Info

此问题已关闭。 请重新打开它进行编辑或回答。

Is there a way to speed up gpuarray loop?

1 次查看（过去 30 天）
Mantas Vaitonis
Mantas Vaitonis 2018-6-18
关闭： Mantas Vaitonis 2018-7-5
Dear All,
I am stuck in the folowing function. I would like to speed up this calculation for GPU, unfortunately now it is very slow.
clear;
p1=[1 2 2 5 4];
idx2=[1 3 4]
idx1=gpuArray(idx2);
p=gpuArray(p1);
tic
for k1=1:1000
cc=p(p);
for i=1:5
k=any(p(i)==idx1);
k1=any(cc(i)==idx1);
k3=idx1(cc(i)==idx1);
%[idxx]=arrayfun(@testas,k,k1,k3,bb(i));
idx1(cc(i)==idx1)=i;
end
end
toc
cc=p(p) is inside the loop, because in my algorithm it would change during every iteration.
  2 个评论
KSSV
KSSV 2018-6-18
You need to check whether using gpu array is necessary? Computations on GPU will be faster if your data is huge. For less data, GPU simulations tend to be slower than CPU. First check the same with CPU and compare times with GPU.
Mantas Vaitonis
Mantas Vaitonis 2018-6-18
The data needed to compute is quite big, for example matrix of size 5000000x100. Thus k1 is the size of 5000000 and i =100. I am trying to move speed sensitive functions to GPU in order to speed up whole algorithm.

此问题已关闭。

采纳的回答

Mantas Vaitonis
Mantas Vaitonis 2018-6-19
Did find a sollution using cellfun.

更多回答（0 个）

此问题已关闭。

标签

Community Treasure Hunt

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

Start Hunting!

Translated by