 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Sun, 11 Aug 2002 18:56:33 +1000, "Chris Cason"
<new### [at] delete thispovrayorg> wrote:
> And I would like to take this opportunity to publicly thank Francois for his
> hard work on the P4 optimizations already in 3.5. I am quite looking forward
> to seeing his hyperthreading implementation.
Since there are some press informations about HyperThreading are there also
some news about optimizations made for POV ?
ABX
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"ABX" <abx### [at] abxartpl> wrote in message
news:c051ou8urot243j2homt8ajerhjjif0c52@4ax.com...
> Since there are some press informations about HyperThreading are there also
> some news about optimizations made for POV ?
No, I know nothing specific.
- Chris
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Fri, 13 Sep 2002 22:10:49 +1000, "Chris Cason"
<new### [at] deletethispovrayorg> wrote:
> > Since there are some press informations about HyperThreading are there also
> > some news about optimizations made for POV ?
>
> No, I know nothing specific.
I have wrote to Francois Piednoel when he asked that one place where threading
could be applied is spline interpolation. Every dimension can be calculated
parallel becouse not influence each other. That's only idea. I can't imagine
how it could be written in the sources to point compiler it should be compiled
with threading.
ABX
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
In article <90m3ou8uubbqfdu6m3ailgpg7n4gtss85u@4ax.com> , ABX
<abx### [at] abxartpl> wrote:
> On Fri, 13 Sep 2002 22:10:49 +1000, "Chris Cason"
> <new### [at] deletethispovrayorg> wrote:
>> > Since there are some press informations about HyperThreading are there also
>> > some news about optimizations made for POV ?
>>
>> No, I know nothing specific.
>
> I have wrote to Francois Piednoel when he asked that one place where threading
> could be applied is spline interpolation. Every dimension can be calculated
> parallel becouse not influence each other. That's only idea. I can't imagine
> how it could be written in the sources to point compiler it should be compiled
> with threading.
Actually, you remind me of something:
While we all usually think more or less of POV-Ray as being hard to modify
for multithreading, this is not really so for one big part of the code. The
intersection tests (and bounding) are with a very few exceptions next to
independent of global variables and very well isolated within the code
already. So one could restrict threading to intersection tests and keep
multiple intersection stacks. Hand the data over to a thread doing the
texturing after all intersections for one pixel have been calculated. This
would still allow things like radiosity to work as they should assuming only
one thread calculates textures. Only finishes (and maybe media) would have
to be taken into account in the other threads. While balancing the load
would be more difficult and depend a lot on the scene, it should be "easy"
to implement in the current codebase and yield reasonable results with two
or maybe even four threads. Of course on real supercomputers it would be
next to useless, but on common desktop multiprocessor systems...
Thorsten
____________________________________________________
Thorsten Froehlich, Duisburg, Germany
e-mail: tho### [at] trfde
Visit POV-Ray on the web: http://mac.povray.org
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
in news:3d833d4b$1@news.povray.org Thorsten Froehlich wrote:
> While we all usually think more or less of POV-Ray as being hard to
> modify for multithreading,
How about microthreading?
http://www.eecs.umich.edu/~stever/pubs/isca99.pdf
There used to be a stackless version of Python that did microthreading
http://www.stackless.com/
http://willware.net:8080/uthread.html
Although I don't understand most of this kind of stuff there are some
aspects to it that made me wonder, would it be possible to do such a thing
to POV-Ray and if so would it be usefull
Ingo
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Sat, 14 Sep 2002 15:44:43 +0200, "Thorsten Froehlich" <tho### [at] trfde>
wrote:
> While we all usually think more or less of POV-Ray as being hard to modify
> for multithreading, this is not really so for one big part of the code. The
> intersection tests (and bounding) are with a very few exceptions next to
> independent of global variables and very well isolated within the code
> already.
Your idea finally convinced me that uv coordinates in parametric object are
not holded as best as it could.
> While balancing the load
> would be more difficult and depend a lot on the scene, it should be "easy"
> to implement in the current codebase and yield reasonable results with two
> or maybe even four threads.
Are there such plans for 4.0?
ABX
Post a reply to this message
|
|
| |
| |
|
|
From: Thorsten Froehlich
Subject: Re: MultiThreaded version of PovRay 3.5
Date: 16 Sep 2002 15:57:04
Message: <3d863790@news.povray.org>
|
|
|
| |
| |
|
|
In article <Xns### [at] povrayorg> , ingo <ing### [at] tagpovrayorg>
wrote:
> How about microthreading?
Well, what is presented in the paper sounds very much like a variant of the
*T machine developed at MIT in the early 1990s and as implemented in the
88110MP (they called what they created microthreads as well). The catch is
that like all fine grain threading architectures it requires hardware
support because the "problem" creating fine grain threads without hardware
switching support is that the required frequent context switching will eat
all of the possible performance gain.
Thorsten
* Sadly, the project was completed in the pre-WWW days and thus not much can
be found online these days...
<http://www.wikipedia.com/wiki/Motorola_88000>
____________________________________________________
Thorsten Froehlich, Duisburg, Germany
e-mail: tho### [at] trfde
Visit POV-Ray on the web: http://mac.povray.org
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
