Nathan Kopp wrote in message <370AE1F3.3DC4D6A3@Kopp.com>...
>Ray Gardener wrote:
>>
>> Agreed, supporting all those CPUs is bothersome. But
>> gcc does it, the templates are mature, and we can tweak
>
>Does gcc support EVERY cpu?  The core of POV is supposted to be completely
>platform independent ...
>
>Therefore any bytecode interpreter that is part of the core must also be
>platform independent... JIT compilers could be extra for each system...
>but the functionality would still exist for EVERY system.



If a JIT is truly fast enough (does anyone have benchmarks
from that JavaRays program?) then that would be great. But
if not... and let's face it, for all of fuss over JITs,
they haven't had any real impact on day-to-day computing.
When I see JITs used for commercial video games, then
I'll be impressed. And how much RAM did Sun say their JIT
needed? Tons and tons? Insane.

The irony is, that if a new instruction set appears, one
still needs to ultimately port something. For the same amount
of work, you may as well just port a native-code compiler.
The bytecode interpreter might be available, but I don't think anyone
would seriously use it. I'd sooner buy hardware supported
by a JIT instead of run bytecode (because hardware is now
so cheap these days). But if I'm doing that, I'll opt
for the hardware that's natively supported, and get even
more speed.

With raytracing, there's never a point where the
system is 'fast enough'. Even if gigahertz machines
were common, we'd still need every last bit of performance.
If the pics we're doing today render quickly, we'll
start doing more complex scenes. And then there's movies,
which need tons of frames, etc. It's not like accounting,
were there's only so much math a transaction can involve.

So Ron's saying gcc's code can't be used to improve POV.
Well, that sure strikes a blow for the open source
movement. What a collosally wasted opportunity.

Ray

Post a reply to this message