uncomplicate / clojurecuda Goto Github PK

Please support my work on Patreon. I'll intive you to a dedicated Discord discussion server. Can't afford to donate? Ask for a free invite.

Are you planning to add any cuRAND APIs to this library? I've started a simple wrapper, in a similar style to your code, for some of my own needs, here. I'm more than happy to contribute.

Solution: simply add explicit dependency to jcuda 0.9.1 or 0.9.2 to your project

At least on Windows, jCUDA 0.9.0d (latest version as of this writing) has a exact dependency on the CUDA Toolkit v.9.0 and ClojureCUDA is therefore incompatible with Toolkit version 9.1 and 9.2.

More specifically, when running the steps from the tutorial at dragan.rocks on Windows 10, when executing the (program kernel-source) form, the following error occurs:

%APPDATA%\Local\Temp\JNvrtc-0.9.0-windows-x86_64.dll: Can't find dependent libraries

Running dumpbin /dependents JNvrtc-0.9.0-windows-x86_64.dll reveals a depency on nvrtc64_90.dll which is located in C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin\

As this seems to be an upstream (downstream?) issue, is there anything you can do?

At the very least, can you add a note about this on Windows?

The blog post says "ClojureCUDA now supports the latest CUDA 9.1."

If this could be re-licensed to use Apache 2.0 License, then a project like Apache Lucene could potentially benefit from this. For more reference, LUCENE-7745.

The info function has been moved to uncomplicate.commons.core namespace. But the rest of the info-functions are in the uncomplicate.clojurecuda.info namespace. I see 2 ways to fix this:

• update the tutorials. I can do that.
• add a stub clojurecuda.info/info which simply calls commons.core/info

So first of all I'm really excited about the stuff you're doing, thanks for sharing all this!

I was following along your tutorial at http://dragan.rocks/articles/18/Interactive-GPU-Programming-1-Hello-CUDA and had a few issues with dependencies. I thought putting them in the README could make things easier for others:

• Java 8 (Java 7 not working due to fluokitten)
• GCC 5 (Standard libraries coming with 4.9 were too old)
• CUDA Toolkit 9 (because of JCuda)

I used version 3.0 from clojars.

This works on 0.4, but doesn't on 0.5(.1)

(require '[uncomplicate.clojurecuda.core :refer :all])
(init)
(uncomplicate.commons.core/info (device 0))

StackOverflowError uncomplicate.clojurecuda.internal.impl/eval11036/fn--11039 (impl.clj:46)