Future plans for cl.oquence
7 views
Skip to first unread message
Cyrus Omar
Feb 24, 2011, 11:46:34 PM2/24/11
to ahh-d...@googlegroups.com
Hi all,
Here is whats going on with cl.oquence over the next few months.
1. I am separating out the cl.oquence device-side language from the host run-time library and creating a standalone compiler for cl.oquence code so it can be integrated into any host language with an OpenCL binding. cl.oquence will only use Python as a compile-time specification language -- so you can still do a lot of the fun stuff (inlining and specialization). But it will be orthogonal to the run-time features completely.
2. I will be writing CUDA and plain-C backends for this compiler. There won't be PyCUDA/Weave bindings from me, but that could be an interesting little project for someone.
3. I am writing a paper that will be submitted to Supercomputing in April about the language design. I'm hopeful that it will be accepted but it might not be, in which case I'll submit to IPDPS in August or so.
4. Before SC in November I will do a 1.0 release of both the standalone language and the Python bindings for it. Expect a beta by June. I'll do a round of heavy advertising (reddit, HN, various relevant mailing lists, etc.) sometime close to that point and expect to get at least a few dozen regular users from that.
5. I am moving the website to http://cl.oquence.org/. The cl.oquence.com, cloquence.org and cloquence.com domains names will also forward to that website. It will likely be redesigned a tiny bit (I need a new logo at least). I am dropping the atomic-hedgehog name entirely (we'll have a new mailing list at some point.)
So by the end of the year you'll have a 1.0 language with a standalone compiler, nice PyOpenCL-based Python bindings, a clean website, a paper (hopefully), complete API documentation, and hopefully some users. Bad news is that I will stop working on new core features at that point. I will keep fixing bugs and responding to support questions on the mailing list and I am willing to mentor anyone who wants to go further with the language design, but don't expect fancy new features beyond the nice ones I've put in there now from me. I'll put some project suggestions on the website in case someone in the OSS community wants to go further.
Also depending on how easy it is, I might do a little mini-startup that owns the IP (alongside CMU) and will relicense the LGPL code under a propietary commercial license if there is any interest. I haven't thought a ton about this so don't quote me.
Let me know if there are any questions or concerns!
Cyrus
Here is whats going on with cl.oquence over the next few months.
1. I am separating out the cl.oquence device-side language from the host run-time library and creating a standalone compiler for cl.oquence code so it can be integrated into any host language with an OpenCL binding. cl.oquence will only use Python as a compile-time specification language -- so you can still do a lot of the fun stuff (inlining and specialization). But it will be orthogonal to the run-time features completely.
2. I will be writing CUDA and plain-C backends for this compiler. There won't be PyCUDA/Weave bindings from me, but that could be an interesting little project for someone.
3. I am writing a paper that will be submitted to Supercomputing in April about the language design. I'm hopeful that it will be accepted but it might not be, in which case I'll submit to IPDPS in August or so.
4. Before SC in November I will do a 1.0 release of both the standalone language and the Python bindings for it. Expect a beta by June. I'll do a round of heavy advertising (reddit, HN, various relevant mailing lists, etc.) sometime close to that point and expect to get at least a few dozen regular users from that.
5. I am moving the website to http://cl.oquence.org/. The cl.oquence.com, cloquence.org and cloquence.com domains names will also forward to that website. It will likely be redesigned a tiny bit (I need a new logo at least). I am dropping the atomic-hedgehog name entirely (we'll have a new mailing list at some point.)
So by the end of the year you'll have a 1.0 language with a standalone compiler, nice PyOpenCL-based Python bindings, a clean website, a paper (hopefully), complete API documentation, and hopefully some users. Bad news is that I will stop working on new core features at that point. I will keep fixing bugs and responding to support questions on the mailing list and I am willing to mentor anyone who wants to go further with the language design, but don't expect fancy new features beyond the nice ones I've put in there now from me. I'll put some project suggestions on the website in case someone in the OSS community wants to go further.
Also depending on how easy it is, I might do a little mini-startup that owns the IP (alongside CMU) and will relicense the LGPL code under a propietary commercial license if there is any interest. I haven't thought a ton about this so don't quote me.
Let me know if there are any questions or concerns!
Cyrus
M.Rule
Feb 25, 2011, 8:33:27 AM2/25/11
to ahh-d...@googlegroups.com, Cyrus Omar
I'm not sure I completely understand [1]. I would welcome a plain C
backend for the AHH simulation framework. Its a good API for networked
neurons even without the GPU acceleration.
backend for the AHH simulation framework. Its a good API for networked
neurons even without the GPU acceleration.
> --
> 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.
>
Reply all
Reply to author
Forward
0 new messages