how to run Gpu and Cpu in parallel in for loop?

1 次查看（过去 30 天）
Naveen kumar Elumalai
评论： Naveen kumar Elumalai 2019-1-14
for i=1:10
A(:,:,i)=rand(5,5);
[q1,r1]=qr(A(:,:,i));
end
I wanted cpu and gpu's to communicate with itself about the loop index. For iteration 1, cpu is working, gpu1 should also simultaneously work on iteration 2. gpu 2 should also simultaneously work on iteration 2. If gpu1 completes its task early, it should start iteration 3. When cpu is done with iteration 1, it should proceed with iteration 4 and so on. How should we implement this. The cpu, gpu1,gpu2 should only communiate about the loop index 'i' and not the A matrix. Once cpu or either of gpu's completes its iteration, it should parallely update the output array. How should this be implemented ?
  6 个评论
显示 4更早的评论
OCDER
OCDER 2019-1-14
Just curious, why do you want to split the job as such? Do you want GPU and CPU to be treated as indistinguishable processors? I foresee an issue where you must continuously exchange an array from a normal array to gpuArray, which can create a lot of overhead and slow things down.
Example of job tasks you want?
Task# CPU1 CPU2 GPU1 GPU2
1 X
2 X
3 X
4 X
5 X
6 X
7 X
8 X
9 X
10 X
11 X
Naveen kumar Elumalai
Naveen kumar Elumalai 2019-1-14
Yeah, the above example you referred, is what I want. The conversion of normalArray to gpuArray does bring a lot of overhead. Thats why, I wanted a work around for executing above task.

请先登录，再进行评论。

请先登录，再回答此问题。

采纳的回答

Walter Roberson
Walter Roberson 2019-1-14
You cannot do that with parfor or spmd.
What you can perhaps do is use parfeval to submit tasks associated with various resources. Monitor task completion, and upon completion upon a resource, collect the results and start a new task on the same resource.

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 GPU Computing in MATLAB 的更多信息

标签

产品

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by