home *** CD-ROM | disk | FTP | other *** search
-
- I was browsing some old DDJs the other day and came across
- a crusty benchmark called the savage floating point benchmark.
-
- Specifically, here is the c src for the 'savage':
-
- #include <stdio.h>
- #include <math.h>
- #define ILOOP 25000
- main()
- {
- int i;
- double a=1;
-
- printf("start\n");
- for(i=0; i<ILOOP; i++)
- a=tan(atan(exp(log(sqrt(a*a)))))+1.0;
- printf("a = %20.14e\n",a);
- printf("done\n");
- }
-
- Not really a serious attempt at a benchmark but it gave those Z80s some-
- thing meaty to chew on. It took several minutes of interpreter time
- to slog through the savage on those venerable cpm machines.
-
- Wondering what modern wonders had wrought, I became curious of the
- speedup affording by present-day hardware. In TC, the c src version
- ran in 9.4 seconds on my 20 MHz 386/387 box. Being something of a speed
- aficianado, I decided to take it 'to the limit'. The enclosed file,
- sav387.asm, is the result. It is an inline 80387- specific version that
- runs at 3.4 seconds/pass on a 20 MHz 30837. Firing up the Micronix 33 MHz
- cached machine brought in a time of 1.62 seconds/pass.
-
- If anyone out there has a Cyrix 387/33, I'd be interested in
- hearing the execution time for this test. Ditto for Weitek (sp?) and
- the 80486.
-
- Ray Berry
- Bellevue WA
- CIS 73407,3152
- exec-bbs
- comp conn pcboard 202 547 2008
- uucp ...ole!ray
-