 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
I've been using Povray since the 2.x days, and I just grabbed 3.1b4
(first new release I've touched since 3.0[12]).... I have two
questions about this 3.1 release:
1. There is no unix beta... does this imply there will be no unix
version period? Or was it just that you didn't want to put out the
source code until after the beta cycle is complete? (which doesn't make
sense to me....) Unix versions need to be beta tested too...
2. I noticed the Windows version indicates that it is compiled with
Watcom "Pentium II Optimized". Since all Pentium II's have MMX
instructions... does this mean it will make use of them when running ona
P2 (or P-MMX)?
I agree with other further back in this list that OpenGL and similar
"3D" technologies don't help POV (although they do help modelling
software). However, despite all the media hype over the Multimedia
benefits of MMX, MMX probably helps povray better than almost any
application. MMX instructions are SIMD math instrucions, allowing fast
multiple operand mathematical operations (add, sub, div, mul, and
logic). Povray could probably see some big speedups by using MMX
instructions deep down in the renderer in a few places for math
calculations.
Going back to the "no unix version?" question.... will the 3.1 source
for the Windows version be released? (and again.. why not in beta...
makes testing much easier when you can see the source)... I'd already
know about the MMX question.. or be able to attempt it myself... if the
source were around..
Brandon
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Wed, 22 Jul 1998 11:35:21 -0500, Photon <pho### [at] dtmf com> wrote:
[...]
> However, despite all the media hype over the Multimedia
>benefits of MMX, MMX probably helps povray better than almost any
>application. MMX instructions are SIMD math instrucions, allowing fast
>multiple operand mathematical operations (add, sub, div, mul, and
>logic). Povray could probably see some big speedups by using MMX
>instructions deep down in the renderer in a few places for math
>calculations.
Unfortunately, MMX and FPU operations, at least on Pentiums, both use the
same registers for different things, and swapping back and forth is costly.
But it doesn't matter much: almost all of the important math in POV is
floating-point, and MMX doesn't help with that.
>Going back to the "no unix version?" question.... will the 3.1 source
>for the Windows version be released? (and again.. why not in beta...
>makes testing much easier when you can see the source)... I'd already
>know about the MMX question.. or be able to attempt it myself... if the
>source were around..
The POV-Team told us during the 3.0 beta that they don't release source
during beta because betas are time-limited, and they don't want people
working around the time limitation, because they don't want to support
beta versions past the beta period. Whatever their reasons, it is their
source code and they can do what they want with it. I'm just happy that
they release it at all.
That said, Chris Young said on cgrr sometime last week that there is no
Linux build because their Linux guy had to leave the team, but they will
have that situation remedied soon. (They already have a list of candidates,
so there's no point in trying to volunteer.) For the actual message, go to
Dejanews and use the advanced search to find postings to cgrr from
Chris Young.
(cgrr is comp.graphics.rendering.raytracing, for those unfamiliar with it.)
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
In article <35B614C9.CEDD9B4A@dtmf.com> , Photon <pho### [at] dtmfcom> wrote:
>I've been using Povray since the 2.x days, and I just grabbed 3.1b4
>(first new release I've touched since 3.0[12]).... I have two
>questions about this 3.1 release:
>
>1. There is no unix beta... does this imply there will be no unix
>version period? Or was it just that you didn't want to put out the
>source code until after the beta cycle is complete? (which doesn't make
>sense to me....) Unix versions need to be beta tested too...
The Mac version contains the following in one of the readmes (BETANOTE.TXT), it refers
to beta 2 (which was only available in the POVRAY forum on CompuServe):
------------ Begin Forwarding --------
A Message from POV-Team Coordinator Chris Young
This beta test version of POV-Ray 3.1 is being distributed only via the CompuServe GO
POVRAY forum. You may use it and share it with friends but do not re-post it on any
web site, newsgroup, online service, bbs, cd rom, or any other form of public mass
distribution.
The POV-Team will only accept bug reports via public forum messages on GO POVRAY on
CompuServe.
This version fails to render any scene which used halo or atmosphere. Keep a copy of
3.02 if you need these features.
The documentation is VERY INCOMPLETE. The new interior{media{...}} feature which
replaced halo and atmosphere is totally undocumented except for the sample scenes.
We're working as fast as we can.
This is an expiring beta with no source code. We will not release any source code
until all platforms are in final release.
At this time only Mac and MS-Dos platforms are being released. Others will follow as
our time and resources permit.
To everyone who complains about this very limited, incomplete release... all we can
say is we're doing our best and we figured a partial release was better than nothing.
------------ End Forwarding --------
>2. I noticed the Windows version indicates that it is compiled with
>Watcom "Pentium II Optimized". Since all Pentium II's have MMX
>instructions... does this mean it will make use of them when running ona
>P2 (or P-MMX)?
>
>I agree with other further back in this list that OpenGL and similar
>"3D" technologies don't help POV (although they do help modelling
>software). However, despite all the media hype over the Multimedia
>benefits of MMX, MMX probably helps povray better than almost any
>application. MMX instructions are SIMD math instrucions, allowing fast
>multiple operand mathematical operations (add, sub, div, mul, and
>logic). Povray could probably see some big speedups by using MMX
>instructions deep down in the renderer in a few places for math
>calculations.
You should not trust Intel and its adds...MMX is SIMD (Single Instruction, Multiple
Data) and the MMX unit uses the FPU registers of the processor, BUT it only offers
integer calculations for 8,16 (and 32?) bit components of a 64 bit register. This
mechanism requires the no change to existing multi-tasking systems (because they save
the state of the FPU registers for each task) but is forces the compiler to switch
between FPU and MMX mode - this switching is (much!) slower than the benefits of MMX
could be for POV-Ray. Additionally, MMX is not implemeted very well in most compilers
- you will usually have to use assembler programming (and loose POV-Rays
cross-platform functions) to get a speed advantage over non MMX code.
But you are not totally wrong: Intel has "MMX-2" (not implmemented in existing
processors yet), which provides floating-point SIMD - but currently the make a big
secret out of it...
However and alliance of the other x86 makers (AMD, Cyrix and Centaur) has created
"3DNow!", which is build into AMDs current K6-II and Cyrix and Centaur will offer it
soon, too. This standard is supported by Microsoft, but currently there is no compiler
out there for it (as far as I know).
If you have not heard about 3DNow! you may want to refer to this article:
<http://www.heise.de/ct/english/98/12/028/> (This is an English translation of a
German computer magazine article! The translations are free and so they are a little
it (two issues) behind...)
>Going back to the "no unix version?" question.... will the 3.1 source
>for the Windows version be released? (and again.. why not in beta...
>makes testing much easier when you can see the source)... I'd already
>know about the MMX question.. or be able to attempt it myself... if the
>source were around..
The source code won't be available until final 3.1. This has been the case for all
previous betas (at least for 3.0) as far as I know.
Thorsten
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Yes, it would have to be done in asm, and yes, it is integer math only..
I'm aware. I was assuming that not everything pov did was float. I
guess I'm wrong there. I didn't think about the FPU/MMX state switching
stuff... I guess that deals it the final death-blow....
What about approximateable calculations? For instance... could you use
approximated integers for bounding box calculations (always rounding
"outwards")... by pushing those to integer... they could be
MMX-ifiyed... but then again.. there's that FPU state saving cost... oh
well...
So much for that :)
And I guess I just didn't read the fine print about source release (I
never read fine print :)
Thanks for your answers,
Brandon
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Shortly after I responded to the two responses to my posting, I
randomly, unintentionally, fell upon new information directly related to
this.
In the Pentium II Deschutes series of processors (P2-333 and up), there
are two new semi-undocumented instructions. These instructions are
primarly meant to be real and documented (people in the know believe...)
when the MMX2/KNI processors come out (which are the Katmai series).
The instructions are FXSAVE and FXRSTOR, which do the same basic
function as FSAVE and FRSTOR, except "faster". Apparently this is
primarily targeted at OS vendors, to speed up FPU state switching when
task switching.... but they are also here to make this state switching
faster for software which makes use of MMX or (sometime soon..) MMX2
instructions.
The presence of the semi-undocmented instructions can be accurately
detected via CPUID.
Here's a link to the site I found this on:
http://www.tbcnet.com/~clive
Look around there for the word "Deschutes".
Given these instructions... that lowers the major barriers to one last
biggy: Finding some intense non-FP math to accelerate in povray :)
Just htought someone might be interested in this stuff anyways, since we
were discussing it though...
Brandon
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Photon wrote:
> Shortly after I responded to the two responses to my posting, I
> randomly, unintentionally, fell upon new information directly related to
> this.
>
> In the Pentium II Deschutes series of processors (P2-333 and up), there
> are two new semi-undocumented instructions. These instructions are
> primarly meant to be real and documented (people in the know believe...)
> when the MMX2/KNI processors come out (which are the Katmai series).
>
> The instructions are FXSAVE and FXRSTOR, which do the same basic
> function as FSAVE and FRSTOR, except "faster". Apparently this is
> primarily targeted at OS vendors, to speed up FPU state switching when
> task switching.... but they are also here to make this state switching
> faster for software which makes use of MMX or (sometime soon..) MMX2
> instructions.
>
> The presence of the semi-undocmented instructions can be accurately
> detected via CPUID.
>
> Here's a link to the site I found this on:
>
> http://www.tbcnet.com/~clive
>
> Look around there for the word "Deschutes".
>
> Given these instructions... that lowers the major barriers to one last
> biggy: Finding some intense non-FP math to accelerate in povray :)
>
> Just htought someone might be interested in this stuff anyways, since we
> were discussing it though...
>
> Brandon
I heard this too, around the time Deschutes was debuting. Apparently
Win98 is supposed to make use of them.
Also, was it mentioned that the 3dnow instructions are supported by DirectX
6? Not that that will matter much, but . . .
Ciao
Willow
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
|
|
