| Summary: | HTMLTreeBuilder::ExternalCharacterTokenBuffer should not need to copy bytes in the CompactHTMLToken case | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Eric Seidel (no email) <eric> |
| Component: | Tools / Tests | Assignee: | Nobody <webkit-unassigned> |
| Status: | RESOLVED CONFIGURATION CHANGED | ||
| Severity: | Normal | CC: | abarth, annevk, tonyg |
| Priority: | P2 | ||
| Version: | 528+ (Nightly build) | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
| Bug Depends on: | |||
| Bug Blocks: | 111645 | ||
It looks like we're doing this today. |
HTMLTreeBuilder::ExternalCharacterTokenBuffer should not need to copy bytes in the CompactHTMLToken case CompactHTMLToken has already allocated a StringImpl of the right size in the common case. We need to plumb this information through AtomicHTMLToken so that ExternalCharacterTokenBuffer can avoid copying the bytes and instead just ref the StringImpl. :) e.g. m_tree.insertTextNode(buffer.takeRemaining()); profiling Parser/html-parser-threaded.html with bug 107236 applied: Running Time Self Symbol Name 84.2ms 4.8% 84.2 WTF::fastMalloc(unsigned long) 19.6ms 1.1% 0.0 WTF::StringImpl::createUninitialized(unsigned int, unsigned char*&) 17.7ms 1.0% 0.0 WTF::String::make8BitFrom16BitSource(unsigned short const*, unsigned long) 8.3ms 0.4% 0.0 WebCore::HTMLTreeBuilder::processCharacterBufferForInBody(WebCore::HTMLTreeBuilder::ExternalCharacterTokenBuffer&) The .4% spent in HTMLTreeBuilder::processCharacterBufferForInBody are likely all unnecessary.