Re: Google Chrome & the V8 Javascript engine

[Mike Pall <mikelu-0809@...>]

Chris wrote:
> On Tue, Sep 2, 2008 at 3:38 PM, David Given <dg@cowlark.com> wrote:
> > LuaJIT still wins, but V8 is definitely catching up --- it's a mere
> > 106kloc too, which competes quite nicely with LuaJIT's 28kloc...
> 
> Depends on the benchmarks used I think.  I have been running through
> some of the benchmarks from the language shootout (both the simple
> ones from the pre-Debian era and the new ones) and V8 is considerably
> faster than even LuaJIT in all of them that I have tried (2x to 4x
> faster for the most part it seems).

This would be highly surprising. I've got no access to my
benchmarking machine right now or I'd run the benchmarks myself.
You've probably forgot to run LuaJIT with -O. And the "old"
shootout benchmarks use completely naive and untuned Lua code.

The JS comparison here shows that V8 is usually slower than
TraceMonkey:

  http://ejohn.org/blog/javascript-performance-rundown/

And I've previously posted results for LuaJIT 1.1 vs. TraceMonkey
where LuaJIT beats TraceMonkey hands down:

  http://www.reddit.com/r/programming/comments/6xkg4/tracemonkey_javascript_lightspeed_firefoxs/c0559p5

So, extrapolating from this, I'd expect V8 to be slower than
LuaJIT except for a few isolated tests (e.g. recursion).

[The focus on recursive benchmarks in V8 is not surprising, given
that these are ported over from Scheme and one of the authors PhD
thesis is about a Scheme VM. But I highly doubt that recursion is
heavily used in JavaScript -- all existing VMs are really bad at it.]

--Mike