> Dissecting PyCuda might give you a headstart :) from everything I can tell, jcuda (http://www.jcuda.de/jcuda/JCuda.html) already contains everything PyCUDA does. That is, both require you to write your GPU kernels in C++. This is what I'm trying to avoid with CUDA Clojure. I want to write my GPU kernels in Clojure.
At this point I have the following roadmap I think I'll start on soon: 1) Wrap jCuda - take the java cuda libs and make them a bit more clojure friendly 2) Integrate Ocelot - this will allow people without a NVIDIA GPU to at least play around with CUDA Clojure (http://code.google.com/p/gpuocelot/) 3) Allow very simple 1:1 Clojure->CUDA translation. That is, you'll use mutable data, and static functions, with tagged function arguments to create CUDA code 4) Implement type inference. CUDAClojure functions will not be able to change their types, but their types will be inferred. Look at something like F# to see how this works 5) Implement deftype and protocol translation 6) Start implementing core.clj functions The nice thing is, 1-3 are fairly simple, and so the project should get off the ground quickly and easily. I'm planning on using Ocelot quickly for one reason....my laptop doesn't have a Nvidia gpu :-(. >From there, it's just a matter of implementing layer after layer. Timothy -- “One of the main causes of the fall of the Roman Empire was that–lacking zero–they had no way to indicate successful termination of their C programs.” (Robert Firth) -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
