On Thu, 03 Feb 2011 12:01:25 +0100, Tomasz Rybak <bogomips(a)post.pl> wrote:
> Sorry if you have received my previous mail (from 2011-02-01 22:52)
> but I have not got any reply - so here I am sending it again.
> I have seen your email from 2010-12-26. I was in Germany, so I was
> able to send reply on 2011-01-11. I have fixed code and removed
My work on your scan implementation is on the scan-from-tomasz branch
. The code as you submitted it was not runnable (it used symbols that
weren't imported). I've fixed that, moved it to a separate file and
matched the code's style to the rest of PyCUDA. I've also folded all the
tests into one simpler routine.
- It's still not working, the tests give large errors.
- Ultimately, the code performs pretty complicated code generation. I
wonder if it might not give more maintainable code to simply use a
template engine. (Mako would be the preferred one for the job.)
Thanks for your submission--I'm looking forward to your comments.
Thank you for quick reply before. This was very helping, I now run things
correctly on that example.
But here is one new question I have. I am trying example
So the error is:
error: identifier "uint" is undefined
(*full error trace is put after email*)
I am trying replacing "uint" in template file by other things, but dont find
something works. During the different tries, either it is give same
"undefined" error or otherwise fails final assertion test (wasn't computed
correctly). So what should I do? Some problem with installation of nvcc?
Sorry for not understanding! (I am immediately new to these thing.)
Traceback (most recent call last):
File "DemoMetaMatrixmulCheetah.py", line 145, in <module>
mat_c, gpu_time = matrixmul_opt(mat_a, mat_b)
File "DemoMetaMatrixmulCheetah.py", line 85, in matrixmul_opt
module = compiler.SourceModule(kernel_code)
line 238, in __init__
arch, code, cache_dir, include_dirs)
line 228, in compile
return compile_plain(source, options, keep, nvcc, cache_dir)
line 107, in compile_plain
cmdline, stdout=stdout, stderr=stderr)
pycuda.driver.CompileError: nvcc compilation of
[command: nvcc --cubin -arch sm_11 -m64
kernel.cu(31): error: identifier "uint" is undefined
kernel.cu(32): error: identifier "uint" is undefined
kernel.cu(35): error: identifier "uint" is undefined
kernel.cu(36): error: identifier "uint" is undefined
kernel.cu(39): error: identifier "uint" is undefined
kernel.cu(41): error: identifier "uint" is undefined
kernel.cu(43): error: identifier "uint" is undefined
kernel.cu(46): error: identifier "uint" is undefined
kernel.cu(48): error: identifier "uint" is undefined
kernel.cu(51): error: identifier "uint" is undefined
kernel.cu(58): error: identifier "uint" is undefined
kernel.cu(59): error: identifier "uint" is undefined
12 errors detected in the compilation of
I'm trying to install PyCUDA in my W7 64 bits computer with Enthought 64
bits too. I have followed the instructions shown in
http://wiki.tiker.net/PyCuda/Installation/Windows for Vista 64 bits and MVS
2008, but I have a problem in the final step.
When I have tried to install pycuda, a message apperared in the screen that
said "error: command 'gcc' failed with exit status 1". Another one that I
think it's important is "gcc.exe: EHsc: No such file or directory".
What should I do for a right installation of PyCUDA?
Francisco Javier García Fernández
looking to install pyCUDA on an Ubuntu box for some academics
interested in investigating its capabilities.
Have a fairly basic Ubuntu 10.04 system without direct access
for pulling in required packages.
Fell foul of the "DistributeVsSetuptools" approach that's been
taken which isn't mentioned until you go looking for it.
Do now have the Python distribute package installed and have seen
most of pyCUDA compile but am seemingly being told it's
Processing dependencies for pycuda==0.94.2
Searching for decorator>=3.2.0
which doesn't seem to be a dependency that one gets informed about
anywhere other than in the build process itself?
Now I am sure I can download the package and build it stand-alone
but is that likely to be a dependency that can be satisfied by a
package installable through the Ubuntu pacjage-management system.
Furthermore, are there going to be similar python dependencies
that are not explicitly listed as dependencies anywhere? A full
list would be most helpful
Any info welcome,
Kevin M. Buckley Room: CO327
School of Engineering and Phone: +64 4 463 5971
Victoria University of Wellington
Is there any way to hold persistent storage of data on the gpu so that we
don't have to repeatedly the data sent each time.
If I do the next for example :
a_gpu = gpuarray.to_gpu(a)
b_gpu = gpuarray.to_gpu(b)
c_gpu = gpuarray.to_gpu(c)
temp_gpu = culinalg.dot(a_gpu, b_gpu)
d_gpu = culinalg.dot(temp_gpu, c_gpu)
To get the var d_gpu wa the matrix temp_gpu resent to the GPU or the result
that already exists on the GPU was used?
View this message in context: http://pycuda.2962900.n2.nabble.com/Persistent-storage-on-GPU-tp6004242p600…
Sent from the PyCuda mailing list archive at Nabble.com.
I'm have some problem with my testing of pycuda, which I don't understand.
I want to understand example of how to pointwise multiply large vector (like
10000 length) use pycuda.
So I am trying the code from here:
about "multiply_them". It work correctly for length 400 vector, but I
replace 400 with larger number, like 550, and now immediately it is not
Here is error trace:
Traceback (most recent call last):
File "test3.py", line 23, in <module>
line 166, in function_call
pycuda._driver.LogicError: cuFuncSetBlockShape failed: invalid value
I am use OSX 10.6, with python 2.7, most recent pycuda, and cuda.
So what is wrong? Is my idea wrong to use larger values, should I split up
problem more into smaller piece?
I may have missed something but I'm getting errors like the following
when running py.test on the git HEAD.
from pytools import match_precision
ImportError: cannot importa name match_precision
The following solved the problem:
diff --git a/setup.py b/setup.py
index 517f127..d4152b8 100644
@@ -316,7 +316,7 @@ def main():
packages=["pycuda", "pycuda.gl", "pycuda.sparse"],