Hi there,
I was wondering if anyone had had any success with PyCUDA on Snow
Leopard. I've installed all the CUDA drivers and Boost-1.39. (I had
to manually use install_name_tool to relink the Boost library against
the framework build of Python I made.) However, on startup I get:
[15:44][...cuda-0.93]$ python2.6 -c "import pycuda.autoinit"
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "pycuda/autoinit.py", line 1, in <module>
import pycuda.driver as cuda
File "pycuda/driver.py", line 1, in <module>
from _driver import *
ImportError: dlopen(pycuda/_driver.so, 2): no suitable image found. Did find:
pycuda/_driver.so: mach-o, but wrong architecture
Checking the _driver.so file, I get:
({1})[15:44][...cuda-0.93]$ file pycuda/_driver.so
pycuda/_driver.so: Mach-O bundle i386
And checking python2.6, I get:
[15:44][...cuda-0.93]$ file `whence python2.6`
/Library/Frameworks/Python.framework/Versions/Current/bin//python2.6:
Mach-O universal binary with 2 architectures
/Library/Frameworks/Python.framework/Versions/Current/bin//python2.6
(for architecture i386): Mach-O executable i386
/Library/Frameworks/Python.framework/Versions/Current/bin//python2.6
(for architecture x86_64): Mach-O 64-bit executable x86_64
I thought this could be the i386/x86_64 problem (and the CUDA forums
assure me that OSX does not yet have 64-bit libraries for CUDA) but
loading python in i386 arch mode doesn't help at all:
[15:44][...cuda-0.93]$ arch -i386 python2.6 -c "import pycuda.autoinit"
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "pycuda/autoinit.py", line 1, in <module>
import pycuda.driver as cuda
File "pycuda/driver.py", line 1, in <module>
from _driver import *
ImportError: dlopen(pycuda/_driver.so, 2): no suitable image found. Did find:
pycuda/_driver.so: mach-o, but wrong architecture
Has anyone been able to get PyCUDA to run on Snow Leopard? Could
something be wrong with my installation? I built my own framework
build of Python, linking against the 10.6 SDK but using both i386 and
x86_64 arches. Any ideas?
Thanks!
-Matt
_______________________________________________
PyCUDA mailing list
[email protected]
http://tiker.net/mailman/listinfo/pycuda_tiker.net
