Re:PyPy.js: Now faster than CPython
by coldcode on Tuesday May 06, 2014 @12:04PM
(#7705291)
Does this work at all in Safari? Is Safari's crappy performance of native->js ever going to catch up to FF and Chrome?
[ Parent
| on HN ]
test.pystone.main()
Chrome on desktop -- 200K/s
pypy on desktop -- 2000K/s
Apple's webkit development has slowed down, as it seems. Only two blog entries in the last half year: https://www.webkit.org/blog/
What a petty, it seems Google's forking of Webkit (named Blink) has hurt Webkit. Google's behavior of removing "CSS regions" was an unacceptable move.
Yes, Safari has been working on an LLVM-based JIT called FTL. It now outperforms v8 on almost all asm.js benchmarks,
http://arewefastyet.com/#machine=12&view=breakdown&suite=asm...
That's not in release versions of Safari yet, it was just enabled on trunk. But it will make Safari very competitive once more.
Re:PyPy.js: Now faster than CPython
by stcredzero on Tuesday May 06, 2014 @01:40PM
(#7705950)
What if LLVM based JITs with NaCL style security were available in all browsers except IE? (I'm just assuming Microsoft will drag their heels.) Then an LLVM IR based bytecode could become a standard bytecode for browsers.
I believe that in many ways, this would be ideal. Not only could any language be used (the implementers being responsible for interfacing with the browser API), but the LLVM pseudo-asm could be pre-optimized saving the browser a lot of work.
In the meantime, browsers would still be free to compete based on their JIT implementations.
With rust and by proxy, Mozilla/Samsung's experimental Servo layout engine targeting the LLVM, I can also see potential synergy in this situation.