PyCUDA November 2013

pycuda@tiker.net

14 participants
14 discussions

by Tomasz Rybak

Hello. I've been packaging PyCUDA for Debian. I run all the tests to ensure that package works on Python 2 and Python 3. All tests pass except for on from test_driver.py: $ python test_driver.py ============================= test session starts ============================== platform linux2 -- Python 2.7.5 -- pytest-2.3.5 collected 21 items test_driver.py ........F............ =================================== FAILURES =================================== _____________________ TestDriver.test_register_host_memory _____________________ args = (<test_driver.TestDriver instance at 0x24e7d88>,), kwargs = {} pycuda = <module 'pycuda' from '/usr/lib/python2.7/dist-packages/pycuda/__init__.pyc'> ctx = <pycuda._driver.Context object at 0x2504488> clear_context_caches = <function clear_context_caches at 0x1dbf848> collect = <built-in function collect> def f(*args, **kwargs): import pycuda.driver # appears to be idempotent, i.e. no harm in calling it more than once pycuda.driver.init() ctx = make_default_context() try: assert isinstance(ctx.get_device().name(), str) assert isinstance(ctx.get_device().compute_capability(), tuple) assert isinstance(ctx.get_device().get_attributes(), dict) > inner_f(*args, **kwargs) /usr/lib/python2.7/dist-packages/pycuda/tools.py:434: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ self = <test_driver.TestDriver instance at 0x24e7d88> @mark_cuda_test def test_register_host_memory(self): if drv.get_version() < (4,): from py.test import skip skip("register_host_memory only exists on CUDA 4.0 and later") import sys if sys.platform == "darwin": from py.test import skip skip("register_host_memory is not supported on OS X") a = drv.aligned_empty((2**20,), np.float64, alignment=4096) > drv.register_host_memory(a) E LogicError: cuMemHostRegister failed: invalid value test_driver.py:559: LogicError ==================== 1 failed, 20 passed in 116.85 seconds ===================== This test fails both on ION (GeForce 9400M, CC 1.1) and GeForce 460 (CC 2.1). I've compiled PyCUDA with gcc 4.8, run with kernel 3.9 and drivers 304.88. Regards. -- Tomasz Rybak GPG/PGP key ID: 2AD5 9860 Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860 http://member.acm.org/~tomaszrybak

6 years, 8 months

sending a device pointer through cython to cuda C

by Rok Roskar

Hello, I've got a host-side CUDA library wrapped in Cython and I'd like to use it on a device-side array that I've allocated in python with PyCuda. However, I'm completely at a loss as to how I should pass the device pointers from the python side of things to the C library. In CUDA the device pointers look like normal pointers, but in PyCuda I get a DeviceAllocation object -- how can I pass the pointer address information to the C library through Cython? What types should I use in the Cython wrapper? thanks! Rok

7 years, 2 months

Passing float3 (or other vector types) to prepared_call CUDA kernels

by Rowland Ben

Hey, Just started working with PyCUDA, and already very taken with it, it makes a whole load of things very simple. Already after a couple of days I have a working program with OpenGL interop using PySide to provide the GUI and CUDA doing 3D texture lookups. One thing I haven’t been able to figure out yet though is how to pass vector arguments to CUDA kernels. I have a function definition like this: __global__ void extract_slice(uchar4 *destPbo, unsigned int width, unsigned int height, float3 center, float3 xDelta, float3 yDelta) but I can’t make this work with pycuda passing in float3 variables like this: self.kernel.prepared_call(grid_dimensions, (16, 16, 1), target_mapping.device_ptr(), self.width(), self.height(), pycuda.gpuarray.vec.make_float3(0.0, 0.0, 0.0), pycuda.gpuarray.vec.make_float3(1.0, 0.0, 0.0), pycuda.gpuarray.vec.make_float3(0.0, 1.0, 0.0) ) so instead I have had to resort to just passing in 9 floats instead of 3 float3 variables, these are then reconstructed into float3s in the kernel and everything works correctly. While this is a solution for the moment I would really like to figure out how to pass vector types like this, and if anyone can point me in the right direction it would be much appreciated. Thanks, Ben

7 years, 4 months

Customize compiler cache directory

by Healther.astro

OHey pycuda-list, I have some kernels which will be used over and over again, but as the compile time of them is pretty high (something like half an hour, for some hundred kernels). I would like to ensure that they are stored permanently on my computer, as opposed to the storage in the tmp folder, where they are deleted at some time. Is there a way to customize the folder in which those compiler caches are stored? And is there a way to find out which of the folders belong to which kernel? Cheers, Andreas Baumbach

7 years, 4 months

ImportError: libcurand.so.5.0: cannot open shared object file

by ggeo

Hello, I receive the above error when I try to run a pycuda code. I use sabayon linux and this happened after an update at which it installed the cuda toolkit 5.5 I tried to install again pycuda through: python configure.py --cuda-root=/opt/cuda/ At start ,it gives me WARNING: nvcc not in path If I do nvcc in terminal it gives "nvcc fatal : No input files specified; use option --help for more information " ,but I can compile and run cu programs. My ~/.bashrc: export PATH=~/bin:$PATH export PATH=/opt/cuda/bin:$PATH export PATH=~/opt/bin:$PATH export LD_LIBRARY_PATH=/opt/cuda/lib64:$LD_LIBRARY_PATH export CUDA_ROOT=/opt/cuda/bin I had a similar thread http://pycuda.2962900.n2.nabble.com/nvcc-not-in-path-td7574709.html <http://pycuda.2962900.n2.nabble.com/nvcc-not-in-path-td7574709.html> but it didn't help now. (If I do echo $LD_LIBRARY_PATH as root it doesn't return anything.) Any help please? Thank you! -- View this message in context: http://pycuda.2962900.n2.nabble.com/ImportError-libcurand-so-5-0-cannot-ope… Sent from the PyCuda mailing list archive at Nabble.com.

7 years, 4 months

Re: [PyCUDA] Thrust interoperability- cgen/ codepy errors on python3.2

by Andreas Kloeckner

Hi Graham, Graham Mills <13gm10(a)queensu.ca> writes: > I looked back about a year in the archives and couldn't find anything on this. I just downloaded and built cgen 2013.1.2 and codepy 2013.1.2 today. When using thrust as in the example at http://wiki.tiker.net/PyCuda/Examples/ThrustInterop , several python version-related errors crop up in cgen and codepy modules, such as > > File "/usr/local/lib/python3.2/dist-packages/codepy-2013.1.2-py3.2.egg/codepy/toolchain.py", line 153, in get_dependencies > lines = join_continued_lines(stdout.split("\n")) > TypeError: Type str doesn't support the buffer API > > and > > File "/usr/local/lib/python3.2/dist-packages/cgen-2013.1.2-py3.2.egg/cgen/__init__.py", line 893, in get_namespace_name > checksum.update(line) > TypeError: Unicode-objects must be encoded before hashing > > for instance. I manually fixed a few of the issues in my installation, but there are a few I cannot figure out. Retracing my steps back to building the installation, when build_py_2to3 gets called in setup.py it misses these lines (and others). Manually calling 2to3 doesn't catch them, either. > > Is there another release of codepy or cgen I should be using with python 3.2? you're probably the first to try this with Py3. :) Python 3 support in PyCUDA is fairly recent after all. If you could post/pull-request your fixes as well as tests for the stuff you're stuck on, I'd be happy to take a look and help get this fixed. Andreas

7 years, 4 months

Thrust interoperability- cgen/ codepy errors on python3.2

by Graham Mills

Hi all, I looked back about a year in the archives and couldn't find anything on this. I just downloaded and built cgen 2013.1.2 and codepy 2013.1.2 today. When using thrust as in the example at http://wiki.tiker.net/PyCuda/Examples/ThrustInterop , several python version-related errors crop up in cgen and codepy modules, such as File "/usr/local/lib/python3.2/dist-packages/codepy-2013.1.2-py3.2.egg/codepy/toolchain.py", line 153, in get_dependencies lines = join_continued_lines(stdout.split("\n")) TypeError: Type str doesn't support the buffer API and File "/usr/local/lib/python3.2/dist-packages/cgen-2013.1.2-py3.2.egg/cgen/__init__.py", line 893, in get_namespace_name checksum.update(line) TypeError: Unicode-objects must be encoded before hashing for instance. I manually fixed a few of the issues in my installation, but there are a few I cannot figure out. Retracing my steps back to building the installation, when build_py_2to3 gets called in setup.py it misses these lines (and others). Manually calling 2to3 doesn't catch them, either. Is there another release of codepy or cgen I should be using with python 3.2? Thanks, Graham

7 years, 4 months

pytools test fail with python-3

by justin

Hi, pytools tests fail with python-3. Is this expected? Are pytools ready for py3? Please find attached a build.log Regards, Justin

7 years, 4 months

Re: [PyCUDA] pycuda compute capability support

by Jerome Kieffer

On Thu, 29 Aug 2013 07:26:40 -0400 Ahmed Fasih <wuzzyview(a)gmail.com> wrote: > Since PyCUDA is just a wrapper to the CUDA C library (with >99.9% > coverage), anything CUDA supports, PyCUDA should support. More directly, > PyCUDA and K20 work fine for me. I am running PyCuda on a debian6 computer with a GeForce Titan. Driver is 319.32 and Cuda 4.2. My only problem is when I try to use scikit.cuda (for CuFFT) where sm_35 is not recognized. The issue is probably located around the old version of cuda. If you stick to pure PyCuda, it runs fine. -- Jérôme Kieffer On-Line Data analysis / Software Group ISDD / ESRF tel +33 476 882 445

7 years, 4 months

Re: [PyCUDA] cuda-gdb doesn't stop at breakpoint

by Tomasz Rybak

I've just tested on GeForce 460 with CUDA 5.5 and drivers 325.15, and your example worked without any problems - kernel was stopped at doublify. I assume that you have installed all programs (drivers, CUDA, PyCUDA) from Debian repositories, or have you downloaded, compiled and installed something yourself? What hardware are you running it on? And, which might be the stupid question, were you running this from X or from console? Anything card below Kepler with Compute Capabilities 3.5 cannot at the same time be used to display X and to debug. So if you have only one card which you use for displaying your X session on monitor, you cannot use the same card for debugging. This might explain problems you described. Best regards. -- Tomasz Rybak GPG/PGP key ID: 2AD5 9860 Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860 http://member.acm.org/~tomaszrybak

7 years, 5 months

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

PyCUDA November 2013