I'm looking for a faster way to roll through a set of indexes into an array by using the bitand() function.

I saw something awhile back that was much faster and as I recall the method was to convert bitand(a,b) a and b to double, then a few multiplies, a floor operation and a divide.

Here's the slow code... first the Test Bench (TB.m), then the real-time task which must be fast:

%% Test Bench
%
clear all
close all
%
b = uint32(0);
figure
for jj = 1:100
    for k = 1:4096
        R(k) = RxTask();
    end

    x = 1:4096;
    R = R.*1e6; % convert to microseconds
    semilogy(x,R)
    hold on
    drawnow
end
%
grid on
grid minor
title('RxTask Timein usec''s')

function [R] = RxTask()
%
persistent c d e
if isempty(c)
    c = uint32(zeros(1,4096));
    d = uint32(0);
    e = uint32(4095);
end
%
tic
   d = d + uint32(1);
   d = bitand(d,e);
st = toc;
% R.d = d;
R = st;
end

Answer 1

[Suraj Kumar]

Hi @Dr W Kurt,

From what I gather, you are trying to find a faster way to roll through a set of indices into an array by using the ‘bitand’ operation.

To optimize the ‘bitand’ operation, you can use the following mathematical operations to replicate the behaviour of the same :

Here is an improved version of the ‘RxTask’ function:

 function [R] = RxTaskOptimized()
        persistent c d e
        if isempty(c)
            c = uint32(zeros(1,4096));
            d = uint32(0);
            e = uint32(4095);
        end
        tic
            d = mod(d + 1, e + 1);  
        st = toc;
        R = st;
    end

Now to further optimize your code, you can check the following steps as well:

• You can recompute ‘x’ outside the loop as it stays the same in each iteration .
• Pre-allocation of ‘R’ can be done outside the loop to increase performance.

b = uint32(0);
x = 1:4096; % Precompute x outside the loop
R = zeros(1, 4096); % Preallocate R outside the loop
figure
for jj = 1:100
    for k = 1:4096
        R(k) = RxTaskOptimized();
            
    end
R = R .* 1e6; % convert to microseconds
semilogy(x, R)
hold on
drawnow
 end

You can monitor the performance by using the MATLAB Profiler for both the methods:

Optimized function:

Original function:

Happy Coding!