🚀 go-pugleaf

In article <315C40EA.41C6@detroit.sgi.com> Joe Landman <landman@detroit.sgi.com> writes:
>Julien Pierre wrote:
>> 
>> In article <4jhbus$dga@nuke.csu.net>,
>> ghansen@isc.sjsu.edu (Gregory L Hansen) wrote:
>> >And the result is:
>> >
>> >Hansen Floating-Point
>> >Machine   OS      Processor   Raw  Normalized  Relative  Compiler
>> >--------  ------  ----------  ---  ----------  --------  --------
>> >Power100  Sys7.5  PPC601/100  82s     82s          1     1,2
>> >==================================================================
>> >1: CodeWarrior 8
>> >2: optimized for speed
>> >
>> 
>> Self-builtPC    OS/2 Warp       Pentium/100     54s
>> 
>> Using Watcom 10.0 compiler and all optimizations
>> 
>
>Note that I spent any time looking at the code to optimize it... but
>here are (unofficial, I am doing nothing official here...  my numbers COULD be wrong) R4.4k at 150MHz,
>{1,2,4,8,16}xsR8k@90MHz, {1,2,4,8}xR10k@196MHz.
>
>Indy  Irix 5.3	R4400@150 MHz   (1 Meg cache)  Raw 44.6  compiler: cc -O3 -mips2 test.c -o test.x  -lm
>
>PowerChallenge 6.1 1xR8k@90 MHz (4 Meg cache)  Raw 19.49 (converted code to fortran to obtain real
>optimization)
>
We are NOT looking to see who can do the task faster!  We are looking to see
whos computer runs the task faster.  By improving calls, stripping assembly,
etc....  maybe I could make it run 10 times as fast.  THAT, however, proves 
NOTHING!  If you change even one line, the whole task could be worthless.

if pie is a routine that produces 3.14, for example, and you change:

x=(pie()*60) to x=(3.14*60), the difference could be ASTOUNDING!  WHY?
the first will make a call, get the value, multiply by 60, and set X to 188.4.
The SECOND will often simply be changed to x=188.4, and the FPU will NEVER
be used!

Steve