Dear all, I attach to this e-mail a sample script for volume rendering. With YT 1.6 I used the old version of volume rendering (also in parallel) without problems, but now (I run on trunk, with an updated revision) I can run this script only on one CPU. If I try a parallel run, I get the following output: -------------- mpirun -np 4 python2.6 ./sample-vr.py --parallel yt INFO 2010-08-20 16:53:08,959 Parallel computation enabled: 0 / 4 yt INFO 2010-08-20 16:53:08,959 Parallel computation enabled: 2 / 4 yt INFO 2010-08-20 16:53:08,959 Parallel computation enabled: 1 / 4 yt INFO 2010-08-20 16:53:08,959 Parallel computation enabled: 3 / 4 ... etc etc... P001 yt INFO 2010-08-20 16:54:51,234 Finishing 'Ray casting ' P002 yt INFO 2010-08-20 16:54:51,383 Finishing 'Ray casting ' P003 yt INFO 2010-08-20 16:55:02,526 Finishing 'Ray casting ' P000 yt INFO 2010-08-20 16:55:02,791 Finishing 'Ray casting ' Traceback (most recent call last): File "./sample-vr.py", line 29, in <module> vp.ray_cast() File "/home/hlrb2/pr95he/lu79qey3/remotefs/yt-trunk-svn/yt/extensions/volume_rendering/software_sampler.py", line 117, in ray_cast if finalize: self._finalize() File "/home/hlrb2/pr95he/lu79qey3/remotefs/yt-trunk-svn/yt/extensions/volume_rendering/software_sampler.py", line 121, in _finalize im, f = self._mpi_catrgb((self.image, self.resolution)) File "/home/hlrb2/pr95he/lu79qey3/remotefs/yt-trunk-svn/yt/lagos/ParallelTools.py", line 165, in passage return func(self, data) File "/home/hlrb2/pr95he/lu79qey3/remotefs/yt-trunk-svn/yt/lagos/ParallelTools.py", line 852, in _mpi_catrgb new_image[0:nsize[0],0:nsize[1],:] = data[:] ValueError: shape mismatch: objects cannot be broadcast to a single shape ---------------- Any suggestion? Thanks in advance, Luigi -- --------------------------------------------------------------- Luigi Iapichino Zentrum fuer Astronomie der Universitaet Heidelberg Institut fuer Theoretische Astrophysik Albert-Ueberle-Str. 2, D-69120 Heidelberg, Germany Tel: +49 6221 548983, Fax: +49 6221 544221 e-mail: luigi@ita.uni-heidelberg.de URL: http://www.ita.uni-heidelberg.de/~luigi/
Hi Luigi,
Yeah, looks like I broke it. The new "camera" interface is far nicer for the user, but required some invasive changes. The old mechanism for parallelizing volume rendering also required image plane decomposition, but Sam Skillman has implemented kD-tree decomposition with post-processing compositing of images, which should be much faster and easier on the system because it allows for a full volume-decomp and better load balancing. Sam, what's the current state of your kD-tree? I think that since the image-plane decomp is currently broken, we can either fix that for both the volume_rendering and camera interfaces and continue to use it (even though it's worse) or we can finish up moving to the kD-tree. Either way, Luigi, this weekend I'll try to fix the script you sent, even if it ends up being kind of a hacky solution... -Matt
Hi Matt and Luigi, I haven't actually used the volume renderer in about 2-3 months at this point so it's been stagnating on my end. I'll take a look this weekend if I get a chance. I'm currently traveling so it might take a bit of extra time. Luigi - If you need the volume rendering capability in the very short term, I'd suggest checking out a separate copy of yt 1.6 and using that when you need the renders, and using 1.7 when you need the new features in 1.7. This isn't sustainable in the long run but is a quick fix if you need things working right away. Best, Sam On Fri, Aug 20, 2010 at 12:53 PM, Matthew Turk <matthewturk@gmail.com>wrote:
-- Samuel W. Skillman DOE Computational Science Graduate Fellow Center for Astrophysics and Space Astronomy University of Colorado at Boulder samuel.skillman[at]colorado.edu
Dear Matt and Sam, thanks for your reply. As suggested, I went back to trunk, r. 1693, and everything works again (with some minor syntax change with respect to the last version of the script) also in parallel. I look forward to the new features for volume rendering, but I'm not in a hurry now! Cheers, Luigi On Friday 20 August 2010, Sam Skillman wrote:
-- --------------------------------------------------------------- Luigi Iapichino Zentrum fuer Astronomie der Universitaet Heidelberg Institut fuer Theoretische Astrophysik Albert-Ueberle-Str. 2, D-69120 Heidelberg, Germany Tel: +49 6221 548983, Fax: +49 6221 544221 e-mail: luigi@ita.uni-heidelberg.de URL: http://www.ita.uni-heidelberg.de/~luigi/
Hi Luigi,
Yeah, looks like I broke it. The new "camera" interface is far nicer for the user, but required some invasive changes. The old mechanism for parallelizing volume rendering also required image plane decomposition, but Sam Skillman has implemented kD-tree decomposition with post-processing compositing of images, which should be much faster and easier on the system because it allows for a full volume-decomp and better load balancing. Sam, what's the current state of your kD-tree? I think that since the image-plane decomp is currently broken, we can either fix that for both the volume_rendering and camera interfaces and continue to use it (even though it's worse) or we can finish up moving to the kD-tree. Either way, Luigi, this weekend I'll try to fix the script you sent, even if it ends up being kind of a hacky solution... -Matt
Hi Matt and Luigi, I haven't actually used the volume renderer in about 2-3 months at this point so it's been stagnating on my end. I'll take a look this weekend if I get a chance. I'm currently traveling so it might take a bit of extra time. Luigi - If you need the volume rendering capability in the very short term, I'd suggest checking out a separate copy of yt 1.6 and using that when you need the renders, and using 1.7 when you need the new features in 1.7. This isn't sustainable in the long run but is a quick fix if you need things working right away. Best, Sam On Fri, Aug 20, 2010 at 12:53 PM, Matthew Turk <matthewturk@gmail.com>wrote:
-- Samuel W. Skillman DOE Computational Science Graduate Fellow Center for Astrophysics and Space Astronomy University of Colorado at Boulder samuel.skillman[at]colorado.edu
Dear Matt and Sam, thanks for your reply. As suggested, I went back to trunk, r. 1693, and everything works again (with some minor syntax change with respect to the last version of the script) also in parallel. I look forward to the new features for volume rendering, but I'm not in a hurry now! Cheers, Luigi On Friday 20 August 2010, Sam Skillman wrote:
-- --------------------------------------------------------------- Luigi Iapichino Zentrum fuer Astronomie der Universitaet Heidelberg Institut fuer Theoretische Astrophysik Albert-Ueberle-Str. 2, D-69120 Heidelberg, Germany Tel: +49 6221 548983, Fax: +49 6221 544221 e-mail: luigi@ita.uni-heidelberg.de URL: http://www.ita.uni-heidelberg.de/~luigi/
participants (3)
-
Luigi Iapichino -
Matthew Turk -
Sam Skillman