If you are running Mac OS 10.6, and want some of that gradient optimization and sparse matrix multiplication goodness that PyCUDA offers that (as far as I can tell) ain't in PyOpenCL, you're in for a fun time. Well, I'm being sarcastic, and not you specifically, more like me, I did most of the hair pulling, and I'd like to save you some hair.

Also, this used to be published at my blog, which was at http://alexspixels.com/, but now I'm trying out this whole "living document" thing at substance.io.

Let's go through this step by step.

You should have Python on your system and NumPy as well. I'd highly recommend downloading the Enthought Python Distribution (EPD), and forgetting about the trials of getting an optimized NumPy installation working. Just don't worry about it, download the EPD, install it, and come back when you're done. 

We need to snag all of the drivers and toolkits that NVidia provides that PyCUDA depends upon. You can find them here, at this linky link. Scroll down to the bottom of that page, under "MAC OS X", and download everything that has a "download" link next to it. Go through all the installations of everything you downloaded. Okay. Now... 

What's that mean? Well, launch Terminal.app, and you'll need to edit your bash profile. You can do this by typing

open ~/.profile  

If you're running a different shell, I'll assume you know where your profile is, and you can open it up on your own. Your profile might look like this (or it might now, who knows):


We're going to have to update this to tell the computer where all the fancy things we installed now live. Add these three lines: 

So, your bash profile should now look like something like this:

Alright, that should do it. We're ready to start rolling. 

 Open up Terminal.app, and issue these commands:

What do those lines do? The first one grabs all the code we want to install. The last line writes up some files that'll help us install everything.

Okay, that primes the pump, but we're not quite ready to install PyCUDA yet. In terminal, run:

That'll open up your favorite text editor. On my system, I use TextMate, and this is what I see:

What's there by default is wrong. We need to add a few lines for our system (otherwise, PyCUDA will complain bitterly that it's not compiling on OS 10.5, and then it'll throw up on itself and crap out). So, we add these lines: 

If you've installed Python 2.7 using Homebrew, then your flags should be something slightly different. Check it;

Thanks to Uznick for the tip!

Alrighty, now we've got to do one last thing. Even though we're grabbing PyCUDA from the freshest source we can find (the active git repository at tiker.net), we still need to do one last update. In your terminal (you didn't close it, right?), run:

In all likelihood, even if you've used git before, you've never used the "submodule" command. I hadn't either. There's some little helper packages within PyCUDA that need some attention. 

Okay, a sidenote: Lewis Anderson has informed me that on Python 2.7, he's found it necessary to force the system to compile all the C and C++ code with gcc and not clang. You can do that quite simply, by typing in your terminal:

Okay, in your terminal, type:

Simple, yeah? If there are no errors, you're going to be waiting for quite some time. I believe it's traditional to say "grab a soda," or something like that. I like orange juice, so that's what I drank. If there are any errors, please do see my little "error disclaimer" below.

When all that goodness of the make command has finished, with one set of keystrokes, enter:

You've now got PyCUDA installed on your system. Convince yourself by launching your Python interpreter (I like iPython best), and import PyCUDA.

Do a jig!

If this doesn't work for you for some reason, feel free to email me at  alex (dot) bw (at) gmail (dot) com. I'd be happy to answer questions. Getting this library installed is kind of a hassle, so anything I can do to make the world a slightly less frustrating place, well, I'd love to do it.