SegFault about scikit-cuda

[andrewcron]

I get the same issue with CULA R12, it seems fine with R11.

[lebedov]

Which test(s) are you alluding to? I'm not sure what "5th" corresponds to.

The function interfaces to the cublas library have changed in CUDA 4.0; I'm currently in the process of rewriting the Python wrappers to take this into account.

[jmcarval]

Thanks for the replies.

I get the same issue with CULA R12, it seems fine with R11.
I'm using CULA R12 too

Which test(s) are you alluding to? I'm not sure what "5th" corresponds to.
bash-4.2$ git clone https://github.com/lebedov/scikits.cuda.git
bash-4.2$ cd scikits.cuda
bash-4.2$ sudo python setup.py install
bash-4.2$ nosetests
....Segmentation fault

Is there any other command or file I can use to provide more information on this error?

Thanks

-- Edit --

bash-4.2$ nosetests -s -v
test_fft_float32_to_complex64 (test_fft.test_fft) ... ok
test_fft_float64_to_complex128 (test_fft.test_fft) ... ok
test_ifft_complex128_to_float64 (test_fft.test_fft) ... ok
test_ifft_complex64_to_float32 (test_fft.test_fft) ... ok
test_trapz2d_complex128 (test_integrate.test_integrate) ... Segmentation fault

[lebedov]

All tests seem to pass with the latest revision in github on my system. I'm using pycuda 2011.1.2, CUDA 4.0, gcc 4.6.1, and CULA R12.

[wslihgt]

Hi all,

I also have the same issue. Actually, I also found out that initializing culinalg (from the dot_demo example) in an interactive python environment (ipython in my case), then leaving python also leads to a segfault... here is what I did:

Welcome to pylab, a matplotlib-based Python environment [backend: MacOSX].
For more information, type 'help(pylab)'.

In [1]: import pycuda.autoinit

In [2]: import pycuda.gpuarray as gpuarray

In [3]: import pycuda.driver as drv

In [4]: import numpy as np

In [5]: import pycuda.autoinit, pycuda.compiler

In [6]: import scikits.cuda.linalg as culinalg

In [7]: import scikits.cuda.misc as cumisc

In [8]: culinalg.init()
c_int(0)
0

In [9]: pycuda.VERSION
Out[9]: (2012, 1)

In [11]: import scikits.cuda

In [12]: scikits.cuda.__version__
Out[12]: '0.042'

In [13]: 
Do you really want to exit ([y]/n)? 
Segmentation fault

Additionally, my CUDA and CULA are the latest, 4.2 and R15 resp..
gcc is 4.2.1, under macosx 10.6.8. Graphic card is an NVIDIA GeForce 9400M, with 256Mo (maybe this could explain the segmentation fault? not enough space there?).
My Python is the one provided by enthought (www.enthought.com) through EPD 7.2 (64 bits).

Running dot_demo also fires a segmentation fault, when calling temp_gpu = culinalg.dot(a_gpu, b_gpu). Any idea what is happening there, and if there is a "quick fix" for solving that?

Thanks in advance!

P.S.: by the way, with CULA R15, it does not seem that libcula.dylib exists, si I replaced it in cula.py by libcula_core.dylib. Not sure whether that was the right thing to do...

[lebedov]

Please reopen if this continues to be a problem with the latest code.