Re: [ahh-dev] cl.oquence

[Cyrus Omar]

I'm pretty sure the pyopencl setup script adds the "lib" prefix? You can look through setup.py I think. Make sure you're following everything on the pyopencl wiki for installing on Windows to a tee.

Before running my test scripts, try to run the test on the pyopencl site. If that works, but one of my test scripts does not, let me know which one it is and if there is any error message (there might not be), what it says.

Cyrus

P.S. These kinds of things should be on ahh-discuss so others with similar problems can find them in the future.

On Sun, Jun 13, 2010 at 20:23, Rick Gerkin <rge...@gmail.com> wrote:

Cyrus,  

I tried installing on Windows, and had one snag, which was that bjam creates Boost libraries with names like "libboost..." while pyopencl wants boost libraries with names like "boost...".  Even after setting the siteconf.py to use "libboost..." type files, the pyopencl build script still somehow sneaks a "boost..." named library in there, so compilation fails unless I just duplicate one of my "libboost..." libraries and rename it to be a "boost..." library.  Anyway, then everything works except that python.exe crashes when I try to run your test scripts.  I didn't bother checking to see where this happens, because the fact that I had to kludge to get pyopencl to compile is a bad sign in and of itself.  Anyway, I guess I'll abandon Windows for now and run in Linux like everyone else.  Although I will probably fail to compile it correctly there, too.  

Rick  

On Wed, May 26, 2010 at 10:09 PM, Cyrus Omar <cyo...@gmail.com> wrote:

cl.oquence is in it's own branch named cl.oquence right now, so you should be able to switch to it (I forget the hg command offhand). I'll merge it once I have a few more tests written.

You need pyopencl and numpy, and a OpenCL platform. Probably the nvidia one for gpu or AMD has an x86 one.

Haven't tested on windows but I can't think of any reason why it wouldn't work. 64-bit Linux and osx does work.

It'd be great to see some analysis code. See if you can get it running, if not I can try to help. I can also set you up a sage acct on tesla if you want.

On May 26, 2010, at 19:26, Rick Gerkin <rge...@gmail.com> wrote:

I read the .pdf and did the Hg checkout.  The default checkout is version 102, btw, which does not have cl.oquence (bad ass name btw).  Before I go further, a few questions:  

What else do I need to install (compilers, APIs, python libraries, etc.).  

Have you tested this in Windows (sorry).  

Have you tested it on 64-bit CPUs/OSs?  

If I can get things working, I can probably contribute analysis code.  That doesn't tax my python skills too much, the ways analysis is parallelized is more intuitive than simulation, and I have a ton of pseudo-code already written for pretty much everything you would ever analyze in spiking neurons.  

Rick  

On Wed, May 26, 2010 at 4:51 PM, Cyrus Omar <cy...@cmu.edu> wrote:

Check it, it works now. Here's a presentation I gave you might wanna read through.

--
You received this message because you are subscribed to the Google Groups "ahh-dev" group.
To post to this group, send email to ahh...@googlegroups.com.
To unsubscribe from this group, send email to ahh-dev+u...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/ahh-dev?hl=en.

--
Richard C. Gerkin
Postdoctoral Fellow, Dept. of Biological Sciences
Carnegie Mellon University

--
Richard C. Gerkin
Postdoctoral Fellow, Dept. of Biological Sciences
Carnegie Mellon University

[Rick Gerkin]

OK, I changed bjam.exe so it creates all possible libraries, which resulted in me getting the ones I needed.  Now all compiles with no errors, although I still cannot import pyopencl (however, at least it doesn't crash python now).  Somehow the _cl.pyd file doesn't have the dependencies it needs to work, but I'm not sure if this is a 64-bit issue or what.  I've posted a message to the pyopencl mailing list.  

Rick  

[Cyrus Omar]

Yeah, I have no idea why it doesn't work if you followed the
directions on the pyopencl wiki and got no errors. Once you figure
that out, ahh stuff should work.

In the meantime, want a sage account setup on the tesla machine to play with it?

> You received this message because you are subscribed to the Google Groups "ahh-discuss" group.
> To post to this group, send email to ahh-d...@googlegroups.com.
> To unsubscribe from this group, send email to ahh-discuss...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/ahh-discuss?hl=en.
>

[Rick Gerkin]

I already have one, and really I should just be using that since you have it working there.  One of the reasons I want it working locally is for web development... bear with me... I started using this new web framework called Django which is like Ruby on Rails but for Python.  I could use it to easily make a web front-end for whatever python code I want, so we could build a model repository in which visitors could play with model parameters and visualize how results change.   

Rick  

[Cyrus Omar]

You can do a lot of that with sage using the @interact command, its pretty nice. Have you played with it?

[Rick Gerkin]

Which of the many python installations is sage using right now on Tesla?  I notice that several users seem to have their own.  Which of the dozens of site-packages folders are in the path of that installation?  The sagenb user apparently does not have pyopencl in the path of its installation.  I presume  if you have gotten it to work you were running sage as some other user that had pyopencl in its path.  

Rick