Fast xor array in matlab

Question

Is there a fatser way to xor every column of a matrix than this?

mod(sum(matrix),2)

It converts from logical to double and uses the expensive modulo.

Update:

According to this source, summing uint's is slower than summing doubles because it involves max clipping and other reasons. Also, note that summing logicals (with 'native') stops at 1.

Answer 1

I tried to avoid the cast to double but it's not better (often worse).

A = rand(2000000, 1) > 0.5;
 class(A)

tic
B = mod(sum(A),2)
toc

tic
C = mod(sum(uint32(A),'native'),2)
toc

tic
D = bitand(sum(uint32(A),'native'),1)
toc

The native option of sum allow you to keep the type of the argument in the result.

Answer 2

In addition to what @ClementJ says, I tried

tic
E = A(1)
for i = 2:numel(A)
    E = xor(y, A(i));
end
E
toc

hoping the accelerator would help, but it doesn't (much), and

tic
F = num2cell(A);
F = xor(F{:})
toc

which doesn't actually work because XOR only allows 2 inputs.

MATLAB's double precision vector arithmetic is about as fast as it gets, so you probably can't do better. If this is really driving your performance, then I suggest writing a C-MEX function: should be easy.