Implement Unicode RegExp support in the YARR JIT
https://bugs.webkit.org/show_bug.cgi?id=174646
Reviewed by Filip Pizlo.
Source/JavaScriptCore:
This support is only implemented for 64 bit platforms. It wouldn't be too hard to add support
for 32 bit platforms with a reasonable number of spare registers. This code slightly refactors
register usage to reduce the number of callee save registers used for non-Unicode expressions.
For Unicode expressions, there are several more registers used to store constants values for
processing surrogate pairs as well as discerning whether a character belongs to the Basic
Multilingual Plane (BMP) or one of the Supplemental Planes.
This implements JIT support for Unicode expressions very similar to how the interpreter works.
Just like in the interpreter, backtracking code uses more space on the stack to save positions.
Moved the BackTrackInfo* structs to YarrPattern as separate functions. Added xxxIndex()
functions to each of these to simplify how the JIT code reads and writes the structure fields.
Given that reading surrogate pairs and transforming them into a single code point takes a
little processing, the code that implements reading a Unicode character is implemented as a
leaf function added to the end of the JIT'ed code. The calling convention for
"tryReadUnicodeCharacterHelper()" is non-standard given that the rest of the code assumes
that argument values stay in argument registers for most of the generated code.
That helper takes the starting character address in one register, regUnicodeInputAndTrail,
and uses another dedicated temporary register, regUnicodeTemp. The result is typically
returned in regT0. If another return register is requested, we'll create an inline copy of
that function.
Added a new flag to CharacterClass to signify if a class has non-BMP characters. This flag
is used in optimizeAlternative() where we swap the order of a fixed character class term with
a fixed character term that immediately follows it. Since the non-BMP character class may
increment "index" when matching, that must be done first before trying to match a fixed
character term later in the string.
Given the usefulness of the LEA instruction on X86 to create a single pointer value from a
base with index and offset, which the YARR JIT uses heavily, I added a new macroAssembler
function, getEffectiveAddress64(), with an ARM64 implementation. It just calls x86Lea64()
on X86-64. Also added an ImplicitAddress version of load16Unaligned().
(JSC::MacroAssemblerARM64::load16Unaligned):
(JSC::MacroAssemblerARM64::getEffectiveAddress64):
- assembler/MacroAssemblerX86Common.h:
(JSC::MacroAssemblerX86Common::load16Unaligned):
(JSC::MacroAssemblerX86Common::load16):
- assembler/MacroAssemblerX86_64.h:
(JSC::MacroAssemblerX86_64::getEffectiveAddress64):
- create_regex_tables:
- runtime/RegExp.cpp:
(JSC::RegExp::compile):
- yarr/YarrInterpreter.cpp:
- yarr/YarrJIT.cpp:
(JSC::Yarr::YarrGenerator::optimizeAlternative):
(JSC::Yarr::YarrGenerator::matchCharacterClass):
(JSC::Yarr::YarrGenerator::tryReadUnicodeCharImpl):
(JSC::Yarr::YarrGenerator::tryReadUnicodeChar):
(JSC::Yarr::YarrGenerator::readCharacter):
(JSC::Yarr::YarrGenerator::jumpIfCharNotEquals):
(JSC::Yarr::YarrGenerator::matchAssertionWordchar):
(JSC::Yarr::YarrGenerator::generateAssertionWordBoundary):
(JSC::Yarr::YarrGenerator::generatePatternCharacterOnce):
(JSC::Yarr::YarrGenerator::generatePatternCharacterFixed):
(JSC::Yarr::YarrGenerator::generatePatternCharacterGreedy):
(JSC::Yarr::YarrGenerator::backtrackPatternCharacterGreedy):
(JSC::Yarr::YarrGenerator::generatePatternCharacterNonGreedy):
(JSC::Yarr::YarrGenerator::backtrackPatternCharacterNonGreedy):
(JSC::Yarr::YarrGenerator::generateCharacterClassOnce):
(JSC::Yarr::YarrGenerator::backtrackCharacterClassOnce):
(JSC::Yarr::YarrGenerator::generateCharacterClassFixed):
(JSC::Yarr::YarrGenerator::generateCharacterClassGreedy):
(JSC::Yarr::YarrGenerator::backtrackCharacterClassGreedy):
(JSC::Yarr::YarrGenerator::generateCharacterClassNonGreedy):
(JSC::Yarr::YarrGenerator::backtrackCharacterClassNonGreedy):
(JSC::Yarr::YarrGenerator::generate):
(JSC::Yarr::YarrGenerator::backtrack):
(JSC::Yarr::YarrGenerator::generateTryReadUnicodeCharacterHelper):
(JSC::Yarr::YarrGenerator::generateEnter):
(JSC::Yarr::YarrGenerator::generateReturn):
(JSC::Yarr::YarrGenerator::YarrGenerator):
(JSC::Yarr::YarrGenerator::compile):
- yarr/YarrJIT.h:
- yarr/YarrPattern.cpp:
(JSC::Yarr::CharacterClassConstructor::CharacterClassConstructor):
(JSC::Yarr::CharacterClassConstructor::reset):
(JSC::Yarr::CharacterClassConstructor::charClass):
(JSC::Yarr::CharacterClassConstructor::addSorted):
(JSC::Yarr::CharacterClassConstructor::addSortedRange):
(JSC::Yarr::CharacterClassConstructor::hasNonBMPCharacters):
(JSC::Yarr::YarrPatternConstructor::setupAlternativeOffsets):
(JSC::Yarr::CharacterClass::CharacterClass):
(JSC::Yarr::BackTrackInfoPatternCharacter::beginIndex):
(JSC::Yarr::BackTrackInfoPatternCharacter::matchAmountIndex):
(JSC::Yarr::BackTrackInfoCharacterClass::beginIndex):
(JSC::Yarr::BackTrackInfoCharacterClass::matchAmountIndex):
(JSC::Yarr::BackTrackInfoBackReference::beginIndex):
(JSC::Yarr::BackTrackInfoBackReference::matchAmountIndex):
(JSC::Yarr::BackTrackInfoAlternative::offsetIndex):
(JSC::Yarr::BackTrackInfoParentheticalAssertion::beginIndex):
(JSC::Yarr::BackTrackInfoParenthesesOnce::beginIndex):
(JSC::Yarr::BackTrackInfoParenthesesTerminal::beginIndex):
LayoutTests:
Updated tests.
- js/regexp-unicode-expected.txt:
- js/script-tests/regexp-unicode.js:
