From multi-core CPU to GPU

Question

Alexandre 2014-7-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/141758-from-multi-core-cpu-to-gpu

编辑： Matt J 2014-7-17

Hi everyone,

I have a code that currently runs on a CPU which has 4 physical cores. This is done by using somewhere in my code a parfor loop.

I wanted to see if i could take any advantage of my GPU by running the code on my 576 Cuda cores instead of my 4 CPU cores. I created/sent all the variables inside the GPU and run the exact same code. However :

- when it comes to the parfor loop, Matlab says :

"Error using parfor_range_check (line 17)
The range of a parfor statement must be numeric.  See Parallel Computing Toolbox, "parfor"."
Here is my parfor initialization :
"n_row_max=gpuArray.colon(1,1,576);
 parfor row_Rubber=n_row_max"

- if i use a mere "for", it works perfectly but sloooowly, so i assume only 1 cuda core is running ?

I would like to do exactly like the parfor, but on the GPU : have 576 workers (instead of 4) doing the job.

Thanks in advance for any help,

Alex

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Matt J 2014-7-14

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/141758-from-multi-core-cpu-to-gpu#answer_145016

编辑：Matt J 2014-7-17

PARFOR is meant for parallelizing tasks on CPU cores, not on the GPU. You cannot use GPU cores to parallelize complex MCode for-loops. They aren't the appropriate hardware for that.