I need to optimize my matlab code through vectorization

I need to vectorize this loop in code.. can someone guide?

function X = IDWT2DCT(x,Dct_matrix,IW_matrix)
[K,MN]=size(x);
block_size = 8;
X = zeros(block_size,block_size,K);
temp = zeros(block_size,block_size);

y = (Dct_matrix'*x);

for k = 1:K
    temp = reshape(y(k,:),block_size,block_size);
    X(:,:,k) = IW_matrix*temp*IW_matrix';    
end

*****Speed is not issue, i just need to vanish this loop****** [EDITED, Jan, Code formatted]

[Jan]

Please provide input data of the usual size e.g. created by rand. You want the vectorization to increase the speed of the function most likely. But then it matters if you operate on 100x100 matrices or on 100'000x1000.

[Abeera Tariq]

I don't want to increase speed, i just want to vanish this loop.. speed does not matter for my prb.. loops does

well

x=rand(10,64);

Dct_matrix=rand(10,10);

IW_matrix=rand(8,8);

[Jan]

Sorry, I cannot imagine that a loop in the code can be a problem. If speed is not your problem, modifying a running code is simply a waste of time. Even with vectorized code the data are processed in loops internally also.

[Abeera Tariq]

can u vectorize this code? @jan Simon?

[Image Analyst]

I agree with Jan that the loop takes virtually no time, unless K is huge, and I'm talking billions. Looping a few hundred thousand or a few million times takes like a microsecond. What is the value of K? What are the dimensions of Y and x?

[Abeera Tariq]

I need to implement the whole code on GPU, issue is that there are many loops inside loops .. so as an initiative i took inner most loop so that i can move to outer most loop..

Answer 1

[Jan]

I hesitate to invest any period of time when I cannot see any benefit. You ask for an "optimization", but the code will not be better in any way if it is vectorized. Therefore I'm convinced, that I cannot "optimize" it, especially when speed does not matter. If you want to have more complex for debugging and most likely faster as a side-effect, try these marvelous functions:

• http://www.mathworks.com/matlabcentral/fileexchange/25977-mtimesx-fast-matrix-multiply-with-multi-dimensional-support
• http://www.mathworks.com/matlabcentral/fileexchange/37515-mmx-multithreaded-matrix-operations-on-n-d-matrices

Please note, that "optimizing" code, which is not a bottleneck of a program is a known anti-pattern of programming. See http://en.wikipedia.org/wiki/Anti-pattern -> "premature optimization" and http://en.wikipedia.org/wiki/Optimization_(computer_science)#When_to_optimize

[Abeera Tariq]

I need to implement this on gpu nd there r many loops inside loops so i started removing these loops through vectorization.. :(

[James Tursa]

If you are implementing this on a GPU then why not work with a BLAS library to do the matrix multiplies? E.g.,

https://developer.nvidia.com/cublas