multi-dimensional gpu array multiplication

2 ビュー (過去 30 日間)
Qiu Xu
Qiu Xu 2021 年 10 月 26 日
コメント済み: Walter Roberson 2021 年 10 月 26 日
Hi,
I am performing the following two GPU calculations:
(1) The first problem
m=100;
n=100;
p=1024;
A=gpuArray(rand(m,n,p)); % 3D matrix
B=gpuArray(rand(m,n,p)); % 3D matrix
C=gpuArray(rand(m,n)+i*rand(m,n)); % 2D matrix with element being complex numbers
D=gpuArray(zeros(p,p));
for j=1:p
for k=1:p
D(j,k)=sum(sum(A(:,:,j).*C.*B(:,:,k)));
end
end
(2) The second problem
m=1024;
n=2048;
p=1024;
a0=2;
a1=3;
u=gpuArray(rand(m,n)+i*rand(m,n)); % 2D complex matrix
v=gpuArray(rand(m,n)+i*rand(m,n)); % 2D complex matrix
B=gpuArray(rand(60,60,m)); % 3D matrix
c=0;
d=0;
for j=1:n
usum=0;
vsum=0;
for k=1:m
usum=usum+u(k,j)*B(:,:,k);
vsum=vsum+v(k,j)*B(:,:,k);
end
c=c+a0*usum.*conj(vsum);
d=d+a1*abs(vsum).^2;
end
In above two problems, the two for-loops are very slowly, are there some good methods to aviod the two for-loops on GPU calculations?
Thanks a lots.
  1 件のコメント
Walter Roberson
Walter Roberson 2021 年 10 月 26 日

サインインしてコメントする。

回答 (0 件)

Community Treasure Hunt

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

Start Hunting!

Translated by