Andreas,
Given your feedback, I was able to compile CUDA and at least execute a
script that simply imports a module from pycuda package.
I do not receive the traceback mentioned earlier. The way that I
"solved" this was by installing the latest Nvidia driver for my GPU
and recompiling the CUDA toolkit and PyCuda.
Unfortunately however, when I execute test_driver.py, the following is
printed to the console:
from: can't read /var/mail/__future__
import: unable to open X server `'.
import: unable to open X server `'.
from: can't read /var/mail/pycuda.tools
./test_driver.py: line 9: syntax error near unexpected token `('
./test_driver.py: line 9: `def have_pycuda():'
The lines after "can't read /var/mail/__future__" only appear if I run
these commands while there is no X server running (logged in via ssh).
If I execute this unit test while an X server is running (logged in
runlevel 3) the cursor turns to a crosshair and all I see is:
from: can't read /var/mail/__future__
However, nothing else. The test appears to hang indefinitely. If I
move the cursor around a lot and click around I receive a message that
the test was unable to capture the mouse.
When running ./test_gpuarray.py however, the following is printed to
the console regardless of how I am logged in:
====================================================================================
test session starts
=====================================================================================
platform linux2 -- Python 2.5.2 -- pytest-1.3.4
test path 1: ./test_gpuarray.py
test_gpuarray.py ...F..Aborted
Are there any suggestions?
Thank you,
Tim
On Wed, Sep 15, 2010 at 1:06 AM, Andreas Kloeckner
<[email protected]> wrote:
> Hi Tim,
>
> On Tue, 14 Sep 2010 23:24:15 -0400, "Timothy O'Keefe"
> <[email protected]> wrote:
>> New user here and anxious to get my feet wet with PyCUDA.
>>
>> I installed v0.94rc today on an Ubuntu 8.10 box w/ a Quadro FX 1700. I
>> also Installed the CUDA toolkit v3.1. Everything appeared to compile
>> fine from PyCUDA, however when I execute the driver unit test
>> `test_driver.py`, I immediately receive the following traceback:
>>
>> Traceback (most recent call last):
>> File "./test_driver.py", line 4, in <module>
>> from pycuda.tools import mark_cuda_test
>> File "/usr/lib/python2.5/site-packages/pycuda/tools.py", line 30, in
>> <module>
>> import pycuda.driver as cuda
>> File "/usr/lib/python2.5/site-packages/pycuda/driver.py", line 1, in
>> <module>
>> from _driver import *
>> ImportError: /usr/lib/python2.5/site-packages/pycuda/_driver.so:
>> undefined symbol: cuMemcpyDtoDAsync
>>
>> I apologize if the mailing list is inappropriate for reporting issues
>> such as these. If not, please advise!
>
> looks like you might've compiled PyCUDA against headers that are newer
> than the libcuda.so that is being found. Check using
>
> ldd /usr/lib/python2.5/site-packages/pycuda/_driver.so
>
> whether it finds the libcuda.so that you're expecting.
>
> HTH,
> Andreas
>
_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda
