3 ms·

It seems kinda weird to tout how great it is that we have CuArrays and CLArrays when CLArrays haven't been updated for 1.0 and only claims experimental support

It seems kinda weird to tout how great it is that we have CuArrays and CLArrays when CLArrays haven't been updated for 1.0 and only claims experimental support for 0.6. Really hoping we see some movement on CLArrays in the near future.

keldaris 8y ago
It is clearly mentioned in a separate paragraph which reads: > For this article I'm going to choose CuArrays, since this article is written for Julia 0.7 / 1.0, which still isn't supported by CLArrays.
- eigenspace 8y ago
  My bad, I should really finish reading things before I comment.
ChrisRackauckas 8y ago
I want to see it updated too. It was really nice to write/test/debug GPU codes on my cheap laptop without an NVIDIA GPU, and then switch over to my desktop and CUDA with single line changes. Even though CuArrays tends to be quite a bit faster, this convenience cannot be overstated. I didn't realize how much I'd miss it until I was at a conference and couldn't just fiddle around with some GPU integration I had to debug. I think the reason CLArrays.jl only claims experimental support on v0.6 is because it uses Transpiler.jl which is quite limited and a direct translation. CUDANative.jl is much more advanced and uses LLVM's .ptx backend. I think Simon mentioned he wanted to do a more direct compilation route, and that would probably be the change to start calling it experimental? I don't know, I'd like to hear about the future of CLArrays as well. If we have both working strong, boy oh boy I'll be happy.
boromi 8y ago
sadly OpenCL seems to always get ignored
- pjmlp 8y ago
  They did that to themselves by sticking originally with C, and forcing the whole read text, compile, link process at runtime, instead of the multiple language support from CUDA. OpenCL has been improved since then, but now it is too late.

3 ms·

It seems kinda weird to tout how great it is that we have CuArrays and CLArrays when CLArrays haven't been updated for 1.0 and only claims experimental support

by eigenspace 8y ago

keldaris 8y ago
It is clearly mentioned in a separate paragraph which reads: > For this article I'm going to choose CuArrays, since this article is written for Julia 0.7 / 1.0, which still isn't supported by CLArrays.
- eigenspace 8y ago
  My bad, I should really finish reading things before I comment.
ChrisRackauckas 8y ago
I want to see it updated too. It was really nice to write/test/debug GPU codes on my cheap laptop without an NVIDIA GPU, and then switch over to my desktop and CUDA with single line changes. Even though CuArrays tends to be quite a bit faster, this convenience cannot be overstated. I didn't realize how much I'd miss it until I was at a conference and couldn't just fiddle around with some GPU integration I had to debug. I think the reason CLArrays.jl only claims experimental support on v0.6 is because it uses Transpiler.jl which is quite limited and a direct translation. CUDANative.jl is much more advanced and uses LLVM's .ptx backend. I think Simon mentioned he wanted to do a more direct compilation route, and that would probably be the change to start calling it experimental? I don't know, I'd like to hear about the future of CLArrays as well. If we have both working strong, boy oh boy I'll be happy.
boromi 8y ago
sadly OpenCL seems to always get ignored
- pjmlp 8y ago
  They did that to themselves by sticking originally with C, and forcing the whole read text, compile, link process at runtime, instead of the multiple language support from CUDA. OpenCL has been improved since then, but now it is too late.