David Airlie's LPC2018 Presentation On An "Open-Source CUDA"

[schmidtbag]

The only reason I especially care about this is because of all the CUDA-based libraries. CUDA's biggest advantage over OpenCL and Vulkan is it's implementation. Nvidia does a great job getting developers started with pre-made libraries and comprehensive documentation, whereas I've found OpenCL is more like "this is just as capable, but you're on your own".
Of course, if you were to write something from scratch and have no intention on using any special libraries, OpenCL makes a lot more sense.

[ms178]

What about the plans for converging OpenCL and Vulkan? For a long time there were no news on that front.

[Michael]

What about the plans for converging OpenCL and Vulkan? For a long time there were no news on that front.

As covered in an earlier Phoronix article, OpenCL will continue on its own, they are not going to merge at this time.

[ms178]

Thanks, Michael. I must have missed that article. Watching airlied's presentation, I wonder what the timeline of his proposal is. He already admitted that in five years down the road GPU support in the C++ standard could be there where he wants it to be (alse he was talking about the competing proposals briefly). On the other hand, I don't get the impression that SYCL has gotten any meaningful traction yet (also AMD seems to have dropped support recently) and I wonder how he wants to change that to be a viable base for his efforts.

Here is the thread on the AMD forum where the dropped support for SPIR / SYCL came up: https://community.amd.com/message/28...omment-2875752

[boboviz]

The only reason I especially care about this is because of all the CUDA-based libraries. CUDA's biggest advantage over OpenCL and Vulkan is it's implementation. Nvidia does a great job getting developers started with pre-made libraries and comprehensive documentation, whereas I've found OpenCL is more like "this is just as capable, but you're on your own".

The force (and the weakness) of OpenCl is that it runs on every kind of hw, from Intel cpu to Amd gpu, from Arm socket to fpga.
Cuda runs only on Nvidia gpu.

[boboviz]

On the other hand, I don't get the impression that SYCL has gotten any meaningful traction yet (also AMD seems to have dropped support recently) and I wonder how he wants to change that to be a viable base for his efforts.

Seems that there is a lot of interest in SPIR-V, despite of SYCL

[Marc Driftmeyer]

And people only need to see the dysfunction that is Khronos [most likely intentional and beneficial to Nvidia] for Apple to create Metal and now Metal 2 API.

[shmerl]

And people only need to see the dysfunction that is Khronos [most likely intentional and beneficial to Nvidia] for Apple to create Metal and now Metal 2 API.

Metal junk is completely political. Nothing stopped Apple from collaborating on Vulkan, and making a higher level API based on it (and call it Metal all the same). They made their Metal NIH out of being their usual lock-in jerks.

[GunpowaderGuy]

Michael khronos biggest blunder since the time they maimed OpenGL 3 for the sake of backwards compatibility

[GunpowaderGuy]

Michael do you have the article at hand ? I already forgot the reasoning given for that desicion , or how final it was