🚀 go-pugleaf

In article <4jpe51$m9v@newsgate.dircon.co.uk> kaoskat@dircon.co.uk (David Franklin) writes:
>Gregory L Hansen (ghansen@isc.sjsu.edu) wrote:
>: If you can supply code for some real benchmarks, then I will heartily 
>: endorse it.  The UseMarks use code that anyone can examine, compile, and 
>: run on their own machines.  They also cover any number of tests the 
>: participants come up with.  My benchmark was only because it is pretty 
>: lame to suggest a course of action and not do anything to get it 
>: started.  But I apparantly know so little about benchmarks, I didn't know 
>: I proposed a bad one.  So I'm counting on help from the participants, 
>: anyone who'se interested.  Even in the "Oh yeah?  Can your machine do 
>: this?" catagory.
>: 
>Here's a slightly more substantial benchmark (is an obfuscated C solution
>to the problem of finding arrangements of the 12 pentominoes that fit
>in a 6 x 10 grid):
>
>#include <stdio.h>
>#define M(x,y) Z=x,x=y,y=Z;
>#define K b[*U++]
>#define R *Q++=*U++-*a
>char W=6,H=10,F[12][32],L[12],S[12],Z,*U,O[300],*B=O+40;long T,i,j;
>void G(int N,char *b){char o,Y,*P;if(N){while(*++b);
>for(Y=12;Y--;)if(L[Y]){L[Y]=0;P=F[Y];for(o=S[Y];U=P,o--;P+=4)
>if(!(K||K||K||K)){U=P;Z=Y+65;K=*b=Z,K=Z,K=Z,K=Z;G(N-1,b);
>U=P;K=*b=0,K=0,K=0,K=0;}L[Y]=1;}}else printf("%s%d\n\n",B,++T);}
>main(){char p,t,k,*Q;for(p=0;Q=F[p],p<12;p++){for(t=8;t--;)
>{char m="737040221000"[p],a[5];if(!(t&m)){for(k=0;k<4;k++){int c=
>"678=23457=>?234:78>?7=>C28>?789=2379278>78>D678>"[k+p*4]-48,
>x=c%6-1,y=c/6;if(t&1)x=-x;if(t&2)y=-y;if(t&4)M(x,y)a[k]=x+y*9;}
>a[4]=0;for(i=300;j=i%4,i--;)if(O[i]=L[i%12]=32-22*(i%9==3),
>a[j+1]<a[j])M(a[j],a[j+1])U=a+1;R;R;R;R;S[p]++;}}}
>for(j=H;j--;)for(i=W;i--;B[i+j*9]=0);B[H*9-1]=0;G(12,B-1);}
>
>N.B. This code assumes char is signed - a global search and replace of
>"char" with "unsigned char" will be sufficient to fix this if your machine
>uses unsigned chars.
>
>I don't claim this is a good benchmark, but I'm interested in timings - 
>the two I have are for my home and work machines (Home is DX-50 and takes
>6 minutes, work is a SGI (R4400 @ 200Mhz) and takes 30 seconds), are a 
>bit more different than I'd expect.
>
>Dave
>
>: 

This only tests loops, and simple logic/math.  The results wouldn't be 
indicative of anything in particular.  On my P5/100, it did this in about
56 seconds.  In a real life app I wrote, the difference between a 486DX/50,
running SCO, and P5/75 DOS, the difference was about the same!

BTW, as a side note, I must say that one person sent me NATIVE 601/604 results
from my benchmark.  It blew the pentium out of the water for everything except
maximum possible disk access.  Of course, I wrote the program to destroy the
cache through random reads in a LARGE sampling.  On the IBM systems, this is
likely to work well, due to small memory banks.  On the PPC, many appear to 
have MORE memory.  THAT could account for the faster disk measurements 
elsewhere(the max disk access was about 1/2 of the pentium, best case).
I SHOULD state, however, that few systems even achieve real throughput equal
to the max worst case.  

When I get to it, I will post the ones I have run, and the ones sent to me.

Incidently, the benches *I* have run on the P5/100 are FAR from optimum, so]
it still shouldn't be taken as gospel.  They are simply ratings I know *I*
can achieve.

Steve