[Radiance-general] running radiance on multiple cores
Gregory J. Ward
gregoryjward at gmail.com
Sun Jun 3 12:24:03 PDT 2012
Quick follow-up to this... I did add a slider to trad to control the number of processes. It's checked into the latest HEAD.
Regarding falsecolor and image size, I recommend using pfilt to correct the image size on the way in.
Cheers,
-Greg
> From: Greg Ward <gward at lmi.net>
> Date: May 23, 2012 9:45:49 PM PDT
>
> Yes to #1, no to #2. This was discussed before, and I added a warning from rpiece when the resolution changes, but I can't keep it from changing. Rpiece (rpict really) needs the tiles to be of equal size, which means both dimensions must be multiples of the respective number of tile divisions.
>
> You can work around the problem with some difficulty by noting what rpiece changes the resolution to and making sure you set -x and -y to this with -pa 0 on any other runs you do. Unfortunately, the actual resolution will change with different -N settings because rad will make a different number of tiles to optimize processor loads.
>
> Regarding #1, I haven't changed trad in ages, so it may take a while to remember enough Tcl/Tk to add this.
>
> Best,
> -Greg
>
>> From: Terrance Mc Minn <t.mcminn at curtin.edu.au>
>> Date: May 23, 2012 9:28:34 PM PDT
>>
>> I can verify that missing sections of images happens on both Linux and MAC computers when an ambient file is used though not consistantly.
>>
>> Often I have suspected/verified that the number of -N specified is greater than the actual number of cores available. For example a i7 Macbook that a student had in my office 10 minutes ago shows only 4 cores hence -N 4 would seem appropriate. Some images would complete while others left black patches. All I did was to delete the offending hdr image and re-issue the rad command exactly to have the image rebuilt - much faster as the ambient file has already been populated.
>>
>> I suspect that the -N number should generally be kept to at least 1 below the max cores available.
>>
>> Questions for the Development List :
>> 1) Can multi-core processing be added to trad (-n for rvu and -N rad)? While it is possible to add a -n 4 in the trad options line for rvu, you cannot use a -N in the option line for rad.
>> 2) Can the image size resolution for single core processing be the same as multi-core to allow further processing with falsecolor. When rad -N processing uses rpiece the image size can be modified/rationalised. This does not happen in single core processing, hence falsecolor would fail when mixing images due to image size mismatch.
>>
>> Terrance Mc Minn
More information about the Radiance-general
mailing list