void JIT::compileSetupVarargsFrame(Instruction* instruction, CallLinkInfo* info)
{
int thisValue = instruction[3].u.operand;
int arguments = instruction[4].u.operand;
int firstFreeRegister = instruction[5].u.operand;
int firstVarArgOffset = instruction[6].u.operand;
emitLoad(arguments, regT1, regT0);
callOperation(operationSizeFrameForVarargs, regT1, regT0, -firstFreeRegister, firstVarArgOffset);
move(TrustedImm32(-firstFreeRegister), regT1);
emitSetVarargsFrame(*this, returnValueGPR, false, regT1, regT1);
addPtr(TrustedImm32(-(sizeof(CallerFrameAndPC) + WTF::roundUpToMultipleOf(stackAlignmentBytes(), 6 * sizeof(void*)))), regT1, stackPointerRegister);
emitLoad(arguments, regT2, regT4);
callOperation(operationSetupVarargsFrame, regT1, regT2, regT4, firstVarArgOffset, regT0);
move(returnValueGPR, regT1);
// Profile the argument count.
load32(Address(regT1, JSStack::ArgumentCount * static_cast<int>(sizeof(Register)) + PayloadOffset), regT2);
load8(info->addressOfMaxNumArguments(), regT0);
Jump notBiggest = branch32(Above, regT0, regT2);
Jump notSaturated = branch32(BelowOrEqual, regT2, TrustedImm32(255));
move(TrustedImm32(255), regT2);
notSaturated.link(this);
store8(regT2, info->addressOfMaxNumArguments());
notBiggest.link(this);
// Initialize 'this'.
emitLoad(thisValue, regT2, regT0);
store32(regT0, Address(regT1, PayloadOffset + (CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))));
store32(regT2, Address(regT1, TagOffset + (CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))));
addPtr(TrustedImm32(sizeof(CallerFrameAndPC)), regT1, stackPointerRegister);
}
static int blake2s_compress( blake2s_state *S, const uint8_t block[BLAKE2S_BLOCKBYTES] )
{
uint32_t m[16];
uint32_t v[16];
size_t i;
for( i = 0; i < 16; ++i )
m[i] = load32( block + i * sizeof( m[i] ) );
for( i = 0; i < 8; ++i )
v[i] = S->h[i];
v[ 8] = blake2s_IV[0];
v[ 9] = blake2s_IV[1];
v[10] = blake2s_IV[2];
v[11] = blake2s_IV[3];
v[12] = S->t[0] ^ blake2s_IV[4];
v[13] = S->t[1] ^ blake2s_IV[5];
v[14] = S->f[0] ^ blake2s_IV[6];
v[15] = S->f[1] ^ blake2s_IV[7];
#define G(r,i,a,b,c,d) \
do { \
a = a + b + m[blake2s_sigma[r][2*i+0]]; \
d = rotr32(d ^ a, 16); \
c = c + d; \
b = rotr32(b ^ c, 12); \
a = a + b + m[blake2s_sigma[r][2*i+1]]; \
d = rotr32(d ^ a, 8); \
c = c + d; \
b = rotr32(b ^ c, 7); \
} while(0)
#define ROUND(r) \
do { \
G(r,0,v[ 0],v[ 4],v[ 8],v[12]); \
G(r,1,v[ 1],v[ 5],v[ 9],v[13]); \
G(r,2,v[ 2],v[ 6],v[10],v[14]); \
G(r,3,v[ 3],v[ 7],v[11],v[15]); \
G(r,4,v[ 0],v[ 5],v[10],v[15]); \
G(r,5,v[ 1],v[ 6],v[11],v[12]); \
G(r,6,v[ 2],v[ 7],v[ 8],v[13]); \
G(r,7,v[ 3],v[ 4],v[ 9],v[14]); \
} while(0)
ROUND( 0 );
ROUND( 1 );
ROUND( 2 );
ROUND( 3 );
ROUND( 4 );
ROUND( 5 );
ROUND( 6 );
ROUND( 7 );
ROUND( 8 );
ROUND( 9 );
for( i = 0; i < 8; ++i )
S->h[i] = S->h[i] ^ v[i] ^ v[i + 8];
#undef G
#undef ROUND
return 0;
}
std::enable_if_t<
Op::opcodeID == op_call_varargs || Op::opcodeID == op_construct_varargs
|| Op::opcodeID == op_tail_call_varargs || Op::opcodeID == op_tail_call_forward_arguments
, void>
JIT::compileSetupFrame(const Op& bytecode, CallLinkInfo* info)
{
OpcodeID opcodeID = Op::opcodeID;
int thisValue = bytecode.m_thisValue.offset();
int arguments = bytecode.m_arguments.offset();
int firstFreeRegister = bytecode.m_firstFree.offset();
int firstVarArgOffset = bytecode.m_firstVarArg;
emitLoad(arguments, regT1, regT0);
Z_JITOperation_EJZZ sizeOperation;
if (Op::opcodeID == op_tail_call_forward_arguments)
sizeOperation = operationSizeFrameForForwardArguments;
else
sizeOperation = operationSizeFrameForVarargs;
callOperation(sizeOperation, JSValueRegs(regT1, regT0), -firstFreeRegister, firstVarArgOffset);
move(TrustedImm32(-firstFreeRegister), regT1);
emitSetVarargsFrame(*this, returnValueGPR, false, regT1, regT1);
addPtr(TrustedImm32(-(sizeof(CallerFrameAndPC) + WTF::roundUpToMultipleOf(stackAlignmentBytes(), 6 * sizeof(void*)))), regT1, stackPointerRegister);
emitLoad(arguments, regT2, regT4);
F_JITOperation_EFJZZ setupOperation;
if (opcodeID == op_tail_call_forward_arguments)
setupOperation = operationSetupForwardArgumentsFrame;
else
setupOperation = operationSetupVarargsFrame;
callOperation(setupOperation, regT1, JSValueRegs(regT2, regT4), firstVarArgOffset, regT0);
move(returnValueGPR, regT1);
// Profile the argument count.
load32(Address(regT1, CallFrameSlot::argumentCount * static_cast<int>(sizeof(Register)) + PayloadOffset), regT2);
load32(info->addressOfMaxNumArguments(), regT0);
Jump notBiggest = branch32(Above, regT0, regT2);
store32(regT2, info->addressOfMaxNumArguments());
notBiggest.link(this);
// Initialize 'this'.
emitLoad(thisValue, regT2, regT0);
store32(regT0, Address(regT1, PayloadOffset + (CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))));
store32(regT2, Address(regT1, TagOffset + (CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))));
addPtr(TrustedImm32(sizeof(CallerFrameAndPC)), regT1, stackPointerRegister);
}
/* init2 xors IV with input parameter block */
int blake2s_init_param( blake2s_state *S, const blake2s_param *P )
{
blake2s_init0( S );
uint32_t *p = ( uint32_t * )( P );
/* IV XOR ParamBlock */
for( size_t i = 0; i < 8; ++i )
S->h[i] ^= load32( &p[i] );
return 0;
}
void JIT::compileSetupVarargsFrame(OpcodeID opcode, Instruction* instruction, CallLinkInfo* info)
{
int thisValue = instruction[3].u.operand;
int arguments = instruction[4].u.operand;
int firstFreeRegister = instruction[5].u.operand;
int firstVarArgOffset = instruction[6].u.operand;
emitGetVirtualRegister(arguments, regT1);
Z_JITOperation_EJZZ sizeOperation;
if (opcode == op_tail_call_forward_arguments)
sizeOperation = operationSizeFrameForForwardArguments;
else
sizeOperation = operationSizeFrameForVarargs;
callOperation(sizeOperation, regT1, -firstFreeRegister, firstVarArgOffset);
move(TrustedImm32(-firstFreeRegister), regT1);
emitSetVarargsFrame(*this, returnValueGPR, false, regT1, regT1);
addPtr(TrustedImm32(-(sizeof(CallerFrameAndPC) + WTF::roundUpToMultipleOf(stackAlignmentBytes(), 5 * sizeof(void*)))), regT1, stackPointerRegister);
emitGetVirtualRegister(arguments, regT2);
F_JITOperation_EFJZZ setupOperation;
if (opcode == op_tail_call_forward_arguments)
setupOperation = operationSetupForwardArgumentsFrame;
else
setupOperation = operationSetupVarargsFrame;
callOperation(setupOperation, regT1, regT2, firstVarArgOffset, regT0);
move(returnValueGPR, regT1);
// Profile the argument count.
load32(Address(regT1, CallFrameSlot::argumentCount * static_cast<int>(sizeof(Register)) + PayloadOffset), regT2);
load32(info->addressOfMaxNumArguments(), regT0);
Jump notBiggest = branch32(Above, regT0, regT2);
store32(regT2, info->addressOfMaxNumArguments());
notBiggest.link(this);
// Initialize 'this'.
emitGetVirtualRegister(thisValue, regT0);
store64(regT0, Address(regT1, CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register))));
addPtr(TrustedImm32(sizeof(CallerFrameAndPC)), regT1, stackPointerRegister);
}
void JIT::compileSetupVarargsFrame(Instruction* instruction, CallLinkInfo* info)
{
int thisValue = instruction[3].u.operand;
int arguments = instruction[4].u.operand;
int firstFreeRegister = instruction[5].u.operand;
int firstVarArgOffset = instruction[6].u.operand;
JumpList slowCase;
JumpList end;
bool canOptimize = m_codeBlock->usesArguments()
&& arguments == m_codeBlock->argumentsRegister().offset()
&& !m_codeBlock->symbolTable()->slowArguments();
if (canOptimize) {
emitGetVirtualRegister(arguments, regT0);
slowCase.append(branch64(NotEqual, regT0, TrustedImm64(JSValue::encode(JSValue()))));
move(TrustedImm32(-firstFreeRegister), regT1);
emitSetupVarargsFrameFastCase(*this, regT1, regT0, regT1, regT2, firstVarArgOffset, slowCase);
end.append(jump());
slowCase.link(this);
}
emitGetVirtualRegister(arguments, regT1);
callOperation(operationSizeFrameForVarargs, regT1, -firstFreeRegister, firstVarArgOffset);
move(TrustedImm32(-firstFreeRegister), regT1);
emitSetVarargsFrame(*this, returnValueGPR, false, regT1, regT1);
addPtr(TrustedImm32(-(sizeof(CallerFrameAndPC) + WTF::roundUpToMultipleOf(stackAlignmentBytes(), 5 * sizeof(void*)))), regT1, stackPointerRegister);
emitGetVirtualRegister(arguments, regT2);
callOperation(operationSetupVarargsFrame, regT1, regT2, firstVarArgOffset, regT0);
move(returnValueGPR, regT1);
if (canOptimize)
end.link(this);
// Profile the argument count.
load32(Address(regT1, JSStack::ArgumentCount * static_cast<int>(sizeof(Register)) + PayloadOffset), regT2);
load8(&info->maxNumArguments, regT0);
Jump notBiggest = branch32(Above, regT0, regT2);
Jump notSaturated = branch32(BelowOrEqual, regT2, TrustedImm32(255));
move(TrustedImm32(255), regT2);
notSaturated.link(this);
store8(regT2, &info->maxNumArguments);
notBiggest.link(this);
// Initialize 'this'.
emitGetVirtualRegister(thisValue, regT0);
store64(regT0, Address(regT1, CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register))));
addPtr(TrustedImm32(sizeof(CallerFrameAndPC)), regT1, stackPointerRegister);
}
/* init2 xors IV with input parameter block */
static void blake2s_init_param(BLAKE2S_CTX *S, const BLAKE2S_PARAM *P)
{
const uint8_t *p = (const uint8_t *)(P);
size_t i;
/* The param struct is carefully hand packed, and should be 32 bytes on
* every platform. */
assert(sizeof(BLAKE2S_PARAM) == 32);
blake2s_init0(S);
/* IV XOR ParamBlock */
for (i = 0; i < 8; ++i) {
S->h[i] ^= load32(&p[i*4]);
}
}
/* init2 xors IV with input parameter block */
int blake2s_init_param( blake2s_state *S, const blake2s_param *P )
{
const unsigned char *p = ( const unsigned char * )( P );
size_t i;
blake2s_init0( S );
/* IV XOR ParamBlock */
for( i = 0; i < 8; ++i )
S->h[i] ^= load32( &p[i * 4] );
S->outlen = P->digest_length;
return 0;
}
// 08080168
u8 *battle_load_arguments(struct battle_config_entry *bce, u8 *cursor) {
while (1) {
switch (bce->target_type) {
case LOAD_8: *(u8 *)bce->target = load8 (cursor); cursor += 1; break;
case LOAD_16: *(u16*)bce->target = load16(cursor); cursor += 2; break;
case LOAD_32: *(u32*)bce->target = load32(cursor); cursor += 4; break;
case ZERO_8: *(u8 *)bce->target = 0; cursor += 1; break;
case ZERO_16: *(u16*)bce->target = 0; cursor += 2; break;
case ZERO_32: *(u32*)bce->target = 0; cursor += 4; break;
case END: *(u32*)bce->target = (u32)cursor; return cursor;
default: break;
}
bce++;
}
}
std::enable_if_t<
Op::opcodeID != op_call_varargs && Op::opcodeID != op_construct_varargs
&& Op::opcodeID != op_tail_call_varargs && Op::opcodeID != op_tail_call_forward_arguments
, void>
JIT::compileSetupFrame(const Op& bytecode, CallLinkInfo*)
{
auto& metadata = bytecode.metadata(m_codeBlock);
int argCount = bytecode.m_argc;
int registerOffset = -static_cast<int>(bytecode.m_argv);
if (Op::opcodeID == op_call && shouldEmitProfiling()) {
emitLoad(registerOffset + CallFrame::argumentOffsetIncludingThis(0), regT0, regT1);
Jump done = branchIfNotCell(regT0);
load32(Address(regT1, JSCell::structureIDOffset()), regT1);
store32(regT1, metadata.m_arrayProfile.addressOfLastSeenStructureID());
done.link(this);
}
addPtr(TrustedImm32(registerOffset * sizeof(Register) + sizeof(CallerFrameAndPC)), callFrameRegister, stackPointerRegister);
store32(TrustedImm32(argCount), Address(stackPointerRegister, CallFrameSlot::argumentCount * static_cast<int>(sizeof(Register)) + PayloadOffset - sizeof(CallerFrameAndPC)));
}
void AssemblyHelpers::callExceptionFuzz()
{
if (!Options::enableExceptionFuzz())
return;
EncodedJSValue* buffer = vm()->exceptionFuzzingBuffer(sizeof(EncodedJSValue) * (GPRInfo::numberOfRegisters + FPRInfo::numberOfRegisters));
for (unsigned i = 0; i < GPRInfo::numberOfRegisters; ++i) {
#if USE(JSVALUE64)
store64(GPRInfo::toRegister(i), buffer + i);
#else
store32(GPRInfo::toRegister(i), buffer + i);
#endif
}
for (unsigned i = 0; i < FPRInfo::numberOfRegisters; ++i) {
move(TrustedImmPtr(buffer + GPRInfo::numberOfRegisters + i), GPRInfo::regT0);
storeDouble(FPRInfo::toRegister(i), Address(GPRInfo::regT0));
}
// Set up one argument.
#if CPU(X86)
poke(GPRInfo::callFrameRegister, 0);
#else
move(GPRInfo::callFrameRegister, GPRInfo::argumentGPR0);
#endif
move(TrustedImmPtr(bitwise_cast<void*>(operationExceptionFuzz)), GPRInfo::nonPreservedNonReturnGPR);
call(GPRInfo::nonPreservedNonReturnGPR);
for (unsigned i = 0; i < FPRInfo::numberOfRegisters; ++i) {
move(TrustedImmPtr(buffer + GPRInfo::numberOfRegisters + i), GPRInfo::regT0);
loadDouble(Address(GPRInfo::regT0), FPRInfo::toRegister(i));
}
for (unsigned i = 0; i < GPRInfo::numberOfRegisters; ++i) {
#if USE(JSVALUE64)
load64(buffer + i, GPRInfo::toRegister(i));
#else
load32(buffer + i, GPRInfo::toRegister(i));
#endif
}
}
static void blake2s_compress( blake2s_state *S, const uint8_t in[BLAKE2S_BLOCKBYTES] )
{
uint32_t m[16];
uint32_t v[16];
size_t i;
for( i = 0; i < 16; ++i ) {
m[i] = load32( in + i * sizeof( m[i] ) );
}
for( i = 0; i < 8; ++i ) {
v[i] = S->h[i];
}
v[ 8] = blake2s_IV[0];
v[ 9] = blake2s_IV[1];
v[10] = blake2s_IV[2];
v[11] = blake2s_IV[3];
v[12] = S->t[0] ^ blake2s_IV[4];
v[13] = S->t[1] ^ blake2s_IV[5];
v[14] = S->f[0] ^ blake2s_IV[6];
v[15] = S->f[1] ^ blake2s_IV[7];
ROUND( 0 );
ROUND( 1 );
ROUND( 2 );
ROUND( 3 );
ROUND( 4 );
ROUND( 5 );
ROUND( 6 );
ROUND( 7 );
ROUND( 8 );
ROUND( 9 );
for( i = 0; i < 8; ++i ) {
S->h[i] = S->h[i] ^ v[i] ^ v[i + 8];
}
}
void JITCompiler::compileFunction()
{
SamplingRegion samplingRegion("DFG Backend");
setStartOfCode();
compileEntry();
// === Function header code generation ===
// This is the main entry point, without performing an arity check.
// If we needed to perform an arity check we will already have moved the return address,
// so enter after this.
Label fromArityCheck(this);
// Plant a check that sufficient space is available in the JSStack.
addPtr(TrustedImm32(virtualRegisterForLocal(m_graph.requiredRegisterCountForExecutionAndExit() - 1).offset() * sizeof(Register)), GPRInfo::callFrameRegister, GPRInfo::regT1);
Jump stackOverflow = branchPtr(Above, AbsoluteAddress(m_vm->addressOfStackLimit()), GPRInfo::regT1);
// Move the stack pointer down to accommodate locals
addPtr(TrustedImm32(m_graph.stackPointerOffset() * sizeof(Register)), GPRInfo::callFrameRegister, stackPointerRegister);
checkStackPointerAlignment();
// === Function body code generation ===
m_speculative = adoptPtr(new SpeculativeJIT(*this));
compileBody();
setEndOfMainPath();
// === Function footer code generation ===
//
// Generate code to perform the stack overflow handling (if the stack check in
// the function header fails), and generate the entry point with arity check.
//
// Generate the stack overflow handling; if the stack check in the function head fails,
// we need to call out to a helper function to throw the StackOverflowError.
stackOverflow.link(this);
emitStoreCodeOrigin(CodeOrigin(0));
if (maxFrameExtentForSlowPathCall)
addPtr(TrustedImm32(-maxFrameExtentForSlowPathCall), stackPointerRegister);
m_speculative->callOperationWithCallFrameRollbackOnException(operationThrowStackOverflowError, m_codeBlock);
// The fast entry point into a function does not check the correct number of arguments
// have been passed to the call (we only use the fast entry point where we can statically
// determine the correct number of arguments have been passed, or have already checked).
// In cases where an arity check is necessary, we enter here.
// FIXME: change this from a cti call to a DFG style operation (normal C calling conventions).
m_arityCheck = label();
compileEntry();
load32(AssemblyHelpers::payloadFor((VirtualRegister)JSStack::ArgumentCount), GPRInfo::regT1);
branch32(AboveOrEqual, GPRInfo::regT1, TrustedImm32(m_codeBlock->numParameters())).linkTo(fromArityCheck, this);
emitStoreCodeOrigin(CodeOrigin(0));
if (maxFrameExtentForSlowPathCall)
addPtr(TrustedImm32(-maxFrameExtentForSlowPathCall), stackPointerRegister);
m_speculative->callOperationWithCallFrameRollbackOnException(m_codeBlock->m_isConstructor ? operationConstructArityCheck : operationCallArityCheck, GPRInfo::regT0);
if (maxFrameExtentForSlowPathCall)
addPtr(TrustedImm32(maxFrameExtentForSlowPathCall), stackPointerRegister);
branchTest32(Zero, GPRInfo::regT0).linkTo(fromArityCheck, this);
emitStoreCodeOrigin(CodeOrigin(0));
move(TrustedImmPtr(m_vm->arityCheckFailReturnThunks->returnPCsFor(*m_vm, m_codeBlock->numParameters())), GPRInfo::regT5);
loadPtr(BaseIndex(GPRInfo::regT5, GPRInfo::regT0, timesPtr()), GPRInfo::regT5);
m_callArityFixup = call();
jump(fromArityCheck);
// Generate slow path code.
m_speculative->runSlowPathGenerators();
compileExceptionHandlers();
linkOSRExits();
// Create OSR entry trampolines if necessary.
m_speculative->createOSREntries();
setEndOfCode();
}
void JIT::compileOpCall(OpcodeID opcodeID, Instruction* instruction, unsigned callLinkInfoIndex)
{
CallLinkInfo* info = m_codeBlock->addCallLinkInfo();
int callee = instruction[2].u.operand;
/* Caller always:
- Updates callFrameRegister to callee callFrame.
- Initializes ArgumentCount; CallerFrame; Callee.
For a JS call:
- Callee initializes ReturnPC; CodeBlock.
- Callee restores callFrameRegister before return.
For a non-JS call:
- Caller initializes ReturnPC; CodeBlock.
- Caller restores callFrameRegister after return.
*/
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_construct), call_and_construct_opcodes_must_be_same_length);
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_call_varargs), call_and_call_varargs_opcodes_must_be_same_length);
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_construct_varargs), call_and_construct_varargs_opcodes_must_be_same_length);
if (opcodeID == op_call_varargs || opcodeID == op_construct_varargs)
compileSetupVarargsFrame(instruction, info);
else {
int argCount = instruction[3].u.operand;
int registerOffset = -instruction[4].u.operand;
if (opcodeID == op_call && shouldEmitProfiling()) {
emitGetVirtualRegister(registerOffset + CallFrame::argumentOffsetIncludingThis(0), regT0);
Jump done = emitJumpIfNotJSCell(regT0);
load32(Address(regT0, JSCell::structureIDOffset()), regT0);
store32(regT0, instruction[OPCODE_LENGTH(op_call) - 2].u.arrayProfile->addressOfLastSeenStructureID());
done.link(this);
}
addPtr(TrustedImm32(registerOffset * sizeof(Register) + sizeof(CallerFrameAndPC)), callFrameRegister, stackPointerRegister);
store32(TrustedImm32(argCount), Address(stackPointerRegister, JSStack::ArgumentCount * static_cast<int>(sizeof(Register)) + PayloadOffset - sizeof(CallerFrameAndPC)));
} // SP holds newCallFrame + sizeof(CallerFrameAndPC), with ArgumentCount initialized.
uint32_t bytecodeOffset = instruction - m_codeBlock->instructions().begin();
uint32_t locationBits = CallFrame::Location::encodeAsBytecodeOffset(bytecodeOffset);
store32(TrustedImm32(locationBits), Address(callFrameRegister, JSStack::ArgumentCount * static_cast<int>(sizeof(Register)) + TagOffset));
emitGetVirtualRegister(callee, regT0); // regT0 holds callee.
store64(regT0, Address(stackPointerRegister, JSStack::Callee * static_cast<int>(sizeof(Register)) - sizeof(CallerFrameAndPC)));
if (opcodeID == op_call_eval) {
compileCallEval(instruction);
return;
}
DataLabelPtr addressOfLinkedFunctionCheck;
Jump slowCase = branchPtrWithPatch(NotEqual, regT0, addressOfLinkedFunctionCheck, TrustedImmPtr(0));
addSlowCase(slowCase);
ASSERT(m_callCompilationInfo.size() == callLinkInfoIndex);
info->callType = CallLinkInfo::callTypeFor(opcodeID);
info->codeOrigin = CodeOrigin(m_bytecodeOffset);
info->calleeGPR = regT0;
m_callCompilationInfo.append(CallCompilationInfo());
m_callCompilationInfo[callLinkInfoIndex].hotPathBegin = addressOfLinkedFunctionCheck;
m_callCompilationInfo[callLinkInfoIndex].callLinkInfo = info;
m_callCompilationInfo[callLinkInfoIndex].hotPathOther = emitNakedCall();
addPtr(TrustedImm32(stackPointerOffsetFor(m_codeBlock) * sizeof(Register)), callFrameRegister, stackPointerRegister);
checkStackPointerAlignment();
sampleCodeBlock(m_codeBlock);
emitPutCallResult(instruction);
}
/* Permute the state while xoring in the block of data. */
static void blake2s_compress(BLAKE2S_CTX *S,
const uint8_t *blocks,
size_t len)
{
uint32_t m[16];
uint32_t v[16];
size_t i;
size_t increment;
/*
* There are two distinct usage vectors for this function:
*
* a) BLAKE2s_Update uses it to process complete blocks,
* possibly more than one at a time;
*
* b) BLAK2s_Final uses it to process last block, always
* single but possibly incomplete, in which case caller
* pads input with zeros.
*/
assert(len < BLAKE2S_BLOCKBYTES || len % BLAKE2S_BLOCKBYTES == 0);
/*
* Since last block is always processed with separate call,
* |len| not being multiple of complete blocks can be observed
* only with |len| being less than BLAKE2S_BLOCKBYTES ("less"
* including even zero), which is why following assignment doesn't
* have to reside inside the main loop below.
*/
increment = len < BLAKE2S_BLOCKBYTES ? len : BLAKE2S_BLOCKBYTES;
for (i = 0; i < 8; ++i) {
v[i] = S->h[i];
}
do {
for (i = 0; i < 16; ++i) {
m[i] = load32(blocks + i * sizeof(m[i]));
}
/* blake2s_increment_counter */
S->t[0] += increment;
S->t[1] += (S->t[0] < increment);
v[ 8] = blake2s_IV[0];
v[ 9] = blake2s_IV[1];
v[10] = blake2s_IV[2];
v[11] = blake2s_IV[3];
v[12] = S->t[0] ^ blake2s_IV[4];
v[13] = S->t[1] ^ blake2s_IV[5];
v[14] = S->f[0] ^ blake2s_IV[6];
v[15] = S->f[1] ^ blake2s_IV[7];
#define G(r,i,a,b,c,d) \
do { \
a = a + b + m[blake2s_sigma[r][2*i+0]]; \
d = rotr32(d ^ a, 16); \
c = c + d; \
b = rotr32(b ^ c, 12); \
a = a + b + m[blake2s_sigma[r][2*i+1]]; \
d = rotr32(d ^ a, 8); \
c = c + d; \
b = rotr32(b ^ c, 7); \
} while (0)
#define ROUND(r) \
do { \
G(r,0,v[ 0],v[ 4],v[ 8],v[12]); \
G(r,1,v[ 1],v[ 5],v[ 9],v[13]); \
G(r,2,v[ 2],v[ 6],v[10],v[14]); \
G(r,3,v[ 3],v[ 7],v[11],v[15]); \
G(r,4,v[ 0],v[ 5],v[10],v[15]); \
G(r,5,v[ 1],v[ 6],v[11],v[12]); \
G(r,6,v[ 2],v[ 7],v[ 8],v[13]); \
G(r,7,v[ 3],v[ 4],v[ 9],v[14]); \
} while (0)
#if defined(OPENSSL_SMALL_FOOTPRINT)
/* almost 3x reduction on x86_64, 4.5x on ARMv8, 4x on ARMv4 */
for (i = 0; i < 10; i++) {
ROUND(i);
}
#else
ROUND(0);
ROUND(1);
ROUND(2);
ROUND(3);
ROUND(4);
ROUND(5);
ROUND(6);
ROUND(7);
ROUND(8);
ROUND(9);
#endif
for (i = 0; i < 8; ++i) {
S->h[i] = v[i] ^= v[i + 8] ^ S->h[i];
}
#undef G
#undef ROUND
blocks += increment;
len -= increment;
} while (len);
}
bool JITCompiler::compileFunction(JITCode& entry, MacroAssemblerCodePtr& entryWithArityCheck)
{
SamplingRegion samplingRegion("DFG Backend");
setStartOfCode();
compileEntry();
// === Function header code generation ===
// This is the main entry point, without performing an arity check.
// If we needed to perform an arity check we will already have moved the return address,
// so enter after this.
Label fromArityCheck(this);
// Plant a check that sufficient space is available in the JSStack.
// FIXME: https://bugs.webkit.org/show_bug.cgi?id=56291
addPtr(TrustedImm32(m_codeBlock->m_numCalleeRegisters * sizeof(Register)), GPRInfo::callFrameRegister, GPRInfo::regT1);
Jump stackCheck = branchPtr(Below, AbsoluteAddress(m_vm->interpreter->stack().addressOfEnd()), GPRInfo::regT1);
// Return here after stack check.
Label fromStackCheck = label();
// === Function body code generation ===
SpeculativeJIT speculative(*this);
compileBody(speculative);
setEndOfMainPath();
// === Function footer code generation ===
//
// Generate code to perform the slow stack check (if the fast one in
// the function header fails), and generate the entry point with arity check.
//
// Generate the stack check; if the fast check in the function head fails,
// we need to call out to a helper function to check whether more space is available.
// FIXME: change this from a cti call to a DFG style operation (normal C calling conventions).
stackCheck.link(this);
move(stackPointerRegister, GPRInfo::argumentGPR0);
poke(GPRInfo::callFrameRegister, OBJECT_OFFSETOF(struct JITStackFrame, callFrame) / sizeof(void*));
CallBeginToken token;
beginCall(CodeOrigin(0), token);
Call callStackCheck = call();
notifyCall(callStackCheck, CodeOrigin(0), token);
jump(fromStackCheck);
// The fast entry point into a function does not check the correct number of arguments
// have been passed to the call (we only use the fast entry point where we can statically
// determine the correct number of arguments have been passed, or have already checked).
// In cases where an arity check is necessary, we enter here.
// FIXME: change this from a cti call to a DFG style operation (normal C calling conventions).
Label arityCheck = label();
compileEntry();
load32(AssemblyHelpers::payloadFor((VirtualRegister)JSStack::ArgumentCount), GPRInfo::regT1);
branch32(AboveOrEqual, GPRInfo::regT1, TrustedImm32(m_codeBlock->numParameters())).linkTo(fromArityCheck, this);
move(stackPointerRegister, GPRInfo::argumentGPR0);
poke(GPRInfo::callFrameRegister, OBJECT_OFFSETOF(struct JITStackFrame, callFrame) / sizeof(void*));
beginCall(CodeOrigin(0), token);
Call callArityCheck = call();
notifyCall(callArityCheck, CodeOrigin(0), token);
move(GPRInfo::regT0, GPRInfo::callFrameRegister);
jump(fromArityCheck);
// Generate slow path code.
speculative.runSlowPathGenerators();
compileExceptionHandlers();
linkOSRExits();
// Create OSR entry trampolines if necessary.
speculative.createOSREntries();
setEndOfCode();
// === Link ===
LinkBuffer linkBuffer(*m_vm, this, m_codeBlock, JITCompilationCanFail);
if (linkBuffer.didFailToAllocate())
return false;
link(linkBuffer);
speculative.linkOSREntries(linkBuffer);
// FIXME: switch the stack check & arity check over to DFGOpertaion style calls, not JIT stubs.
linkBuffer.link(callStackCheck, cti_stack_check);
linkBuffer.link(callArityCheck, m_codeBlock->m_isConstructor ? cti_op_construct_arityCheck : cti_op_call_arityCheck);
if (shouldShowDisassembly())
m_disassembler->dump(linkBuffer);
if (m_graph.m_compilation)
m_disassembler->reportToProfiler(m_graph.m_compilation.get(), linkBuffer);
entryWithArityCheck = linkBuffer.locationOf(arityCheck);
entry = JITCode(
linkBuffer.finalizeCodeWithoutDisassembly(),
JITCode::DFGJIT);
return true;
}
void JITCompiler::compileFunction()
{
setStartOfCode();
compileEntry();
// === Function header code generation ===
// This is the main entry point, without performing an arity check.
// If we needed to perform an arity check we will already have moved the return address,
// so enter after this.
Label fromArityCheck(this);
// Plant a check that sufficient space is available in the JSStack.
addPtr(TrustedImm32(virtualRegisterForLocal(m_graph.requiredRegisterCountForExecutionAndExit() - 1).offset() * sizeof(Register)), GPRInfo::callFrameRegister, GPRInfo::regT1);
Jump stackOverflow = branchPtr(Above, AbsoluteAddress(m_vm->addressOfStackLimit()), GPRInfo::regT1);
// Move the stack pointer down to accommodate locals
addPtr(TrustedImm32(m_graph.stackPointerOffset() * sizeof(Register)), GPRInfo::callFrameRegister, stackPointerRegister);
checkStackPointerAlignment();
compileSetupRegistersForEntry();
compileEntryExecutionFlag();
// === Function body code generation ===
m_speculative = std::make_unique<SpeculativeJIT>(*this);
compileBody();
setEndOfMainPath();
// === Function footer code generation ===
//
// Generate code to perform the stack overflow handling (if the stack check in
// the function header fails), and generate the entry point with arity check.
//
// Generate the stack overflow handling; if the stack check in the function head fails,
// we need to call out to a helper function to throw the StackOverflowError.
stackOverflow.link(this);
emitStoreCodeOrigin(CodeOrigin(0));
if (maxFrameExtentForSlowPathCall)
addPtr(TrustedImm32(-maxFrameExtentForSlowPathCall), stackPointerRegister);
m_speculative->callOperationWithCallFrameRollbackOnException(operationThrowStackOverflowError, m_codeBlock);
// The fast entry point into a function does not check the correct number of arguments
// have been passed to the call (we only use the fast entry point where we can statically
// determine the correct number of arguments have been passed, or have already checked).
// In cases where an arity check is necessary, we enter here.
// FIXME: change this from a cti call to a DFG style operation (normal C calling conventions).
m_arityCheck = label();
compileEntry();
load32(AssemblyHelpers::payloadFor((VirtualRegister)JSStack::ArgumentCount), GPRInfo::regT1);
branch32(AboveOrEqual, GPRInfo::regT1, TrustedImm32(m_codeBlock->numParameters())).linkTo(fromArityCheck, this);
emitStoreCodeOrigin(CodeOrigin(0));
if (maxFrameExtentForSlowPathCall)
addPtr(TrustedImm32(-maxFrameExtentForSlowPathCall), stackPointerRegister);
m_speculative->callOperationWithCallFrameRollbackOnException(m_codeBlock->m_isConstructor ? operationConstructArityCheck : operationCallArityCheck, GPRInfo::regT0);
if (maxFrameExtentForSlowPathCall)
addPtr(TrustedImm32(maxFrameExtentForSlowPathCall), stackPointerRegister);
branchTest32(Zero, GPRInfo::returnValueGPR).linkTo(fromArityCheck, this);
emitStoreCodeOrigin(CodeOrigin(0));
move(GPRInfo::returnValueGPR, GPRInfo::argumentGPR0);
m_callArityFixup = call();
jump(fromArityCheck);
// Generate slow path code.
m_speculative->runSlowPathGenerators(m_pcToCodeOriginMapBuilder);
m_pcToCodeOriginMapBuilder.appendItem(label(), PCToCodeOriginMapBuilder::defaultCodeOrigin());
compileExceptionHandlers();
linkOSRExits();
// Create OSR entry trampolines if necessary.
m_speculative->createOSREntries();
setEndOfCode();
// === Link ===
auto linkBuffer = std::make_unique<LinkBuffer>(*m_vm, *this, m_codeBlock, JITCompilationCanFail);
if (linkBuffer->didFailToAllocate()) {
m_graph.m_plan.finalizer = std::make_unique<FailedFinalizer>(m_graph.m_plan);
return;
}
link(*linkBuffer);
m_speculative->linkOSREntries(*linkBuffer);
m_jitCode->shrinkToFit();
codeBlock()->shrinkToFit(CodeBlock::LateShrink);
linkBuffer->link(m_callArityFixup, FunctionPtr((m_vm->getCTIStub(arityFixupGenerator)).code().executableAddress()));
disassemble(*linkBuffer);
MacroAssemblerCodePtr withArityCheck = linkBuffer->locationOf(m_arityCheck);
m_graph.m_plan.finalizer = std::make_unique<JITFinalizer>(
m_graph.m_plan, m_jitCode.release(), WTFMove(linkBuffer), withArityCheck);
}
void JITCompiler::compileFunction()
{
SamplingRegion samplingRegion("DFG Backend");
setStartOfCode();
compileEntry();
// === Function header code generation ===
// This is the main entry point, without performing an arity check.
// If we needed to perform an arity check we will already have moved the return address,
// so enter after this.
Label fromArityCheck(this);
// Plant a check that sufficient space is available in the JSStack.
// FIXME: https://bugs.webkit.org/show_bug.cgi?id=56291
addPtr(TrustedImm32(-m_codeBlock->m_numCalleeRegisters * sizeof(Register)), GPRInfo::callFrameRegister, GPRInfo::regT1);
Jump stackCheck = branchPtr(Above, AbsoluteAddress(m_vm->interpreter->stack().addressOfEnd()), GPRInfo::regT1);
// Return here after stack check.
Label fromStackCheck = label();
// === Function body code generation ===
m_speculative = adoptPtr(new SpeculativeJIT(*this));
compileBody();
setEndOfMainPath();
// === Function footer code generation ===
//
// Generate code to perform the slow stack check (if the fast one in
// the function header fails), and generate the entry point with arity check.
//
// Generate the stack check; if the fast check in the function head fails,
// we need to call out to a helper function to check whether more space is available.
// FIXME: change this from a cti call to a DFG style operation (normal C calling conventions).
stackCheck.link(this);
move(stackPointerRegister, GPRInfo::argumentGPR0);
poke(GPRInfo::callFrameRegister, OBJECT_OFFSETOF(struct JITStackFrame, callFrame) / sizeof(void*));
emitStoreCodeOrigin(CodeOrigin(0));
m_callStackCheck = call();
jump(fromStackCheck);
// The fast entry point into a function does not check the correct number of arguments
// have been passed to the call (we only use the fast entry point where we can statically
// determine the correct number of arguments have been passed, or have already checked).
// In cases where an arity check is necessary, we enter here.
// FIXME: change this from a cti call to a DFG style operation (normal C calling conventions).
m_arityCheck = label();
compileEntry();
load32(AssemblyHelpers::payloadFor((VirtualRegister)JSStack::ArgumentCount), GPRInfo::regT1);
branch32(AboveOrEqual, GPRInfo::regT1, TrustedImm32(m_codeBlock->numParameters())).linkTo(fromArityCheck, this);
move(stackPointerRegister, GPRInfo::argumentGPR0);
poke(GPRInfo::callFrameRegister, OBJECT_OFFSETOF(struct JITStackFrame, callFrame) / sizeof(void*));
emitStoreCodeOrigin(CodeOrigin(0));
m_callArityCheck = call();
branchTest32(Zero, GPRInfo::regT0).linkTo(fromArityCheck, this);
emitStoreCodeOrigin(CodeOrigin(0));
m_callArityFixup = call();
jump(fromArityCheck);
// Generate slow path code.
m_speculative->runSlowPathGenerators();
compileExceptionHandlers();
linkOSRExits();
// Create OSR entry trampolines if necessary.
m_speculative->createOSREntries();
setEndOfCode();
}
/* only 2 valids values are 256 (32) and 128 (16) */
void cryptonite_chacha_init_core(cryptonite_chacha_state *st,
uint32_t keylen, const uint8_t *key,
uint32_t ivlen, const uint8_t *iv)
{
const uint8_t *constants = (keylen == 32) ? sigma : tau;
int i;
st->d[0] = load32(constants + 0);
st->d[1] = load32(constants + 4);
st->d[2] = load32(constants + 8);
st->d[3] = load32(constants + 12);
st->d[4] = load32(key + 0);
st->d[5] = load32(key + 4);
st->d[6] = load32(key + 8);
st->d[7] = load32(key + 12);
/* we repeat the key on 128 bits */
if (keylen == 32)
key += 16;
st->d[8] = load32(key + 0);
st->d[9] = load32(key + 4);
st->d[10] = load32(key + 8);
st->d[11] = load32(key + 12);
st->d[12] = 0;
switch (ivlen) {
case 8:
st->d[13] = 0;
st->d[14] = load32(iv + 0);
st->d[15] = load32(iv + 4);
break;
case 12:
st->d[13] = load32(iv + 0);
st->d[14] = load32(iv + 4);
st->d[15] = load32(iv + 8);
default:
return;
}
}
void JIT::compileLoadVarargs(Instruction* instruction)
{
int thisValue = instruction[3].u.operand;
int arguments = instruction[4].u.operand;
int firstFreeRegister = instruction[5].u.operand;
JumpList slowCase;
JumpList end;
bool canOptimize = m_codeBlock->usesArguments()
&& VirtualRegister(arguments) == m_codeBlock->argumentsRegister()
&& !m_codeBlock->symbolTable()->slowArguments();
if (canOptimize) {
emitLoadTag(arguments, regT1);
slowCase.append(branch32(NotEqual, regT1, TrustedImm32(JSValue::EmptyValueTag)));
load32(payloadFor(JSStack::ArgumentCount), regT2);
slowCase.append(branch32(Above, regT2, TrustedImm32(Arguments::MaxArguments + 1)));
// regT2: argumentCountIncludingThis
move(regT2, regT3);
neg32(regT3);
add32(TrustedImm32(firstFreeRegister - JSStack::CallFrameHeaderSize), regT3);
lshift32(TrustedImm32(3), regT3);
addPtr(callFrameRegister, regT3);
// regT3: newCallFrame
slowCase.append(branchPtr(Above, AbsoluteAddress(m_vm->interpreter->stack().addressOfEnd()), regT3));
// Initialize ArgumentCount.
store32(regT2, payloadFor(JSStack::ArgumentCount, regT3));
// Initialize 'this'.
emitLoad(thisValue, regT1, regT0);
store32(regT0, Address(regT3, OBJECT_OFFSETOF(JSValue, u.asBits.payload) + (CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))));
store32(regT1, Address(regT3, OBJECT_OFFSETOF(JSValue, u.asBits.tag) + (CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))));
// Copy arguments.
end.append(branchSub32(Zero, TrustedImm32(1), regT2));
// regT2: argumentCount;
Label copyLoop = label();
load32(BaseIndex(callFrameRegister, regT2, TimesEight, OBJECT_OFFSETOF(JSValue, u.asBits.payload) +(CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))), regT0);
load32(BaseIndex(callFrameRegister, regT2, TimesEight, OBJECT_OFFSETOF(JSValue, u.asBits.tag) +(CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))), regT1);
store32(regT0, BaseIndex(regT3, regT2, TimesEight, OBJECT_OFFSETOF(JSValue, u.asBits.payload) +(CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))));
store32(regT1, BaseIndex(regT3, regT2, TimesEight, OBJECT_OFFSETOF(JSValue, u.asBits.tag) +(CallFrame::thisArgumentOffset() * static_cast<int>(sizeof(Register)))));
branchSub32(NonZero, TrustedImm32(1), regT2).linkTo(copyLoop, this);
end.append(jump());
}
if (canOptimize)
slowCase.link(this);
emitLoad(thisValue, regT1, regT0);
emitLoad(arguments, regT3, regT2);
callOperation(operationLoadVarargs, regT1, regT0, regT3, regT2, firstFreeRegister);
move(returnValueRegister, regT3);
if (canOptimize)
end.link(this);
}
void JIT::compileOpCall(OpcodeID opcodeID, Instruction* instruction, unsigned callLinkInfoIndex)
{
int callee = instruction[2].u.operand;
/* Caller always:
- Updates callFrameRegister to callee callFrame.
- Initializes ArgumentCount; CallerFrame; Callee.
For a JS call:
- Callee initializes ReturnPC; CodeBlock.
- Callee restores callFrameRegister before return.
For a non-JS call:
- Caller initializes ReturnPC; CodeBlock.
- Caller restores callFrameRegister after return.
*/
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_construct), call_and_construct_opcodes_must_be_same_length);
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_call_varargs), call_and_call_varargs_opcodes_must_be_same_length);
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_construct_varargs), call_and_construct_varargs_opcodes_must_be_same_length);
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_tail_call), call_and_tail_call_opcodes_must_be_same_length);
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_tail_call_varargs), call_and_tail_call_varargs_opcodes_must_be_same_length);
COMPILE_ASSERT(OPCODE_LENGTH(op_call) == OPCODE_LENGTH(op_tail_call_forward_arguments), call_and_tail_call_forward_arguments_opcodes_must_be_same_length);
CallLinkInfo* info = nullptr;
if (opcodeID != op_call_eval)
info = m_codeBlock->addCallLinkInfo();
if (opcodeID == op_call_varargs || opcodeID == op_construct_varargs || opcodeID == op_tail_call_varargs || opcodeID == op_tail_call_forward_arguments)
compileSetupVarargsFrame(opcodeID, instruction, info);
else {
int argCount = instruction[3].u.operand;
int registerOffset = -instruction[4].u.operand;
if (opcodeID == op_call && shouldEmitProfiling()) {
emitGetVirtualRegister(registerOffset + CallFrame::argumentOffsetIncludingThis(0), regT0);
Jump done = emitJumpIfNotJSCell(regT0);
load32(Address(regT0, JSCell::structureIDOffset()), regT0);
store32(regT0, instruction[OPCODE_LENGTH(op_call) - 2].u.arrayProfile->addressOfLastSeenStructureID());
done.link(this);
}
addPtr(TrustedImm32(registerOffset * sizeof(Register) + sizeof(CallerFrameAndPC)), callFrameRegister, stackPointerRegister);
store32(TrustedImm32(argCount), Address(stackPointerRegister, CallFrameSlot::argumentCount * static_cast<int>(sizeof(Register)) + PayloadOffset - sizeof(CallerFrameAndPC)));
} // SP holds newCallFrame + sizeof(CallerFrameAndPC), with ArgumentCount initialized.
uint32_t bytecodeOffset = instruction - m_codeBlock->instructions().begin();
uint32_t locationBits = CallSiteIndex(bytecodeOffset).bits();
store32(TrustedImm32(locationBits), Address(callFrameRegister, CallFrameSlot::argumentCount * static_cast<int>(sizeof(Register)) + TagOffset));
emitGetVirtualRegister(callee, regT0); // regT0 holds callee.
store64(regT0, Address(stackPointerRegister, CallFrameSlot::callee * static_cast<int>(sizeof(Register)) - sizeof(CallerFrameAndPC)));
if (opcodeID == op_call_eval) {
compileCallEval(instruction);
return;
}
DataLabelPtr addressOfLinkedFunctionCheck;
Jump slowCase = branchPtrWithPatch(NotEqual, regT0, addressOfLinkedFunctionCheck, TrustedImmPtr(0));
addSlowCase(slowCase);
ASSERT(m_callCompilationInfo.size() == callLinkInfoIndex);
info->setUpCall(CallLinkInfo::callTypeFor(opcodeID), CodeOrigin(m_bytecodeOffset), regT0);
m_callCompilationInfo.append(CallCompilationInfo());
m_callCompilationInfo[callLinkInfoIndex].hotPathBegin = addressOfLinkedFunctionCheck;
m_callCompilationInfo[callLinkInfoIndex].callLinkInfo = info;
if (opcodeID == op_tail_call) {
CallFrameShuffleData shuffleData;
shuffleData.tagTypeNumber = GPRInfo::tagTypeNumberRegister;
shuffleData.numLocals =
instruction[4].u.operand - sizeof(CallerFrameAndPC) / sizeof(Register);
shuffleData.args.resize(instruction[3].u.operand);
for (int i = 0; i < instruction[3].u.operand; ++i) {
shuffleData.args[i] =
ValueRecovery::displacedInJSStack(
virtualRegisterForArgument(i) - instruction[4].u.operand,
DataFormatJS);
}
shuffleData.callee =
ValueRecovery::inGPR(regT0, DataFormatJS);
shuffleData.setupCalleeSaveRegisters(m_codeBlock);
info->setFrameShuffleData(shuffleData);
CallFrameShuffler(*this, shuffleData).prepareForTailCall();
m_callCompilationInfo[callLinkInfoIndex].hotPathOther = emitNakedTailCall();
return;
}
if (opcodeID == op_tail_call_varargs || opcodeID == op_tail_call_forward_arguments) {
emitRestoreCalleeSaves();
prepareForTailCallSlow();
m_callCompilationInfo[callLinkInfoIndex].hotPathOther = emitNakedTailCall();
return;
}
m_callCompilationInfo[callLinkInfoIndex].hotPathOther = emitNakedCall();
addPtr(TrustedImm32(stackPointerOffsetFor(m_codeBlock) * sizeof(Register)), callFrameRegister, stackPointerRegister);
checkStackPointerAlignment();
sampleCodeBlock(m_codeBlock);
emitPutCallResult(instruction);
}
TrackFrames buildTrackFramesFromMidi(const std::string& filename)
{
TrackFrames trackFrames;
FILE* pFic;
fopen_s(&pFic, filename.c_str(), "rb");
// Load header
struct HeaderChunk
{
char HThd[4];
uint32_t size;
uint16_t format;
uint16_t nbTrack;
uint16_t division;
} headerChunk;
loadStr(headerChunk.HThd, 4, pFic);
headerChunk.size = load32(pFic);
headerChunk.format = load16(pFic);
headerChunk.nbTrack = load16(pFic);
headerChunk.division = load16(pFic);
uint32_t maxTime = 0;
std::set<int> channelSet;
//int midiToChannels[] = { // mario
// 5, 0, 1, 2,
// 5, 5, 5, 5,
// 5, 5, 5, 5,
// 5, 5, 5, 5,
//};
//int midiToChannels[] = { // sm2
// 0, 2, 1, 5,
// 5, 5, 5, 5,
// 5, 3, 5, 5,
// 5, 5, 5, 5,
//};
int midiToChannels[] = {
0, 1, 2, 5,
5, 5, 5, 5,
5, 5, 5, 5,
5, 5, 5, 5,
};
//int midiToChannels[] = { // zelda
// 0, 1, 2, 5,
// 5, 5, 5, 5,
// 5, 3, 5, 5,
// 5, 5, 5, 5,
//};
// Load tracks
for (auto i = 0u; i < headerChunk.nbTrack; ++i)
{
char MTrk[4];
loadStr(MTrk, 4, pFic);
auto len = load32(pFic);
auto start = ftell(pFic);
auto cur = start;
uint32_t curTime = 0;
while (cur - start < static_cast<decltype(cur)>(len))
{
auto v_time = loadLength(pFic);
auto event = load8(pFic);
curTime += v_time;// / (headerChunk.division / 24);
maxTime = std::max(maxTime, curTime);
trackFrames.resize(maxTime * 4 + 16);
if (event == 0xFF)
{
// Meta event
event = load8(pFic);
if (event == 0x58)
{
// Time Signature
event = load8(pFic);
assert(event == 0x04);
load32(pFic); // Ignore who f*****g cares
}
else if (event == 0x7F)
{
// Sequencer Specific Meta-Event
auto dataLen = load8(pFic); // Len, ignore
while (dataLen)
{
load8(pFic);
--dataLen;
}
}
else if (event == 0x51)
{
// Set Tempo
event = load8(pFic);
assert(event == 0x03);
load8(pFic);
load8(pFic);
load8(pFic);
}
else if (event == 0x2F)
{
// End of Track
event = load8(pFic);
assert(event == 0x00);
}
else if (event == 0x03)
{
// Sequence/Track Name
auto textLen = load8(pFic);
char text[257] = {0};
loadStr(text, textLen, pFic);
printf("Track found: %i, %s\n", i, text);
}
else if (event == 0x59)
{
// Key Signature
event = load8(pFic);
assert(event == 0x02);
auto sf = load8(pFic);
auto mi = load8(pFic);
}
else if (event == 0x21)
{
// End of Track
event = load8(pFic);
assert(event == 0x01);
load8(pFic);
}
else if (event == 0x20)
{
// MIDI Channel Prefix
event = load8(pFic);
assert(event == 0x01);
auto channel = load8(pFic);
}
else
{
assert(false);
}
}
else if (event & 0x80)
{
auto channelNo = event & 0x0F;
channelSet.insert(channelNo);
if (channelNo < sizeof(midiToChannels) / sizeof(int)) channelNo = midiToChannels[channelNo];
event = event & 0xF0;
if (event == 0xB0)
{
// Control Change
auto controllerNumber = load8(pFic) & 0x7F;
auto controllerValue = load8(pFic) & 0x7F;
}
else if (event == 0xC0)
{
// Program Change
auto programNumber = load8(pFic) & 0x7F;
}
else if (event == 0x90)
{
// Note On event
auto note = load8(pFic) & 0x7F;
auto vol = load8(pFic) & 0x7F;
if (channelNo < 3)
{
auto& trackFrame = trackFrames[curTime * 4 + channelNo];
trackFrame = NOTE(noteTable[note], channelNo, (vol * 15) / 127);
}
else if (channelNo == 3)
{
auto& trackFrame = trackFrames[curTime * 4 + channelNo];
trackFrame = NOTE(noteTable[note + 12], channelNo, (vol * 15) / 127);
}
}
else if (event == 0x80)
{
// Note Off event
auto note = load8(pFic) & 0x7F;
auto vol = load8(pFic) & 0x7F;
if (channelNo == 2)
{
//auto& trackFrame = trackFrames[curTime * 4 + channelNo];
//trackFrame = NOTE(noteTable[note + 12], channelNo, 0);
}
if (channelNo == 3)
{
auto& trackFrame = trackFrames[curTime * 4 + channelNo];
trackFrame = NOTE(noteTable[note], channelNo, 0);
}
}
else if (event == 0xE0)
{
// Pitch Wheel Change
auto l = load8(pFic) & 0x7F;
auto m = load8(pFic) & 0x7F;
}
else
{
assert(false);
}
}
else if (event <= 0x7F)
{
load8(pFic);
}
else
{
assert(false);
}
cur = ftell(pFic);
}
}
fclose(pFic);
for (auto& channelNo : channelSet)
{
printf("%i\n", channelNo);
}
return trackFrames;
}
void JITCompiler::compileFunction(JITCode& entry, MacroAssemblerCodePtr& entryWithArityCheck)
{
compileEntry();
// === Function header code generation ===
// This is the main entry point, without performing an arity check.
// If we needed to perform an arity check we will already have moved the return address,
// so enter after this.
Label fromArityCheck(this);
// Plant a check that sufficient space is available in the RegisterFile.
// FIXME: https://bugs.webkit.org/show_bug.cgi?id=56291
addPtr(Imm32(m_codeBlock->m_numCalleeRegisters * sizeof(Register)), GPRInfo::callFrameRegister, GPRInfo::regT1);
Jump registerFileCheck = branchPtr(Below, AbsoluteAddress(m_globalData->interpreter->registerFile().addressOfEnd()), GPRInfo::regT1);
// Return here after register file check.
Label fromRegisterFileCheck = label();
// === Function body code generation ===
SpeculativeJIT speculative(*this);
compileBody(speculative);
// === Function footer code generation ===
//
// Generate code to perform the slow register file check (if the fast one in
// the function header fails), and generate the entry point with arity check.
//
// Generate the register file check; if the fast check in the function head fails,
// we need to call out to a helper function to check whether more space is available.
// FIXME: change this from a cti call to a DFG style operation (normal C calling conventions).
registerFileCheck.link(this);
move(stackPointerRegister, GPRInfo::argumentGPR0);
poke(GPRInfo::callFrameRegister, OBJECT_OFFSETOF(struct JITStackFrame, callFrame) / sizeof(void*));
Call callRegisterFileCheck = call();
jump(fromRegisterFileCheck);
// The fast entry point into a function does not check the correct number of arguments
// have been passed to the call (we only use the fast entry point where we can statically
// determine the correct number of arguments have been passed, or have already checked).
// In cases where an arity check is necessary, we enter here.
// FIXME: change this from a cti call to a DFG style operation (normal C calling conventions).
Label arityCheck = label();
compileEntry();
load32(Address(GPRInfo::callFrameRegister, RegisterFile::ArgumentCount * static_cast<int>(sizeof(Register))), GPRInfo::regT1);
branch32(Equal, GPRInfo::regT1, Imm32(m_codeBlock->m_numParameters)).linkTo(fromArityCheck, this);
move(stackPointerRegister, GPRInfo::argumentGPR0);
poke(GPRInfo::callFrameRegister, OBJECT_OFFSETOF(struct JITStackFrame, callFrame) / sizeof(void*));
Call callArityCheck = call();
move(GPRInfo::regT0, GPRInfo::callFrameRegister);
jump(fromArityCheck);
// === Link ===
LinkBuffer linkBuffer(*m_globalData, this);
link(linkBuffer);
speculative.linkOSREntries(linkBuffer);
// FIXME: switch the register file check & arity check over to DFGOpertaion style calls, not JIT stubs.
linkBuffer.link(callRegisterFileCheck, cti_register_file_check);
linkBuffer.link(callArityCheck, m_codeBlock->m_isConstructor ? cti_op_construct_arityCheck : cti_op_call_arityCheck);
entryWithArityCheck = linkBuffer.locationOf(arityCheck);
entry = JITCode(linkBuffer.finalizeCode(), JITCode::DFGJIT);
}
