Thanks to Friedrich Hertweck for developing a little timing test
routine to examine how different argument passing strategies perform.
Here are results for Lahey/Fujitsu Express for Linux, v 5.5 on a
450 MHz Pentium III Xeon:
1. -g option:
f77 loop, contiguous array, sum =32927.359400 time = 2.630000
f77 loop, stride 2x2 array, sum =32927.359400 time = 4.620000
f90 loop, contiguous array, sum =32927.359400 time = 2.930000
f90 loop, stride 2x2 array, sum =32927.359400 time = 4.950000
f90 SUM(stride 2x2 array), sum =32927.359400 time = 4.110001
2. -O option:
f77 loop, contiguous array, sum =32927.371100 time = 0.510000
f77 loop, stride 2x2 array, sum =32927.371100 time = 1.160000
f90 loop, contiguous array, sum =32927.371100 time = 0.540000
f90 loop, stride 2x2 array, sum =32927.371100 time = 1.570000
f90 SUM(stride 2x2 array), sum =32927.351600 time = 1.590000
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|