POV-Ray: Newsgroups: povray.programming: here is a profile of the 3.6 code, FPGA round 2

POV-Ray : Newsgroups : povray.programming : here is a profile of the 3.6 code, FPGA round 2		Server Time 22 Dec 2024 07:34:28 EST (-0500)

From: Brannon
Subject: here is a profile of the 3.6 code, FPGA round 2
Date: 26 Apr 2005 01:00:00
Message: <web.426dca6c971bf69c9df08e0@news.povray.org>

Here's a profile of the 3.6 windows code running the demo:

http://garrionent.com/pov/

I ran the profile to see if I could accelerate it with an FPGA module.
Unfortunately, I don't see a whole lot of ways I could help. I could
eliminate the multiplies and adds in MInvTransPoint and MInvTransDirection.
The DMA overhead kills us, though, on such a small piece of data. It would
be no improvement. What I need is a function similar to those that
opperates on half a megabyte of data and performs about three times as many
math operations. Those are the kind of functions I could really accelerate.

Post a reply to this message

From: Peter Duthie
Subject: Re: here is a profile of the 3.6 code, FPGA round 2
Date: 27 Apr 2005 04:33:34
Message: <426f4e5e@news.povray.org>

Interesting.  What scene did you use to gather the stats?  How long did 
the parse and render take?

Peter D.

Brannon wrote:
> Here's a profile of the 3.6 windows code running the demo:
> 
> http://garrionent.com/pov/
> 
> I ran the profile to see if I could accelerate it with an FPGA module.
> Unfortunately, I don't see a whole lot of ways I could help. I could
> eliminate the multiplies and adds in MInvTransPoint and MInvTransDirection.
> The DMA overhead kills us, though, on such a small piece of data. It would
> be no improvement. What I need is a function similar to those that
> opperates on half a megabyte of data and performs about three times as many
> math operations. Those are the kind of functions I could really accelerate.
> 
>

Post a reply to this message