[Webkit-unassigned] [Bug 258706] New: JS markdown parser performs 50x slower in JSC compared to V8, likely due to regex

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=258706

            Bug ID: 258706
           Summary: JS markdown parser performs 50x slower in JSC compared
                    to V8, likely due to regex
           Product: WebKit
           Version: WebKit Nightly Build
          Hardware: Unspecified
                OS: Unspecified
            Status: NEW
          Severity: Normal
          Priority: P2
         Component: JavaScriptCore
          Assignee: webkit-unassigned at lists.webkit.org
          Reporter: jarred at jarredsumner.com

Created attachment 466880

  --> https://bugs.webkit.org/attachment.cgi?id=466880&action=review

Reproduction that jsc shell can run

I've attached a repro that can be executed in jsc shell directly. It prints out the length of the output file after parsing it.

❯ hyperfine "node out.mjs" "jsc -m out.mjs"
Benchmark 1: node out.mjs
  Time (mean ± σ):     123.8 ms ±  21.1 ms    [User: 94.7 ms, System: 11.0 ms]
  Range (min … max):    77.9 ms … 153.5 ms    16 runs

Benchmark 2: jsc -m out.mjs
  Time (mean ± σ):      6.803 s ±  0.211 s    [User: 2.511 s, System: 0.025 s]
  Range (min … max):    6.507 s …  7.196 s    10 runs

Summary
  'node out.mjs' ran
   54.95 ± 9.53 times faster than 'jsc -m out.mjs'

If we profile it with Instruments, we can see that majority of time is spent in JSC::Yarr

2.16 s   97.6%  2.15 s           JSC::Yarr::Interpreter<char16_t>::matchDisjunction(JSC::Yarr::ByteDisjunction*, JSC::Yarr::Interpreter<char16_t>::DisjunctionContext*, bool)
3.00 ms    0.1% 3.00 ms          void std::__1::__introsort<std::__1::_ClassicAlgPolicy, JSC::ARM64Assembler::jumpsToLink()::'lambda'(auto&, auto&)&, JSC::ARM64Assembler::LinkRecord*>(JSC::ARM64Assembler::LinkRecord*, JSC::ARM64Assembler::LinkRecord*, auto, std::__1::iterator_traits<JSC::ARM64Assembler::LinkRecord*>::difference_type)
2.00 ms    0.0% 2.00 ms          JSC::BytecodeGenerator::variable(JSC::Identifier const&, JSC::ThisResolutionType)
5.00 ms    0.2% 2.00 ms          void JSC::LinkBuffer::copyCompactAndLinkCode<unsigned int>(JSC::MacroAssembler&, JSC::JITCompilationEffort)
1.00 ms    0.0% 1.00 ms          sys_icache_invalidate
3.00 ms    0.1% 1.00 ms          JSC::ASTBuilder::SourceElements JSC::Parser<JSC::Lexer<unsigned char>>::parseModuleSourceElements<JSC::ASTBuilder>(JSC::ASTBuilder&)
1.00 ms    0.0% 1.00 ms          ash_map.HashMapUnmanaged([]const u8,void,src.bun.StringHashMapContext,80).growIfNeeded
1.00 ms    0.0% 1.00 ms          _platform_memset
1.00 ms    0.0% 1.00 ms          __bzero
1.00 ms    0.0% 1.00 ms          JSC::DFG::DefMethodClobberize<JSC::DFG::(anonymous namespace)::LocalCSEPhase::BlockCSE<JSC::DFG::(anonymous namespace)::LocalCSEPhase::LargeMaps>>::operator()(JSC::DFG::PureValue) const


https://github.com/oven-sh/bun/assets/709451/f12bdd48-4bbe-4dd7-83f8-94d7dd920204

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20230630/77f6f018/attachment-0001.htm>