PyCUDA March 2011

pycuda@tiker.net

17 participants
29 discussions

by Tomasz Rybak

Hello. I attach patch updating pycuda.tools.DeviceData and pycuda.tools.OccupancyRecord to take new devices into consideration. I have tried to maintain "style" of those classes and introduced changes only when necessary. I have done changes using my old notes and NVIDIA Occupancy Calculator. Unfortunately I currently do not have access to Fermi to test those fully. Best regards. Tomasz Rybak

10 years, 1 month

Cannot import both pycuda and pyopencl in the same program

by Bogdan Opanchuk

Hi all, I'm observing the following behavior with latest (git-fetched today) pycuda and opencl versions on Snow Leopard 10.6.4: $ python >>> import pycuda.driver >>> import pyopencl Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/Library/Python/2.6/site-packages/pyopencl-0.92beta-py2.6-macosx-10.6-i386.egg/pyopencl/__init__.py", line 3, in <module> import pyopencl._cl as _cl AttributeError: 'NoneType' object has no attribute '__dict__' $ python >>> import pyopencl >>> import pycuda.driver Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/Library/Python/2.6/site-packages/pycuda-0.94rc-py2.6-macosx-10.6-i386.egg/pycuda/driver.py", line 1, in <module> from _driver import * AttributeError: 'NoneType' object has no attribute '__dict__' This worked with stable versions. Does anyone know why is this happening? (One may ask why do I need both libraries in the same program. I have the set of tests for my module, which can use both Cuda and OpenCL, and it is convenient to run all the tests using the single file. Although it is not a critical issue, I'm just curious). Best regards, Bogdan

10 years, 2 months

Slicing

by Thomas Wiecki

Hi, I want to run element-wise computations on different parts of an array. Loading each part of the array to device mem when needed turned out to use up a lot of time and not really speed things up compared to cpu. Instead, I want to once load the data array into device mem and provide pointers to which elements to look at (I do have the numpy view/slice of the array). I looked into different ways of doing this but can't seem to find the right approach, any help would be appreciated. ElementwiseKernel seems to support range and slicing now, however, my code is (cuda) c and I import it as a SourceModule which probably means I can't use the ElementwiseKernel approach. -Thomas

10 years, 2 months

pycuda install help

by nhieu duong

hi, ive been endlessly trying to install pycuda on a red hat dist. machine, but to no avail. it would be much appreciated if i could get some help. i am able to get past the configure part of the installation, but the when i "make" , the problem occurs. here is my siteconf.py file BOOST_INC_DIR = ['/usr/local/include/boost/'] BOOST_LIB_DIR = ['/usr/lib'] BOOST_COMPILER = 'gcc4.1.2' BOOST_PYTHON_LIBNAME = ['boost_python'] BOOST_THREAD_LIBNAME = ['boost_thread'] CUDA_TRACE = False CUDA_ROOT = '/usr/local/cuda/' CUDA_ENABLE_GL = False CUDADRV_LIB_DIR = ['/usr/lib'] CUDADRV_LIBNAME = ['cuda'] CXXFLAGS = ['-DBOOST_PYTHON_NO_PY_SIGNATURES'] LDFLAGS = [] i beleive i built boost with gcc version 4.1.2 the error im getting is..... /usr/local/include/boost/type_traits/remove_const.hpp:61: instantiated from ‘boost::remove_const<<unnamed>::pooled_host_allocation>’ /usr/local/include/boost/python/object/pointer_holder.hpp:127: instantiated from ‘void* boost::python::objects::pointer_holder<Pointer, Value>::holds(boost::python::type_info, bool) [with Pointer = std::auto_ptr<<unnamed>::pooled_host_allocation>, Value = <unnamed>::pooled_host_allocation]’ src/wrapper/mempool.cpp:278: instantiated from here /usr/local/include/boost/type_traits/detail/cv_traits_impl.hpp:38: internal compiler error: in make_rtl_for_nonlocal_decl, at cp/decl.c:5067 i only included the ends. if you want the entire thing let me know. but the error seems to point to a gcc problem. ive read thru your archives but doesnt seem to solve this problem if someone could shed some light on this issue, i would very appreciate it. thanks -nhieu

10 years, 5 months

Pycuda.GpuArray.get() failed at GeForce GT 425M

by Summit

HI,everyone. I installed the pycuda followed by the "Installing PyCuda on Windows - Windows 7 64-bit with Visual Studio Professional 2008 (Strictly Binary Versions) ". The process seems too success. The base demo(hello_demo) which not used Pycuda.GPUArray can passed. But when I run all program by pycuda.Gpuarray, I recived this exception message: " Traceback (most recent call last): File "C:\Users\summit\workspace\JCudaIP\src\pycudatest.py", line 11, in <module> a_doubled = (2*a_gpu).get() File "C:\Python27\lib\site-packages\pycuda\gpuarray.py", line 285, in __rmul__ return self._axpbz(scalar, 0, result) File "C:\Python27\lib\site-packages\pycuda\gpuarray.py", line 164, in _axpbz other, out.gpudata, self.mem_size) File "C:\Python27\lib\site-packages\pycuda\driver.py", line 273, in function_prepared_async_call func.launch_grid(*grid) pycuda._driver.LaunchError: cuLaunchGrid failed: launch out of resources " My card is "GeForce GT 425M", more detail relate to the properties of the CUDA devices : CUDA Driver Version: 3.20 CUDA Runtime Version: 3.20 CUDA Capability Major/Minor version number: 2.1 Total amount of global memory: 1041825792 bytes Multiprocessors x Cores/MP = Cores: 2 (MP) x 48 (Cores/MP) = 96 (Co res) Total amount of constant memory: 65536 bytes Total amount of shared memory per block: 49152 bytes Total number of registers available per block: 32768 Warp size: 32 Maximum number of threads per block: 1024 Maximum sizes of each dimension of a block: 1024 x 1024 x 64 Maximum sizes of each dimension of a grid: 65535 x 65535 x 1 Maximum memory pitch: 2147483647 bytes Texture alignment: 512 bytes Clock rate: 1.12 GHz Concurrent copy and execution: Yes Run time limit on kernels: Yes Integrated: No Support host page-locked memory mapping: Yes Compute mode: Default (multiple host threads can use this device simultaneously) Concurrent kernel execution: Yes Device has ECC support enabled: No Device is using TCC driver mode: No deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 3.20, CUDA Runtime Vers ion = 3.20, NumDevs = 1, Device = GeForce GT 425M My OS is Win7 x64, and VC 9.0 These problem trouble me a long time, wait for your help! Thanks

10 years, 5 months

ImportError: DLL load failed: The specified procedure could not be found

by Ilya V. Goz

I'm trying to install PyCUDA 0-94.2 under Windows 7, using MSVC 2008 Express, Python 2.6.2, boost 1.46.1 and Nvidia driver 196.21 (CUDA 3.0.1). Everything successfully compiled, however, importing pycuda does not work. Traceback (most recent call last): File "test_driver.py", line 4, in <module> from pycuda.tools import mark_cuda_test File "C:\Python26\lib\site-packages\pycuda\tools.py", line 30, in <module> import pycuda.driver as cuda File "C:\Python26\lib\site-packages\pycuda\driver.py", line 1, in <module> from pycuda._driver import * ImportError: DLL load failed: The specified procedure could not be found. Note that this isn't "The specified module..." error. I tried different boost versions, binary installers for both boost and pycuda, and I am still getting the same error. What could be the problem?

10 years, 5 months

Re: [PyCUDA] printf

by Bergtholdt, Martin

Tesla 2070C From: Bryan Catanzaro [mailto:bryan.catanzaro@gmail.com] Sent: 25 March 2011 19:07 To: Bergtholdt, Martin Cc: pycuda(a)tiker.net Subject: Re: [PyCUDA] printf What device are you running this on? - bryan On Mar 25, 2011, at 11:05 AM, "Bergtholdt, Martin" <martin.bergtholdt(a)philips.com<mailto:martin.bergtholdt@philips.com>> wrote: Hi, I'm trying to run the printf example in the wiki: >>> import pycuda.driver as cuda >>> import pycuda.autoinit >>> from pycuda.compiler import SourceModule >>> >>> mod = SourceModule(""" ... #include <stdio.h> ... ... __global__ void say_hi() ... { ... printf("I am %d.%d\\n", threadIdx.x, threadIdx.y); ... } ... """) >>> >>> func = mod.get_function("say_hi") >>> func(block=(4,4,1)) >>> Everything works fine, but I do not get any output on the last line. Any ideas what can go wrong here? Cuda 3.2 Windows-7 pycuda.VERSION (2011, 1) MSVC 9.0 2008 (32-bit) ________________________________ The information contained in this message may be confidential and legally protected under applicable law. The message is intended solely for the addressee(s). If you are not the intended recipient, you are hereby notified that any use, forwarding, dissemination, or reproduction of this message is strictly prohibited and may be unlawful. If you are not the intended recipient, please contact the sender by return e-mail and destroy all copies of the original message. _______________________________________________ PyCUDA mailing list PyCUDA(a)tiker.net<mailto:PyCUDA@tiker.net> http://lists.tiker.net/listinfo/pycuda

10 years, 5 months

printf

by Bergtholdt, Martin

Hi, I'm trying to run the printf example in the wiki: >>> import pycuda.driver as cuda >>> import pycuda.autoinit >>> from pycuda.compiler import SourceModule >>> >>> mod = SourceModule(""" ... #include <stdio.h> ... ... __global__ void say_hi() ... { ... printf("I am %d.%d\\n", threadIdx.x, threadIdx.y); ... } ... """) >>> >>> func = mod.get_function("say_hi") >>> func(block=(4,4,1)) >>> Everything works fine, but I do not get any output on the last line. Any ideas what can go wrong here? Cuda 3.2 Windows-7 pycuda.VERSION (2011, 1) MSVC 9.0 2008 (32-bit) ________________________________ The information contained in this message may be confidential and legally protected under applicable law. The message is intended solely for the addressee(s). If you are not the intended recipient, you are hereby notified that any use, forwarding, dissemination, or reproduction of this message is strictly prohibited and may be unlawful. If you are not the intended recipient, please contact the sender by return e-mail and destroy all copies of the original message.

10 years, 5 months

index multiple blocks and grids

by Mike Tischler

Hi, I'm new to CUDA and PyCUDA, and have having a problem indexing multiple grids. I'm using an older CUDA enabled card (Quadro FX 1700) before I begin writing for a larger GPU. I've been trying to understand the relationship between threads, blocks, and grids in the context of my individual card. To do so, I've set up a simple script. The following code will work just fine, printing out an array of values 0-99 ---------------------------------------------------------------------------------------------- import pycuda.gpuarray as gpuarray import pycuda.driver as drv import pycuda.autoinit def testgpu2(): from pycuda.compiler import SourceModule mod = SourceModule(""" __global__ void kernel1(float *z1) { const int i = (blockIdx.x * blockDim.x) + threadIdx.x; const int j = (blockIdx.y * blockDim.y) + threadIdx.y; z1[i*j]=i*j; } """) kernel1 = mod.get_function("kernel1") z1 = numpy.zeros((100)).astype(numpy.float32) kernel1(drv.Out(z1),block=(10,10,1),grid=(1,1)) print z1 return z1 ---------------------------------------------------------------------------------------------- However, what if I have an array that's 1024 in length? If I understand the documentation correctly, block=(16,16,1) is the max value (256 threads) allowed for my hardware, which means I have to increase the number of grids. If I change the parameters of my script to: z1 = numpy.zeros((1024)).astype(numpy.float32) kernel1(drv.Out(z1),block=(16,16,1),grid=(2,2)) How do I correctly index the array locations in my kernel function given multiple grids (z1[???]=???) ? There is a gridDim property, but not gridIdx property, like with threads and blocks. Thanks! Mike

10 years, 5 months

CURAND - Sobol initialisation

by Tomasz Rybak

Hello. I attach patch containing code that is supposed to initialise Sobol32 direction vectors. Initially I wanted to implement it myself based on article, but after looking at CUDA 4.0 (and new functions that would also need to be reimplemented) I decided just to call existing function in library. Few remarks about implementation: 1) I have added new option to configure.py and new configuration variable HAVE_LIBRARIES It might not be ideal for CURAND, but I am also thinking about additional libraries that are available in CUDA 4.0 in cudatoolssdk_4.0_linux_64.run. I believe that it might be possible to include CURAND in the PyCUDA core (and hence add dependency on libcurand), but I also believe that dependency on e.g. libcupti should be optional. In such case I would propose changing name from HAVE_LIBRARIES to HAVE_TOOLS and changing patch not to surround CURAND-related code with ifdefs. 2) I have added enum direction_vector_set; currently it has only one value, but it will have more in CUDA 4.0 - so please leave it as is. 3) Although there is ability to call function get_direction_vectors32 and to create object Sobol32* there is no code that joins those two. I am thinking whether this code should be in C or in Python - so for now please just apply patch to curand branch, and after I have good implementation I will send it to the list. 4) Sorry - no documentation for C code yet. Thanks. Best regards.

10 years, 6 months

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

PyCUDA March 2011