Changeset 60273 in webkit for trunk/JavaScriptCore/yarr/RegexPattern.h

Timestamp:

May 26, 2010, 11:25:56 PM (15 years ago)

Author:

[email protected]

Message:

Bug 39795 - Add support for YARR JIT generation of greedy quantified parens at the end of the main disjunction.
(relanding r60267)

Reviewed by Oliver Hunt.

If the last item in a main disjunction is a quantified set of parentheses,
this is easier to code generate for than the general case for quantified
parentheses. This is because we never need to backtrack into the parentheses

• the first match will be the final and accepted match.

This patch also somewhat reverts a recent change to when fallback to PCRE
occurs. At the minute the compiler is tracking on patterns which will
require JIT fallback. This is handy from a performance perspective (it saves
the failed attempt at JIT compilation), but it means introducing knowledge
of the JITs capabilities into the other layers of the regex compilers. For
the specific feature of back-references, add a flag tracking their presence
on the pattern, and make these expressions fallback without attempting to
JIT. For parentheses, return to detecting which cases are have or have not
been handled during JIT compilation.

18% progression on tagcloud, ~1.5% overall on sunspidey.

• yarr/RegexCompiler.cpp:

(JSC::Yarr::RegexPatternConstructor::atomBackReference):
(JSC::Yarr::RegexPatternConstructor::quantifyAtom):

• yarr/RegexJIT.cpp:

(JSC::Yarr::RegexGenerator::TermGenerationState::isLastTerm):
(JSC::Yarr::RegexGenerator::TermGenerationState::isMainDisjunction):
(JSC::Yarr::RegexGenerator::generateParenthesesGreedyNoBacktrack):
(JSC::Yarr::RegexGenerator::generateTerm):
(JSC::Yarr::RegexGenerator::RegexGenerator):
(JSC::Yarr::RegexGenerator::shouldFallBack):
(JSC::Yarr::jitCompileRegex):

• yarr/RegexPattern.h:

(JSC::Yarr::RegexPattern::RegexPattern):
(JSC::Yarr::RegexPattern::reset):

File:

• trunk/JavaScriptCore/yarr/RegexPattern.h (modified) (3 diffs)

trunk/JavaScriptCore/yarr/RegexPattern.h

@@ -272 +272 @@
         , m_numSubpatterns(0)
         , m_maxBackReference(0)
-        , m_shouldFallBack(false)
+        , m_containsBackreferences(false)
         , newlineCached(0)
         , digitsCached(0)
@@ -294 +294 @@
         m_maxBackReference = 0;
 
-        m_shouldFallBack = false;
+        m_containsBackreferences = false;
 
         newlineCached = 0;
@@ -362 +362 @@
     unsigned m_numSubpatterns;
     unsigned m_maxBackReference;
-    bool m_shouldFallBack;
+    bool m_containsBackreferences;
     PatternDisjunction* m_body;
     Vector<PatternDisjunction*, 4> m_disjunctions;