[Date Prev][Date Next][Thread Prev][Thread Next]
- Subject: Re: New LuaJIT benchmark results (was Re: [ANN] LuaJIT-2.0.0-beta3)
- From: Mike Pall <mikelu-1003@...>
- Date: Wed, 10 Mar 2010 14:56:09 +0100
Mark Hamburg wrote:
> On Mar 9, 2010, at 4:00 PM, Mike Pall wrote:
> > Sure. I've previously mentioned that even an interpreter-only port
> > of LuaJIT to PPC (PS3, XBox 360 and Wii) might be interesting:
> > http://lua-users.org/lists/lua-l/2009-11/msg00195.html
> > Find the sponsors and I'll do it.
> I'll probably let you focus for a bit on the x86 work, but I
> might be able to find sponsorship for an ARM port.
Thank you, that sounds good! :-)
Hey, I could start an auction: bid-for-your-CPU and the winner
gets it first ... but seriously:
The difficulty with ARM is the wide variety of devices out there.
Some don't have hardware floating-point at all, some have only
single-precision FP (can't use that) and few have double-precision
FP support. FPU performance on ARM is another sad story.
LuaJIT is currently very FP-heavy. Porting this directly to ARM is
straightforward, but may lead to unacceptable performance. Another
approach would be to prototype an integer-only variant of LuaJIT
on x86 and then port this to ARM. A third approach is to do more
aggressive narrowing. But there's a limit to that and it requires
some major design changes.
So any potential sponsors of the ARM port need to indicate the
requirements for FP vs. int and the intended target devices. This
influences which path the development should take.