search

preda / gpuowl

GPU Mersenne primality test.
GNU General Public License v3.0
176 stars 39 forks source link

Will gpuOwl be converted to OpenCL 3.0 ? #191

Closed valeriob01 closed 4 years ago

valeriob01 commented 4 years ago

Announcement from Khronos today: https://www.khronos.org/blog/opencl-3.0-specification-finalized-and-initial-khronos-open-source-opencl-sdk-released

baryluk commented 4 years ago

Why / What for exactly?

valeriob01 commented 4 years ago

evolution ?

preda commented 4 years ago

I don't see an important benefit ATM. Also AFAIK there are no implementations on AMD hardware, so that's that.

As I understand it, OpenCL 3.0 does not add things, it mainly makes the various parts optional/configurable.

baryluk commented 4 years ago

Yes, I think the same. There is nothing interesting in OpenCL 3.0 really. It might be good for new project, but there is nothing wrong sticking to OpenCL 2.x indefinitively. Porting to OpenCL 3.x should be hard, but as you said, there are no implementations, so that would make it less accessible. OpenCL 3.x will probably get some interesting features in the future, so then things can be re-evaluated.

valeriob01 commented 4 years ago

Maybe you both should read the full announcement: OpenCL 3.0 brings C plus plus for OpenCL to write kernels directly. And, Some of the longer-term design directions being considered by the OpenCL Working Group are: Recordable Command buffers Machine Learning Primitives Indirect Dispatch Device Topology Unified Shared Memory Global Barriers

Notably the Device Topology feature can help you implement the long missing feature of correctly detecting the GPU UUID.

ATM I have to apply a patch to correctly detect my GPU UUIDs which makes the gpuOwl version dirty.

valeriob01 commented 4 years ago

and btw, I was talking "will be" that refers to the future, mind how you read what I write.