llvm-project/clang/lib/CodeGen/CGBuilder.h

285 lines
12 KiB
C
Raw Normal View History

//===-- CGBuilder.h - Choose IRBuilder implementation ----------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
#ifndef LLVM_CLANG_LIB_CODEGEN_CGBUILDER_H
#define LLVM_CLANG_LIB_CODEGEN_CGBUILDER_H
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/IRBuilder.h"
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
#include "Address.h"
#include "CodeGenTypeCache.h"
namespace clang {
namespace CodeGen {
class CodeGenFunction;
/// \brief This is an IRBuilder insertion helper that forwards to
2014-09-11 00:59:01 +08:00
/// CodeGenFunction::InsertHelper, which adds necessary metadata to
/// instructions.
class CGBuilderInserter : protected llvm::IRBuilderDefaultInserter {
public:
CGBuilderInserter() = default;
explicit CGBuilderInserter(CodeGenFunction *CGF) : CGF(CGF) {}
protected:
/// \brief This forwards to CodeGenFunction::InsertHelper.
void InsertHelper(llvm::Instruction *I, const llvm::Twine &Name,
llvm::BasicBlock *BB,
llvm::BasicBlock::iterator InsertPt) const;
private:
CodeGenFunction *CGF = nullptr;
};
typedef CGBuilderInserter CGBuilderInserterTy;
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
typedef llvm::IRBuilder<llvm::ConstantFolder, CGBuilderInserterTy>
CGBuilderBaseTy;
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
class CGBuilderTy : public CGBuilderBaseTy {
/// Storing a reference to the type cache here makes it a lot easier
/// to build natural-feeling, target-specific IR.
const CodeGenTypeCache &TypeCache;
public:
CGBuilderTy(const CodeGenTypeCache &TypeCache, llvm::LLVMContext &C)
: CGBuilderBaseTy(C), TypeCache(TypeCache) {}
CGBuilderTy(const CodeGenTypeCache &TypeCache,
llvm::LLVMContext &C, const llvm::ConstantFolder &F,
const CGBuilderInserterTy &Inserter)
: CGBuilderBaseTy(C, F, Inserter), TypeCache(TypeCache) {}
CGBuilderTy(const CodeGenTypeCache &TypeCache, llvm::Instruction *I)
: CGBuilderBaseTy(I), TypeCache(TypeCache) {}
CGBuilderTy(const CodeGenTypeCache &TypeCache, llvm::BasicBlock *BB)
: CGBuilderBaseTy(BB), TypeCache(TypeCache) {}
llvm::ConstantInt *getSize(CharUnits N) {
return llvm::ConstantInt::get(TypeCache.SizeTy, N.getQuantity());
}
llvm::ConstantInt *getSize(uint64_t N) {
return llvm::ConstantInt::get(TypeCache.SizeTy, N);
}
// Note that we intentionally hide the CreateLoad APIs that don't
// take an alignment.
llvm::LoadInst *CreateLoad(Address Addr, const llvm::Twine &Name = "") {
return CreateAlignedLoad(Addr.getPointer(),
Addr.getAlignment().getQuantity(),
Name);
}
llvm::LoadInst *CreateLoad(Address Addr, const char *Name) {
// This overload is required to prevent string literals from
// ending up in the IsVolatile overload.
return CreateAlignedLoad(Addr.getPointer(),
Addr.getAlignment().getQuantity(),
Name);
}
llvm::LoadInst *CreateLoad(Address Addr, bool IsVolatile,
const llvm::Twine &Name = "") {
return CreateAlignedLoad(Addr.getPointer(),
Addr.getAlignment().getQuantity(),
IsVolatile,
Name);
}
using CGBuilderBaseTy::CreateAlignedLoad;
llvm::LoadInst *CreateAlignedLoad(llvm::Value *Addr, CharUnits Align,
const llvm::Twine &Name = "") {
return CreateAlignedLoad(Addr, Align.getQuantity(), Name);
}
llvm::LoadInst *CreateAlignedLoad(llvm::Value *Addr, CharUnits Align,
const char *Name) {
return CreateAlignedLoad(Addr, Align.getQuantity(), Name);
}
llvm::LoadInst *CreateAlignedLoad(llvm::Type *Ty, llvm::Value *Addr,
CharUnits Align,
const llvm::Twine &Name = "") {
assert(Addr->getType()->getPointerElementType() == Ty);
return CreateAlignedLoad(Addr, Align.getQuantity(), Name);
}
// Note that we intentionally hide the CreateStore APIs that don't
// take an alignment.
llvm::StoreInst *CreateStore(llvm::Value *Val, Address Addr,
bool IsVolatile = false) {
return CreateAlignedStore(Val, Addr.getPointer(),
Addr.getAlignment().getQuantity(), IsVolatile);
}
using CGBuilderBaseTy::CreateAlignedStore;
llvm::StoreInst *CreateAlignedStore(llvm::Value *Val, llvm::Value *Addr,
CharUnits Align, bool IsVolatile = false) {
return CreateAlignedStore(Val, Addr, Align.getQuantity(), IsVolatile);
}
// FIXME: these "default-aligned" APIs should be removed,
// but I don't feel like fixing all the builtin code right now.
llvm::StoreInst *CreateDefaultAlignedStore(llvm::Value *Val,
llvm::Value *Addr,
bool IsVolatile = false) {
return CGBuilderBaseTy::CreateStore(Val, Addr, IsVolatile);
}
/// Emit a load from an i1 flag variable.
llvm::LoadInst *CreateFlagLoad(llvm::Value *Addr,
const llvm::Twine &Name = "") {
assert(Addr->getType()->getPointerElementType() == getInt1Ty());
return CreateAlignedLoad(getInt1Ty(), Addr, CharUnits::One(), Name);
}
/// Emit a store to an i1 flag variable.
llvm::StoreInst *CreateFlagStore(bool Value, llvm::Value *Addr) {
assert(Addr->getType()->getPointerElementType() == getInt1Ty());
return CreateAlignedStore(getInt1(Value), Addr, CharUnits::One());
}
using CGBuilderBaseTy::CreateBitCast;
Address CreateBitCast(Address Addr, llvm::Type *Ty,
const llvm::Twine &Name = "") {
return Address(CreateBitCast(Addr.getPointer(), Ty, Name),
Addr.getAlignment());
}
/// Cast the element type of the given address to a different type,
/// preserving information like the alignment and address space.
Address CreateElementBitCast(Address Addr, llvm::Type *Ty,
const llvm::Twine &Name = "") {
auto PtrTy = Ty->getPointerTo(Addr.getAddressSpace());
return CreateBitCast(Addr, PtrTy, Name);
}
using CGBuilderBaseTy::CreatePointerBitCastOrAddrSpaceCast;
Address CreatePointerBitCastOrAddrSpaceCast(Address Addr, llvm::Type *Ty,
const llvm::Twine &Name = "") {
llvm::Value *Ptr =
CreatePointerBitCastOrAddrSpaceCast(Addr.getPointer(), Ty, Name);
return Address(Ptr, Addr.getAlignment());
}
using CGBuilderBaseTy::CreateStructGEP;
Address CreateStructGEP(Address Addr, unsigned Index, CharUnits Offset,
const llvm::Twine &Name = "") {
return Address(CreateStructGEP(Addr.getElementType(),
Addr.getPointer(), Index, Name),
Addr.getAlignment().alignmentAtOffset(Offset));
}
Address CreateStructGEP(Address Addr, unsigned Index,
const llvm::StructLayout *Layout,
const llvm::Twine &Name = "") {
auto Offset = CharUnits::fromQuantity(Layout->getElementOffset(Index));
return CreateStructGEP(Addr, Index, Offset, Name);
}
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
/// Given
/// %addr = [n x T]* ...
/// produce
/// %name = getelementptr inbounds %addr, i64 0, i64 index
/// where i64 is actually the target word size.
///
/// This API assumes that drilling into an array like this is always
/// an inbounds operation.
///
/// \param EltSize - the size of the type T in bytes
Address CreateConstArrayGEP(Address Addr, uint64_t Index, CharUnits EltSize,
const llvm::Twine &Name = "") {
return Address(CreateInBoundsGEP(Addr.getPointer(),
{getSize(CharUnits::Zero()),
getSize(Index)},
Name),
Addr.getAlignment().alignmentAtOffset(Index * EltSize));
}
/// Given
/// %addr = T* ...
/// produce
/// %name = getelementptr inbounds %addr, i64 index
/// where i64 is actually the target word size.
///
/// \param EltSize - the size of the type T in bytes
Address CreateConstInBoundsGEP(Address Addr, uint64_t Index,
CharUnits EltSize,
const llvm::Twine &Name = "") {
return Address(CreateInBoundsGEP(Addr.getElementType(), Addr.getPointer(),
getSize(Index), Name),
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
Addr.getAlignment().alignmentAtOffset(Index * EltSize));
}
/// Given
/// %addr = T* ...
/// produce
/// %name = getelementptr inbounds %addr, i64 index
/// where i64 is actually the target word size.
///
/// \param EltSize - the size of the type T in bytes
Address CreateConstGEP(Address Addr, uint64_t Index, CharUnits EltSize,
const llvm::Twine &Name = "") {
return Address(CreateGEP(Addr.getElementType(), Addr.getPointer(),
getSize(Index), Name),
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
Addr.getAlignment().alignmentAtOffset(Index * EltSize));
}
/// Given a pointer to i8, adjust it by a given constant offset.
Address CreateConstInBoundsByteGEP(Address Addr, CharUnits Offset,
const llvm::Twine &Name = "") {
assert(Addr.getElementType() == TypeCache.Int8Ty);
return Address(CreateInBoundsGEP(Addr.getPointer(), getSize(Offset), Name),
Addr.getAlignment().alignmentAtOffset(Offset));
}
Address CreateConstByteGEP(Address Addr, CharUnits Offset,
const llvm::Twine &Name = "") {
assert(Addr.getElementType() == TypeCache.Int8Ty);
return Address(CreateGEP(Addr.getPointer(), getSize(Offset), Name),
Addr.getAlignment().alignmentAtOffset(Offset));
}
llvm::Value *CreateConstInBoundsByteGEP(llvm::Value *Ptr, CharUnits Offset,
const llvm::Twine &Name = "") {
assert(Ptr->getType()->getPointerElementType() == TypeCache.Int8Ty);
return CreateInBoundsGEP(Ptr, getSize(Offset), Name);
}
llvm::Value *CreateConstByteGEP(llvm::Value *Ptr, CharUnits Offset,
const llvm::Twine &Name = "") {
assert(Ptr->getType()->getPointerElementType() == TypeCache.Int8Ty);
return CreateGEP(Ptr, getSize(Offset), Name);
}
using CGBuilderBaseTy::CreateMemCpy;
llvm::CallInst *CreateMemCpy(Address Dest, Address Src, llvm::Value *Size,
bool IsVolatile = false) {
auto Align = std::min(Dest.getAlignment(), Src.getAlignment());
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
return CreateMemCpy(Dest.getPointer(), Src.getPointer(), Size,
Align.getQuantity(), IsVolatile);
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
}
llvm::CallInst *CreateMemCpy(Address Dest, Address Src, uint64_t Size,
bool IsVolatile = false) {
auto Align = std::min(Dest.getAlignment(), Src.getAlignment());
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
return CreateMemCpy(Dest.getPointer(), Src.getPointer(), Size,
Align.getQuantity(), IsVolatile);
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
}
using CGBuilderBaseTy::CreateMemMove;
llvm::CallInst *CreateMemMove(Address Dest, Address Src, llvm::Value *Size,
bool IsVolatile = false) {
auto Align = std::min(Dest.getAlignment(), Src.getAlignment());
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
return CreateMemMove(Dest.getPointer(), Src.getPointer(), Size,
Align.getQuantity(), IsVolatile);
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
}
using CGBuilderBaseTy::CreateMemSet;
llvm::CallInst *CreateMemSet(Address Dest, llvm::Value *Value,
llvm::Value *Size, bool IsVolatile = false) {
return CreateMemSet(Dest.getPointer(), Value, Size,
Dest.getAlignment().getQuantity(), IsVolatile);
}
};
} // end namespace CodeGen
} // end namespace clang
#endif