PyCUDA July 2009

pycuda@tiker.net

14 participants
20 discussions

by Chris Heuser

Hey everyone, Out of curiosity, has anyone had any success with OpenGL in pycuda? I have searched through the past messages on the list that I could find, and it did not seem that the issue was ever resolved. Does anyone have any insight? Thanks, >>>Chris Heuser

11 years, 8 months

NVIDIA CUDA Visual Profiler?

by Ahmed Fasih

Hi, I'm very surprised that google isn't turning up something about this topic because I thought it's been previously discussed, so my apologies if it has. I'm trying the NVIDIA CUDA Visual Profiler (v 2.2.05) in Windows XP with a fairly recent PyCUDA git, on CUDA 2.2 (pycuda.driver.get_driver_version() returns 2020). I provide the Visual Profiler with a Windows batch file that calls "python my_pycuda_script.py -some -flags", but the Visual Profiler (after running the script 4 times) only reports two methods, "memcopy". All other counters are zero (so it doesn't display them in the table). Manipulating the counters enabled doesn't change this. Any assistance would be much appreciated. My application runs only ~10% faster on a Tesla C1060 than a G80 Quadro (despite having twice as many MPs) so I'm hoping the profiler will help me understand why. Thanks! Ahmed

11 years, 8 months

Version Mismatch on OS X

by Chris Heuser

I am trying to get .94 beta up on my Mac, and despite a seemingly successful install, I get the following error: Fatal Python error: Interpreter not initialized (version mismatch?) Abort trap After some googling, I recognize that it is a common error with Mac applications of programs, due to the system python. But despite trying a few suggested methods, I have not been able to fix the problem. Any Suggestions? Thank you, >>>Chris Heuser

11 years, 8 months

Fwd: Re: about pyopencl

by Andreas Klöckner

Alan, I'm forwarding your message to the PyCUDA mailing list, maybe someone there has an idea. Andreas ---------- Forwarded Message ---------- Betreff: Re: about pyopencl Datum: Montag 27 Juli 2009 Von: Alan <alanwilter(a)gmail.com> An: Andreas Klöckner <kloeckner(a)dam.brown.edu> HI Andreas, Thank you very much for your comments. I did try pyCuda first. After some struggle to get boost to work (I use Fink x86_64), I did this for pycuda: ./configure.py --boost-inc-dir=/sw/include --boost-lib-dir=/sw/lib \ --boost-thread-libname=boost_thread-mt --boost-python-libname=boost_python-mt ( edited mine siteconf.py: BOOST_INC_DIR = ['/sw/include'] BOOST_LIB_DIR = ['/sw/lib'] BOOST_COMPILER = 'gcc-4.2' BOOST_PYTHON_LIBNAME = ['boost_python-mt'] BOOST_THREAD_LIBNAME = ['boost_thread-mt'] CUDA_TRACE = False CUDA_ENABLE_GL = False CUDADRV_LIB_DIR = [] CUDADRV_LIBNAME = ['cuda'] CXXFLAGS = ['-m64'] LDFLAGS = [] ) make [snip] g++ -L/sw/lib -bundle -L/sw/lib/python2.6/config -lpython2.6 build/temp.macosx-10.5-i386-2.6/src/cpp/cuda.o build/temp.macosx-10.5-i386-2.6/src/cpp/bitlog.o build/temp.macosx-10.5-i386-2.6/src/wrapper/wrap_cudadrv.o build/temp.macosx-10.5-i386-2.6/src/wrapper/mempool.o -L/sw/lib -L/usr/local/cuda/lib -lboost_python-mt -lboost_thread-mt -lcuda -o build/lib.macosx-10.5-i386-2.6/pycuda/_driver.so -arch i386 ld warning: in /sw/lib/python2.6/config/libpython2.6.dylib, file is not of required architecture ld warning: in build/temp.macosx-10.5-i386-2.6/src/cpp/cuda.o, file is not of required architecture ld warning: in build/temp.macosx-10.5-i386-2.6/src/cpp/bitlog.o, file is not of required architecture ld warning: in build/temp.macosx-10.5-i386-2.6/src/wrapper/wrap_cudadrv.o, file is not of required architecture ld warning: in build/temp.macosx-10.5-i386-2.6/src/wrapper/mempool.o, file is not of required architecture ld warning: in /sw/lib/libboost_python-mt.dylib, file is not of required architecture ld warning: in /sw/lib/libboost_thread-mt.dylib, file is not of required architecture I know quite well the warnings above and hence, make tests failed. It's because: otool -L build/lib.macosx-10.5-i386-2.6/pycuda/_driver.so build/lib.macosx-10.5-i386-2.6/pycuda/_driver.so: /usr/local/cuda/lib/libcuda.dylib (compatibility version 1.1.0, current version 2.3.0) /usr/lib/libstdc++.6.dylib (compatibility version 7.0.0, current version 7.4.0) /usr/lib/libgcc_s.1.dylib (compatibility version 1.0.0, current version 1.0.0) /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 111.1.4) file /usr/local/cuda/lib/libcuda.dylib /usr/local/cuda/lib/libcuda.dylib: Mach-O dynamically linked shared library i386 All others are x86_64. So basically my problem is libcuda is 32 bits. And I guess I will be in this dead end until Nvidia release Cuda in 64bits for Mac. And maybe this will happen only when Mac OSX becomes only 64 bits... Cheers, Alan On Fri, Jul 24, 2009 at 10:50, Andreas Klöckner <kloeckner(a)dam.brown.edu>wrote: > Hi Alan, > > On Freitag 24 Juli 2009, you wrote: > > I am about to start to play with pycuda but I am keen to consider > > opencl and I was trying to find any information about pyopencl and got > > nothing. And since your name is behind both project I was wondering if > > pyopencl is not developed anymore. > > It's not that PyOpenCL isn't developed any more--I just haven't found time > to > grow it beyond its embryonic stage just yet, but it'll happen. Funnily, a > different project called PyOpenCL has cropped up [1], which is a minimal, > but > functional, ctypes-based wrapper. I've gotten in touch with its author > regarding working together or at least changing one of the names, but so > far > haven't gotten any response. Since I personally don't like ctypes for > big-ish > wrappers, I'll continue to develop "my" PyOpenCL as-is. Contributions to > the > code would be more than welcome. > > [1] http://pyopencl.next-touch.com/ > > > I see CUDA has been ahead but for the future and the good of GPU > > applications I believe opencl has to be the way and I am afraid of > > developing my applications in CUDA only. > > While OpenCL promises vendor independence, I believe that switching devices > will still involve significant code changes, so in that sense you're > committing yourself to a device (and hence a vendor) anyway. Relatedly, it > appears that Nvidia is the only vendor with a credible CL story right now, > and > it's plausible that they'll have CUDA stays 'ahead' in some sense on their > own > hardware--be that in terms of features or performance. If "just wait for a > year or two and see what happens" is a valid choice for you, that may be > what > you want to do, but if you want to do GPU computing right now, CUDA is > probably the best choice. > > HTH > Andreas > > -- > Andreas Kloeckner > Applied Mathematics, Brown University > http://www.dam.brown.edu/people/kloeckner > +1-401-648-0599 > -- Alan Wilter S. da Silva, D.Sc. - CCPN Research Associate Department of Biochemistry, University of Cambridge. 80 Tennis Court Road, Cambridge CB2 1GA, UK. >>http://www.bio.cam.ac.uk/~awd28<< -------------------------------------------------------

11 years, 8 months

Out of memory (and possibly other errors) when running test_driver on Quadro NVS 140M

by Philipp Lies

Hi, first of all: nice piece of work Andreas! Now to my problem: I installed CUDA 2.1 on my T61 laptop computer with a Nvidia Quadro NVS 140M graphics card w/ 128MB memory. git version of PyCUDA (0.93 beta I assume) installs fine but I get some errors when I try to run tests. First test_driver: /opt/PyCUDA/pycuda-git/test$python test_driver.py ...E....Eterminate called after throwing an instance of 'cuda::error' what(): cuMemFree failed: invalid context Aborted Or runnung gpuarray speed test: /opt/PyCUDA/pycuda-git/test$python undistributed/test_gpuarray_speed.py 1024 2048 4096 8192 16384 32768 65536 131072 262144 524288 1048576 2097152 Traceback (most recent call last): File "undistributed/test_gpuarray_speed.py", line 83, in <module> main() File "undistributed/test_gpuarray_speed.py", line 27, in main b = gpuarray.zeros((size,), dtype=numpy.float32) File "/usr/lib/python2.5/site-packages/pycuda-0.93beta-py2.5-linux-x86_64.egg/pycuda/gpuarray.py", line 409, in zeros result = GPUArray(shape, dtype, stream, allocator) File "/usr/lib/python2.5/site-packages/pycuda-0.93beta-py2.5-linux-x86_64.egg/pycuda/gpuarray.py", line 75, in __init__ self.gpudata = self.allocator(self.size * self.dtype.itemsize) pycuda._driver.MemoryError: cuMemAlloc failed: out of memory ----------------------------------------------------------- PyCUDA WARNING: I'm being asked to destroy a context that's part of the current context stack. ----------------------------------------------------------- I will pick the next lower active context from the context stack. Since this choice is happening at an unspecified point in time, your code may be making false assumptions about which context is active at what point. Call Context.pop() to avoid this warning. ----------------------------------------------------------- If Python is terminating abnormally (eg. exiting upon an unhandled exception), you may ignore this. ----------------------------------------------------------- test_gpuarray only threw some warnings: python test_gpuarray.py /usr/lib/python2.5/site-packages/pycuda-0.93beta-py2.5-linux-x86_64.egg/pycuda/gpuarray.py:498: UserWarning: behavior change: arange guessed dtype other than float32. suggest specifying explicit dtype. warn("behavior change: arange guessed dtype other than float32. " .................. ---------------------------------------------------------------------- Ran 18 tests in 14.756s OK Here are some debug information you might need: In [61]: pycuda.autoinit.device.get_attributes() Out[62]: {pycuda._driver.device_attribute.MAX_THREADS_PER_BLOCK: 512, pycuda._driver.device_attribute.MAX_BLOCK_DIM_X: 512, pycuda._driver.device_attribute.MAX_BLOCK_DIM_Y: 512, pycuda._driver.device_attribute.MAX_BLOCK_DIM_Z: 64, pycuda._driver.device_attribute.MAX_GRID_DIM_X: 65535, pycuda._driver.device_attribute.MAX_GRID_DIM_Y: 65535, pycuda._driver.device_attribute.MAX_GRID_DIM_Z: 1, pycuda._driver.device_attribute.MAX_SHARED_MEMORY_PER_BLOCK: 16384, pycuda._driver.device_attribute.TOTAL_CONSTANT_MEMORY: 65536, pycuda._driver.device_attribute.WARP_SIZE: 32, pycuda._driver.device_attribute.MAX_PITCH: 262144, pycuda._driver.device_attribute.MAX_REGISTERS_PER_BLOCK: 8192, pycuda._driver.device_attribute.CLOCK_RATE: 800000, pycuda._driver.device_attribute.TEXTURE_ALIGNMENT: 256, pycuda._driver.device_attribute.GPU_OVERLAP: 1, pycuda._driver.device_attribute.MULTIPROCESSOR_COUNT: 2} In [63]: pycuda.autoinit.device.compute_capability() Out[63]: (1, 1) In [65]: pycuda.autoinit.device.total_memory() Out[65]: 133496832 In [66]: pycuda.autoinit.device.make_context() terminate called after throwing an instance of 'cuda::error' what(): cuMemFree failed: invalid context Aborted All the other tests were fine, so PyCUDA works nicely. Could you please check where the "invalid context" error comes from? And maybe add a memory checker to your tests so we low-mem GPU user are not discriminated ;-). Keep up the good work! Philipp

11 years, 8 months

PyCUDA at SciPy 2009?

by Paul Rigor (uci)

Hi Andreas,Any interest in giving a talk at scipy this year? http://conference.scipy.org/ Thanks, Paul Paul Rigor http://www.paulrigor.net/ http://www.igb.uci.edu/~prigor

11 years, 8 months

Mac Installation

by Chris Heuser

Hey everybody, I am trying to install .94 beta on a MacBook Pro. I went through the installation instructions on the wiki, and everything went fairly well. It seemed to have all worked and returned 0 errors. but when I run test_driver.py, this error comes up: Traceback (most recent call last): File "test/test_driver.py", line 472, in <module> import pycuda.autoinit File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/pycuda-0.94beta-py2.5-macosx-10.3-i386.egg/pycuda/autoinit.py", line 1, in <module> import pycuda.driver as cuda File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/pycuda-0.94beta-py2.5-macosx-10.3-i386.egg/pycuda/driver.py", line 1, in <module> from _driver import * ImportError: dlopen(/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/pycuda-0.94beta-py2.5-macosx-10.3-i386.egg/pycuda/_driver.so, 2): Library not loaded: libboost_python-xgcc40-mt-1_39.dylib Referenced from: /Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/pycuda-0.94beta-py2.5-macosx-10.3-i386.egg/pycuda/_driver.so Reason: image not found Any ideas how I could fix this? Thanks! >>>Chris Heuser

11 years, 9 months

my quest for re-usable auto-tuning code via templating engines

by Andrew Wagner

Hi Guys- I'm trying to write re-usable code for automatically tuning (CUDA) kernels and their associated (python) calling functions for optimal execution time. My main goals: 1. Write re-usable code to handle the optimization of kernels (and their associated support code), given templates that contain: 2. Inline definition of logic for conditional unrolling of loops, disabling sections of code, etc... 3. Inline definition of the tunable parameters, along with the sets of possible values they can take I've found two templating languages that I think are promising for this goal: Cheetah http://www.cheetahtemplate.org and Mako http://www.makotemplates.org Both are pretty rich and Pythonic in terms of the logic you can incorporate into your templates, so that covers 2. pretty well. I'm stuck on 3. To give you some idea what I'm talking about, here's a dummy template that generates python code that just waits different amounts of time based on the (after the template is rendered) hard coded values of x1, x2, y1, y2. __________contents of mainFileName__________________ ## This is a mako template demonstrating how to make a template for use with the AutoTunedFunction class <%! tuneableParameterSets = [] %> import time def main: ## Two tuneable parameters that must be optimized together <%! tuneableParameterSets.append({}) tuneableParameterSets[0]['x1'] = [0, 3, 2] tuneableParameterSets[0]['x2'] = [0, 5, 4] %> time.sleep(.01 * ${x1} * ${x2}) ## Another set of parameters that should be optimized together. <%! tuneableParameterSets.append({}) tuneableParameterSets[1]['y1'] = [True, False] tuneableParameterSets[1]['y2'] = [False, True] %> if ${y1} or ${y2}: time.sleep(.02) ____________________________ So there is a list of dictionaries, tuneableParameterSets, that I want to be able to access from outside the template. When you do: >>> from mako.template import Template >>> t = Template(filename=mainTemplateFileName) the template source gets compiled to template code. I was hoping there'd be a nice way to inspect the resulting class instance t to get the tuneableParameterSets variable, and use that to iterate through the possible values for the tuned parameters. Unfortunately, there doesn't seem to be a nice clean way to do this. One ugly hack would be to scrape the code that sets the variables out of the compiled (but not rendered/filled) template code, perhaps with the help of some added delimiters. This is really ugly, but so far it's the only thing that I've come up with that would work. In Cheetah there is a syntax #attr that lets you set attributes of the resulting class (after the template is compiled, before it is rendered/filled). That almost worked, but unfortunately it doesn't seem like you can set a dictionary, only string and numeric literals. Anyone familiar with these templating engines have any clever ideas? Is anyone else using generic tuning code, or is everyone writing a separate auto-tuning script for each module? I also looked a bit a codepy, but I don't quite understand what to do with it; it doesn't seem suitable for manipulating chunks of code written in another language. I may be misunderstanding it completely though, and there's not much documentation. Thanks! Drew

11 years, 9 months

Yet another thread about installing on WinXP

by Ahmed Fasih

Hey everyone, it turns out PyCUDA on Linux was such a bit hit that some colleagues are asking for the same thing for their Windows boxes also. I have made sure the NVIDIA CUDA SDK examples work and am following the wiki installation for MSVC 9.0 with some trouble. I used the BoostPro 1.38 installer to install default variants for Visual C++ 9.0 (Visual Studio 2008), with variants Multithread, DLL; Multithread; Multithread, static runtime; and Single thread, static runtime. I install the default components as well as Boost DateTime, Boost Python, and Boost Threads, without which the installation of PyCUDA won't happen. I've tried various other combinations of variants and compilers but test_driver.py fails with the same error message: C:\home\fasiha\pycuda\test>python test_driver.py Traceback (most recent call last): File "test_driver.py", line 472, in <module> import pycuda.autoinit File "C:\Python26\lib\site-packages\pycuda-0.94beta-py2.6-win32.egg\pycuda\autoinit.py", line 1, in <module> import pycuda.driver as cuda File "C:\Python26\lib\site-packages\pycuda-0.94beta-py2.6-win32.egg\pycuda\driver.py", line 1, in <module> from _driver import * ImportError: DLL load failed: The specified module could not be found. I was slightly confused with the instruction to "set HOME=123456", I'm not too familiar with Windows, so I tried that as well as "set HOME=c:/home/fasiha" (and with backslashes), the same error. I'm also slightly confused by this sentence in the wiki: "To work with pycuda You need to have Microsoft Visual Studio 2005 (VS 8) included in the PATH variable (for example c:\mvs8\vc\bin, where mvs8 - is your vc8 installation part) (otherwise, CUDA complains about version 6 and 7.1 !)" I assume that this is not a problem for me since PyCUDA doesn't seem to complain about MSVC. I am so sorry to unload yet another plea for installation assistance, I never ever thought I'd be installing it in Windows but I guess it's a small price for success :) I'll keep trying some stuff and report any further info. Thanks all, Ahmed

11 years, 9 months

Re: [PyCUDA] Wiki installation updated

by Vincent Favre-Nicolin

On mercredi 15 juillet 2009, Andreas Klöckner wrote: > On Mittwoch 15 Juli 2009, you wrote: > > Andreas, > > > > I added a quick installation tutorial for Ubuntu 9.04 (8.10 is > > similar, only the libboost version changes). > > http://wiki.tiker.net/PyCuda/Installation/Linux#Installing_PyCUDA_on_Ubun > >tu _9.04 Is there any reason this tutorial installs numpy 1.3.0 from source ? the python-numpy package (in 9.0.4) already provides numpy 1.2.1. Moreover atlas is not a requirement either ? -- Vincent Favre-Nicolin CEA/Grenoble Institut Nanosciences & Cryogénie - http://inac.cea.fr/ Université Joseph Fourier - http://physique-eea.ujf-grenoble.fr/ tél: (+33) 4 38 78 95 40 fax: (+33) 4 38 78 51 38

11 years, 9 months

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

PyCUDA July 2009