Hi Frank, Oh ok, only the examples fail, and the error messages are pretty clear. I thought you were speaking of the automated tests, which definitely should run on every card. The examples serve mainly as a place to copy code from, and the errors below should be easy enough to fix.
Thanks for sending this, though!
Andreas
On Dienstag 07 Juli 2009, you wrote:
> Thank you for creating PyCUDA. Here is the output from dump_properties.py,
> and some of the demos that had issues:
>
> Device #0: Quadro NVS 135M
> Compute Capability: 1.1
> Total Memory: 130752 KB
> CAN_MAP_HOST_MEMORY: 0
> CLOCK_RATE: 800000
> COMPUTE_MODE: DEFAULT
> GPU_OVERLAP: 1
> INTEGRATED: 0
> KERNEL_EXEC_TIMEOUT: 1
> MAX_BLOCK_DIM_X: 512
> MAX_BLOCK_DIM_Y: 512
> MAX_BLOCK_DIM_Z: 64
> MAX_GRID_DIM_X: 65535
> MAX_GRID_DIM_Y: 65535
> MAX_GRID_DIM_Z: 1
> MAX_PITCH: 262144
> MAX_REGISTERS_PER_BLOCK: 8192
> MAX_SHARED_MEMORY_PER_BLOCK: 16384
> MAX_THREADS_PER_BLOCK: 512
> MULTIPROCESSOR_COUNT: 1
> TEXTURE_ALIGNMENT: 256
> TOTAL_CONSTANT_MEMORY: 65536
> WARP_SIZE: 32
>
>
>
> C:\downloads\cuda\pycuda\examples>matrix-transpose.py
> Traceback (most recent call last):
> File "C:\downloads\cuda\pycuda\examples\matrix-transpose.py", line 218,
> in <module>
> run_benchmark()
> File "C:\downloads\cuda\pycuda\examples\matrix-transpose.py", line 176,
> in run_benchmark
> target = gpuarray.empty((size, size), dtype=source.dtype)
> File
> "c:\python25\lib\site-packages\pycuda-0.94beta-py2.5-win32.egg\pycuda\gpuar
>ray.py", line 81, in __init__
> self.gpudata = self.allocator(self.size * self.dtype.itemsize)
> pycuda._driver.MemoryError: cuMemAlloc failed: out of memory
>
>
>
> C:\downloads\cuda\pycuda\examples>select-to-list.py
> kernel.cu
> tmpxft_000003bc_00000000-3_kernel.cudafe1.gpu
> tmpxft_000003bc_00000000-8_kernel.cudafe2.gpu
> kernel.cu
> tmpxft_0000073c_00000000-3_kernel.cudafe1.gpu
> tmpxft_0000073c_00000000-8_kernel.cudafe2.gpu
> ptxas
> C:\DOCUME~1\fbuckle\LOCALS~1\Temp/tmpxft_0000073c_00000000-4_kernel.ptx,
> line 103; error : Shared-space reduction operations require SM 1.2 or
> higher
> ptxas fatal : Ptx assembly aborted due to errors
> Traceback (most recent call last):
> File "C:\downloads\cuda\pycuda\examples\select-to-list.py", line 100, in
> <module>
> """ % {"block_size": block_size, "el_per_thread": el_per_thread})
> File
> "c:\python25\lib\site-packages\pycuda-0.94beta-py2.5-win32.egg\pycuda\compi
>ler.py", line 180, in __init__
> arch, code, cache_dir, include_dirs)
> File
> "c:\python25\lib\site-packages\pycuda-0.94beta-py2.5-win32.egg\pycuda\compi
>ler.py", line 170, in compile
> return compile_plain(source, options, keep, nvcc, cache_dir)
> File
> "c:\python25\lib\site-packages\pycuda-0.94beta-py2.5-win32.egg\pycuda\compi
>ler.py", line 79, in compile_plain
> raise CompileError, "nvcc compilation of %s failed" % cu_file_path
> pycuda.driver.CompileError: nvcc compilation of
> c:\docume~1\fbuckle\locals~1\temp\tmpfbxzhp\kernel.cu failed
>
> C:\downloads\cuda\pycuda\examples>measure_gpuarray_speed_random.py
> 1024
> 2048
> 4096
> 8192
> 16384
> 32768
> 65536
> 131072
> 262144
> 524288
> 1048576
> 2097152
> 4194304
> 8388608
> 16777216
> Traceback (most recent call last):
> File
> "C:\downloads\cuda\pycuda\examples\measure_gpuarray_speed_random.py", line
> 88, in <module>
> main()
> File
> "C:\downloads\cuda\pycuda\examples\measure_gpuarray_speed_random.py", line
> 39, in main
> curandom.rand((size, ))
> File
> "c:\python25\lib\site-packages\pycuda-0.94beta-py2.5-win32.egg\pycuda\curan
>dom.py", line 182, in rand
> result = GPUArray(shape, dtype)
> File
> "c:\python25\lib\site-packages\pycuda-0.94beta-py2.5-win32.egg\pycuda\gpuar
>ray.py", line 81, in __init__
> self.gpudata = self.allocator(self.size * self.dtype.itemsize)
> pycuda._driver.MemoryError: cuMemAlloc failed: out of memory
>
>
> On Mon, Jul 6, 2009 at 11:29 AM, Andreas Klöckner
>
> <[email protected]>wrote:
> > On Samstag 04 Juli 2009, Frank Buckle wrote:
> > > I am new to CUDA, and just have a little python experience. I was able
> >
> > to
> >
> > > get the PyCUDA demos to run on Windows using MinGW. I added to the
> >
> > Windows
> >
> > > installation instructions at
> > > http://wiki.tiker.net/PyCuda/Installation/Windows.
> > >
> > > I am not sure if there are any python or PyCUDA issues that might arise
> > > from instructing distutils to use the mingw32 compiler. The demos
> > > appear to work OK, except for some out of memory errors that I think
> > > are related to my puny laptop graphics card. For now, MinGW seems to
> > > be a decent option.
> > >
> > > Let me know if you have any questions, or can foresee any issues with
> >
> > this
> >
> > > option.
> >
> > Sweet! Thank you for taking the time to write this up.
> >
> > Also, I'd be happy if you could copy and paste the output of the
> > tests--I'd like to see which ones I need to fix to run with smaller
> > cards.
> >
> > Andreas
> >
> >
> > _______________________________________________
> > PyCUDA mailing list
> > [email protected]
> > http://tiker.net/mailman/listinfo/pycuda_tiker.net
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ PyCUDA mailing list [email protected] http://tiker.net/mailman/listinfo/pycuda_tiker.net
