Reduce memory requirement of some vectorized operations

Question

Christopher 2015-11-27

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/257839-reduce-memory-requirement-of-some-vectorized-operations

编辑： Christopher 2015-11-27

In the following code, i perform some vectorized operations which happen to be highly memory intensive because my actual value of 'num' is 250. This is close to my limit, with 64 GB RAM. However, i would like to use a higher value. The goal of the code is to obtain the variable here called 'Energy_ind', which is only about 10 GB for num=250.

The main problem is the line:

Esiteindices = eindices(:,wbrdindic(:));

How can I do this same operation without the huge memory burden?

 num        = 100;
 elnum      = num^3;
 brd        = 4;
 P          = randi(4,[num num num]);
 P_wbrd     = padarray(P,[brd brd brd], 'circular');
 elnum_wbrd = (num+brd*2)^3;
 eoffsets   = int32(randperm(171)');
 %%%%%
 allind        = reshape(uint32(1:elnum),size(P));
 allind_wbrd   = padarray(allind,[brd brd brd],'circular');
 eindices      = repmat(eoffsets,1,elnum_wbrd);
     for i=1:elnum_wbrd
         eindices(:,i) = eindices(:,i)+i;
     end
 % get indices of O inside O_wbrd
 wbrdindic_i = reshape(uint32(1:elnum_wbrd),size(P_wbrd));
 wbrdindic   = wbrdindic_i(1+brd:end-brd,1+brd:end-brd,1+brd:end-brd);
 % Energy indices for elements in Owbrd(O)
 Esiteindices = eindices(:,wbrdindic(:));  % BIG
     clear eindices wbrdindic_i wbrdindic
 Energy_ind   = vec2mat(uint32(allind_wbrd(Esiteindices(:))),numel(eoffsets))';
     clear Esiteindices