For indexing in gpuArray, how can I improve my code?

1 回表示 (過去 30 日間)
zihao chang
zihao chang 2021 年 3 月 16 日
編集済み: Joss Knight 2021 年 3 月 18 日
function dists = compute_distances(obj,X)%X is a matrix with 10000 X 1024
num_test=size(X, 1);
num_train=size(obj.x_train,1);
temp_dists=double(zeros(num_test,num_train));
gpu_X=gpuArray(X);
gpu_xtrain=gpuArray(obj.x_train);%obj.x_train is a matrix with 1000 X 1024
for i = 1:num_test
for j =1:num_train
tmp=X(i,:)-obj.x_train(j,:);
temp_dists(i,j)=sqrt(sum(tmp.*tmp));
end
end
dists=temp_dists;
end
My CPU code executed in 5 mins.
The GPU code executed in 25mins.
I am trying to calculate the distance, but I noticed that I can use gpuArray to improve the speed.
I have no idea why it doesn't work.

採用された回答

Joss Knight
Joss Knight 2021 年 3 月 18 日
編集済み: Joss Knight 2021 年 3 月 18 日
Firstly, use pdist2.
Secondly, say no to loops:
Y = reshape(obj.x_train',1,[],num_train);
delta = X-Y;
dists = squeeze(sqrt(sum(delta.*delta,2)));

その他の回答 (0 件)

カテゴリ

Help Center および File ExchangeGPU Computing についてさらに検索

タグ

Community Treasure Hunt

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

Start Hunting!

Translated by