Hi Karl,
First of all, I thank you very much for your time and efforts again. I also
thank you very much for the link to your blog.
1. The integer issue is related to the array indexing. I would not know the
type of integer that was used in the array indexing. In C/C++ you can use
any C-type integer and I do not know whether the Boost is mapping the
intrinsic C-types to python, completely. This is the portion of the python
code I still have from PyViennaCL test:
if (a != 0.0):
ki = int (GI)
kj = int (GJ)
self.matrix[ki, kj] += float(a)
As you can see, it is very frustrating to find the correct type of integer
for the indices. There is also an effort to make the np.float64 to python
float.
2. I agree that there shouldn't be any problem having both openCL and
PyViennaCL at the same time in the same system, but I did not know what the
problem was. However, eliminating uncertainties always isolates core of the
issue. Now I have installed and tested the pyopenCL, and confirm that it is
working.
The next step is reinstalling PyViennaCL (and probably sending you error
messages again!) :o)
Have a nice evening.
Kind regards,
Berat
2015-05-23 19:41 GMT+02:00 Karl Rupp <[email protected]>:
> Hi Berat,
>
> > 1. Error message : The same as the one with the double precision
>
>> arithmetic even if it was np.float32. Then I thought that perhaps there
>> is no compatibility between ctype float and numpy float. I got a similar
>> message with the np.int64 in the same routine as pyvienna did not accept
>> either np.int64 or np.int32. I resorted to python and casted all my
>> integer variables with int(). pyvienna accepted python int.
>>
>
> Hmm, I'm not sure what you are referring to. 'float' (32 bit) and 'double'
> (64 bit) are standardized data types, so unlike compiler-specific lengths
> of 'long' and 'int', there should be no issues.
>
>
> 2. the print gave a "*None*" error message
>>
>> I just wanted to be sure that my devices are compatible with opencl and
>> I downloaded pyOpenCL today. I run a simple test and pyOpenCL detected
>> all of my devices.
>>
>
> Ok, this should be fine. If there's an issue with the OpenCL installation,
> one usually doesn't even get to this step.
>
>
> I just wonder if you could provide - for everybody - a simple python
>> code that checks and verifies the installation. It may include the name
>> of device(s), platform compatibility, etc. If this test succeeds, then
>> the issue lies in pyvienna.
>>
>
> You could just try the steps from my blog post here:
>
> https://www.karlrupp.net/2014/02/pyviennacl-gpu-accelerated-linear-algebra-for-python/
>
> This is verified to work, so you should be able to reproduce.
>
>
>> In order to install the pyOpenCL, I downloaded pyvienna at the moment.
>>
>
> Hm? What are you trying to achieve? You don't need PyViennaCL in order to
> use/install PyOpenCL...
>
> Best regards,
> Karli
>
>
>
------------------------------------------------------------------------------
One dashboard for servers and applications across Physical-Virtual-Cloud
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
_______________________________________________
ViennaCL-support mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/viennacl-support
