Hey, I am one of the maintainers of OpenCL.jl and I would recommend you to look at https://github.com/JuliaGPU/OpenCL.jl, https://github.com/JuliaGPU/CLFFT.jl, and https://github.com/JuliaGPU/CLBLAS.jl. We welcome any improvements and help. I can only speak for myself in that most of the improvements happen when somebody feels that the current state is inadequate and something needs to be done about it. I am personally not using OpenCL as much as I used to so I have less of these moments :) Otherwise look at the issues and if you want I have a personal wishlist of things that need to be done at some point ;)
Best, Valentin On Wednesday, 27 April 2016 03:36:27 UTC+9, Michael Jin wrote: > > I've noticed that there are computer owners who only have access to an > AMD GPU (Mac Pro 2013, 2015 iMac) and wish to tap into their GPUs for extra > processing power for machine learning applications. > > cuBLAS won't help them and it appears that the state of out of the box > (without much customizations required by the user) OpenCL tool development > seems to lacking progress. I wish to change that. > > On Tuesday, April 26, 2016 at 7:20:52 AM UTC-4, Chris Rackauckas wrote: >> >> I think the GPU integration libraries for Julia are already really good >> if you're using CUDA. CUDArt.jl and Arrayfire.jl work quite well. I don't >> know too much about the OpenCL side, but I don't tend to have a use for it. >> >> On Monday, April 25, 2016 at 6:06:10 PM UTC-7, Michael Jin wrote: >>> >>> (Reposted from julia-dev I was told that julia-user was a >>> more appropriate place to have this thread.) >>> >>> Hi, I'm an undergraduate student and I've been using Julia since 2013. >>> I've been trying to use the GPU seamlessly for projects involving Julia >>> matrices. For that end, I have started working on my own OpenCL BLAS Julia >>> library to test the clBLAS library at the lowest level possible for the GPU >>> with the OpenCL C library. >>> >>> Here's a link to my project: https://github.com/mikhail-j/OpenCLBLAS.jl >>> >>> This project has been tested on a NVIDIA GTX 780 Ti. >>> >>> Any suggestions on what I can do to improve the state of GPU integration >>> with Julia? >>> >>
