pjs/js/tamarin/codegen/CodegenMIR.h

1646 строки
48 KiB
C++

/* ***** BEGIN LICENSE BLOCK *****
* Version: MPL 1.1/GPL 2.0/LGPL 2.1
*
* The contents of this file are subject to the Mozilla Public License Version 1.1 (the
* "License"); you may not use this file except in compliance with the License. You may obtain
* a copy of the License at http://www.mozilla.org/MPL/
*
* Software distributed under the License is distributed on an "AS IS" basis, WITHOUT
* WARRANTY OF ANY KIND, either express or implied. See the License for the specific
* language governing rights and limitations under the License.
*
* The Original Code is [Open Source Virtual Machine.]
*
* The Initial Developer of the Original Code is Adobe System Incorporated. Portions created
* by the Initial Developer are Copyright (C)[ 2004-2006 ] Adobe Systems Incorporated. All Rights
* Reserved.
*
* Contributor(s): Adobe AS3 Team
*
* Alternatively, the contents of this file may be used under the terms of either the GNU
* General Public License Version 2 or later (the "GPL"), or the GNU Lesser General Public
* License Version 2.1 or later (the "LGPL"), in which case the provisions of the GPL or the
* LGPL are applicable instead of those above. If you wish to allow use of your version of this
* file only under the terms of either the GPL or the LGPL, and not to allow others to use your
* version of this file under the terms of the MPL, indicate your decision by deleting provisions
* above and replace them with the notice and other provisions required by the GPL or the
* LGPL. If you do not delete the provisions above, a recipient may use your version of this file
* under the terms of any one of the MPL, the GPL or the LGPL.
*
***** END LICENSE BLOCK ***** */
#ifndef __avmplus_CodegenMIR__
#define __avmplus_CodegenMIR__
namespace avmplus
{
// bit insert and extraction macros.
// usage v - variable, h - high bit number. l - lower bit number, q qunatity to insert
// E.g. pull out bits 6..2 of a variable BIT_EXTRACT(v,6,2).... 0xxx xx00
// wanring bit insert does not mask off upper bits of q, so you must make sure that
// the value of q is legal given h and l.
// BIT_VALUE_FIT can be used to determine if a value can be represeted by the
// specified number of bits.
#define BIT_INSERT(v,h,l,q) ( ( ( ((unsigned)v)&~(((1L<<((h+1)-(l))) - 1L)<<(l)) ) | ((q)<<(l)) ) )
#define BIT_EXTRACT(v,h,l) ( ( ((unsigned)v)&(((1L<<((h+1)-(l))) - 1L)<<(l)) ) >>(l) )
#define BIT_VALUE_FITS(v,n) ( BIT_EXTRACT(v,n-1,0) == (v) )
// rounding v up to the given 2^ quantity
#define BIT_ROUND_UP(v,q) ( (((uintptr)v)+(q)-1) & ~((q)-1) )
/**
* The CodegenMIR class is a dynamic code generator which translates
* AVM+ bytecodes into an architecture neutral intermediate representation
* suitable for common subexpression elimination, inlining, and register
* allocation.
*/
class CodegenMIR
#ifdef AVMPLUS_ARM
: public ArmAssembler
#endif
{
AvmCore * const core;
PoolObject * const pool;
MethodInfo * const info;
FrameState *state;
#ifdef AVMPLUS_MAC_CARBON
static int setjmpDummy(jmp_buf buf);
static int setjmpAddress;
static void setjmpInit();
#endif
public:
/** set to true if the no more memory. */
bool overflow;
static const int MIR_float = 0x20; // result is double
static const int MIR_oper = 0x40; // eligible for cse
enum MirOpcode
{
MIR_bb = 2,
MIR_jmpt = 3, // offset, size
MIR_cm = 4, // count, imm32 - call method
MIR_cs = 5, // count, imm32 - call static
MIR_ci = 6, // count, ptr - call indirect
MIR_icmp = 7, // ptr, imm32
MIR_ucmp = 8, // ptr, imm32
MIR_fcmp = 9, // ptr, imm32
MIR_jeq = 10, // ptr, imm32
MIR_jne = 11, // ptr, imm32
MIR_ret = 12, // ptr
MIR_jmp = 13, // ptr
MIR_jmpi = 14, // ptr, disp
MIR_st = 15, // ptr, disp, ptr
MIR_arg = 16, // pos | reg - defines methods incoming arguments
MIR_def = 17, //
MIR_use = 18, //
MIR_usea = 19, // imm32
MIR_alloc = 20,
// = 21,
MIR_ld = 22, // non-optimizable load
MIR_jlt = 23,
MIR_jle = 24,
MIR_jnlt = 25,
MIR_jnle = 26,
MIR_imm = 1 | MIR_oper, // 0,imm32
MIR_imul = 2 | MIR_oper,
MIR_neg = 3 | MIR_oper, // ptr, ptr
MIR_cmop = 4 | MIR_oper, // MIR_cm|oper, call method w/out side effects
MIR_csop = 5 | MIR_oper, // MIR_cs|oper, call static method w/out side effects
MIR_lsh = 6 | MIR_oper, // <<
MIR_rsh = 7 | MIR_oper, // >>
MIR_ush = 8 | MIR_oper, // >>>
MIR_and = 9 | MIR_oper,
MIR_or = 10 | MIR_oper,
MIR_xor = 11 | MIR_oper,
MIR_add = 12 | MIR_oper,
MIR_sub = 13 | MIR_oper, // ptr, ptr
MIR_eq = 14 | MIR_oper,
MIR_le = 15 | MIR_oper,
MIR_lt = 16 | MIR_oper,
MIR_ne = 17 | MIR_oper,
MIR_lea = 18 | MIR_oper, // ptr, disp
MIR_ldop = 22 | MIR_oper, // ptr, disp (optimizable)
// After this point are all instructions that return a double-sized
// result.
MIR_fcm = 4 | MIR_float, // count, imm32 - call method, float return
MIR_fcs = 5 | MIR_float, // count, imm32 - call static, float return
MIR_fci = 6 | MIR_float, // count, addr - call indirect, float return
MIR_fdef = 17 | MIR_float, // defines value of variable as floating point
MIR_fuse = 18 | MIR_float, //
MIR_fld = 22 | MIR_float, // float load
MIR_i2d = 1 | MIR_float | MIR_oper, // ptr
MIR_fneg = 2 | MIR_float | MIR_oper, // ptr
MIR_u2d = 3 | MIR_float | MIR_oper,
MIR_fcmop = 4 | MIR_float | MIR_oper,
MIR_fcsop = 5 | MIR_float | MIR_oper,
MIR_fadd = 6 | MIR_float | MIR_oper, // ptr, ptr
MIR_fsub = 7 | MIR_float | MIR_oper, // ptr, ptr
MIR_fmul = 8 | MIR_float | MIR_oper, // ptr, ptr
MIR_fdiv = 9 | MIR_float | MIR_oper, // ptr, ptr
MIR_fldop = 22 | MIR_float | MIR_oper, // ptr, disp (optimizable load)
MIR_last = 23 | MIR_float | MIR_oper // highest ordinal value possible
};
enum ArgNumber {
_env = 0,
_argc = 1,
_ap = 2
};
// ia32 offset = 4*ArgNumber+4
/**
* Constructor. Initializes the generator for building a method
*/
CodegenMIR(MethodInfo* info);
/**
* Constructor. Initializes the generator for native method thunks only.
*/
CodegenMIR(NativeMethod* m);
/**
* Constructor for use with emitImtThunk
*/
CodegenMIR(PoolObject* pool);
~CodegenMIR();
/**
* Generates code for the method info. The bytecode is translated
* into native machine code and the TURBO flag is set
* on the MethodInfo. The original bytecode is retained for debugging.
* @param info method to compile
*/
void epilogue(FrameState* state);
bool prologue(FrameState* state);
void emitCall(FrameState* state, AbcOpcode opcode, int method_id, int argc, Traits* result);
void emit(FrameState* state, AbcOpcode opcode, uintptr op1=0, uintptr op2=0, Traits* result=NULL);
void emitIf(FrameState* state, AbcOpcode opcode, int target, int lhs, int rhs);
void emitSwap(FrameState* state, int i, int j);
void emitCopy(FrameState* state, int src, int dest);
void emitGetscope(FrameState* state, int scope, int dest);
void emitKill(FrameState* state, int i);
void emitBlockStart(FrameState* state);
void emitBlockEnd(FrameState* state);
void emitIntConst(FrameState* state, int index, uintptr c);
void emitDoubleConst(FrameState* state, int index, double* pd);
void emitCoerce(FrameState* state, int index, Traits* type);
void emitCheckNull(FrameState* state, int index);
void emitSetContext(FrameState* state, AbstractFunction* f);
void emitSetDxns(FrameState* state);
void merge(const Value& current, Value& target);
#ifdef AVMPLUS_VERBOSE
bool verbose();
#endif
#ifndef FEATURE_BUFFER_GUARD
bool checkOverflow();
#endif
/**
* Generates code for a native method thunk.
*/
void emitNativeThunk(NativeMethod* info);
void* emitImtThunk(ImtBuilder::ImtEntry *e);
private:
void emitPrep(AbcOpcode opcode);
const byte *abcStart;
const byte *abcEnd;
//
// helpers called by generated code
//
static Atom coerce_o(Atom v);
#ifdef AVMPLUS_ARM
//
// ARM needs helpers for floating point
//
static double fadd(double x, double y);
static double fsub(double x, double y);
static double fmul(double x, double y);
static double fdiv(double x, double y);
static int fcmp(double x, double y);
static double i2d(int i);
static double u2d(uint32 i);
#endif
public:
//
// Register allocation stuff follows
// Encoded in 7b quantity
//
#ifdef AVMPLUS_PPC
#ifdef AVMPLUS_VERBOSE
static const char *gpregNames[];
static const char *fpregNames[];
#endif
typedef enum {
CR0 = 0,
CR1 = 1,
CR2 = 2,
CR3 = 3,
CR4 = 4,
CR5 = 5,
CR6 = 6,
CR7 = 7
} ConditionRegister;
// These are used as register numbers in various parts of the code
static const int MAX_REGISTERS = 32;
enum
{
// general purpose 32bit regs
R0 = 0, // scratch or the value 0
SP = 1, // stack pointer
RTOC= 2, // toc pointer
R3 = 3, // this, return value
R4 = 4,
R5 = 5,
R6 = 6,
R7 = 7,
R8 = 8,
R9 = 9,
R10 = 10,
R11 = 11,
R12 = 12,
R13 = 13,
R14 = 14,
R15 = 15,
R16 = 16,
R17 = 17,
R18 = 18,
R19 = 19,
R20 = 20,
R21 = 21,
R22 = 22,
R23 = 23,
R24 = 24,
R25 = 25,
R26 = 26,
R27 = 27,
R28 = 28,
R29 = 29,
R30 = 30,
R31 = 31,
// FP regs
F0 = 0,
F1 = 1,
F2 = 2,
F3 = 3,
F4 = 4,
F5 = 5,
F6 = 6,
F7 = 7,
F8 = 8,
F9 = 9,
F10 = 10,
F11 = 11,
F12 = 12,
F13 = 13,
F14 = 14,
F15 = 15,
F16 = 16,
F17 = 17,
F18 = 18,
F19 = 19,
F20 = 20,
F21 = 21,
F22 = 22,
F23 = 23,
F24 = 24,
F25 = 25,
F26 = 26,
F27 = 27,
F28 = 28,
F29 = 29,
F30 = 30,
F31 = 31,
// special purpose registers (SPR)
LR = 8,
CTR = 9,
Unknown = 0x7f
};
typedef unsigned Register;
#endif
#ifdef AVMPLUS_ARM
static const int MAX_REGISTERS = 16;
#endif
#ifdef AVMPLUS_IA32
#ifdef AVMPLUS_VERBOSE
static const char *gpregNames[];
static const char *xmmregNames[];
static const char *x87regNames[];
#endif
#ifdef _MAC
byte *patch_esp_padding;
#endif
// These are used as register numbers in various parts of the code
static const int MAX_REGISTERS = 8;
typedef enum
{
// general purpose 32bit regs
EAX = 0, // return value, scratch
ECX = 1, // this, scratch
EDX = 2, // scratch
EBX = 3,
ESP = 4, // stack pointer
EBP = 5, // frame pointer
ESI = 6,
EDI = 7,
SP = 4, // alias SP to ESP to match PPC name
// SSE regs
XMM0 = 0,
XMM1 = 1,
XMM2 = 2,
XMM3 = 3,
XMM4 = 4,
XMM5 = 5,
XMM6 = 6,
XMM7 = 7,
// X87 regs
FST0 = 0,
FST1 = 1,
FST2 = 2,
FST3 = 3,
FST4 = 4,
FST5 = 5,
FST6 = 6,
FST7 = 7,
Unknown = -1
}
Register;
#endif
#if defined (AVMPLUS_AMD64)
#ifdef AVMPLUS_VERBOSE
static const char *gpregNames[];
static const char *xmmregNames[];
static const char *x87regNames[];
#endif
#ifdef _MAC
byte *patch_esp_padding;
#endif
// These are used as register numbers in various parts of the code
static const int MAX_REGISTERS = 16;
typedef enum
{
// 64bit - at this point, I'm not sure if we'll be using EAX, etc as registers
// or whether this numbering scheme is correct. For some generated ASM, usage
// of RAX implies a REX prefix byte to define the 64-bit operand. But the ModRM
// byte stays the same (EAX=RAX in the encoding)
// general purpose 32bit regs
EAX = 0, // return value, scratch
ECX = 1, // this, scratch
EDX = 2, // scratch
EBX = 3,
ESP = 4, // stack pointer
EBP = 5, // frame pointer
ESI = 6,
EDI = 7,
RAX = 8, // return value, scratch
RCX = 9, // this, scratch
RDX = 10, // scratch
RBX = 11,
RSP = 12, // stack pointer
RBP = 13, // frame pointer
RSI = 14,
RDI = 15,
R8 = 16,
R9 = 17,
R10 = 18,
R11 = 19,
R12 = 20,
R13 = 21,
R14 = 22,
R15 = 23,
SP = 12, // alias SP to RSP to match PPC name
// SSE regs
XMM0 = 0,
XMM1 = 1,
XMM2 = 2,
XMM3 = 3,
XMM4 = 4,
XMM5 = 5,
XMM6 = 6,
XMM7 = 7,
// !!@ remove all float support? SSE2 is always available
// X87 regs
FST0 = 0,
FST1 = 1,
FST2 = 2,
FST3 = 3,
FST4 = 4,
FST5 = 5,
FST6 = 6,
FST7 = 7,
Unknown = -1
}
Register;
#endif // AVMPLUS_AMD64
#define rmask(r) (1 << (r))
/**
* MIR instruction
*
* This structure contains the definition of the instruction format.
* Generally instructions take 1 or 2 operands. MIR_st is the
* the ONLY instruction that takes 3 operands. The form
* of the operands (or type) is instruction dependent. Each instruction
* expects the operands to be in a specific form. For example, MIR_add
* expects 2 operands which are pointers to other instructions. Debug
* builds will check operand types for conformance, but Release builds
* will have no type checking; consider yourself warned!
*
* During MIR generation we keep track of various details about the use
* of the instruction. Such as whether its result is needed across a
* function call and which instruction last uses the result of each
* instruction; lastUse. The lastUse field of every instruction
* is updated as instructions are generated. In the case of MIR_st
* no result is generated and thus lastUse can be overlayed with oprnd3.
*
* During machine dependent code generation oprnd2 is clobbered and
* the current register / stack position is maintained within this field.
*/
class OP
{
public:
/** opcode */
MirOpcode code:8;
/** register assigned to this expr, or Unknown */
Register reg:7;
/** flag indicating if result is used after call, so we prefer callee-saved regs */
uint32 liveAcrossCall:1;
/** link to a previous expr with the same opcode, for finding CSE's */
uint32 prevcse:16;
union
{
OP* oprnd1; // 1st operand of instruction
OP* base; // base ptr for load/store/lea/jmpi
uint32 argc; // arg count, for calls
int pos; // position of spilled value, or position of label
};
union
{
OP* oprnd2; // 2nd operand of instruction
int32 addr; // call addr or pc addr
int32 size; // alloca size, table size
int32 imm; // imm value if const
int disp; // immediate signed displacement for load/store/lea/jmpi
OP* target; // branch target
OP* join; // def joined to
uint32 *nextPatch;
};
union
{
OP* value; // value to store in MIR_st
OP* lastUse; // OP that uses this expr furthest in the future (end of live range)
OP* args[1]; // arg array, for call ops. arg[1] is the
// first arg, which will run over into the
// space of subsequent ops.
};
int isDouble() const {
return code & MIR_float;
}
int spillSize() const {
return code == MIR_alloc ? size : (code & MIR_float) ? 8 : 4;
}
int isDirty() const {
return code != MIR_imm && pos == InvalidPos;
}
};
class MirLabel
{
public:
OP* bb;
OP** nextPatchIns;
MirLabel() {
bb = 0;
nextPatchIns = 0;
}
};
class MdLabel
{
public:
int value;
uint32* nextPatch; /* linked list of patch locations, where next address is an offset in the instruction */
MdLabel() {
value = 0;
nextPatch = 0;
}
};
#ifdef AVMPLUS_VERBOSE
void buildFlowGraph();
static void formatOpcode(PrintWriter& buffer, OP *ipStart, OP* op, PoolObject* pool, MMgc::GCHashtable* names);
static void formatInsOperand(PrintWriter& buffer, OP* oprnd, OP* ipStart);
static void formatOperand(PrintWriter& buffer, OP* oprnd, OP* ipStart);
static MMgc::GCHashtable* initMethodNames(AvmCore* core);
#endif //AVMPLUS_VERBOSE
void emitMD();
OP* exAtom;
private:
#define COREADDR(f) coreAddr((int (AvmCore::*)())(&f))
#define GCADDR(f) gcAddr((int (MMgc::GC::*)())(&f))
#define ENVADDR(f) envAddr((int (MethodEnv::*)())(&f))
#define TOPLEVELADDR(f) toplevelAddr((int (Toplevel::*)())(&f))
#define CALLSTACKADDR(f) callStackAddr((int (CallStackNode::*)())(&f))
#define SCRIPTADDR(f) scriptAddr((int (ScriptObject::*)())(&f))
#define ARRAYADDR(f) arrayAddr((int (ArrayObject::*)())(&f))
#define EFADDR(f) efAddr((int (ExceptionFrame::*)())(&f))
#define DEBUGGERADDR(f) debuggerAddr((int (Debugger::*)())(&f))
static sintptr coreAddr( int (AvmCore::*f)() );
static sintptr gcAddr( int (MMgc::GC::*f)() );
static sintptr envAddr( int (MethodEnv::*f)() );
static sintptr toplevelAddr( int (Toplevel::*f)() );
#ifdef DEBUGGER
static sintptr callStackAddr( int (CallStackNode::*f)() );
static sintptr debuggerAddr( int (Debugger::*f)() );
#endif
static sintptr efAddr( int (ExceptionFrame::*f)() );
static sintptr scriptAddr( int (ScriptObject::*f)() );
static sintptr arrayAddr( int (ArrayObject::*f)() );
friend class Verifier;
// MIR instruction buffer
OP* ip;
OP* ipStart;
OP* ipEnd;
uintptr mirBuffSize;
int expansionFactor;
GrowableBuffer* mirBuffer;
byte* code;
uint32* casePtr;
int case_count;
#ifdef AVMPLUS_PPC
typedef uint32 MDInstruction;
#define PREV_MD_INS(m) (m-1)
#endif
#ifdef AVMPLUS_ARM
#define PREV_MD_INS(m) (m-1)
#endif
#ifdef AVMPLUS_IA32
typedef byte MDInstruction;
#define PREV_MD_INS(m) (m-4)
// for intel and our purposes previous instruction is 4 bytes prior to m; used for patching 32bit target addresses
#endif
// 64bit - need to verify
#ifdef AVMPLUS_AMD64
typedef byte MDInstruction;
#define PREV_MD_INS(m) (m-4)
// for intel and our purposes previous instruction is 4 bytes prior to m; used for patching 32bit target addresses
#endif
// machine specific instruction buffer
#ifndef AVMPLUS_ARM
MDInstruction* mip;
MDInstruction* mipStart;
#endif
uint32 arg_index;
void mirLabel(MirLabel& l, OP* bb);
void mirPatchPtr(OP** targetp, int pc); /* patch the location 'where' with the 32b value of the label */
void mirPatchPtr(OP** targetp, MirLabel& l);
void mirPatch(OP* i, int pc);
void mdLabel(MdLabel* l, void* v); /* machine specific version of position label (will trigger patching) */
void mdLabel(OP* l, void* v); /* machine specific version of position label (will trigger patching) */
void mdPatch(void* where, MdLabel* l); /* machine specific version for patch the location 'where' with the 32b value of the label */
void mdPatch(void* where, OP* l); /* machine specific version for patch the location 'where' with the 32b value of the label */
void mdApplyPatch(uint32* where, int labelvalue); /* patch label->value into where */
void mdPatchPreviousOp(OP* ins) {
mdPatch( PREV_MD_INS(mip), ins );
}
void mdPatchPrevious(MdLabel* l) {
mdPatch( PREV_MD_INS(mip), l );
}
class StackCheck
{
public:
uint32 *patchStackSize;
MdLabel resumeLabel;
MdLabel overflowLabel;
StackCheck() {
patchStackSize = NULL;
}
};
StackCheck stackCheck;
MdLabel patch_end;
MirLabel npe_label;
MirLabel interrupt_label;
bool interruptable;
#if defined (AVMPLUS_IA32) || defined(AVMPLUS_AMD64)
// stack probe for win32
void emitAllocaProbe(int growthAmt, MdLabel* returnTo);
#endif /* AVMPLUS_IA32 or AVMPLUS_AMD64 */
#ifdef AVMPLUS_ARM
uint32 *patch_stmfd;
uint32 *patch_frame_size;
/**
* Compute the size of the ARM callee area. Must be done
* after the generation of MIR opcodes.
*/
int calleeAreaSize() const { return 8*maxArgCount; }
int countBits(uint32);
#endif
#ifdef AVMPLUS_PPC
uint32 *patch_stmw;
uint32 *patch_stwu;
/**
* Compute the size of the PPC callee area. Must be done
* after the generation of MIR opcodes.
*/
int calleeAreaSize() const { return 8*maxArgCount; }
// This helper exists on PPC only to minimize the code size
// of the generated code for stack overflow checks
static void stackOverflow(MethodEnv *env);
#endif
uint32 maxArgCount; // most number of arguments used in a call
#ifdef AVMPLUS_PROFILE
// register allocator stats
int fullyUsedCount; // number of times all registers fully used
int longestSpan; // most number of instructions that a register is used
int spills; // number of spills required
int steals; // number of spills due to a register being stolen.
int remats; // number of rematerializations
// profiler stats
uint64 verifyStartTime; // note the time we started verification
uint64 mdStartTime; // note the time we started MD generation
#ifndef AVMPLUS_ARM
int mInstructionCount; // number of machine instructions
#define incInstructionCount() mInstructionCount++
#endif
#ifdef _DEBUG
// buffer tuning information
enum { SZ_ABC, SZ_MIR, SZ_MD, SZ_MIRWASTE, SZ_MDWASTE, SZ_MIREPI, SZ_MDEPI, SZ_MIRPRO, SZ_MDPRO, SZ_MIREXP, SZ_MDEXP, SZ_LAST };
double sizingStats[SZ_LAST];
#endif /* _DEBUG */
#else
#define incInstructionCount()
#endif /* AVMPLUS_PROFILE */
// pointer to list of argument definitions
OP* methodArgs;
OP* calleeVars;
Register framep;
// stack space reserved by MIR_alloca instruction in prologue for locals, exception frame and Multiname
OP* _save_eip;
OP* _ef;
OP* dxns;
OP* dxnsAddrSave; // methods that set dxns need to save/restore
#ifdef DEBUGGER
OP* localPtrs; // array of local_count + max_scope (holds locals and scopes)
OP* localTraits; // array of local_count (holds snapshot of Traits* per local)
OP* _callStackNode;
#endif
// track last function call issued
OP* lastFunctionCall;
// track last pc value we generated a store for
int lastPcSave;
// cse table which prevents duplicate instructions in the same bb
OP* cseTable[MIR_last];
OP* firstCse;
#ifdef AVMPLUS_PROFILE
int cseHits; // measure effectiveness of cse optimizer
#define incCseHits() cseHits++
int dceHits; // measure effectiveness of dead code eliminator
#define incDceHits() dceHits++
#else
#define incCseHits()
#define incDceHits()
#endif
#ifdef DEBUGGER
void extendDefLifetime(OP* current);
#endif
void saveState();
OP* defIns(OP* v);
OP* useIns(OP* def, int i);
OP* undefConst;
// frame state helpers
OP* localGet(uint32 i);
void localSet(uint32 i, OP* o);
OP* loadAtomRep(uint32 i);
OP* InsAt(int nbr) { return ipStart+nbr; }
int InsNbr(OP* ins) { AvmAssert(ins >= ipStart); return (ins-ipStart); }
OP* InsConst(uintptr value) { return Ins(MIR_imm, value); }
OP* InsAlloc(size_t s) { return Ins(MIR_alloc, (int32)s); }
void InsDealloc(OP* alloc);
OP* ldargIns(ArgNumber arg) { return &methodArgs[arg]; }
OP* Ins(MirOpcode code, uintptr v=0);
OP* Ins(MirOpcode code, OP* a1, uintptr a2);
OP* Ins(MirOpcode code, OP* a1, OP* a2=0);
OP* defineArgInsPos(int spOffset);
OP* defineArgInsReg(Register r);
OP* binaryIns(MirOpcode code, OP* a1, OP* a2);
OP* loadIns(MirOpcode _code, int _disp, OP* _base)
{
AvmAssert((_code & ~MIR_float & ~MIR_oper) == MIR_ld);
return Ins(_code, _base, (int32)_disp);
}
OP* cmpOptimization (int lhs, int rhs);
OP* i2dIns(OP* v);
OP* u2dIns(OP* v);
OP* fcmpIns(OP* a1, OP* a2);
OP* binaryFcmpIns(OP* a1, OP* a2);
OP* cmpLt(int lhs, int rhs);
OP* cmpLe(int lhs, int rhs);
OP* cmpEq(int funcaddr, int lhs, int rhs);
void storeIns(OP* v, uintptr disp, OP* base);
OP* leaIns(int disp, OP* base);
OP* callIns(int32 addr, uint32 argCount, MirOpcode code);
OP* callIndirect(MirOpcode code, OP* target, uint32 argCount, ...);
OP* callIns(MirOpcode code, int32 addr, uint32 argCount, ...);
OP* promoteNumberIns(Traits* t, int i);
OP* loadVTable(int base_index);
OP* loadToplevel(OP* env);
// simple cse within BBs
OP* cseMatch(MirOpcode code, OP* a1, OP* a2=0);
// dead code search
void markDead(OP* ins);
bool usedInState(OP* ins);
void argIns(OP* a);
OP* storeAtomArgs(int count, int startAt);
OP* storeAtomArgs(OP* receiver, int count, int startAt);
void updateUse(OP* currentIns, OP* op, Register hint=Unknown);
void extendLastUse(OP* use, int targetpc);
void extendLastUse(OP* ins, OP* use, OP* target);
OP* atomToNativeRep(Traits* t, OP* atom);
OP* atomToNativeRep(int i, OP* atom);
bool isPointer(int i);
bool isDouble(int i);
OP* ptrToNativeRep(Traits*, OP* ptr);
OP* initMultiname(Multiname* multiname, int& csp, bool isDelete = false);
#ifdef DEBUGGER
void emitSampleCheck();
#endif
//
// -- MD Specific stuff
//
void generate();
void generatePrologue();
void generateEpilogue();
void bindMethod(AbstractFunction* f);
#ifdef AVMPLUS_JIT_READONLY
void makeCodeExecutable();
#endif /* AVMPLUS_JIT_READONLY */
bool ensureMDBufferCapacity(PoolObject* pool, size_t s); // only if buffer guard is not used
byte* getMDBuffer(PoolObject* pool); //
size_t estimateMDBufferReservation(PoolObject* pool); //
/**
* Information about the activation record for the method is built up
* as we generate machine code. As part of the prologue, we issue
* a stack adjustment instruction and then later patch the adjustment
* value. Temporary values can be placed into the AR as method calls
* are issued. Also MIR_alloca instructions will consume space.
*/
class AR
{
public:
AR(MMgc::GC *gc) : temps(gc) {}
List<OP*,LIST_NonGCObjects> temps; /* list of active temps */
int size; /* current # of bytes consumed by the temps */
int highwatermark; /* max size of temps */
MDInstruction* adjustmentIns; /* AR sizing instruction to patch */
};
AR activation;
static const int InvalidPos = -1; /* invalid spill position */
void reserveStackSpace(OP* ins);
#ifdef AVMPLUS_VERBOSE
void displayStackTable();
#endif //AVMPLUS_VERBOSE
#ifdef AVMPLUS_PPC
static const int kLinkageAreaSize = 24;
#endif
// converts an instructions 'pos' field to a stack pointer relative displacement
int stackPos(OP* ins);
// structures for register allocator
class RegInfo
{
public:
uint32 free;
uint32 calleeSaved; // 1 = callee-saved, 0=caller-saved
#ifdef AVMPLUS_PPC
unsigned LowerBound;
#endif
#ifdef AVMPLUS_ARM
unsigned nonVolatileMask;
#endif
OP* active[MAX_REGISTERS]; // active[r] = OP that defines r
OP* findLastActive(int set);
void flushCallerActives(uint32 flushmask);
RegInfo()
{
clear();
}
void clear()
{
free = 0;
memset(active, 0, MAX_REGISTERS * sizeof(OP*));
}
uint32 isFree(Register r)
{
return free & rmask(r);
}
/**
* Add a register to the free list for the allocator.
*
* IMPORTANT
*
* This is a necessary call when freeing a register.
*/
void addFree(Register r)
{
AvmAssert(!isFree(r));
free |= rmask(r);
}
void removeFree(Register r)
{
AvmAssert(isFree(r));
free &= ~rmask(r);
}
// if an instruction is no longer in use retire it
void expire(OP* ins, OP* currentIns)
{
if (ins->reg != Unknown && ins->lastUse <= currentIns)
retire(ins);
}
void expireAll(OP* currentIns)
{
for(int i=0; i<MAX_REGISTERS; i++)
{
OP* ins = active[i];
if (ins && ins->lastUse <= currentIns)
retire(ins);
}
}
// instruction gives up the register it is currently bound to, adding it back to the free list,
// removing it from the active list
void retire(OP* ins)
{
AvmAssert(active[ins->reg] == ins);
retire(ins->reg);
ins->reg = Unknown;
}
void retire(Register r)
{
AvmAssert(r != Unknown);
AvmAssert(active[r] != NULL);
active[r] = NULL;
free |= rmask(r);
}
/**
* add the register provided in v->reg to the active list
* IMPORTANT: this is a necesary call when allocating a register
*/
void addActive(OP* v)
{
//addActiveCount++;
AvmAssert(active[v->reg] == NULL);
active[v->reg] = v;
}
/**
* Remove a register from the active list without
* spilling its contents.
*
* The caller MUST eventually call addFree()
* or regs.addActive() in order to add the register
* back to the freelist or back onto the active list.
*
* It is the responsibility of the caller to update
* the reg field of the instruction.
*
* @param r - register to remove from free list
* @param forAlloctor - DO NOT USE, for allocator only
*
* @return the instruction for which the register
* was retaining a value.
*/
void removeActive(Register r)
{
//registerReleaseCount++;
AvmAssert(r != Unknown);
AvmAssert(active[r] != NULL);
// remove the given register from the active list
active[r] = NULL;
}
#ifdef _DEBUG
int count;
int countFree()
{
int _count = 0;
for(int i=0;i<MAX_REGISTERS; i++)
_count += ( (free & rmask(i)) == 0) ? 0 : 1;
return _count;
}
int countActive()
{
int _count = 0;
for(int i=0;i<MAX_REGISTERS; i++)
_count += (active[i] == 0) ? 0 : 1;
return _count;
}
void checkCount()
{
AvmAssert(count == (countActive() + countFree()));
}
void checkActives(OP* current);
#endif //_DEBUG
};
// all registers that we use
RegInfo gpregs;
RegInfo fpregs;
#ifdef AVMPLUS_VERBOSE
void showRegisterStats(RegInfo& regs);
#endif //AVMPLUS_VERBOSE
// register active/freelist routines
Register registerAllocSpecific(RegInfo& regs, Register r);
Register registerAllocAny(RegInfo& regs, OP* ins);
Register registerAllocFromSet(RegInfo& regs, int set);
Register registerAlloc(RegInfo& regs, OP* ins, Register r)
{
return r == Unknown
? registerAllocAny(regs, ins)
: registerAllocSpecific(regs, r);
}
uint32 spillCallerRegisters(OP* ins, RegInfo& regs, uint32 live);
void spillTmps(OP* target);
Register InsPrepResult(RegInfo& regs, OP* ins, int exclude=0);
void setResultReg(RegInfo& regs, OP* ins, Register reg)
{
ins->reg = reg;
ins->pos = InvalidPos;
regs.addActive(ins);
}
// MD instruction generators for spill/remat
void spill(OP* ins);
void rematerialize(OP* ins);
void copyToStack(OP* ins, Register r);
// spill and argument code for MD calls
int prepCallArgsOntoStack(OP* ip, OP* postCall);
void InsRegisterPrepA(OP* insRes, RegInfo& regsA, OP* insA, Register& reqdA);
void InsRegisterPrepAB(OP* insRes, RegInfo& regsA, OP* insA, Register& reqdA, RegInfo& regsB, OP* insB, Register& reqdB);
// prepare the registers in the best possible fashion for issuing the given instruction
inline void InsPrep_A_IN_REQ_B_IN_WRONG(RegInfo& regs, OP* insA, Register& reqdA, OP* insB, Register& reqdB);
inline void InsPrep_A_IN_REQ_B_OUT_REQ(RegInfo& regs, OP* insA, Register& reqdA, OP* insB, Register& reqdB);
inline void InsPrep_A_IN_REQ_B_OUT_ANY(RegInfo& regs, OP* insA, Register& reqdA, OP* insB, Register& reqdB);
#if 0
inline void InsPrep_A_IN_WRONG_B_IN_WRONG(RegInfo& regs, OP* insA, Register& reqdA, OP* insB, Register& reqdB);
#endif
inline void InsPrep_A_IN_WRONG_B_IN_ANY(RegInfo& regs, OP* insA, Register& reqdA, OP* insB, Register& reqdB);
inline void InsPrep_A_IN_WRONG_B_OUT(RegInfo& regs, OP* insA, Register& reqdA, OP* insB, Register& reqdB);
inline void InsPrep_A_IN_ANY_B_OUT(RegInfo& regs, OP* insA, Register& reqdA, OP* insB, Register& reqdB);
inline void InsPrep_A_OUT_B_OUT(RegInfo& regs, OP* insA, Register& reqdA, OP* insB, Register& reqdB);
void moveR2R(OP* ins, Register src, Register dst);
// Returns true if immediate folding can be used
bool canImmFold(OP *ins, OP *imm) const;
#ifdef AVMPLUS_PPC
//
// PowerPC code generation
//
void GpBinary(int op, Register dst, Register src1, Register src2);
void ADD(Register dst, Register src1, Register src2) { GpBinary(0x214, dst, src1, src2); }
void MULLW(Register dst, Register src1, Register src2) { GpBinary(0x1D6, dst, src1, src2); }
void AND(Register dst, Register src1, Register src2) { GpBinary(0x038, src1, dst, src2); }
void SUB(Register dst, Register src1, Register src2) { GpBinary(0x050, dst, src2, src1); } // z = x - y is encoded as subf z,y,x
void SLW(Register dst, Register src, Register shift) { GpBinary(0x030, src, dst, shift); }
void FpUnary(int op, Register dst, Register src);
void FMR(Register dst, Register src) { FpUnary(0x90, dst, src); }
void FNEG(Register dst, Register src) { FpUnary(0x50, dst, src); }
void MR(Register dst, Register src, bool dot=false);
void MR_dot(Register dst, Register src) { MR(dst,src,true); }
void NEG(Register dst, Register src);
void LWZ(Register dst, int disp, Register base);
void LWZX(Register dst, Register rA, Register rB);
void LFD(Register dst, int disp, Register base);
void LFDX(Register dst, Register rA, Register rB);
void STFD(Register src, int disp, Register dst);
void STFDX(Register src, Register rA, Register rB);
void LMW(Register dst, int disp, Register base);
void STW(Register src, int disp, Register base);
void STWX(Register src, Register rA, Register rB);
void RLWINM(Register dst, Register src, int shift, int begin, int end);
void SLWI(Register dst, Register src, int shift);
void SRW(Register dst, Register src, Register shift);
void SRWI(Register dst, Register src, int shift);
void SRAW(Register dst, Register src, Register shift);
void SRAWI(Register dst, Register src, int shift);
void ADDI(Register dst, Register src, int imm16);
void ADDIS(Register dst, Register src, int imm16);
void SI(Register dst, Register src, int imm16);
void CMP(ConditionRegister CR, Register regA, Register regB);
void CMPI(ConditionRegister CR, Register src, int imm16);
void LI(Register reg, int imm16);
void LIS(Register reg, int imm16);
void ORI(Register dst, Register src, int imm16);
void ORIS(Register dst, Register src, int imm16);
void ANDI_dot(Register dst, Register src, int imm16);
void CMPL(ConditionRegister CR, Register regA, Register regB);
void CMPLI(ConditionRegister CR, Register src, int imm16);
void FCMPU(ConditionRegister CR, Register regA, Register regB);
void MFCR(Register dst);
void CROR(int crbD, int crbA, int crbB);
void CRNOT(int crbD, int crbA);
void XORI(Register dst, Register src, int imm16);
void XORIS(Register dst, Register src, int imm16);
void XOR(Register dst, Register src1, Register src2);
void FADD(Register dst, Register src1, Register src2);
void FSUB(Register dst, Register src1, Register src2);
void FMUL(Register dst, Register src1, Register src2);
void FDIV(Register dst, Register src1, Register src2);
void OR(Register dst, Register src1, Register src2);
void STMW(Register start, int disp, Register base);
void STWU(Register dst, int disp, Register base);
void STWUX(Register rS, Register rA, Register rB);
void Movspr(int op, Register r, Register spr);
void MTCTR(Register reg) { Movspr(0x7C0003A6, reg, CTR); }
void MTLR(Register reg) { Movspr(0x7C0003A6, reg, LR); }
void MFLR(Register reg) { Movspr(0x7C0002A6, reg, LR); }
void BR(int op, int addr);
void B(int offset) { BR(0, offset); }
void BL(int offset) { BR(1, offset); }
void BA(int addr) { BR(2, addr); }
void BLA(int addr) { BR(3, addr); }
void Bcc(int op, ConditionRegister cr, int offset, bool hint=false);
void BEQ(ConditionRegister cr, int offset, bool hint=false) { Bcc(0x4182, cr, offset, hint); }
void BNE(ConditionRegister cr, int offset, bool hint=false) { Bcc(0x4082, cr, offset, hint); }
void BLE(ConditionRegister cr, int offset, bool hint=false) { Bcc(0x4081, cr, offset, hint); }
void BGT(ConditionRegister cr, int offset, bool hint=false) { Bcc(0x4181, cr, offset, hint); }
void BLT(ConditionRegister cr, int offset, bool hint=false) { Bcc(0x4180, cr, offset, hint); }
void BGE(ConditionRegister cr, int offset, bool hint=false) { Bcc(0x4080, cr, offset, hint); }
void BDNZ(ConditionRegister cr, int offset, bool hint=false){ Bcc(0x4200, cr, offset, hint); }
void Simple(int op);
void BLR() { Simple(0x4E800020); }
void BCTR() { Simple(0x4E800420); }
void BCTRL() { Simple(0x4E800421); }
void NOP() { Simple(0x60000000); }
void patchRelativeBranch(uint32 *patch_ip, uint32 *target_ip)
{
int offset = (target_ip-patch_ip)*4;
*patch_ip &= ~0xFFFC;
*patch_ip |= offset;
}
// Load and store macros that take 32-bit offsets.
// R0 will be used for the offset if needed.
inline bool IsSmallDisplacement(int disp) const
{
return (uint32)(disp+0x8000) < 0x10000;
}
// Returns whether displacement fits in 26-bit
// branch displacement (B, BL, BLA instructions)
inline bool IsBranchDisplacement(int disp) const
{
return (uint32)(disp+0x2000000) < 0x4000000;
}
// Returns true if imm will fit in the SIMM field of
// a PowerPC instruction (16-bit signed immediate)
inline bool isSIMM(int imm) const
{
return !((imm + 0x8000) & 0xFFFF0000);
}
// Returns true if imm will fit in the UIMM field of
// a PowerPC instruction (16-bit unsigned immediate)
inline bool isUIMM(int imm) const
{
return !(imm & 0xFFFF0000);
}
void LWZ32(Register dst, int disp, Register base)
{
if (IsSmallDisplacement(disp))
{
LWZ(dst, disp, base);
}
else
{
LI32(R0, disp);
LWZX(dst, base, R0);
}
}
void STW32(Register src, int disp, Register dst)
{
if (IsSmallDisplacement(disp))
{
STW(src, disp, dst);
}
else
{
LI32(R0, disp);
STWX(src, dst, R0);
}
}
void LFD32(Register dst, int disp, Register base)
{
if (IsSmallDisplacement(disp))
{
LFD(dst, disp, base);
}
else
{
LI32(R0, disp);
LFDX(dst, base, R0);
}
}
void STFD32(Register src, int disp, Register dst)
{
if (IsSmallDisplacement(disp))
{
STFD(src, disp, dst);
}
else
{
LI32(R0, disp);
STFDX(src, dst, R0);
}
}
void ADDI32(Register dst, Register src, int imm32)
{
if (IsSmallDisplacement(imm32))
{
ADDI(dst, src, imm32);
}
else
{
LI32(R0, imm32);
ADD(dst, src, R0);
}
}
int HIWORD(int value) { return (value>>16)&0xFFFF; }
int LOWORD(int value) { return value&0xFFFF; }
// Macro to emit the right sequence for
// loading an immediate 32-bit value
void LI32(Register reg, int imm32)
{
if (imm32 & 0xFFFF0000) {
LIS(reg, HIWORD(imm32));
ORI(reg, reg, LOWORD(imm32));
} else {
if (imm32 & 0x8000) {
// todo
LI(reg, 0);
ORI(reg, reg, imm32);
} else {
LI(reg, imm32);
}
}
}
//
// End PowerPC code generation
//
// 64bit - enabled this for 64-bit to get compiling. There
// are certain differences between 32 and 64-bit bytecode
#elif defined(AVMPLUS_IA32) || defined(AVMPLUS_AMD64)
// ---------------------------------------------------
//
// IA32 specific stuff follows
//
// ---------------------------------------------------
bool x87Dirty;
#ifdef AVMPLUS_VERBOSE
unsigned x87Top:3;
#endif
bool is8bit(int i)
{
return ((signed char)i) == i;
}
void IMM32(int imm32)
{
*(int*)mip = imm32;
mip += 4;
}
void MODRM(Register reg, sintptr disp, Register base, int lshift, Register index);
void MODRM(Register reg, sintptr disp, Register base);
void MODRM(Register reg, Register operand);
void ALU(int op);
void RET() { ALU(0xc3); }
void NOP() { ALU(0x90); }
#ifdef AVMPLUS_64BIT
void PUSH(Register r);
void POP(Register r);
#else
void PUSH(Register r) { ALU(0x50|r); }
void POP(Register r) { ALU(0x58|r); }
#endif
void LAHF() { ALU(0x9F); }
void PUSH(sintptr imm);
void MOV (Register dest, sintptr imm32);
void MOV (sintptr disp, Register base, sintptr imm);
// sse data transfer
// sse math
void SSE(int op, Register dest, Register src);
void ADDSD(Register dest, Register src) { SSE(0xf20f58, dest, src); }
void SUBSD(Register dest, Register src) { SSE(0xf20f5c, dest, src); }
void MULSD(Register dest, Register src) { SSE(0xf20f59, dest, src); }
void DIVSD(Register dest, Register src) { SSE(0xf20f5e, dest, src); }
void CVTSI2SD(Register dest, Register src) { SSE(0xf20f2a, dest, src); }
void UCOMISD(Register xmm1, Register xmm2) { SSE(0x660f2e, xmm1, xmm2); }
void MOVAPD(Register dest, Register src) { SSE(0x660f28, dest, src); }
void XORPD(Register dest, uintptr src);
void SSE(int op, Register r, sintptr disp, Register base);
void ADDSD(Register r, sintptr disp, Register base) { SSE(0xf20f58, r, disp, base); }
void SUBSD(Register r, sintptr disp, Register base) { SSE(0xf20f5C, r, disp, base); }
void MULSD(Register r, sintptr disp, Register base) { SSE(0xf20f59, r, disp, base); }
void DIVSD(Register r, sintptr disp, Register base) { SSE(0xf20f5E, r, disp, base); }
void MOVSD(Register r, sintptr disp, Register base) { SSE(0xf20f10, r, disp, base); }
void MOVSD(sintptr disp, Register base, Register r) { SSE(0xf20f11, r, disp, base); }
void CVTSI2SD(Register r, sintptr disp, Register base) { SSE(0xf20f2a, r, disp, base); }
void ALU (byte op, Register reg, sintptr imm);
void ADD(Register reg, sintptr imm) { ALU(0x05, reg, imm); }
void SUB(Register reg, sintptr imm) { ALU(0x2d, reg, imm); }
void AND(Register reg, sintptr imm) { ALU(0x25, reg, imm); }
void XOR(Register reg, sintptr imm) { ALU(0x35, reg, imm); }
void OR (Register reg, sintptr imm) { ALU(0x0d, reg, imm); }
void CMP(Register reg, sintptr imm) { ALU(0x3d, reg, imm); }
void IMUL(Register dst, sintptr imm);
void ALU (int op, Register lhs_dest, Register rhs);
void OR (Register lhs, Register rhs) { ALU(0x0b, lhs, rhs); }
void AND(Register lhs, Register rhs) { ALU(0x23, lhs, rhs); }
void XOR(Register lhs, Register rhs) { ALU(0x33, lhs, rhs); }
void ADD(Register lhs, Register rhs) { ALU(0x03, lhs, rhs); }
void CMP(Register lhs, Register rhs) { ALU(0x3b, lhs, rhs); }
void SUB(Register lhs, Register rhs) { ALU(0x2b, lhs, rhs); }
void TEST(Register lhs, Register rhs) { ALU(0x85, lhs, rhs); }
void NEG(Register reg) { ALU(0xf7, (Register)3, reg); }
void SHR(Register reg, Register /*amt*/) { ALU(0xd3, (Register)5, reg); } // unsigned >> ecx
void SAR(Register reg, Register /*amt*/) { ALU(0xd3, (Register)7, reg); } // signed >> ecx
void SHL(Register reg, Register /*amt*/) { ALU(0xd3, (Register)4, reg); } // unsigned << ecx
void XCHG(Register rA, Register rB) { ALU(0x87, rA, rB); }
void MOV (Register dest, Register src) { ALU(0x8b, dest, src); }
void ALU2(int op, Register lhs_dest, Register rhs);
void IMUL(Register lhs, Register rhs) { ALU2(0x0faf, lhs, rhs); }
void SETB (Register reg) { ALU2(0x0f92, reg, reg); }
void SETNB (Register reg) { ALU2(0x0f93, reg, reg); }
void SETE (Register reg) { ALU2(0x0f94, reg, reg); }
void SETNE (Register reg) { ALU2(0x0f95, reg, reg); }
void SETBE (Register reg) { ALU2(0x0f96, reg, reg); }
void SETNBE(Register reg) { ALU2(0x0f97, reg, reg); }
void SETNP (Register reg) { ALU2(0x0f9b, reg, reg); }
void SETP (Register reg) { ALU2(0x0f9a, reg, reg); }
void SETL (Register reg) { ALU2(0x0f9C, reg, reg); }
void SETLE (Register reg) { ALU2(0x0f9E, reg, reg); }
void MOVZX_r8 (Register dest, Register src) { ALU2(0x0fb6, dest, src); }
void ALU(int op, Register r, sintptr disp, Register base);
void TEST(sintptr disp, Register base, Register r) { ALU(0x85, r, disp, base); }
void LEA(Register r, sintptr disp, Register base) { ALU(0x8d, r, disp, base); }
void CALL(sintptr disp, Register base) { ALU(0xff, (Register)2, disp, base); }
void JMP(sintptr disp, Register base) { ALU(0xff, (Register)4, disp, base); }
void PUSH(sintptr disp, Register base) { ALU(0xff, (Register)6, disp, base); }
void MOV (sintptr disp, Register base, Register r) { ALU(0x89, r, disp, base); }
void MOV (Register r, sintptr disp, Register base) { ALU(0x8b, r, disp, base); }
void SHIFT(int op, Register reg, int imm8);
void SAR(Register reg, int imm8) { SHIFT(7, reg, imm8); } // signed >> imm
void SHR(Register reg, int imm8) { SHIFT(5, reg, imm8); } // unsigned >> imm
void SHL(Register reg, int imm8) { SHIFT(4, reg, imm8); } // unsigned << imm
void TEST_AH(uint8 imm8);
void JCC (byte op, sintptr offset);
void JB (sintptr offset) { JCC(0x02, offset); }
void JNB (sintptr offset) { JCC(0x03, offset); }
void JE (sintptr offset) { JCC(0x04, offset); }
void JNE (sintptr offset) { JCC(0x05, offset); }
void JBE (sintptr offset) { JCC(0x06, offset); }
void JNBE(sintptr offset) { JCC(0x07, offset); }
void JP (sintptr offset) { JCC(0x0A, offset); }
void JNP (sintptr offset) { JCC(0x0B, offset); }
void JL (sintptr offset) { JCC(0x0C, offset); }
void JNL (sintptr offset) { JCC(0x0D, offset); }
void JLE (sintptr offset) { JCC(0x0E, offset); }
void JNLE(sintptr offset) { JCC(0x0F, offset); }
void JMP (sintptr offset);
void CALL(sintptr offset);
void FPU(int op, sintptr disp, Register base);
void FSTQ(sintptr disp, Register base) { FPU(0xdd02, disp, base); }
void FSTPQ(sintptr disp, Register base) { FPU(0xdd03, disp, base); }
void FCOM(sintptr disp, Register base) { FPU(0xdc02, disp, base); }
void FLDQ(sintptr disp, Register base) { x87Dirty=true; FPU(0xdd00, disp, base); }
void FILDQ(sintptr disp, Register base) { x87Dirty=true; FPU(0xdf05, disp, base); }
void FILD(sintptr disp, Register base) { x87Dirty=true; FPU(0xdb00, disp, base); }
void FADDQ(sintptr disp, Register base) { FPU(0xdc00, disp, base); }
void FSUBQ(sintptr disp, Register base) { FPU(0xdc04, disp, base); }
void FMULQ(sintptr disp, Register base) { FPU(0xdc01, disp, base); }
void FDIVQ(sintptr disp, Register base) { FPU(0xdc06, disp, base); }
void FPU(int op, Register r);
void FFREE(Register r) { FPU(0xddc0, r); }
void FSTP(Register r) { FPU(0xddd8, r); }
void FADDQ(Register r) { FPU(0xd8c0, r); }
void FSUBQ(Register r) { FPU(0xd8e0, r); }
void FMULQ(Register r) { FPU(0xd8c8, r); }
void FDIVQ(Register r) { FPU(0xd8f0, r); }
void FPU(int op);
void FUCOMP() { FPU(0xdde9); }
void FCHS() { FPU(0xd9e0); }
void FNSTSW_AX() { FPU(0xdfe0); }
void EMMS() { FPU(0x0f77); x87Dirty=false; }
#endif // IA32 or AMD64
#ifdef AVMPLUS_AMD64
void IMM64(int64 imm64)
{
*(int64*)mip = imm64;
mip += 8;
}
void REX(Register a, Register b=EAX);
bool is32bit(sintptr i)
{
return ((int32)i) == i;
}
#endif
// macros
// windows ia32 calling conventions
// - args pushed right-to-left
// - EAX, ECX, EDX are scratch registers
// - result in EAX (32bit) or EAX:EDX (64bit)
// cdecl calling conventions:
// - caller pops args
// thiscall calling conventions:
// - this in ECX
// - callee pops args
// stdcall calling conventions
// - callee pops args
// fastcall calling conventions
// - first 2 args in ECX, EDX
// - callee pops args
/** call a method using a relative offset */
void thincall(sintptr addr)
{
#ifdef AVMPLUS_PPC
int disp = addr - (int)mip;
if (IsBranchDisplacement(disp)) {
// Use relative branch if possible
BL(disp);
} else if (IsBranchDisplacement(addr)) {
// Use absolute branch if possible
// Note: address will be sign-extended.
BLA (addr);
} else {
// Otherwise, use register-based branched
LI32(R12, addr);
MTCTR(R12);
BCTRL();
}
#endif
#ifdef AVMPLUS_ARM
int disp = (MDInstruction*)addr - mip - 2;
if (!(disp & 0xFF000000))
{
// Branch displacement fits in 24-bits, use BL instruction.
BL(addr - (int)mip);
}
else
{
// Branch displacement doesn't fit
MOV_imm32(IP, addr);
MOV(LR, PC);
MOV(PC, IP);
}
#endif
#ifdef AVMPLUS_IA32
CALL (addr - (5+(int)mip));
#endif
#ifdef AVMPLUS_AMD64
CALL (addr - (5+(uintptr)mip));
#endif
}
bool isCodeContextChanged() const;
};
// machine dependent buffer sizing information try to use 4B aligned values
#ifdef AVMPLUS_PPC
static const int md_prologue_size = 96;
static const int md_epilogue_size = 280;
static const int md_native_thunk_size = 1024;
#endif
#ifdef AVMPLUS_ARM
static const int md_prologue_size = 256;
static const int md_epilogue_size = 128;
static const int md_native_thunk_size = 1024;
#endif /* AVMPLUS_ARM */
#ifdef AVMPLUS_IA32
static const int md_prologue_size = 32;
static const int md_epilogue_size = 128;
static const int md_native_thunk_size = 256;
#endif /* AVMPLUS_PPC */
#ifdef AVMPLUS_AMD64
// 64bit - may need adjustment
static const int md_prologue_size = 32;
static const int md_epilogue_size = 128;
static const int md_native_thunk_size = 256;
#endif /* AVMPLUS_PPC */
}
#endif /* __avmplus_CodegenMIR__ */