2008-08-23 11:10:25 +08:00
|
|
|
//===-- CGValue.h - LLVM CodeGen wrappers for llvm::Value* ------*- C++ -*-===//
|
|
|
|
//
|
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
|
|
|
// This file is distributed under the University of Illinois Open Source
|
|
|
|
// License. See LICENSE.TXT for details.
|
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//
|
|
|
|
// These classes implement wrappers around llvm::Value in order to
|
|
|
|
// fully represent the range of values for C L- and R- values.
|
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
#ifndef CLANG_CODEGEN_CGVALUE_H
|
|
|
|
#define CLANG_CODEGEN_CGVALUE_H
|
|
|
|
|
2010-08-21 10:53:44 +08:00
|
|
|
#include "clang/AST/ASTContext.h"
|
2008-08-23 11:10:25 +08:00
|
|
|
#include "clang/AST/Type.h"
|
|
|
|
|
2008-09-09 09:06:48 +08:00
|
|
|
namespace llvm {
|
|
|
|
class Constant;
|
|
|
|
class Value;
|
|
|
|
}
|
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
namespace clang {
|
2008-08-29 16:11:39 +08:00
|
|
|
class ObjCPropertyRefExpr;
|
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
namespace CodeGen {
|
2010-04-06 05:36:35 +08:00
|
|
|
class CGBitFieldInfo;
|
2008-08-23 11:10:25 +08:00
|
|
|
|
|
|
|
/// RValue - This trivial value class is used to represent the result of an
|
|
|
|
/// expression that is evaluated. It can be one of three things: either a
|
|
|
|
/// simple LLVM SSA value, a pair of SSA values for complex numbers, or the
|
|
|
|
/// address of an aggregate value in memory.
|
|
|
|
class RValue {
|
2010-05-02 22:59:10 +08:00
|
|
|
enum Flavor { Scalar, Complex, Aggregate };
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2010-05-02 22:59:10 +08:00
|
|
|
// Stores first value and flavor.
|
|
|
|
llvm::PointerIntPair<llvm::Value *, 2, Flavor> V1;
|
|
|
|
// Stores second value and volatility.
|
|
|
|
llvm::PointerIntPair<llvm::Value *, 1, bool> V2;
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2010-05-02 22:59:10 +08:00
|
|
|
public:
|
|
|
|
bool isScalar() const { return V1.getInt() == Scalar; }
|
|
|
|
bool isComplex() const { return V1.getInt() == Complex; }
|
|
|
|
bool isAggregate() const { return V1.getInt() == Aggregate; }
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2010-05-02 22:59:10 +08:00
|
|
|
bool isVolatileQualified() const { return V2.getInt(); }
|
2009-05-24 04:21:36 +08:00
|
|
|
|
2009-11-04 00:11:57 +08:00
|
|
|
/// getScalarVal() - Return the Value* of this scalar value.
|
2008-08-23 11:10:25 +08:00
|
|
|
llvm::Value *getScalarVal() const {
|
|
|
|
assert(isScalar() && "Not a scalar!");
|
2010-05-02 22:59:10 +08:00
|
|
|
return V1.getPointer();
|
2008-08-23 11:10:25 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/// getComplexVal - Return the real/imag components of this complex value.
|
|
|
|
///
|
|
|
|
std::pair<llvm::Value *, llvm::Value *> getComplexVal() const {
|
2010-05-02 22:59:10 +08:00
|
|
|
return std::make_pair(V1.getPointer(), V2.getPointer());
|
2008-08-23 11:10:25 +08:00
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
/// getAggregateAddr() - Return the Value* of the address of the aggregate.
|
|
|
|
llvm::Value *getAggregateAddr() const {
|
|
|
|
assert(isAggregate() && "Not an aggregate!");
|
2010-05-02 22:59:10 +08:00
|
|
|
return V1.getPointer();
|
2008-08-23 11:10:25 +08:00
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
static RValue get(llvm::Value *V) {
|
|
|
|
RValue ER;
|
2010-05-02 22:59:10 +08:00
|
|
|
ER.V1.setPointer(V);
|
|
|
|
ER.V1.setInt(Scalar);
|
|
|
|
ER.V2.setInt(false);
|
2008-08-23 11:10:25 +08:00
|
|
|
return ER;
|
|
|
|
}
|
|
|
|
static RValue getComplex(llvm::Value *V1, llvm::Value *V2) {
|
|
|
|
RValue ER;
|
2010-05-02 22:59:10 +08:00
|
|
|
ER.V1.setPointer(V1);
|
|
|
|
ER.V2.setPointer(V2);
|
|
|
|
ER.V1.setInt(Complex);
|
|
|
|
ER.V2.setInt(false);
|
2008-08-23 11:10:25 +08:00
|
|
|
return ER;
|
|
|
|
}
|
|
|
|
static RValue getComplex(const std::pair<llvm::Value *, llvm::Value *> &C) {
|
2010-05-02 22:59:10 +08:00
|
|
|
return getComplex(C.first, C.second);
|
2008-08-23 11:10:25 +08:00
|
|
|
}
|
2009-05-24 04:21:36 +08:00
|
|
|
// FIXME: Aggregate rvalues need to retain information about whether they are
|
|
|
|
// volatile or not. Remove default to find all places that probably get this
|
|
|
|
// wrong.
|
2010-05-02 22:59:10 +08:00
|
|
|
static RValue getAggregate(llvm::Value *V, bool Volatile = false) {
|
2008-08-23 11:10:25 +08:00
|
|
|
RValue ER;
|
2010-05-02 22:59:10 +08:00
|
|
|
ER.V1.setPointer(V);
|
|
|
|
ER.V1.setInt(Aggregate);
|
|
|
|
ER.V2.setInt(Volatile);
|
2008-08-23 11:10:25 +08:00
|
|
|
return ER;
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
/// LValue - This represents an lvalue references. Because C/C++ allow
|
|
|
|
/// bitfields, this is not a simple LLVM pointer, it may be a pointer plus a
|
|
|
|
/// bitrange.
|
|
|
|
class LValue {
|
|
|
|
// FIXME: alignment?
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
enum {
|
|
|
|
Simple, // This is a normal l-value, use getAddress().
|
|
|
|
VectorElt, // This is a vector element l-value (V[i]), use getVector*
|
|
|
|
BitField, // This is a bitfield l-value, use getBitfield*.
|
2008-08-29 16:11:39 +08:00
|
|
|
ExtVectorElt, // This is an extended vector subset, use getExtVectorComp
|
2008-11-23 06:30:21 +08:00
|
|
|
PropertyRef, // This is an Objective-C property reference, use
|
2008-08-29 16:11:39 +08:00
|
|
|
// getPropertyRefExpr
|
2008-11-23 06:30:21 +08:00
|
|
|
KVCRef // This is an objective-c 'implicit' property ref,
|
|
|
|
// use getKVCRefExpr
|
2008-08-23 11:10:25 +08:00
|
|
|
} LVType;
|
2008-11-19 08:59:10 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
llvm::Value *V;
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
union {
|
|
|
|
// Index into a vector subscript: V[i]
|
|
|
|
llvm::Value *VectorIdx;
|
|
|
|
|
|
|
|
// ExtVector element subset: V.xyx
|
|
|
|
llvm::Constant *VectorElts;
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
// BitField start bit and size
|
2010-04-06 05:36:35 +08:00
|
|
|
const CGBitFieldInfo *BitFieldInfo;
|
2008-08-29 16:11:39 +08:00
|
|
|
|
|
|
|
// Obj-C property reference expression
|
|
|
|
const ObjCPropertyRefExpr *PropertyRefExpr;
|
2008-08-23 11:10:25 +08:00
|
|
|
};
|
|
|
|
|
2009-09-25 03:53:00 +08:00
|
|
|
// 'const' is unused here
|
|
|
|
Qualifiers Quals;
|
2008-11-19 08:59:10 +08:00
|
|
|
|
2010-08-21 10:39:23 +08:00
|
|
|
/// The alignment to use when accessing this lvalue.
|
2010-08-26 14:02:12 +08:00
|
|
|
unsigned short Alignment;
|
2010-08-21 10:39:23 +08:00
|
|
|
|
2008-11-21 04:53:20 +08:00
|
|
|
// objective-c's ivar
|
|
|
|
bool Ivar:1;
|
2009-09-18 08:04:00 +08:00
|
|
|
|
|
|
|
// objective-c's ivar is an array
|
2009-09-22 02:54:29 +08:00
|
|
|
bool ObjIsArray:1;
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2009-02-21 08:30:43 +08:00
|
|
|
// LValue is non-gc'able for any reason, including being a parameter or local
|
|
|
|
// variable.
|
|
|
|
bool NonGC: 1;
|
2008-11-21 04:53:20 +08:00
|
|
|
|
2009-05-05 07:27:20 +08:00
|
|
|
// Lvalue is a global reference of an objective-c object
|
|
|
|
bool GlobalObjCRef : 1;
|
2010-07-21 04:30:03 +08:00
|
|
|
|
|
|
|
// Lvalue is a thread local reference
|
|
|
|
bool ThreadLocalRef : 1;
|
2009-05-05 07:27:20 +08:00
|
|
|
|
2009-09-25 06:25:38 +08:00
|
|
|
Expr *BaseIvarExp;
|
2010-10-15 07:06:10 +08:00
|
|
|
|
|
|
|
/// TBAAInfo - TBAA information to attach to dereferences of this LValue.
|
|
|
|
llvm::MDNode *TBAAInfo;
|
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
private:
|
2010-10-15 07:06:10 +08:00
|
|
|
void Initialize(Qualifiers Quals, unsigned Alignment = 0,
|
|
|
|
llvm::MDNode *TBAAInfo = 0) {
|
2009-09-25 03:53:00 +08:00
|
|
|
this->Quals = Quals;
|
2010-08-21 10:39:23 +08:00
|
|
|
this->Alignment = Alignment;
|
|
|
|
assert(this->Alignment == Alignment && "Alignment exceeds allowed max!");
|
|
|
|
|
|
|
|
// Initialize Objective-C flags.
|
2009-09-25 03:53:00 +08:00
|
|
|
this->Ivar = this->ObjIsArray = this->NonGC = this->GlobalObjCRef = false;
|
2010-07-21 04:30:03 +08:00
|
|
|
this->ThreadLocalRef = false;
|
2009-09-25 06:25:38 +08:00
|
|
|
this->BaseIvarExp = 0;
|
2010-10-15 07:06:10 +08:00
|
|
|
this->TBAAInfo = TBAAInfo;
|
2008-08-23 11:10:25 +08:00
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
public:
|
|
|
|
bool isSimple() const { return LVType == Simple; }
|
|
|
|
bool isVectorElt() const { return LVType == VectorElt; }
|
2010-04-06 05:36:35 +08:00
|
|
|
bool isBitField() const { return LVType == BitField; }
|
2008-08-23 11:10:25 +08:00
|
|
|
bool isExtVectorElt() const { return LVType == ExtVectorElt; }
|
2008-08-29 16:11:39 +08:00
|
|
|
bool isPropertyRef() const { return LVType == PropertyRef; }
|
2008-11-23 06:30:21 +08:00
|
|
|
bool isKVCRef() const { return LVType == KVCRef; }
|
2008-08-29 16:11:39 +08:00
|
|
|
|
2009-09-25 03:53:00 +08:00
|
|
|
bool isVolatileQualified() const { return Quals.hasVolatile(); }
|
|
|
|
bool isRestrictQualified() const { return Quals.hasRestrict(); }
|
|
|
|
unsigned getVRQualifiers() const {
|
|
|
|
return Quals.getCVRQualifiers() & ~Qualifiers::Const;
|
2009-02-17 06:25:49 +08:00
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-11-21 04:53:20 +08:00
|
|
|
bool isObjCIvar() const { return Ivar; }
|
2010-08-21 11:51:29 +08:00
|
|
|
void setObjCIvar(bool Value) { Ivar = Value; }
|
|
|
|
|
2009-09-22 02:54:29 +08:00
|
|
|
bool isObjCArray() const { return ObjIsArray; }
|
2010-08-21 11:51:29 +08:00
|
|
|
void setObjCArray(bool Value) { ObjIsArray = Value; }
|
2010-08-21 11:22:38 +08:00
|
|
|
|
2009-02-21 08:30:43 +08:00
|
|
|
bool isNonGC () const { return NonGC; }
|
2010-08-21 11:22:38 +08:00
|
|
|
void setNonGC(bool Value) { NonGC = Value; }
|
|
|
|
|
2009-05-05 07:27:20 +08:00
|
|
|
bool isGlobalObjCRef() const { return GlobalObjCRef; }
|
2010-08-21 11:51:29 +08:00
|
|
|
void setGlobalObjCRef(bool Value) { GlobalObjCRef = Value; }
|
|
|
|
|
2010-07-21 04:30:03 +08:00
|
|
|
bool isThreadLocalRef() const { return ThreadLocalRef; }
|
2010-08-21 11:51:29 +08:00
|
|
|
void setThreadLocalRef(bool Value) { ThreadLocalRef = Value;}
|
|
|
|
|
|
|
|
bool isObjCWeak() const {
|
|
|
|
return Quals.getObjCGCAttr() == Qualifiers::Weak;
|
|
|
|
}
|
|
|
|
bool isObjCStrong() const {
|
|
|
|
return Quals.getObjCGCAttr() == Qualifiers::Strong;
|
|
|
|
}
|
2009-09-25 06:25:38 +08:00
|
|
|
|
|
|
|
Expr *getBaseIvarExp() const { return BaseIvarExp; }
|
|
|
|
void setBaseIvarExp(Expr *V) { BaseIvarExp = V; }
|
2009-07-22 11:08:17 +08:00
|
|
|
|
2010-10-15 07:06:10 +08:00
|
|
|
llvm::MDNode *getTBAAInfo() const { return TBAAInfo; }
|
|
|
|
void setTBAAInfo(llvm::MDNode *N) { TBAAInfo = N; }
|
|
|
|
|
2010-08-21 11:29:54 +08:00
|
|
|
const Qualifiers &getQuals() const { return Quals; }
|
|
|
|
Qualifiers &getQuals() { return Quals; }
|
|
|
|
|
2009-09-25 03:53:00 +08:00
|
|
|
unsigned getAddressSpace() const { return Quals.getAddressSpace(); }
|
2009-07-22 11:08:17 +08:00
|
|
|
|
2010-08-21 10:39:23 +08:00
|
|
|
unsigned getAlignment() const { return Alignment; }
|
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
// simple lvalue
|
|
|
|
llvm::Value *getAddress() const { assert(isSimple()); return V; }
|
2010-04-06 05:36:35 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
// vector elt lvalue
|
|
|
|
llvm::Value *getVectorAddr() const { assert(isVectorElt()); return V; }
|
|
|
|
llvm::Value *getVectorIdx() const { assert(isVectorElt()); return VectorIdx; }
|
2010-04-06 05:36:35 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
// extended vector elements.
|
|
|
|
llvm::Value *getExtVectorAddr() const { assert(isExtVectorElt()); return V; }
|
|
|
|
llvm::Constant *getExtVectorElts() const {
|
|
|
|
assert(isExtVectorElt());
|
|
|
|
return VectorElts;
|
|
|
|
}
|
2010-04-06 05:36:35 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
// bitfield lvalue
|
2010-04-08 10:59:45 +08:00
|
|
|
llvm::Value *getBitFieldBaseAddr() const {
|
2010-04-06 05:36:35 +08:00
|
|
|
assert(isBitField());
|
|
|
|
return V;
|
2008-08-23 11:10:25 +08:00
|
|
|
}
|
2010-04-06 05:36:35 +08:00
|
|
|
const CGBitFieldInfo &getBitFieldInfo() const {
|
|
|
|
assert(isBitField());
|
|
|
|
return *BitFieldInfo;
|
2008-08-23 11:10:25 +08:00
|
|
|
}
|
2010-04-06 05:36:35 +08:00
|
|
|
|
2008-08-29 16:11:39 +08:00
|
|
|
// property ref lvalue
|
|
|
|
const ObjCPropertyRefExpr *getPropertyRefExpr() const {
|
|
|
|
assert(isPropertyRef());
|
|
|
|
return PropertyRefExpr;
|
|
|
|
}
|
2008-08-23 11:10:25 +08:00
|
|
|
|
2008-11-23 06:30:21 +08:00
|
|
|
// 'implicit' property ref lvalue
|
2010-12-02 09:19:52 +08:00
|
|
|
const ObjCPropertyRefExpr *getKVCRefExpr() const {
|
2008-11-23 06:30:21 +08:00
|
|
|
assert(isKVCRef());
|
2010-12-02 09:19:52 +08:00
|
|
|
return PropertyRefExpr;
|
2008-11-23 06:30:21 +08:00
|
|
|
}
|
|
|
|
|
2010-08-21 11:58:45 +08:00
|
|
|
static LValue MakeAddr(llvm::Value *V, QualType T, unsigned Alignment,
|
2010-10-15 07:06:10 +08:00
|
|
|
ASTContext &Context,
|
|
|
|
llvm::MDNode *TBAAInfo = 0) {
|
2010-08-21 11:58:45 +08:00
|
|
|
Qualifiers Quals = Context.getCanonicalType(T).getQualifiers();
|
|
|
|
Quals.setObjCGCAttr(Context.getObjCGCAttrKind(T));
|
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
LValue R;
|
|
|
|
R.LVType = Simple;
|
|
|
|
R.V = V;
|
2010-10-15 07:06:10 +08:00
|
|
|
R.Initialize(Quals, Alignment, TBAAInfo);
|
2008-08-23 11:10:25 +08:00
|
|
|
return R;
|
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
static LValue MakeVectorElt(llvm::Value *Vec, llvm::Value *Idx,
|
2009-09-25 03:53:00 +08:00
|
|
|
unsigned CVR) {
|
2008-08-23 11:10:25 +08:00
|
|
|
LValue R;
|
|
|
|
R.LVType = VectorElt;
|
|
|
|
R.V = Vec;
|
|
|
|
R.VectorIdx = Idx;
|
2010-08-21 10:31:58 +08:00
|
|
|
R.Initialize(Qualifiers::fromCVRMask(CVR));
|
2008-08-23 11:10:25 +08:00
|
|
|
return R;
|
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
static LValue MakeExtVectorElt(llvm::Value *Vec, llvm::Constant *Elts,
|
2009-09-25 03:53:00 +08:00
|
|
|
unsigned CVR) {
|
2008-08-23 11:10:25 +08:00
|
|
|
LValue R;
|
|
|
|
R.LVType = ExtVectorElt;
|
|
|
|
R.V = Vec;
|
|
|
|
R.VectorElts = Elts;
|
2010-08-21 10:31:58 +08:00
|
|
|
R.Initialize(Qualifiers::fromCVRMask(CVR));
|
2008-08-23 11:10:25 +08:00
|
|
|
return R;
|
|
|
|
}
|
|
|
|
|
2010-04-08 10:59:45 +08:00
|
|
|
/// \brief Create a new object to represent a bit-field access.
|
|
|
|
///
|
|
|
|
/// \param BaseValue - The base address of the structure containing the
|
|
|
|
/// bit-field.
|
|
|
|
/// \param Info - The information describing how to perform the bit-field
|
|
|
|
/// access.
|
|
|
|
static LValue MakeBitfield(llvm::Value *BaseValue, const CGBitFieldInfo &Info,
|
2010-04-06 09:07:44 +08:00
|
|
|
unsigned CVR) {
|
2008-08-23 11:10:25 +08:00
|
|
|
LValue R;
|
|
|
|
R.LVType = BitField;
|
2010-04-08 10:59:45 +08:00
|
|
|
R.V = BaseValue;
|
2010-04-06 05:36:35 +08:00
|
|
|
R.BitFieldInfo = &Info;
|
2010-08-21 10:31:58 +08:00
|
|
|
R.Initialize(Qualifiers::fromCVRMask(CVR));
|
2008-08-23 11:10:25 +08:00
|
|
|
return R;
|
|
|
|
}
|
2008-08-29 16:11:39 +08:00
|
|
|
|
2009-05-16 15:57:57 +08:00
|
|
|
// FIXME: It is probably bad that we aren't emitting the target when we build
|
|
|
|
// the lvalue. However, this complicates the code a bit, and I haven't figured
|
|
|
|
// out how to make it go wrong yet.
|
2008-08-29 16:11:39 +08:00
|
|
|
static LValue MakePropertyRef(const ObjCPropertyRefExpr *E,
|
2009-09-25 03:53:00 +08:00
|
|
|
unsigned CVR) {
|
2008-08-29 16:11:39 +08:00
|
|
|
LValue R;
|
|
|
|
R.LVType = PropertyRef;
|
|
|
|
R.PropertyRefExpr = E;
|
2010-08-21 10:31:58 +08:00
|
|
|
R.Initialize(Qualifiers::fromCVRMask(CVR));
|
2008-08-29 16:11:39 +08:00
|
|
|
return R;
|
|
|
|
}
|
2009-09-09 23:08:12 +08:00
|
|
|
|
2010-12-02 09:19:52 +08:00
|
|
|
static LValue MakeKVCRef(const ObjCPropertyRefExpr *E,
|
2009-09-25 03:53:00 +08:00
|
|
|
unsigned CVR) {
|
2008-11-23 06:30:21 +08:00
|
|
|
LValue R;
|
|
|
|
R.LVType = KVCRef;
|
2010-12-02 09:19:52 +08:00
|
|
|
R.PropertyRefExpr = E;
|
2010-08-21 10:31:58 +08:00
|
|
|
R.Initialize(Qualifiers::fromCVRMask(CVR));
|
2008-11-23 06:30:21 +08:00
|
|
|
return R;
|
|
|
|
}
|
2008-08-23 11:10:25 +08:00
|
|
|
};
|
|
|
|
|
2010-09-15 18:14:12 +08:00
|
|
|
/// An aggregate value slot.
|
|
|
|
class AggValueSlot {
|
2010-09-16 08:20:07 +08:00
|
|
|
/// The address.
|
2010-09-16 11:16:41 +08:00
|
|
|
llvm::Value *Addr;
|
2010-09-16 08:20:07 +08:00
|
|
|
|
|
|
|
// Associated flags.
|
|
|
|
bool VolatileFlag : 1;
|
|
|
|
bool LifetimeFlag : 1;
|
|
|
|
bool RequiresGCollection : 1;
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
|
|
|
|
/// IsZeroed - This is set to true if the destination is known to be zero
|
|
|
|
/// before the assignment into it. This means that zero fields don't need to
|
|
|
|
/// be set.
|
|
|
|
bool IsZeroed : 1;
|
2010-09-15 18:14:12 +08:00
|
|
|
|
|
|
|
public:
|
|
|
|
/// ignored - Returns an aggregate value slot indicating that the
|
|
|
|
/// aggregate value is being ignored.
|
|
|
|
static AggValueSlot ignored() {
|
|
|
|
AggValueSlot AV;
|
2010-09-16 11:16:41 +08:00
|
|
|
AV.Addr = 0;
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
AV.VolatileFlag = AV.LifetimeFlag = AV.RequiresGCollection = AV.IsZeroed =0;
|
2010-09-15 18:14:12 +08:00
|
|
|
return AV;
|
|
|
|
}
|
|
|
|
|
|
|
|
/// forAddr - Make a slot for an aggregate value.
|
|
|
|
///
|
|
|
|
/// \param Volatile - true if the slot should be volatile-initialized
|
|
|
|
/// \param LifetimeExternallyManaged - true if the slot's lifetime
|
|
|
|
/// is being externally managed; false if a destructor should be
|
|
|
|
/// registered for any temporaries evaluated into the slot
|
2010-09-16 11:13:23 +08:00
|
|
|
/// \param RequiresGCollection - true if the slot is located
|
|
|
|
/// somewhere that ObjC GC calls should be emitted for
|
2010-09-15 18:14:12 +08:00
|
|
|
static AggValueSlot forAddr(llvm::Value *Addr, bool Volatile,
|
2010-09-16 08:20:07 +08:00
|
|
|
bool LifetimeExternallyManaged,
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
bool RequiresGCollection = false,
|
|
|
|
bool IsZeroed = false) {
|
2010-09-15 18:14:12 +08:00
|
|
|
AggValueSlot AV;
|
2010-09-16 11:16:41 +08:00
|
|
|
AV.Addr = Addr;
|
2010-09-16 11:13:23 +08:00
|
|
|
AV.VolatileFlag = Volatile;
|
|
|
|
AV.LifetimeFlag = LifetimeExternallyManaged;
|
|
|
|
AV.RequiresGCollection = RequiresGCollection;
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
AV.IsZeroed = IsZeroed;
|
2010-09-15 18:14:12 +08:00
|
|
|
return AV;
|
|
|
|
}
|
|
|
|
|
2010-09-16 08:20:07 +08:00
|
|
|
static AggValueSlot forLValue(LValue LV, bool LifetimeExternallyManaged,
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
bool RequiresGCollection = false) {
|
2010-09-15 18:14:12 +08:00
|
|
|
return forAddr(LV.getAddress(), LV.isVolatileQualified(),
|
2010-09-16 08:20:07 +08:00
|
|
|
LifetimeExternallyManaged, RequiresGCollection);
|
2010-09-15 18:14:12 +08:00
|
|
|
}
|
2010-10-23 06:05:03 +08:00
|
|
|
|
2010-09-15 18:14:12 +08:00
|
|
|
bool isLifetimeExternallyManaged() const {
|
2010-09-16 08:20:07 +08:00
|
|
|
return LifetimeFlag;
|
2010-09-15 18:14:12 +08:00
|
|
|
}
|
2010-11-17 08:07:33 +08:00
|
|
|
void setLifetimeExternallyManaged(bool Managed = true) {
|
|
|
|
LifetimeFlag = Managed;
|
2010-09-15 18:14:12 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
bool isVolatile() const {
|
2010-09-16 08:20:07 +08:00
|
|
|
return VolatileFlag;
|
2010-09-15 18:14:12 +08:00
|
|
|
}
|
|
|
|
|
2010-09-16 11:13:23 +08:00
|
|
|
bool requiresGCollection() const {
|
2010-09-16 08:20:07 +08:00
|
|
|
return RequiresGCollection;
|
|
|
|
}
|
|
|
|
|
2010-09-15 18:14:12 +08:00
|
|
|
llvm::Value *getAddr() const {
|
2010-09-16 11:16:41 +08:00
|
|
|
return Addr;
|
2010-09-15 18:14:12 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
bool isIgnored() const {
|
2010-09-16 11:16:41 +08:00
|
|
|
return Addr == 0;
|
2010-09-15 18:14:12 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
RValue asRValue() const {
|
|
|
|
return RValue::getAggregate(getAddr(), isVolatile());
|
|
|
|
}
|
|
|
|
|
Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch). This has three pieces:
1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
whether they are profitable to emit a memset + inividual stores vs
stores for everything.
The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform. The two testcases
are illustrative of the scenarios this catches. We now codegen test9 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
%.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
%tmp = load i32* %X.addr, align 4
store i32 %tmp, i32* %.array
and test10 into:
call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
%tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
%tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
%tmp2 = load i32* %X.addr, align 4
store i32 %tmp2, i32* %tmp1, align 4
%tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
%tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
%tmp11 = load i32* %X.addr, align 4
store i32 %tmp11, i32* %tmp10, align 4
Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases. This resolves
PR279.
llvm-svn: 120692
2010-12-02 15:07:26 +08:00
|
|
|
void setZeroed(bool V = true) { IsZeroed = V; }
|
|
|
|
bool isZeroed() const {
|
|
|
|
return IsZeroed;
|
|
|
|
}
|
2010-09-15 18:14:12 +08:00
|
|
|
};
|
|
|
|
|
2008-08-23 11:10:25 +08:00
|
|
|
} // end namespace CodeGen
|
|
|
|
} // end namespace clang
|
|
|
|
|
|
|
|
#endif
|