Sure, sorry about that. Here is the full message:
The following simple script generates and multiplies two random 1000-by-1000 matrices for 20 iterations and measures how long (in terms of both the wall-clock time and the cpu time) each iteration was executing.
Here is the code:
for i = 1 : 20
t0 = tic;
t1 = cputime;
rand(1000) * rand(1000);
printf('wall time: %.6f | cpu time: % .6f\n', toc(t0), cputime-t1);
endfor
And here is the output that I get:
wall time: 0.071920 | cpu time: 0.031200
wall time: 0.071924 | cpu time: 0.124800
wall time: 0.062927 | cpu time: 0.062402
wall time: 0.056930 | cpu time: 0.062400
wall time: 0.055934 | cpu time: 1.000000
wall time: 0.056937 | cpu time: 0.000000
wall time: 0.059940 | cpu time: 0.093600
wall time: 0.055943 | cpu time: 0.000000
wall time: 0.057947 | cpu time: 1.000000
wall time: 0.068950 | cpu time: 0.062400
wall time: 0.072955 | cpu time: -0.781598
wall time: 0.062958 | cpu time: 0.062400
wall time: 0.056961 | cpu time: 0.000000
wall time: 0.055964 | cpu time: 1.000000
wall time: 0.055968 | cpu time: 0.000000
wall time: 0.060971 | cpu time: 0.062400
wall time: 0.055974 | cpu time: 1.000000
wall time: 0.068978 | cpu time: 0.062402
wall time: 0.076982 | cpu time: 0.156000
wall time: 0.074987 | cpu time: 0.312002
The wall-clock time measured by tic-toc is believable but the cpu time is clearly false: sometimes it is negative and sometimes it is exactly 0 or 1.
These particular results were obtained in Octave 4.2.2 on Windows 7 machine with Intel i7-3610QM cpu but I observed the similar output on Ubuntu 16.04 with Intel Xeon E3-1535M v6.
Coincidentally, in Matlab the same code on the same machines returns reasonable cpu time.
Is this a bug or am I doing something terribly wrong?
Thanks,
Anton