corda/src/codegen/assembler.h

474 lines
11 KiB
C
Raw Normal View History

2012-05-11 23:43:27 +00:00
/* Copyright (c) 2008-2012, Avian Contributors
Permission to use, copy, modify, and/or distribute this software
for any purpose with or without fee is hereby granted, provided
that the above copyright notice and this permission notice appear
in all copies.
There is NO WARRANTY for this software. See license.txt for
details. */
2008-02-11 17:21:41 +00:00
#ifndef ASSEMBLER_H
#define ASSEMBLER_H
#include "system.h"
#include "zone.h"
namespace vm {
#ifdef AVIAN_TAILS
const bool TailCalls = true;
#else
const bool TailCalls = false;
#endif
#if (defined AVIAN_USE_FRAME_POINTER) || (defined ARCH_powerpc)
support stack unwinding without using a frame pointer Previously, we unwound the stack by following the chain of frame pointers for normal returns, stack trace creation, and exception unwinding. On x86, this required reserving EBP/RBP for frame pointer duties, making it unavailable for general computation and requiring that it be explicitly saved and restored on entry and exit, respectively. On PowerPC, we use an ABI that makes the stack pointer double as a frame pointer, so it doesn't cost us anything. We've been using the same convention on ARM, but it doesn't match the native calling convention, which makes it unusable when we want to call native code from Java and pass arguments on the stack. So far, the ARM calling convention mismatch hasn't been an issue because we've never passed more arguments from Java to native code than would fit in registers. However, we must now pass an extra argument (the thread pointer) to e.g. divideLong so it can throw an exception on divide by zero, which means the last argument must be passed on the stack. This will clobber the linkage area we've been using to hold the frame pointer, so we need to stop using it. One solution would be to use the same convention on ARM as we do on x86, but this would introduce the same overhead of making a register unavailable for general use and extra code at method entry and exit. Instead, this commit removes the need for a frame pointer. Unwinding involves consulting a map of instruction offsets to frame sizes which is generated at compile time. This is necessary because stack trace creation can happen at any time due to Thread.getStackTrace being called by another thread, and the frame size varies during the execution of a method. So far, only x86(_64) is working, and continuations and tail call optimization are probably broken. More to come.
2011-01-17 02:05:05 +00:00
const bool UseFramePointer = true;
#else
const bool UseFramePointer = false;
#endif
2008-02-11 17:21:41 +00:00
enum Operation {
2009-03-03 03:18:15 +00:00
Return,
LoadBarrier,
StoreStoreBarrier,
fix a couple of subtle Thread.getStackTrace bugs The first problem was that, on x86, we failed to properly keep track of whether to expect the return address to be on the stack or not when unwinding through a frame. We were relying on a "stackLimit" pointer to tell us whether we were looking at the most recently-called frame by comparing it with the stack pointer for that frame. That was inaccurate in the case of a thread executing at the beginning of a method before a new frame is allocated, in which case the most recent two frames share a stack pointer, confusing the unwinder. The solution involves keeping track of how many frames we've looked at while walking the stack. The other problem was that compareIpToMethodBounds assumed every method was followed by at least one byte of padding before the next method started. That assumption was usually valid because we were storing the size following method code prior to the code itself. However, the last method of an AOT-compiled code image is not followed by any such method header and may instead be followed directly by native code with no intervening padding. In that case, we risk interpreting that native code as part of the preceding method, with potentially bizarre results. The reason for the compareIpToMethodBounds assumption was that methods which throw exceptions as their last instruction generate a non-returning call, which nonetheless push a return address on the stack which points past the end of the method, and the unwinder needs to know that return address belongs to that method. A better solution is to add an extra trap instruction to the end of such methods, which is what this patch does.
2012-05-05 00:35:13 +00:00
StoreLoadBarrier,
Trap
2008-02-11 17:21:41 +00:00
};
fix a couple of subtle Thread.getStackTrace bugs The first problem was that, on x86, we failed to properly keep track of whether to expect the return address to be on the stack or not when unwinding through a frame. We were relying on a "stackLimit" pointer to tell us whether we were looking at the most recently-called frame by comparing it with the stack pointer for that frame. That was inaccurate in the case of a thread executing at the beginning of a method before a new frame is allocated, in which case the most recent two frames share a stack pointer, confusing the unwinder. The solution involves keeping track of how many frames we've looked at while walking the stack. The other problem was that compareIpToMethodBounds assumed every method was followed by at least one byte of padding before the next method started. That assumption was usually valid because we were storing the size following method code prior to the code itself. However, the last method of an AOT-compiled code image is not followed by any such method header and may instead be followed directly by native code with no intervening padding. In that case, we risk interpreting that native code as part of the preceding method, with potentially bizarre results. The reason for the compareIpToMethodBounds assumption was that methods which throw exceptions as their last instruction generate a non-returning call, which nonetheless push a return address on the stack which points past the end of the method, and the unwinder needs to know that return address belongs to that method. A better solution is to add an extra trap instruction to the end of such methods, which is what this patch does.
2012-05-05 00:35:13 +00:00
const unsigned OperationCount = Trap + 1;
2008-02-11 17:21:41 +00:00
enum UnaryOperation {
2008-02-17 22:29:04 +00:00
Call,
LongCall,
AlignedLongCall,
2008-02-17 22:29:04 +00:00
AlignedCall,
Jump,
LongJump,
AlignedLongJump,
AlignedJump,
2009-10-07 00:50:32 +00:00
NoUnaryOperation = -1
2008-02-11 17:21:41 +00:00
};
2009-10-07 00:50:32 +00:00
const unsigned UnaryOperationCount = AlignedJump + 1;
2008-02-11 17:21:41 +00:00
enum BinaryOperation {
Move,
2009-10-04 19:56:48 +00:00
MoveLow,
MoveHigh,
MoveZ,
Negate,
FloatNegate,
Float2Float,
Float2Int,
Int2Float,
FloatSquareRoot,
FloatAbsolute,
Absolute,
NoBinaryOperation = -1
2008-08-16 17:45:36 +00:00
};
const unsigned BinaryOperationCount = Absolute + 1;
2008-08-16 17:45:36 +00:00
enum TernaryOperation {
2008-02-11 17:21:41 +00:00
Add,
Subtract,
Multiply,
Divide,
Remainder,
ShiftLeft,
ShiftRight,
UnsignedShiftRight,
And,
Or,
Xor,
FloatAdd,
FloatSubtract,
FloatMultiply,
FloatDivide,
FloatRemainder,
FloatMax,
FloatMin,
2009-10-07 00:50:32 +00:00
JumpIfLess,
JumpIfGreater,
JumpIfLessOrEqual,
JumpIfGreaterOrEqual,
JumpIfEqual,
JumpIfNotEqual,
JumpIfFloatEqual,
JumpIfFloatNotEqual,
JumpIfFloatLess,
JumpIfFloatGreater,
JumpIfFloatLessOrEqual,
JumpIfFloatGreaterOrEqual,
JumpIfFloatLessOrUnordered,
JumpIfFloatGreaterOrUnordered,
JumpIfFloatLessOrEqualOrUnordered,
JumpIfFloatGreaterOrEqualOrUnordered,
NoTernaryOperation = -1
2008-02-11 17:21:41 +00:00
};
2009-10-07 00:50:32 +00:00
const unsigned TernaryOperationCount
= JumpIfFloatGreaterOrEqualOrUnordered + 1;
2009-10-19 16:31:34 +00:00
const unsigned NonBranchTernaryOperationCount = FloatMin + 1;
const unsigned BranchOperationCount
= JumpIfFloatGreaterOrEqualOrUnordered - FloatMin;
2008-02-11 17:21:41 +00:00
enum OperandType {
2008-04-17 02:55:38 +00:00
ConstantOperand,
AddressOperand,
RegisterOperand,
2008-04-18 03:47:42 +00:00
MemoryOperand
2008-02-11 17:21:41 +00:00
};
2009-09-26 19:43:44 +00:00
enum ValueType {
ValueGeneral,
ValueFloat
};
2008-04-17 22:07:32 +00:00
const unsigned OperandTypeCount = MemoryOperand + 1;
2008-02-11 17:21:41 +00:00
const int NoRegister = -1;
class Promise {
public:
class Listener {
public:
virtual bool resolve(int64_t value, void** location) = 0;
2008-12-02 02:38:00 +00:00
Listener* next;
};
2008-02-11 17:21:41 +00:00
virtual int64_t value() = 0;
virtual bool resolved() = 0;
virtual Listener* listen(unsigned) { return 0; }
2008-02-11 17:21:41 +00:00
};
class ResolvedPromise: public Promise {
public:
ResolvedPromise(int64_t value): value_(value) { }
virtual int64_t value() {
return value_;
}
virtual bool resolved() {
return true;
}
int64_t value_;
};
class ShiftMaskPromise: public Promise {
public:
ShiftMaskPromise(Promise* base, unsigned shift, int64_t mask):
base(base), shift(shift), mask(mask)
{ }
virtual int64_t value() {
return (base->value() >> shift) & mask;
}
virtual bool resolved() {
return base->resolved();
}
Promise* base;
unsigned shift;
int64_t mask;
};
class CombinedPromise: public Promise {
public:
CombinedPromise(Promise* low, Promise* high):
low(low), high(high)
{ }
virtual int64_t value() {
return low->value() | (high->value() << 32);
}
virtual bool resolved() {
return low->resolved() and high->resolved();
}
Promise* low;
Promise* high;
};
class OffsetPromise: public Promise {
public:
OffsetPromise(Promise* base, int64_t offset):
base(base), offset(offset)
{ }
virtual int64_t value() {
return base->value() + offset;
}
virtual bool resolved() {
return base->resolved();
}
Promise* base;
int64_t offset;
};
class ListenPromise: public Promise {
2008-11-23 23:58:01 +00:00
public:
ListenPromise(System* s, Allocator* allocator):
s(s), allocator(allocator), listener(0)
{ }
2008-11-23 23:58:01 +00:00
virtual int64_t value() {
abort(s);
}
virtual bool resolved() {
return false;
}
virtual Listener* listen(unsigned sizeInBytes) {
2008-12-02 02:38:00 +00:00
Listener* l = static_cast<Listener*>(allocator->allocate(sizeInBytes));
l->next = listener;
listener = l;
return l;
2008-11-23 23:58:01 +00:00
}
System* s;
Allocator* allocator;
Listener* listener;
Promise* promise;
};
class DelayedPromise: public ListenPromise {
public:
DelayedPromise(System* s, Allocator* allocator, Promise* basis,
DelayedPromise* next):
ListenPromise(s, allocator), basis(basis), next(next)
{ }
virtual int64_t value() {
abort(s);
}
virtual bool resolved() {
return false;
}
virtual Listener* listen(unsigned sizeInBytes) {
Listener* l = static_cast<Listener*>(allocator->allocate(sizeInBytes));
l->next = listener;
listener = l;
return l;
}
Promise* basis;
DelayedPromise* next;
2008-11-23 23:58:01 +00:00
};
2008-02-11 17:21:41 +00:00
class Assembler {
public:
class Operand { };
class Constant: public Operand {
public:
Constant(Promise* value): value(value) { }
Promise* value;
};
class Address: public Operand {
public:
Address(Promise* address): address(address) { }
Promise* address;
};
class Register: public Operand {
public:
Register(int low, int high = NoRegister): low(low), high(high) { }
int low;
int high;
};
class Memory: public Operand {
public:
Memory(int base, int offset, int index = NoRegister, unsigned scale = 1):
base(base), offset(offset), index(index), scale(scale)
2008-02-11 17:21:41 +00:00
{ }
int base;
int offset;
int index;
unsigned scale;
};
class Client {
public:
2008-05-06 21:13:02 +00:00
virtual int acquireTemporary
(uint32_t mask = ~static_cast<uint32_t>(0)) = 0;
virtual void releaseTemporary(int r) = 0;
virtual void save(int r) = 0;
};
2008-09-07 20:12:11 +00:00
class Block {
public:
virtual unsigned resolve(unsigned start, Block* next) = 0;
};
2008-08-17 19:32:40 +00:00
class Architecture {
public:
2009-09-26 19:43:44 +00:00
virtual unsigned floatRegisterSize() = 0;
virtual uint32_t generalRegisterMask() = 0;
virtual uint32_t floatRegisterMask() = 0;
virtual int scratch() = 0;
2008-08-17 19:32:40 +00:00
virtual int stack() = 0;
virtual int thread() = 0;
virtual int returnLow() = 0;
virtual int returnHigh() = 0;
2009-05-03 20:57:11 +00:00
virtual int virtualCallTarget() = 0;
virtual int virtualCallIndex() = 0;
virtual bool bigEndian() = 0;
2009-09-26 19:43:44 +00:00
virtual uintptr_t maximumImmediateJump() = 0;
virtual bool alwaysCondensed(BinaryOperation op) = 0;
virtual bool alwaysCondensed(TernaryOperation op) = 0;
virtual bool reserved(int register_) = 0;
2008-02-11 17:21:41 +00:00
virtual unsigned frameFootprint(unsigned footprint) = 0;
virtual unsigned argumentFootprint(unsigned footprint) = 0;
support stack unwinding without using a frame pointer Previously, we unwound the stack by following the chain of frame pointers for normal returns, stack trace creation, and exception unwinding. On x86, this required reserving EBP/RBP for frame pointer duties, making it unavailable for general computation and requiring that it be explicitly saved and restored on entry and exit, respectively. On PowerPC, we use an ABI that makes the stack pointer double as a frame pointer, so it doesn't cost us anything. We've been using the same convention on ARM, but it doesn't match the native calling convention, which makes it unusable when we want to call native code from Java and pass arguments on the stack. So far, the ARM calling convention mismatch hasn't been an issue because we've never passed more arguments from Java to native code than would fit in registers. However, we must now pass an extra argument (the thread pointer) to e.g. divideLong so it can throw an exception on divide by zero, which means the last argument must be passed on the stack. This will clobber the linkage area we've been using to hold the frame pointer, so we need to stop using it. One solution would be to use the same convention on ARM as we do on x86, but this would introduce the same overhead of making a register unavailable for general use and extra code at method entry and exit. Instead, this commit removes the need for a frame pointer. Unwinding involves consulting a map of instruction offsets to frame sizes which is generated at compile time. This is necessary because stack trace creation can happen at any time due to Thread.getStackTrace being called by another thread, and the frame size varies during the execution of a method. So far, only x86(_64) is working, and continuations and tail call optimization are probably broken. More to come.
2011-01-17 02:05:05 +00:00
virtual bool argumentAlignment() = 0;
2011-08-11 03:33:56 +00:00
virtual bool argumentRegisterAlignment() = 0;
2008-08-17 19:32:40 +00:00
virtual unsigned argumentRegisterCount() = 0;
virtual int argumentRegister(unsigned index) = 0;
2008-02-11 17:21:41 +00:00
virtual bool hasLinkRegister() = 0;
virtual unsigned stackAlignmentInWords() = 0;
virtual bool matchCall(void* returnAddress, void* target) = 0;
virtual void updateCall(UnaryOperation op, void* returnAddress,
void* newTarget) = 0;
2008-02-11 17:21:41 +00:00
virtual void setConstant(void* dst, uint64_t constant) = 0;
2009-03-10 00:52:09 +00:00
2008-08-17 19:32:40 +00:00
virtual unsigned alignFrameSize(unsigned sizeInWords) = 0;
virtual void nextFrame(void* start, unsigned size, unsigned footprint,
fix a couple of subtle Thread.getStackTrace bugs The first problem was that, on x86, we failed to properly keep track of whether to expect the return address to be on the stack or not when unwinding through a frame. We were relying on a "stackLimit" pointer to tell us whether we were looking at the most recently-called frame by comparing it with the stack pointer for that frame. That was inaccurate in the case of a thread executing at the beginning of a method before a new frame is allocated, in which case the most recent two frames share a stack pointer, confusing the unwinder. The solution involves keeping track of how many frames we've looked at while walking the stack. The other problem was that compareIpToMethodBounds assumed every method was followed by at least one byte of padding before the next method started. That assumption was usually valid because we were storing the size following method code prior to the code itself. However, the last method of an AOT-compiled code image is not followed by any such method header and may instead be followed directly by native code with no intervening padding. In that case, we risk interpreting that native code as part of the preceding method, with potentially bizarre results. The reason for the compareIpToMethodBounds assumption was that methods which throw exceptions as their last instruction generate a non-returning call, which nonetheless push a return address on the stack which points past the end of the method, and the unwinder needs to know that return address belongs to that method. A better solution is to add an extra trap instruction to the end of such methods, which is what this patch does.
2012-05-05 00:35:13 +00:00
void* link, bool mostRecent,
unsigned targetParameterFootprint, void** ip,
void** stack) = 0;
virtual void* frameIp(void* stack) = 0;
virtual unsigned frameHeaderSize() = 0;
2008-11-08 22:36:38 +00:00
virtual unsigned frameReturnAddressSize() = 0;
virtual unsigned frameFooterSize() = 0;
2009-04-22 01:39:25 +00:00
virtual int returnAddressOffset() = 0;
virtual int framePointerOffset() = 0;
2008-08-17 19:32:40 +00:00
virtual void plan
(UnaryOperation op,
unsigned aSize, uint8_t* aTypeMask, uint64_t* aRegisterMask,
bool* thunk) = 0;
virtual void planSource
2008-08-17 19:32:40 +00:00
(BinaryOperation op,
unsigned aSize, uint8_t* aTypeMask, uint64_t* aRegisterMask,
unsigned bSize, bool* thunk) = 0;
virtual void planDestination
(BinaryOperation op,
2009-10-04 19:56:48 +00:00
unsigned aSize, uint8_t aTypeMask, uint64_t aRegisterMask,
unsigned bSize, uint8_t* bTypeMask, uint64_t* bRegisterMask) = 0;
2008-08-17 19:32:40 +00:00
2009-10-04 19:56:48 +00:00
virtual void planMove
(unsigned size, uint8_t* srcTypeMask, uint64_t* srcRegisterMask,
uint8_t* tmpTypeMask, uint64_t* tmpRegisterMask,
uint8_t dstTypeMask, uint64_t dstRegisterMask) = 0;
2009-10-04 19:56:48 +00:00
virtual void planSource
2008-08-17 19:32:40 +00:00
(TernaryOperation op,
unsigned aSize, uint8_t* aTypeMask, uint64_t* aRegisterMask,
unsigned bSize, uint8_t* bTypeMask, uint64_t* bRegisterMask,
unsigned cSize, bool* thunk) = 0;
virtual void planDestination
(TernaryOperation op,
2009-10-04 19:56:48 +00:00
unsigned aSize, uint8_t aTypeMask, uint64_t aRegisterMask,
unsigned bSize, uint8_t bTypeMask, uint64_t bRegisterMask,
unsigned cSize, uint8_t* cTypeMask, uint64_t* cRegisterMask) = 0;
virtual Assembler* makeAssembler(Allocator*, Zone*) = 0;
2008-08-16 18:46:14 +00:00
virtual void acquire() = 0;
virtual void release() = 0;
2008-08-17 19:32:40 +00:00
};
2008-07-05 20:21:13 +00:00
2008-08-17 19:32:40 +00:00
virtual void setClient(Client* client) = 0;
virtual Architecture* arch() = 0;
virtual void checkStackOverflow(uintptr_t handler,
unsigned stackLimitOffsetFromThread) = 0;
virtual void saveFrame(unsigned stackOffset, unsigned ipOffset) = 0;
virtual void pushFrame(unsigned argumentCount, ...) = 0;
virtual void allocateFrame(unsigned footprint) = 0;
support stack unwinding without using a frame pointer Previously, we unwound the stack by following the chain of frame pointers for normal returns, stack trace creation, and exception unwinding. On x86, this required reserving EBP/RBP for frame pointer duties, making it unavailable for general computation and requiring that it be explicitly saved and restored on entry and exit, respectively. On PowerPC, we use an ABI that makes the stack pointer double as a frame pointer, so it doesn't cost us anything. We've been using the same convention on ARM, but it doesn't match the native calling convention, which makes it unusable when we want to call native code from Java and pass arguments on the stack. So far, the ARM calling convention mismatch hasn't been an issue because we've never passed more arguments from Java to native code than would fit in registers. However, we must now pass an extra argument (the thread pointer) to e.g. divideLong so it can throw an exception on divide by zero, which means the last argument must be passed on the stack. This will clobber the linkage area we've been using to hold the frame pointer, so we need to stop using it. One solution would be to use the same convention on ARM as we do on x86, but this would introduce the same overhead of making a register unavailable for general use and extra code at method entry and exit. Instead, this commit removes the need for a frame pointer. Unwinding involves consulting a map of instruction offsets to frame sizes which is generated at compile time. This is necessary because stack trace creation can happen at any time due to Thread.getStackTrace being called by another thread, and the frame size varies during the execution of a method. So far, only x86(_64) is working, and continuations and tail call optimization are probably broken. More to come.
2011-01-17 02:05:05 +00:00
virtual void adjustFrame(unsigned difference) = 0;
virtual void popFrame(unsigned footprint) = 0;
virtual void popFrameForTailCall(unsigned footprint, int offset,
int returnAddressSurrogate,
int framePointerSurrogate) = 0;
support stack unwinding without using a frame pointer Previously, we unwound the stack by following the chain of frame pointers for normal returns, stack trace creation, and exception unwinding. On x86, this required reserving EBP/RBP for frame pointer duties, making it unavailable for general computation and requiring that it be explicitly saved and restored on entry and exit, respectively. On PowerPC, we use an ABI that makes the stack pointer double as a frame pointer, so it doesn't cost us anything. We've been using the same convention on ARM, but it doesn't match the native calling convention, which makes it unusable when we want to call native code from Java and pass arguments on the stack. So far, the ARM calling convention mismatch hasn't been an issue because we've never passed more arguments from Java to native code than would fit in registers. However, we must now pass an extra argument (the thread pointer) to e.g. divideLong so it can throw an exception on divide by zero, which means the last argument must be passed on the stack. This will clobber the linkage area we've been using to hold the frame pointer, so we need to stop using it. One solution would be to use the same convention on ARM as we do on x86, but this would introduce the same overhead of making a register unavailable for general use and extra code at method entry and exit. Instead, this commit removes the need for a frame pointer. Unwinding involves consulting a map of instruction offsets to frame sizes which is generated at compile time. This is necessary because stack trace creation can happen at any time due to Thread.getStackTrace being called by another thread, and the frame size varies during the execution of a method. So far, only x86(_64) is working, and continuations and tail call optimization are probably broken. More to come.
2011-01-17 02:05:05 +00:00
virtual void popFrameAndPopArgumentsAndReturn(unsigned frameFootprint,
unsigned argumentFootprint)
= 0;
support stack unwinding without using a frame pointer Previously, we unwound the stack by following the chain of frame pointers for normal returns, stack trace creation, and exception unwinding. On x86, this required reserving EBP/RBP for frame pointer duties, making it unavailable for general computation and requiring that it be explicitly saved and restored on entry and exit, respectively. On PowerPC, we use an ABI that makes the stack pointer double as a frame pointer, so it doesn't cost us anything. We've been using the same convention on ARM, but it doesn't match the native calling convention, which makes it unusable when we want to call native code from Java and pass arguments on the stack. So far, the ARM calling convention mismatch hasn't been an issue because we've never passed more arguments from Java to native code than would fit in registers. However, we must now pass an extra argument (the thread pointer) to e.g. divideLong so it can throw an exception on divide by zero, which means the last argument must be passed on the stack. This will clobber the linkage area we've been using to hold the frame pointer, so we need to stop using it. One solution would be to use the same convention on ARM as we do on x86, but this would introduce the same overhead of making a register unavailable for general use and extra code at method entry and exit. Instead, this commit removes the need for a frame pointer. Unwinding involves consulting a map of instruction offsets to frame sizes which is generated at compile time. This is necessary because stack trace creation can happen at any time due to Thread.getStackTrace being called by another thread, and the frame size varies during the execution of a method. So far, only x86(_64) is working, and continuations and tail call optimization are probably broken. More to come.
2011-01-17 02:05:05 +00:00
virtual void popFrameAndUpdateStackAndReturn(unsigned frameFootprint,
unsigned stackOffsetFromThread)
= 0;
2008-02-11 17:21:41 +00:00
virtual void apply(Operation op) = 0;
2008-08-16 17:45:36 +00:00
virtual void apply(UnaryOperation op,
unsigned aSize, OperandType aType, Operand* aOperand) = 0;
2008-02-11 17:21:41 +00:00
2008-08-16 17:45:36 +00:00
virtual void apply(BinaryOperation op,
unsigned aSize, OperandType aType, Operand* aOperand,
unsigned bSize, OperandType bType, Operand* bOperand) = 0;
virtual void apply(TernaryOperation op,
unsigned aSize, OperandType aType, Operand* aOperand,
unsigned bSize, OperandType bType, Operand* bOperand,
unsigned cSize, OperandType cType, Operand* cOperand) = 0;
2008-02-11 17:21:41 +00:00
virtual void setDestination(uint8_t* dst) = 0;
virtual void write() = 0;
2008-02-11 17:21:41 +00:00
virtual Promise* offset(bool forTrace = false) = 0;
2008-08-30 20:12:27 +00:00
virtual Block* endBlock(bool startNew) = 0;
2008-08-30 20:12:27 +00:00
virtual void endEvent() = 0;
2008-02-11 17:21:41 +00:00
virtual unsigned length() = 0;
2009-10-20 14:20:49 +00:00
virtual unsigned footerSize() = 0;
support stack unwinding without using a frame pointer Previously, we unwound the stack by following the chain of frame pointers for normal returns, stack trace creation, and exception unwinding. On x86, this required reserving EBP/RBP for frame pointer duties, making it unavailable for general computation and requiring that it be explicitly saved and restored on entry and exit, respectively. On PowerPC, we use an ABI that makes the stack pointer double as a frame pointer, so it doesn't cost us anything. We've been using the same convention on ARM, but it doesn't match the native calling convention, which makes it unusable when we want to call native code from Java and pass arguments on the stack. So far, the ARM calling convention mismatch hasn't been an issue because we've never passed more arguments from Java to native code than would fit in registers. However, we must now pass an extra argument (the thread pointer) to e.g. divideLong so it can throw an exception on divide by zero, which means the last argument must be passed on the stack. This will clobber the linkage area we've been using to hold the frame pointer, so we need to stop using it. One solution would be to use the same convention on ARM as we do on x86, but this would introduce the same overhead of making a register unavailable for general use and extra code at method entry and exit. Instead, this commit removes the need for a frame pointer. Unwinding involves consulting a map of instruction offsets to frame sizes which is generated at compile time. This is necessary because stack trace creation can happen at any time due to Thread.getStackTrace being called by another thread, and the frame size varies during the execution of a method. So far, only x86(_64) is working, and continuations and tail call optimization are probably broken. More to come.
2011-01-17 02:05:05 +00:00
2008-02-11 17:21:41 +00:00
virtual void dispose() = 0;
};
} // namespace vm
#endif//ASSEMBLER_H