3 ms·
I'd love to see someone compare the performance of generated C/C++ code https://code.google.com/p/emscripten/wiki/GeneratedCodeComparison#Emscriptened_Function
by mtigas 16y ago
I'd love to see someone compare the performance of generated C/C++ code https://code.google.com/p/emscripten/wiki/GeneratedCodeComparison#Emscriptened_Function https://code.google.com/p/emscripten/wiki/GeneratedCodeCompa... versus a (possibly non-trivial) native Javascript implementation in say, V8, Spidermonkey, or Nitro.
It'd be interesting to see how decent performance can get between the LLVM optimizations and the tracing/JIT of the newer JS engines.
- Zev 16y agoFWIW, the author did somewhat touch on the issue of performance at the end of the article: Of course speed is an issue here. Emscriptened benchmarks currently run at about 1/20th the speed of gcc -O0 run on the original C++. and goes on to explain where the performance loss comes from.
- pohl 16y agoNot terribly useful as comparisons go. GWT likes to compare itself to hand-written JavaScript, which tells a more practical story.
- nl 16y agoI don't understand what you are asking for here. Emscriptem doesn't generate C/C++ code, it generates Javascript, which then runs on top of a Javascript execution environment (V8 etc). The real comparison would be some C/C++ code run though this and running in Javascript vs handwritten Javascript. (It's possible what you mean is a comparison of the generated machine code from LLVM vs that from the Javascript engine. I agree that would be interesting, but I think you'd be better off comparing the machine code generated by a Javascript-on-LLVM implementation such as http://github.com/omo/jsllc http://github.com/omo/jsllc vs that generated by a conventional Javascript runtime.)