% A long chain of elementwise math on 8 million elements, fused into ONE
% GPU kernel per source line and replayed 600 times (about 5 billion updates).
% Paste this whole script into MATLAB to compare (lower niter if slow).
n = 8000000;
niter = 600;
x = rand(n, 1);
y = zeros(n, 1);
tic;
for k = 1:niter
  y = y + 0.1*sin(x + k) .* exp(-x) + x.^2;
end
t = toc;
fprintf('%.0f million element-updates/sec\n', niter*n/1e6/t);
fprintf('checksum %.4f (rand differs from MATLAB; expect ~equal, not equal)\n', mean(y));