PyCUDA May 2012

pycuda@tiker.net

22 participants
70 discussions

by Andreas Kloeckner

On Tue, 1 May 2012 16:38:42 -0400, Frédéric Bastien <nouiz(a)nouiz.org> wrote: > If I do not call pycuda.autoinit, I get this error: > > LogicError: explicit_context_dependent failed: invalid context - no > currently active context? > > > If I try to manually init the driver like show in the doc like this: > > import pycuda.driver as cuda > cuda.init() #init pycuda driver > current_dev = cuda.Device(device_nr) #device we are working on > ctx = current_dev.make_context() #make a working context > ctx.push() #let context make the lead > > I get this error: > > LogicError: cuCtxCreate failed: invalid device > > > I'm triing to force to always initialize pycuda gpu first. I can do it > , but it is not easy. So if someone remember how or can find where we > explained how to tell pycuda to reuse the current open context, it > would be great. There's a new method Context.attach(), which attaches to (and returns) the currently active context. This should do what you need, I think. Please let me know whether this works--otherwise I'll revert this change. Thanks, Andreas

8 years, 8 months

Re: [PyCUDA] Please permit to open translated PyCUDA documentation site

by Andreas Kloeckner

On Fri, 4 May 2012 21:12:07 +0900, Norio Saito <no5ri1o5(a)gmail.com> wrote: > My name is Norio Saito and I think PyCUDA is one of amazing libraries in > Python. > So I would like to expand recognition of this library in Japan, > and then I have translated PyCUDA documentation < > http://documen.tician.de/pycuda/> into Japanese. > I would like to open this translated site to the public. > Please give me a permission for it. By all means, go ahead. How did you do this technically? Did you start From the Sphinx/ReST source or from the HTML? If you'll share the URL, I'd be happy to point to it from the 'official' (English) docs. Andreas

8 years, 8 months

Please permit to open translated PyCUDA documentation site

by Norio Saito

Hello Andreas, My name is Norio Saito and I think PyCUDA is one of amazing libraries in Python. So I would like to expand recognition of this library in Japan, and then I have translated PyCUDA documentation < http://documen.tician.de/pycuda/> into Japanese. I would like to open this translated site to the public. Please give me a permission for it. Sincerely, -------------------- Norio Saito

8 years, 8 months

Re: [PyCUDA] Problem setting GPUArray with numpy array with FORTRAN flag

by Andreas Kloeckner

On Fri, 27 Apr 2012 15:52:03 -0400, Lars Pastewka <pas(a)pha.jhu.edu> wrote: > Hello Everybody, > > I ran into the following problem with PuCUDA. When I calling for example > > import pycuda.gpuarray as gpu > a = np.array([ [ 0, 1, 2 ], [ 1, 2, 3 ] ], dtype=np.float64) > a_gpu = gpu.to_gpu(np.transpose(a)) > > a_gpu does not contain the transposed, but the original array. > > When looking into the numpy source I find that the transpose operation > just flips the FORTRAN flag. This can be easily checked: > > >>> a.flags.fortran > False > >>> b=np.transpose(a) > >>> b.flags.fortran > True > >>> c=np.transpose(b) > >>> c.flags.fortran > False > >>> > > GPUArray.set just copies the raw data. So if the FORTRAN flag is set, > the array on the GPU will end up transposed. Does anybody know an easy > fix for this behavior? You should now get a warning if you try this. I didn't quite dare make this an error just yet, but it is now deprecated behavior and will go away in 2013.x. Andreas

8 years, 8 months

Re: [PyCUDA] Problem setting GPUArray with numpy array with FORTRAN flag

by Lars Pastewka

Wonderful, thanks! On Wed, May 2, 2012 at 11:47 AM, Andreas Kloeckner <kloeckner(a)cims.nyu.edu> wrote: > On Fri, 27 Apr 2012 15:52:03 -0400, Lars Pastewka <pas(a)pha.jhu.edu> wrote: >> Hello Everybody, >> >> I ran into the following problem with PuCUDA. When I calling for example >> >> import pycuda.gpuarray as gpu >> a = np.array([ [ 0, 1, 2 ], [ 1, 2, 3 ] ], dtype=np.float64) >> a_gpu = gpu.to_gpu(np.transpose(a)) >> >> a_gpu does not contain the transposed, but the original array. >> >> When looking into the numpy source I find that the transpose operation >> just flips the FORTRAN flag. This can be easily checked: >> >> >>> a.flags.fortran >> False >> >>> b=np.transpose(a) >> >>> b.flags.fortran >> True >> >>> c=np.transpose(b) >> >>> c.flags.fortran >> False >> >>> >> >> GPUArray.set just copies the raw data. So if the FORTRAN flag is set, >> the array on the GPU will end up transposed. Does anybody know an easy >> fix for this behavior? > > That's a bug--.set() is missing checks for matching strides. I'll add > that and let you know when it's done. > > Andreas > > -- > Andreas Kloeckner > Room 1105A (Warren Weaver Hall), Courant Institute, NYU > http://www.cims.nyu.edu/~kloeckner/ > +1-401-648-0599

8 years, 8 months

Re: [PyCUDA] install issue on OS X 10.7.3

by Serge Rey

On Tue, May 1, 2012 at 1:02 PM, Eli Stevens (Gmail) <wickedgrey(a)gmail.com> wrote: > I recall the pci bus ID behavior changing between... 4.0 and 4.1 > maybe? It swapped from being a property-like thing to being a > callable, with only changing the underlying cuda version (no PyCUDA > update). > > What version of cuda are you using? 4.2.9 But, your question prompted me to look more closely and it turns out I had followed the link on the pycuda line to the 4.0 drivers, so when went back to download the 4.2 drivers from NVIDIA's site and rebuilt pycuda (and rebooted the box as some of the cuda tests on the nvidia side didn't pass without a reboot), then all is well. So put this one down to pilot error. Thanks for the quick feedback that led me to discover my error! -- Sergio (Serge) Rey Professor, School of Geographical Sciences and Urban Planning Arizona State University http://geoplan.asu.edu/rey Editor, International Regional Science Review http://irx.sagepub.com

8 years, 8 months

Re: [PyCUDA] device already initialized

by Andreas Kloeckner

On Mon, 30 Apr 2012 16:58:59 -0400, Frédéric Bastien <nouiz(a)nouiz.org> wrote: > Hi, > > I have a the driver already initialized and I want pycuda to reuse > that context. How can I make this? I didn't found in the doc or in > previous email. But I remember having it somewhere. Can someone find > this information again? Just start using CUDA, I think. Since context state is implicit, you'll be good to go as soon as a context exists... ? Andreas

8 years, 8 months

Re: [PyCUDA] install issue on OS X 10.7.3

by Eli Stevens (Gmail)

I recall the pci bus ID behavior changing between... 4.0 and 4.1 maybe? It swapped from being a property-like thing to being a callable, with only changing the underlying cuda version (no PyCUDA update). What version of cuda are you using? Eli On Tue, May 1, 2012 at 11:50 AM, Serge Rey <srey(a)asu.edu> wrote: > Hi, > > Following the install instructions for PyCuda on Lion I'm able to > build and install > without any errors. However if I run the test I'm getting: > > Traceback (most recent call last): > File "test_driver.py", line 4, in <module> > from pycuda.tools import mark_cuda_test > File "/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/tools.py", > line 30, in <module> > import pycuda.driver as cuda > File "/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/driver.py", > line 2, in <module> > from pycuda._driver import * > ImportError: dlopen(/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so, > 2): Symbol not found: _cuDeviceGetByPCIBusId > Referenced from: > /Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so > Expected in: flat namespace > in /Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so > > > I've got the CUDA tookit and drivers installed and paths dialed in > accordingly, so I'm wondering if anyone has an idea of what might be going > wrong? > > Thanks, > > > -- > Sergio (Serge) Rey > Professor, School of Geographical Sciences and Urban Planning > GeoDa Center for Geospatial Analysis and Computation > Arizona State University > http://geoplan.asu.edu/rey > > Editor, International Regional Science Review > http://irx.sagepub.com > > _______________________________________________ > PyCUDA mailing list > PyCUDA(a)tiker.net > http://lists.tiker.net/listinfo/pycuda

8 years, 8 months

PyCUDA install issue on OS X 10.7.3

by Serge Rey

Hi, Following the install instructions for PyCuda on Lion I'm able to build and install without any errors. However if I run the test I'm getting: Traceback (most recent call last): File "test_driver.py", line 4, in <module> from pycuda.tools import mark_cuda_test File "/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/tools.py", line 30, in <module> import pycuda.driver as cuda File "/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/driver.py", line 2, in <module> from pycuda._driver import * ImportError: dlopen(/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so, 2): Symbol not found: _cuDeviceGetByPCIBusId Referenced from: /Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so Expected in: flat namespace in /Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so I've got the CUDA tookit and drivers installed and paths dialed in accordingly, so I'm wondering if anyone has an idea of what might be going wrong? Thanks -- Sergio (Serge) Rey Professor, School of Geographical Sciences and Urban Planning Arizona State University http://geoplan.asu.edu/rey Editor, International Regional Science Review http://irx.sagepub.com

8 years, 8 months

install issue on OS X 10.7.3

by Serge Rey

Hi, Following the install instructions for PyCuda on Lion I'm able to build and install without any errors. However if I run the test I'm getting: Traceback (most recent call last): File "test_driver.py", line 4, in <module> from pycuda.tools import mark_cuda_test File "/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/tools.py", line 30, in <module> import pycuda.driver as cuda File "/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/driver.py", line 2, in <module> from pycuda._driver import * ImportError: dlopen(/Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so, 2): Symbol not found: _cuDeviceGetByPCIBusId Referenced from: /Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so Expected in: flat namespace in /Library/Python/2.7/site-packages/pycuda-2011.2.2-py2.7-macosx-10.7-intel.egg/pycuda/_driver.so I've got the CUDA tookit and drivers installed and paths dialed in accordingly, so I'm wondering if anyone has an idea of what might be going wrong? Thanks, -- Sergio (Serge) Rey Professor, School of Geographical Sciences and Urban Planning GeoDa Center for Geospatial Analysis and Computation Arizona State University http://geoplan.asu.edu/rey Editor, International Regional Science Review http://irx.sagepub.com

8 years, 8 months

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

PyCUDA May 2012