Google Looks To Open Up StreamExecutor To Make GPGPU Programming Easier

[phoronix]

Phoronix: Google Looks To Open Up StreamExecutor To Make GPGPU Programming Easier

Google developers are looking at starting a new LLVM sub-project around parallel runtime and support libraries for GPUs, CPUs, and other platforms. As part of it, they are also looking to open-source their StreamExecutor that wraps around the CUDA and OpenCL runtimes...

http://www.phoronix.com/scan.php?page=news_item&px=Google-StreamExec-Parallel

[kloumpt]

Why do we need StreamExecutor when we have collenchyma, the Rust library that powers the fastest Maching Learning framework and allows easy hardware agnostic parallel computations ? (((((:

https://github.com/autumnai/collenchyma

[kloumpt]

Why do we need StreamExecutor when we have collenchyma, the Rust library that powers the fastest Maching Learning framework and allows easy hardware agnostic parallel computations ? (((((:

https://github.com/autumnai/collenchyma

[darkbasic]

* abstracts the underlying accelerator platform (avoids locking you into a single vendor, and lets you write code without thinking about which platform you'll be running on).

Isn't OpenCL alone more than enough to avoid getting stuck in such issues?

[rob11311]

Isn't OpenCL alone more than enough to avoid getting stuck in such issues?

I think the problem is that CUDA is an entrenched proprietary de-facto standard in areas like HPC.
Being able to use CUDA rather than OpenCL is likely reassuring.

[rob11311]

Why do we need StreamExecutor when we have collenchyma, the Rust library that powers the fastest Maching Learning framework and allows easy hardware agnostic parallel computations ? (((((:
https://github.com/autumnai/collenchyma

How many Rust compilers/interpeters can you use? What large projects outside Mozilla use Rust?
What major piece of software by a well resourced company is written & supported in Rust?
How much software & other languages interface with C?

Answer these questions and you'll know why this makes sense to Google

[rob11311]

Why do we need StreamExecutor when we have collenchyma, the Rust library that powers the fastest Maching Learning framework and allows easy hardware agnostic parallel computations ? (((((:
https://github.com/autumnai/collenchyma

Rust isn't well proven and mature enough (yet) to stop ppl targetting other languages

[elldekaa]

Their StreamExecutor seems indeed quite close to the OpenMP offloading library (libomptarget), currently under review:

http://reviews.llvm.org/D14031