wgilbert@watmath.uwaterloo.ca (William Gilbert) (04/15/91)
When I was upgrading, I timed a few long programs that I use.
One program ran seven times faster with the 040 chip.
Here are the results.
PROGRAM CPU OS COMPILED TIME RELATIVE TIME
------------------------------------------------------------
1 030 1.0a on 1.0 with -O 2:49 min 1.00
030 2.0 on 1.0 with -O 2:00 min 0.71
030 2.0 on 2.0 with -O 3:04 min 1.09
040 2.0 on 1.0 with -O ***
040 2.0 on 2.0 with -O ***
040 2.0 on 2.0 without O 1:21 min 0.48
2 030 1.0a on 1.0 with -O 77 min 1.00
030 2.0 on 1.0 with -O 47 min 0.61
030 2.0 on 2.0 with -O 87 min 1.13
040 2.0 on 1.0 with -O ***
040 2.0 on 2.0 with -O ***
040 2.0 on 2.0 without O 45 min 0.58
3 030 1.0a on 1.0 with -O 10:62 min 1.00
030 2.0 on 1.0 with -O 10:29 min 0.96
030 2.0 on 2.0 with -O 10:44 min 0.99
040 2.0 on 1.0 with -O 1:29 min 0.14
040 2.0 on 2.0 with -O 1:29 min 0.14
040 2.0 on 2.0 without O 2:00 min 0.18
4 030 1.0a on 1.0 with -O 46 min 1.00
030 2.0 on 1.0 with -O 28 min 0.61
030 2.0 on 2.0 with -O 48 min 1.04
040 2.0 on 1.0 with -O 7 min 0.15
040 2.0 on 2.0 with -O 7 min 0.15
040 2.0 on 2.0 without O 9 min 0.20
PROGRAM 1 was a modification of Julia (on sonata in 1.0 sources directory)
that produced a Julia set by iterating each pixel in a window, up to 100 times.
However it contained many log and exponential calculations.
PROGRAM 2 was similar to 1 except that it wrote to a file on /tmp rather than
the screen and with much higher resolution (more points).
PROGRAM 3 was a program to display the basins of attraction of Newton's
Method, again by iterating each pixel in a window. It did not have any logs
in it.
PROGRAM 4 was similar to 3 except that it wrote to a file on the /tmp.
*** Programs 1 and 2 crashed with a known bug on the 040 when the compiler
optimization was turned on and the program went into a very tight loop
(see a previous posting re: cc compiler bug).
I noticed that the size of the executable program is much larger when compiled
under 2.0 (0.437 MB under 2.0 and 0.181 MB under 1.0a). Why is this?
____________________________________________________________________
William Gilbert, Pure Mathematics Department, University of Waterloo
wgilbert@math.UWaterloo.ca NeXT mail: wgilbert@fatou.UWaterloo.ca