Intel Open-Sources Its OpenCL CPU-Based Runtime

[phoronix]

Phoronix: Intel Open-Sources Its OpenCL CPU-Based Runtime

As 718,996 lines of newly open-sourced code, Intel recently began opening up their previously proprietary CPU-based OpenCL run-time...

https://www.phoronix.com/news/Intel-OpenCL-CPU-Open-Source

[Lbibass]

HOT DAMN. That's a lot of code.

[vegabook]

HOT DAMN. That's a lot of code.

That's what I thought. 700k+lines? What the hell are they doing? Re-writing Windows?? lol

Then I thought okay let's compare with something very roughly in the same domain. LLVM. Friends: 2.48 million lines of C and 7.75m lines of C++ (and 1.98m lines of header file lines). Astounding.

If you had asked me to wild guess before seeing the numbers I would have thought 150k / 2m lines respectively. Turns out it's 5x more.

(Yeah I know I know Windows is reputed to be 50 million lines of code)

[Jumbotron]

I do wonder how much of this move by Intel is their desire at the organizational level to be a good progenitor and steward of open source software or is in fact a cost cutting move to transition paid developers from this project and more to the lesser paid, if paid at all, open source community. I mean, Intel has completely lost the plot, as the Brits say, when it comes to innovating and in printing money from selling underperforming and overpriced CPUs much less the unmitigated disaster that is their discreet GPU division even after poaching half of AMD's top talent and former head architect of AMD's GPU division.

That said, I'm glad to see this move by Intel. It continues to bolster the slow growing and long suffering OpenCL standard as an alternative to Nvidia and Cuda.

[Quackdoc]

interesting, wonder how this will compare to rusticl

[mdedetrich]

I do wonder how much of this move by Intel is their desire at the organizational level to be a good progenitor and steward of open source software or is in fact a cost cutting move to transition paid developers from this project and more to the lesser paid, if paid at all, open source community.

In reality with companies its a mixture of both to varying degrees.

[Developer12]

interesting, wonder how this will compare to rusticl

Probably quite poorly, considering rusticl run on top of GPUs and this is a CPU-only runtime. It's basically LLVMpipe but for openCL.

[Quackdoc]

Probably quite poorly, considering rusticl run on top of GPUs and this is a CPU-only runtime. It's basically LLVMpipe but for openCL.

im talking specifically software rendering

[RonanKeryell]

It continues to bolster the slow growing and long suffering OpenCL standard as an alternative to Nvidia and Cuda.

I think this is a misconception to consider OpenCL as a direct CUDA alternative and this might explain the suffering.
According to https://en.wikipedia.org/wiki/CUDA "CUDA provides both a low level API (CUDA Driver API, non single-source) and a higher level API (CUDA Runtime API, single-source)." While OpenCL is similar to lower-level CUDA Driver API, usual programmers use higher-level single-source CUDA Runtime API which is difficult to adapt to non-single source OpenCL. But the good news is that Khronos has another standard for this purpose, SYCL, which is single-source and C++-based and can use OpenCL as a back-end among others. :-) Cf the comparison in https://en.wikipedia.org/wiki/SYCL#CUDA

[kieffer]

I find this move pretty exciting: the CPU driver by Intel is still my favorite one when it comes to validate OpenCL code (designed for GPU) in various platforms for CI. PoCL is very interesting but much less reliable (for their CPU/pthreads implementation).