/*
- * Copyright (C) 2009 Apple Inc. All rights reserved.
+ * Copyright (C) 2009, 2013 Apple Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
#include <wtf/ASCIICType.h>
#include "LinkBuffer.h"
+#include "Options.h"
#include "Yarr.h"
#include "YarrCanonicalizeUCS2.h"
template<YarrJITCompileMode compileMode>
class YarrGenerator : private MacroAssembler {
- friend void jitCompile(JSGlobalData*, YarrCodeBlock& jitObject, const UString& pattern, unsigned& numSubpatterns, const char*& error, bool ignoreCase, bool multiline);
+ friend void jitCompile(VM*, YarrCodeBlock& jitObject, const String& pattern, unsigned& numSubpatterns, const char*& error, bool ignoreCase, bool multiline);
#if CPU(ARM)
static const RegisterID input = ARMRegisters::r0;
static const RegisterID index = ARMRegisters::r1;
static const RegisterID length = ARMRegisters::r2;
- static const RegisterID output = ARMRegisters::r4;
+ static const RegisterID output = ARMRegisters::r3;
- static const RegisterID regT0 = ARMRegisters::r5;
- static const RegisterID regT1 = ARMRegisters::r6;
+ static const RegisterID regT0 = ARMRegisters::r4;
+ static const RegisterID regT1 = ARMRegisters::r5;
static const RegisterID returnRegister = ARMRegisters::r0;
static const RegisterID returnRegister2 = ARMRegisters::r1;
+#elif CPU(ARM64)
+ static const RegisterID input = ARM64Registers::x0;
+ static const RegisterID index = ARM64Registers::x1;
+ static const RegisterID length = ARM64Registers::x2;
+ static const RegisterID output = ARM64Registers::x3;
+
+ static const RegisterID regT0 = ARM64Registers::x4;
+ static const RegisterID regT1 = ARM64Registers::x5;
+
+ static const RegisterID returnRegister = ARM64Registers::x0;
+ static const RegisterID returnRegister2 = ARM64Registers::x1;
#elif CPU(MIPS)
static const RegisterID input = MIPSRegisters::a0;
static const RegisterID index = MIPSRegisters::a1;
static const RegisterID returnRegister = X86Registers::eax;
static const RegisterID returnRegister2 = X86Registers::edx;
#elif CPU(X86_64)
+#if !OS(WINDOWS)
static const RegisterID input = X86Registers::edi;
static const RegisterID index = X86Registers::esi;
static const RegisterID length = X86Registers::edx;
static const RegisterID output = X86Registers::ecx;
+#else
+ // If the return value doesn't fit in 64bits, its destination is pointed by rcx and the parameters are shifted.
+ // http://msdn.microsoft.com/en-us/library/7572ztz4.aspx
+ COMPILE_ASSERT(sizeof(MatchResult) > sizeof(void*), MatchResult_does_not_fit_in_64bits);
+ static const RegisterID input = X86Registers::edx;
+ static const RegisterID index = X86Registers::r8;
+ static const RegisterID length = X86Registers::r9;
+ static const RegisterID output = X86Registers::r10;
+#endif
static const RegisterID regT0 = X86Registers::eax;
static const RegisterID regT1 = X86Registers::ebx;
void matchCharacterClass(RegisterID character, JumpList& matchDest, const CharacterClass* charClass)
{
if (charClass->m_table) {
- ExtendedAddress tableEntry(character, reinterpret_cast<intptr_t>(charClass->m_table->m_table));
- matchDest.append(branchTest8(charClass->m_table->m_inverted ? Zero : NonZero, tableEntry));
+ ExtendedAddress tableEntry(character, reinterpret_cast<intptr_t>(charClass->m_table));
+ matchDest.append(branchTest8(charClass->m_tableInverted ? Zero : NonZero, tableEntry));
return;
}
Jump unicodeFail;
jump(Address(stackPointerRegister, frameLocation * sizeof(void*)));
}
+ unsigned alignCallFrameSizeInBytes(unsigned callFrameSize)
+ {
+ callFrameSize *= sizeof(void*);
+ if (callFrameSize / sizeof(void*) != m_pattern.m_body->m_callFrameSize)
+ CRASH();
+ callFrameSize = (callFrameSize + 0x3f) & ~0x3f;
+ if (!callFrameSize)
+ CRASH();
+ return callFrameSize;
+ }
void initCallFrame()
{
unsigned callFrameSize = m_pattern.m_body->m_callFrameSize;
if (callFrameSize)
- subPtr(Imm32(callFrameSize * sizeof(void*)), stackPointerRegister);
+ subPtr(Imm32(alignCallFrameSizeInBytes(callFrameSize)), stackPointerRegister);
}
void removeCallFrame()
{
unsigned callFrameSize = m_pattern.m_body->m_callFrameSize;
if (callFrameSize)
- addPtr(Imm32(callFrameSize * sizeof(void*)), stackPointerRegister);
+ addPtr(Imm32(alignCallFrameSizeInBytes(callFrameSize)), stackPointerRegister);
}
// Used to record subpatters, should only be called if compileMode is IncludeSubpatterns.
const RegisterID character = regT0;
int maxCharactersAtOnce = m_charSize == Char8 ? 4 : 2;
unsigned ignoreCaseMask = 0;
+#if CPU(BIG_ENDIAN)
+ int allCharacters = ch << (m_charSize == Char8 ? 24 : 16);
+#else
int allCharacters = ch;
+#endif
int numberCharacters;
int startTermPosition = term->inputPosition;
ASSERT(!m_pattern.m_ignoreCase || isASCIIAlpha(ch) || isCanonicallyUnique(ch));
if (m_pattern.m_ignoreCase && isASCIIAlpha(ch))
+#if CPU(BIG_ENDIAN)
+ ignoreCaseMask |= 32 << (m_charSize == Char8 ? 24 : 16);
+#else
ignoreCaseMask |= 32;
+#endif
for (numberCharacters = 1; numberCharacters < maxCharactersAtOnce && nextOp->m_op == OpTerm; ++numberCharacters, nextOp = &m_ops[opIndex + numberCharacters]) {
PatternTerm* nextTerm = nextOp->m_term;
nextOp->m_isDeadCode = true;
+#if CPU(BIG_ENDIAN)
+ int shiftAmount = (m_charSize == Char8 ? 24 : 16) - ((m_charSize == Char8 ? 8 : 16) * numberCharacters);
+#else
int shiftAmount = (m_charSize == Char8 ? 8 : 16) * numberCharacters;
+#endif
UChar currentCharacter = nextTerm->patternCharacter;
case PatternTerm::TypeParenthesesSubpattern:
case PatternTerm::TypeParentheticalAssertion:
- ASSERT_NOT_REACHED();
+ RELEASE_ASSERT_NOT_REACHED();
case PatternTerm::TypeBackReference:
m_shouldFallBack = true;
break;
case PatternTerm::TypeParenthesesSubpattern:
case PatternTerm::TypeParentheticalAssertion:
- ASSERT_NOT_REACHED();
+ RELEASE_ASSERT_NOT_REACHED();
case PatternTerm::TypeDotStarEnclosure:
backtrackDotStarEnclosure(opIndex);
const RegisterID indexTemporary = regT0;
ASSERT(term->quantityCount == 1);
-#ifndef NDEBUG
// Runtime ASSERT to make sure that the nested alternative handled the
// "no input consumed" check.
- if (term->quantityType != QuantifierFixedCount && !term->parentheses.disjunction->m_minimumSize) {
+ if (!ASSERT_DISABLED && term->quantityType != QuantifierFixedCount && !term->parentheses.disjunction->m_minimumSize) {
Jump pastBreakpoint;
pastBreakpoint = branch32(NotEqual, index, Address(stackPointerRegister, term->frameLocation * sizeof(void*)));
- breakpoint();
+ abortWithReason(YARRNoInputConsumed);
pastBreakpoint.link(this);
}
-#endif
// If the parenthese are capturing, store the ending index value to the
// captures array, offsetting as necessary.
}
case OpParenthesesSubpatternTerminalEnd: {
YarrOp& beginOp = m_ops[op.m_previousOp];
-#ifndef NDEBUG
- PatternTerm* term = op.m_term;
-
- // Runtime ASSERT to make sure that the nested alternative handled the
- // "no input consumed" check.
- Jump pastBreakpoint;
- pastBreakpoint = branch32(NotEqual, index, Address(stackPointerRegister, term->frameLocation * sizeof(void*)));
- breakpoint();
- pastBreakpoint.link(this);
-#endif
+ if (!ASSERT_DISABLED) {
+ PatternTerm* term = op.m_term;
+
+ // Runtime ASSERT to make sure that the nested alternative handled the
+ // "no input consumed" check.
+ Jump pastBreakpoint;
+ pastBreakpoint = branch32(NotEqual, index, Address(stackPointerRegister, term->frameLocation * sizeof(void*)));
+ abortWithReason(YARRNoInputConsumed);
+ pastBreakpoint.link(this);
+ }
// We know that the match is non-zero, we can accept it and
// loop back up to the head of the subpattern.
m_ops.append(alternativeBeginOpCode);
m_ops.last().m_previousOp = notFound;
m_ops.last().m_term = term;
- Vector<PatternAlternative*>& alternatives = term->parentheses.disjunction->m_alternatives;
+ Vector<OwnPtr<PatternAlternative>>& alternatives = term->parentheses.disjunction->m_alternatives;
for (unsigned i = 0; i < alternatives.size(); ++i) {
size_t lastOpIndex = m_ops.size() - 1;
- PatternAlternative* nestedAlternative = alternatives[i];
+ PatternAlternative* nestedAlternative = alternatives[i].get();
opCompileAlternative(nestedAlternative);
size_t thisOpIndex = m_ops.size();
m_ops.append(OpSimpleNestedAlternativeBegin);
m_ops.last().m_previousOp = notFound;
m_ops.last().m_term = term;
- Vector<PatternAlternative*>& alternatives = term->parentheses.disjunction->m_alternatives;
+ Vector<OwnPtr<PatternAlternative>>& alternatives = term->parentheses.disjunction->m_alternatives;
for (unsigned i = 0; i < alternatives.size(); ++i) {
size_t lastOpIndex = m_ops.size() - 1;
- PatternAlternative* nestedAlternative = alternatives[i];
+ PatternAlternative* nestedAlternative = alternatives[i].get();
opCompileAlternative(nestedAlternative);
size_t thisOpIndex = m_ops.size();
// to return the failing result.
void opCompileBody(PatternDisjunction* disjunction)
{
- Vector<PatternAlternative*>& alternatives = disjunction->m_alternatives;
+ Vector<OwnPtr<PatternAlternative>>& alternatives = disjunction->m_alternatives;
size_t currentAlternativeIndex = 0;
// Emit the 'once through' alternatives.
do {
size_t lastOpIndex = m_ops.size() - 1;
- PatternAlternative* alternative = alternatives[currentAlternativeIndex];
+ PatternAlternative* alternative = alternatives[currentAlternativeIndex].get();
opCompileAlternative(alternative);
size_t thisOpIndex = m_ops.size();
m_ops.last().m_previousOp = notFound;
do {
size_t lastOpIndex = m_ops.size() - 1;
- PatternAlternative* alternative = alternatives[currentAlternativeIndex];
+ PatternAlternative* alternative = alternatives[currentAlternativeIndex].get();
ASSERT(!alternative->onceThrough());
opCompileAlternative(alternative);
push(X86Registers::ebp);
move(stackPointerRegister, X86Registers::ebp);
push(X86Registers::ebx);
+ // The ABI doesn't guarantee the upper bits are zero on unsigned arguments, so clear them ourselves.
+ zeroExtend32ToPtr(index, index);
+ zeroExtend32ToPtr(length, length);
+#if OS(WINDOWS)
+ if (compileMode == IncludeSubpatterns)
+ loadPtr(Address(X86Registers::ebp, 6 * sizeof(void*)), output);
+#endif
#elif CPU(X86)
push(X86Registers::ebp);
move(stackPointerRegister, X86Registers::ebp);
if (compileMode == IncludeSubpatterns)
loadPtr(Address(X86Registers::ebp, 2 * sizeof(void*)), output);
#endif
+#elif CPU(ARM64)
+ // The ABI doesn't guarantee the upper bits are zero on unsigned arguments, so clear them ourselves.
+ zeroExtend32ToPtr(index, index);
+ zeroExtend32ToPtr(length, length);
#elif CPU(ARM)
push(ARMRegisters::r4);
push(ARMRegisters::r5);
push(ARMRegisters::r6);
-#if CPU(ARM_TRADITIONAL)
- push(ARMRegisters::r8); // scratch register
-#endif
- if (compileMode == IncludeSubpatterns)
- move(ARMRegisters::r3, output);
#elif CPU(SH4)
push(SH4Registers::r11);
push(SH4Registers::r13);
void generateReturn()
{
#if CPU(X86_64)
+#if OS(WINDOWS)
+ // Store the return value in the allocated space pointed by rcx.
+ store64(returnRegister, Address(X86Registers::ecx));
+ store64(returnRegister2, Address(X86Registers::ecx, sizeof(void*)));
+ move(X86Registers::ecx, returnRegister);
+#endif
pop(X86Registers::ebx);
pop(X86Registers::ebp);
#elif CPU(X86)
pop(X86Registers::ebx);
pop(X86Registers::ebp);
#elif CPU(ARM)
-#if CPU(ARM_TRADITIONAL)
- pop(ARMRegisters::r8); // scratch register
-#endif
pop(ARMRegisters::r6);
pop(ARMRegisters::r5);
pop(ARMRegisters::r4);
{
}
- void compile(JSGlobalData* globalData, YarrCodeBlock& jitObject)
+ void compile(VM* vm, YarrCodeBlock& jitObject)
{
generateEnter();
backtrack();
// Link & finalize the code.
- LinkBuffer linkBuffer(*globalData, this, REGEXP_CODE_ID);
+ LinkBuffer linkBuffer(*vm, *this, REGEXP_CODE_ID);
m_backtrackingState.linkDataLabels(linkBuffer);
if (compileMode == MatchOnly) {
if (m_charSize == Char8)
- jitObject.set8BitCodeMatchOnly(linkBuffer.finalizeCode());
+ jitObject.set8BitCodeMatchOnly(FINALIZE_CODE(linkBuffer, ("Match-only 8-bit regular expression")));
else
- jitObject.set16BitCodeMatchOnly(linkBuffer.finalizeCode());
+ jitObject.set16BitCodeMatchOnly(FINALIZE_CODE(linkBuffer, ("Match-only 16-bit regular expression")));
} else {
if (m_charSize == Char8)
- jitObject.set8BitCode(linkBuffer.finalizeCode());
+ jitObject.set8BitCode(FINALIZE_CODE(linkBuffer, ("8-bit regular expression")));
else
- jitObject.set16BitCode(linkBuffer.finalizeCode());
+ jitObject.set16BitCode(FINALIZE_CODE(linkBuffer, ("16-bit regular expression")));
}
jitObject.setFallBack(m_shouldFallBack);
}
BacktrackingState m_backtrackingState;
};
-void jitCompile(YarrPattern& pattern, YarrCharSize charSize, JSGlobalData* globalData, YarrCodeBlock& jitObject, YarrJITCompileMode mode)
+void jitCompile(YarrPattern& pattern, YarrCharSize charSize, VM* vm, YarrCodeBlock& jitObject, YarrJITCompileMode mode)
{
if (mode == MatchOnly)
- YarrGenerator<MatchOnly>(pattern, charSize).compile(globalData, jitObject);
+ YarrGenerator<MatchOnly>(pattern, charSize).compile(vm, jitObject);
else
- YarrGenerator<IncludeSubpatterns>(pattern, charSize).compile(globalData, jitObject);
+ YarrGenerator<IncludeSubpatterns>(pattern, charSize).compile(vm, jitObject);
}
}}