Hi. I built Python 2.6.4 and boost 1.40 using gcc 4.4.2.
I also installed NumPy 1.3.0.
I downloaded PyCuda 0.93, couldn't get it to install,
read the posts, and then downloaded PyCuda 0.93.1rc2.
I used the siteconf.py from the 0.93 directory, which I
think is okay:
BOOST_INC_DIR = ['/usr/common/usg/boost/1_40_0/pool/include']
BOOST_LIB_DIR = ['/usr/common/usg/boost/1_40_0/pool/lib']
BOOST_COMPILER = 'gcc442'
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/lib64']
CUDADRV_LIBNAME = ['cuda']
CXXFLAGS = []
LDFLAGS = []
However, I don't have sudo privileges on the machine
with the GPUs, so I tried "python setup.py install".
It looks like the install was okay - lots of running,
writing, copying, building, and creating lines. At the
end:
Installed
/global/common/tesla/usg/python/2.6.4/lib/python2.6/site-packages/pycuda-0.93.1rc2-py2.6-linux-x86_64.egg
Processing dependencies for pycuda==0.93.1rc2
Searching for pytools>=8
Reading http://pypi.python.org/simple/pytools/
Reading http://mathema.tician.de/software/pytools
Best match: pytools 9
Downloading
http://pypi.python.org/packages/source/p/pytools/pytools-9.tar.gz#md5=5cf7944bba89d8bf7d7187b702fbf587
Processing pytools-9.tar.gz
Running pytools-9/setup.py -q bdist_egg --dist-dir
/tmp/easy_install-LsZWTT/pytools-9/egg-dist-tmp-a3SJHd
Adding pytools 9 to easy-install.pth file
Installing logtool script to /usr/common/usg/python/2.6.4/bin
Installing runalyzer-gather script to /usr/common/usg/python/2.6.4/bin
Installing runalyzer script to /usr/common/usg/python/2.6.4/bin
Installed
/global/common/tesla/usg/python/2.6.4/lib/python2.6/site-packages/pytools-9-py2.6.egg
Finished processing dependencies for pycuda==0.93.1rc2
Am I missing something? Did I omit a step?
I then tried the first line of the tutorial:
Python 2.6.4 (r264:75706, Nov 16 2009, 18:40:23)
[GCC 4.4.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import pycuda.driver as cuda
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "pycuda/driver.py", line 1, in <module>
from _driver import *
ImportError: No module named _driver
(In one of the archive posts, the same error was
reported, but in the next post, he said that he had
resolved the error, but didn't say how.)
So then I tried running test_driver.py. The first time I
got an undefined lib file that went away after I set
my LD_LIBRARY_PATH to include
/usr/common/usg/boost/1_40_0/pool/lib
Then I tried running test_driver.py again. This is what
I got.
% python test_driver.py
Traceback (most recent call last):
File "test_driver.py", line 3, in <module>
import pycuda.autoinit
File
"/usr/common/usg/python/2.6.4/lib/python2.6/site-packages/pycuda-0.93.1rc2-py2.6-linux-x86_64.egg/pycuda/autoinit.py",
line 1, in <module>
import pycuda.driver as cuda
File
"/usr/common/usg/python/2.6.4/lib/python2.6/site-packages/pycuda-0.93.1rc2-py2.6-linux-x86_64.egg/pycuda/driver.py",
line 1, in <module>
from _driver import *
ImportError:
/usr/common/usg/boost/1_40_0/pool/lib/libboost_python.so.1.40.0:
undefined symbol: PyUnicodeUCS4_FromEncodedObject
What am I missing/need to do? The machine (x86_64) I am
working on is running Linux. The sys admin installed all
of the NVIDA and CUDA stuff (2.3). Others in the group
have been able to run CUDA programs.
Thank you for your help,
Janet
_______________________________________________
PyCUDA mailing list
[email protected]
http://tiker.net/mailman/listinfo/pycuda_tiker.net
