method_check should repatch itself if it finds that the new structure(s)
are the result of transitions from the old structure(s)
https://bugs.webkit.org/show_bug.cgi?id=68294
Reviewed by Gavin Barraclough.
Previously a patched method_check would slow-path to get_by_id. Now it
slow-paths to method_check_update, which attempts to correct the
method_check due to structure transitions before bailing to get_by_id.
This is a 1-2% speed-up on some benchmarks and is not a slow-down
anywhere, leading to a 0.6% speed-up on the Kraken geomean.
* jit/JITPropertyAccess.cpp:
(JSC::JIT::patchMethodCallProto):
* jit/JITStubs.cpp:
(JSC::DEFINE_STUB_FUNCTION):
* jit/JITStubs.h:
* runtime/Structure.h:
(JSC::Structure::transitivelyTransitionedFrom):
git-svn-id: http://svn.webkit.org/repository/webkit/trunk@95388 268f45cc-cd09-0410-ab3c-d52691b4dbfc
diff --git a/Source/JavaScriptCore/runtime/Structure.h b/Source/JavaScriptCore/runtime/Structure.h
index a86a1df..8607c45 100644
--- a/Source/JavaScriptCore/runtime/Structure.h
+++ b/Source/JavaScriptCore/runtime/Structure.h
@@ -133,6 +133,7 @@
void visitChildren(SlotVisitor&);
Structure* previousID() const { ASSERT(structure()->classInfo() == &s_info); return m_previous.get(); }
+ bool transitivelyTransitionedFrom(Structure* structureToFind);
void growPropertyStorageCapacity();
unsigned propertyStorageCapacity() const { ASSERT(structure()->classInfo() == &s_info); return m_propertyStorageCapacity; }
@@ -352,6 +353,15 @@
return Hash::Key(structure->m_nameInPrevious.get(), +structure->m_attributesInPrevious);
}
+ inline bool Structure::transitivelyTransitionedFrom(Structure* structureToFind)
+ {
+ for (Structure* current = this; current; current = current->previousID()) {
+ if (current == structureToFind)
+ return true;
+ }
+ return false;
+ }
+
} // namespace JSC
#endif // Structure_h
