Merged the global function cache into the source code cache
https://bugs.webkit.org/show_bug.cgi?id=108660
Reviewed by Sam Weinig.
This has a few benefits:
(*) Saves a few kB by removing a second cache data structure.
(*) Reduces the worst case memory usage of the cache by 1.75X. (Heavy
use of 'new Function' and other techniques could cause us to fill
both root caches, and they didn't trade off against each other.)
(*) Paves the way for future improvements based on a non-trivial
cache key (for example, shrinkable pointer to the key string, and
more precise cache size accounting).
Also cleaned up the cache implementation and simplified it a bit.
(HandleBase):
(Strong): Build!
(JSC):
(JSC::CodeCache::getCodeBlock):
(JSC::CodeCache::generateFunctionCodeBlock):
(JSC::CodeCache::getFunctionExecutableFromGlobalCode):
(JSC::CodeCache::usedFunctionCode): Updated for three interface changes:
(*) SourceCodeKey is a class, not a pair.
(*) Table values are abstract pointers, since they can be executables
or code blocks. (In a future patch, I'd like to change this so we
always store only code blocks. But that's too much for one patch.)
(*) The cache function is named "set" because it always overwrites
unconditionally.
(CacheMap):
(JSC::CacheMap::find):
(JSC::CacheMap::set):
(JSC::CacheMap::clear): Added support for specifying hash traits, so we
can use a SourceCodeKey.
Removed side table and random number generator to save space and reduce
complexity. Hash tables are already random, so we don't need another source
of randomness.
(SourceCodeKey):
(JSC::SourceCodeKey::SourceCodeKey):
(JSC::SourceCodeKey::isHashTableDeletedValue):
(JSC::SourceCodeKey::hash):
(JSC::SourceCodeKey::isNull):
(JSC::SourceCodeKey::operator==):
(JSC::SourceCodeKeyHash::hash):
(JSC::SourceCodeKeyHash::equal):
(SourceCodeKeyHash):
(SourceCodeKeyHashTraits):
(JSC::SourceCodeKeyHashTraits::isEmptyValue): A SourceCodeKey is just a
fancy triplet: source code string; function name (or null, for non-functions);
and flags. Flags and function name distinguish between functions and programs
with identical code, so they can live in the same cache.
I chose to use the source code string as the primary hashing reference
because it's likely to be unique. We can use profiling to choose another
technique in future, if collisions between functions and programs prove
to be hot. I suspect they won't.
(JSC::CodeCache::clear):
(CodeCache): Removed the second cache.
(HandleBase):
(Strong):
(JSC):
(JSC::CodeCache::getCodeBlock):
(JSC::CodeCache::generateFunctionCodeBlock):
(JSC::CodeCache::getFunctionExecutableFromGlobalCode):
(JSC::CodeCache::usedFunctionCode):
(JSC):
(CacheMap):
(JSC::CacheMap::find):
(JSC::CacheMap::set):
(JSC::CacheMap::clear):
(SourceCodeKey):
(JSC::SourceCodeKey::SourceCodeKey):
(JSC::SourceCodeKey::isHashTableDeletedValue):
(JSC::SourceCodeKey::hash):
(JSC::SourceCodeKey::isNull):
(JSC::SourceCodeKey::operator==):
(JSC::SourceCodeKeyHash::hash):
(JSC::SourceCodeKeyHash::equal):
(SourceCodeKeyHash):
(SourceCodeKeyHashTraits):
(JSC::SourceCodeKeyHashTraits::isEmptyValue):
(JSC::CodeCache::clear):
(CodeCache):
