Propose parallel path#4
Conversation
DanNixon
left a comment
DanNixon
left a comment
There was a problem hiding this comment.
I think the other thing missing from the background information is the relative differences in workload expected on different systems. e.g. how much should someone be able to do on a laptop?
| moves the code base into a position that will be easier to parallelize | ||
| through one of the GPU libraries | ||
|
|
||
| ### 3) CyCL |
There was a problem hiding this comment.
| ### 3) CyCL | |
| ### 3) SYCL |
There was a problem hiding this comment.
And throughout the following paragraphs.
| 3. [OpenMP](https://openmp.org) | ||
| 4. [CUDA](https://developer.nvidia.com/cuda-zone) | ||
| 5. [OpenCL](https://www.khronos.org/opencl/) | ||
| 6. [CyCL](https://www.khronos.org/sycl/) |
There was a problem hiding this comment.
| 6. [CyCL](https://www.khronos.org/sycl/) | |
| 6. [SYCL](https://www.khronos.org/sycl/) |
| @@ -0,0 +1,137 @@ | |||
| # Print Formatting | |||
| - Good, supported by most common C++ compilers | ||
| - Good, support for multiple cores and vector instructions | ||
| - Good, requires simple annotations of existing for-loops | ||
| - Bad, weak support for advanced hardware (GPU code) |
There was a problem hiding this comment.
This is not the case, OpenMP can target GPUs. It is only in the latest spec IIRC but there is support for at least some Nvidia GPUs from what I recall.
| moves the code base into a position that will be easier to parallelize | ||
| through one of the GPU libraries | ||
|
|
||
| ### 3) CyCL |
There was a problem hiding this comment.
And throughout the following paragraphs.
Fair comment. Depends how much detail we want to go in to here, but happy to chip in on this bit. |
On second thought, it is more the kind of information I would expect in a design document and not too relevant here. |
3 participants
Does everyone feel this matches our discussion from the sprint review?