Dear Pradeep,

to participate in the Summer of Code, you have to write a proposal and
submit it before the deadline.
https://developers.google.com/open-source/gsoc/timeline
Before writing the proposal, you may want to look at tips on our website:
https://github.com/STEllAR-GROUP/hpx/wiki/Hints-for-Successful-Proposals
There is not a lot of time for writing the proposal but I think that with
your background in OpenCL, it should not be hard to understand basic
concepts of SYCL and prepare a good project proposal. I'm cc-ing this to
the mailing list, please keep the conversation on the public list. I'm
traveling a lot right now and I might reply to messages with a significant
delay. Hopefully, other community members will help you with questions and
doubts.

The main goal of the SYCL project is to update the existing implementation
and extend it with a full support for parallel algorithms.
1. The first important task is to provide a C++ standard layout
implementation of std::tuple and integrate it with HPX. This step is
crucial since it allows direct capture of such tuple in kernel code. The
currently used hpx::tuple does not work because of multiple inheritance.
There are different ways of implementing it, ComputeCpp devs promised
publishing their tuple implementation but I don't think they did it.
3. This should be used to write a generic executor which can accept tuple
of any type and transform compute iterators to SYCL pointers. This would
replace the manual tuple transformation in the current executor.
https://github.com/mcopik/hpx/tree/compute_sycl/hpx/compute/sycl
4. Several new algorithms should be tested and evaluated, with a focus on
hpx::for_loop.
5. SYCL API does not support a non-blocking copy operations. It could be
emulated with less obvious approaches such as enqueueing copy kernel on the
default host queue. This should be the basis for an alternative
implementation of copy operations.
6. Furthermore, the existing code should be extended with a well-tested
support for multiple devices.

Best,
Marcin

czw., 15 mar 2018 o 15:10 użytkownik Pradeep Kumar <[email protected]>
napisał:

> Dear Mr. Copik,
>     I am Pradeep Kumar S. doing my 3rd year of 5 year integrated M.Sc.
> Theoretical Computer Science at PSG College of Technology, Coimbatore,
> India. I am interested to work on SYCL backend for HPX Compute. I am
> working on optimising algorithms for Nvidia GPU architecture. I have
> experience in working on CUDA and OpenCL. Recently, I started learning LLVM
> in the view of generating efficient PTX. I also watched C++ Summer Lecture
> Series online by the Stellar group. Please provide me guide lines on where
> to start. I am interested in contributing to the HPX library.
>
> Regards,
> Pradeep Kumar.
>
_______________________________________________
hpx-users mailing list
[email protected]
https://mail.cct.lsu.edu/mailman/listinfo/hpx-users

Reply via email to