The libraries are not as mature but OpenCL now has an open source BLAS / FFT library from Amd and a recently updated OpenCL version of Magma so the gap is closing. All major OpenCL compilers are built on LLVM and can take advantage of this though SPIR compilation. I think a great project would be to wrap and provide a nice interface over clMagma, or to work on a common GPU array interface similar to python's compyte<https://github.com/inducer/compyte>library.
-Jake there's an opencl package at https://github.com/jakebolewski/OpenCL.jlg i am very much looking forwards to, i don't know of any reason to favour cuda at this point. there's an opencl package at https://github.com/jakebolewski/OpenCL.jlbrary, lik On Sunday, February 16, 2014 3:22:14 PM UTC-5, Dahua Lin wrote:I think what would be nice is a commI think what would be nice is a common array interface library, like on arI think what would be nice is a common array interface library, like ray interface library, like > OpenCL is definitely more open (without vendor lock-in). > > However, in practice, there are several aspects that make CUDA more > appealing for scientific computing: > > - A number of mature libraries for various computation purpose: > cuBLAS, cuFFT, cuRand, CULA, Magma, etc. > - CUDA LLVM <https://developer.nvidia.com/cuda-llvm-compiler> > > - Dahua > > > On Sunday, February 16, 2014 2:13:22 PM UTC-6, andrew cooke wrote: >> >> >> is the emphasis on cuda over opencl just an oversight? while julia + gpu >> is something i am very much looking forwards to, i don't know of any reason >> to favour cuda at this point. >> >> there's an opencl package at https://github.com/jakebolewski/OpenCL.jl >> >> andrew >> >> >> On Sunday, 16 February 2014 15:50:06 UTC-3, Mike Innes wrote: >>> >>> We've published a project ideas list for GSoC here: >>> >>> http://julialang.org/gsoc/2014/ >>> >>> We'd like our ideas page to be as healthy and diverse as possible, so >>> please do make your suggestions. Projects can include things like new >>> packages, specific language/package features, or something more >>> experimental; really, there's scope for any kind of coding project here, >>> but those which fit roughly three months of work and have a clear, tangible >>> benefit are best. >>> >>> If you maintain or use a package which is missing key features, now >>> would be a great time to ask for them! >>> >>> You're welcome to add project descriptions via github, but if you want >>> to suggest something more informally you can do so here - I'll continue to >>> write up as many as I can. >>> >>> Thanks, >>> Mike >>> >>>