Intel SYCL Compiler + Runtimes 2019-09 Released

[phoronix]

Phoronix: Intel SYCL Compiler + Runtimes 2019-09 Released

Moving towards their oneAPI beta release next quarter, the Intel developers are as busy as ever advancing their LLVM-based SYCL compiler and run-times for Windows and Linux...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-SYCL-Compiler-2019-09

[coder]

Now, they should get busy and port it to ROCm and maybe Mali or Adreno. I'm not about to adopt SYCL, so long as the only GPUs supporting it are from Intel. So, if those guys are dragging their feet on SYCL, maybe Intel should help 'em out!

Sorry, Intel, but the oneAPI that I want will support hardware products from more than oneVendor.

[boxie]

Now, they should get busy and port it to ROCm and maybe Mali or Adreno. I'm not about to adopt SYCL, so long as the only GPUs supporting it are from Intel. So, if those guys are dragging their feet on SYCL, maybe Intel should help 'em out!

Sorry, Intel, but the oneAPI that I want will support hardware products from more than oneVendor.

I was just going to ask what the state of play is with AMD/Nvidia and their OneAPI SYCL initiatives. If they can get AMD on board that would be amazing (as it is something they have been trying to get off the ground for many years).

In general I love the idea of SYCL and hope it can be a proper industry wide spec (and not a oneVendor)!

[coder]

In general I love the idea of SYCL and hope it can be a proper industry wide spec (and not a oneVendor)!

The nice thing about Intel embracing SYCL is that other vendors can adopt it, in theory. However, this should not be taken for granted. History has shown a surprising amount of complacency, in the industry.

I expect a lot of software developers will continue to target the lowest-common-denominator, which is currently OpenCL 1.2, in spite of all its shortcomings.

[chuckatkins]

If they can get AMD on board that would be amazing (as it is something they have been trying to get off the ground for many years).
In general I love the idea of SYCL and hope it can be a proper industry wide spec (and not a oneVendor)!

The sycl effort is a great idea but crippling it from the get-go to only a "subset" of C++ is a major downside and hampers a lot of the transition efforts from cuda. OpenMP v5+,which is primarily centered around targeting accelerators, has a much better likelihood of success as a "universal" programming model given the broader language support via Fortran 03, 08, C11, C++ 11, 14, and 17. Plus all of the vendors are getting behind it since it will be the preferred portable API and programming model for a fairly diverse set of GPU based exascale and pre-exascale supercomputing systems (Summit is IBM CPUs and NVidia GPUs, Aurora will be Intel CPUs and GPUs, Frontier will be AMD CPUs GPUs)

[RonanKeryell]

The nice thing about Intel embracing SYCL is that other vendors can adopt it, in theory. However, this should not be taken for granted. History has shown a surprising amount of complacency, in the industry.

I expect a lot of software developers will continue to target the lowest-common-denominator, which is currently OpenCL 1.2, in spite of all its shortcomings.

SYCL is an open standard with various vendors, academic members or individual contributors contributing. Nice to see Intel contributing an open-source version to it! There are also several non OpenCL implementations of SYCL. You should look for example at https://github.com/illuhad/hipSYCL