + if (Has64BitPointers) { + Lo = Hi = Integer; + } else { + uint64_t EB_FuncPtr = OffsetBase / 64; + uint64_t EB_ThisAdj = (OffsetBase + 64 - 1) / 64; + if (EB_FuncPtr != EB_ThisAdj) { + Lo = Hi = Integer; + } else + Current = Integer; + } + } else + Current = Integer; + return; + } + + if (const auto *VT = dyn_cast<VectorType>(T)) { + auto Size = VT->getSizeInBits().getFixedValue(); + const Type *ElementType = VT->getElementType(); + + if (Size == 1 || Size == 8 || Size == 16 || Size == 32) { + // gcc passes the following as integer: + // 4 bytes - <4 x char>, <2 x short>, <1 x int>, <1 x float> + // 2 bytes - <2 x char>, <1 x short> + // 1 byte - <1 x char> + Current = Integer; + // If this type crosses an eightbyte boundary, it should be + // split. + uint64_t EB_Lo = (OffsetBase) / 64; + uint64_t EB_Hi = (OffsetBase + Size - 1) / 64; + if (EB_Lo != EB_Hi) + Hi = Lo; + } else if (Size == 64) { + if (const auto *FT = dyn_cast<FloatType>(ElementType)) { + // gcc passes <1 x double> in memory. :( + if (FT->getSemantics() == &llvm::APFloat::IEEEdouble()) + return; + } + + // gcc passes <1 x long long> as SSE but clang used to unconditionally + // pass them as integer. For platforms where clang is the de facto + // platform compiler, we must continue to use integer. + if (const auto *IT = dyn_cast<IntegerType>(ElementType)) { + uint64_t ElemBits = IT->getSizeInBits().getFixedValue(); + if (!getABICompatInfo().Flags.ClassifyIntegerMMXAsSSE && + (ElemBits == 64 || ElemBits == 32)) { + Current = Integer; + } else + Current = SSE; + } else + Current = SSE; + // If this type crosses an eightbyte boundary, it should be + // split. + if (OffsetBase && OffsetBase != 64) + Hi = Lo; + } else if (Size == 128 || + (IsNamedArg && Size <= getNativeVectorSizeForAVXABI(AVXLevel))) { + if (const auto *IT = dyn_cast<IntegerType>(ElementType)) { + uint64_t ElemBits = IT->getSizeInBits().getFixedValue(); + // gcc passes 256 and 512 bit <X x __int128> vectors in memory. :( + if (getABICompatInfo().Flags.PassInt128VectorsInMem && Size != 128 && + ElemBits == 128) + return; + } + + // Arguments of 256-bits are split into four eightbyte chunks. The + // least significant one belongs to class SSE and all the others to class + // SSEUP. The original Lo and Hi design considers that types can't be + // greater than 128-bits, so a 64-bit split in Hi and Lo makes sense. + // This design isn't correct for 256-bits, but since there're no cases + // where the upper parts would need to be inspected, avoid adding + // complexity and just consider Hi to match the 64-256 part. + // + // Note that per 3.5.7 of AMD64-ABI, 256-bit args are only passed in + // registers if they are "named", i.e. not part of the "..." of a + // variadic function. + // + // Similarly, per 3.2.3. of the AVX512 draft, 512-bits ("named") args are + // split into eight eightbyte chunks, one SSE and seven SSEUP. + Lo = SSE; + Hi = SSEUp; + } + return; + } + + if (const auto *CT = dyn_cast<ComplexType>(T)) { + const Type *ElementType = CT->getElementType(); + uint64_t Size = T->getSizeInBits().getFixedValue(); + + if (const auto *EIT = dyn_cast<IntegerType>(ElementType)) { ---------------- nikic wrote:
```suggestion if (isa<IntegerType>(ElementType)) { ``` To avoid warning https://github.com/llvm/llvm-project/pull/140112 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits