Conference hydra::amiga_v1

    I'll try to run it on my Amiga 1000 with a 68010, 512k, external
    3.5 " drive.  That is assuming that it makes it to my local host
    by tonight.
    
     - Chris
    

re:	.1

Great!	Now we need a 68000 taker...


P.S. I'm getting some very wierd VAX numbers (O/S dependent):

			single	double
	uVAX-II	ULTRIX	57	46
		VMS	88	69
	8350	ULTRIX	34	28
	8650	VMS	18	15
	8700	ULTRIX	10	 9
		VMS	18	15

Is VMS REALLY 60% slower than ULTRIX (i.e. What am I doing wrong?) ???

Here is the VMS DESCRIP.MMS file and the PI_VAX.C program:

####################################################################
CC	= cc
LINK	= link
CFLAGS	=
LFLAGS	=
LIBS	=

all : pi_vax.exe pi_vaxD.exe

pi_vax.exe :
	$(CC) $(CFLAGS) /define=(DODOUBLE=0) pi_vax.c
	$(LINK) $(LFLAGS) /exe=$@ pi_vax.obj $(LIBS)

pi_vaxD.exe :
	$(CC) $(CFLAGS) /define=(DODOUBLE=1) pi_vax.c
	$(LINK) $(LFLAGS) /exe=$@ pi_vax.obj $(LIBS)


/******************************************************************/
#include <stdio.h>
#include <math.h>
#include <types.h>
#include <time.h>

struct  tbuffer	etime;
long		t1,t2;

#if DODOUBLE
#   define PRECISION double
#else
#   define PRECISION float
#   endif

main(argc,argv)
char	**argv;
    {
    int		i,last,incr,sgn;
    PRECISION   pi,div;

    if (argc > 1)   last = atoi(argv[1]);
    else	    last = 100000;
    if (argc > 2)   incr = atoi(argv[2]);
    else	    incr = 10000;

    pi  = (PRECISION)0.0;
    div = (PRECISION)1.0;
    sgn = 0;

    times(&etime);
    t1 = etime.proc_user_time;
    for (i=0; i<last; i++) {
	if (sgn) pi -= (PRECISION)4.0  / div;
	else     pi += (PRECISION)4.0  / div;
	sgn = 1 - sgn;
	div += (PRECISION)2.0;
	if (!(i%incr)) {
#	    if DODOUBLE
		printf("%6d: %lf\n",i,(double)pi);
#	    else
		printf("%6d: %f\n",i,(float)pi);
#		endif
	    }
	}
    times(&etime);
    t2	= etime.proc_user_time;
    t2 -= t1;
    t2 *= 1000000;
    t2 /= 60;
    t2 /= last;
    printf("%d usecs per loop\n",t2);
    }

    
    Stock amiga 2000 1 meg memory 1 flopy 1 40 meg HD.

    brad

re:	.3

Great! Now we just need some 68881s (any starboards out there?)

BTW:	The reason for the rush is that I'm demoing the board at the monthly
	club meeting this week and would like to show some numbers for it.
	Thanks to everyone who's helping out!

Regards,
dave

    
re: .2
    
    were the numbers, cpu seconds used or were those the total times??
    were other people using the VMS vaxes?
    
    brad
    
    perf. results follow:
    
--------------------------------------------------------------------------------
    
Amiga 2000 with 68000 @ 7.16 mhz

	pi_single		      pi_double
run	  usecs		time		usecs		time
---	---------     ----------       ---------      ----------
1	397		39.7 sec	1530		2 min 33.0 sec
2	397		39.7 sec	1536		2 min 33.6 sec
3	397		39.7 sec	1534		2 min 33.4 sec
4 	397		39.7 sec	1530		2 min 33.0 sec
5	397		39.7 sec	1532		2 min 33.2 sec

-------

Best Times:

pi_single 39.7 seconds       pi_double 2 minutes 33.0 seconds

-------

Best PI Aproximations:

pi_single @ 50k loop 3.141640

pi_double @ 90k loop 3.141604
    

    
    Hardware:	A2000, 3 megs RAM
    		CMI 14.32 MHz 68000
    		12 MHz 68881
    
    Software:   v 1.3 Gamma-something-or-other, WB 34.4
    
    PI_SINGLE	PI_DOUBLE	PI_68881
    ------------------------------------
    337 uSecs	436 uSecs	GURU (under 1.2 & 1.3)
    
    Comparing my results with PI_DOUBLE with Brad's stock A2000, it
    looks like the CMI board provides some genuine speed with large
    numbers.
    
    I was puzzled by the GURUs when running PI_68881, since I know for
    a fact that my math chip works fine.  I ran some CMI benchmarks
    with and without the math chip installed, and it works like a champ.
    
    As an interesting note, I got virtually the same results (to within a
    microsecond) with MACHII and SNIPIT installed as background tasks.  I
    did not change their priority levels.
    
    Ed.
    
    
    PS:  This is fun stuff! I think I'll run these benchmarks on a CSA
    	 Amiga at the Software Shop. 

    Here are some numbers to sow confusion for Amiga pi benchmark results. 
    I ran 3 trials of each test-- times are usec per loop.

    Configuration:  A1000, 512K, 2 drives, OS v1.2, CLI, DMouse 

    Pi_single	 397	 397	 397 
    Pi_double	1527	1529	1530 

    Configuration:  same, except OS v1.3, and AmigaShell replaces CLI 

    Pi_single	 398	 397	 397 
    Pi_double	 777	 761	 760	(!) 

    Looks like the mathieeedoubbas.library has definitely been sped up.  So
    the question I have is, which version of the operating system do you
    use to get fair results? 
    
    Wes

    
    
    I too was running dmouse whilst generating my numbers.  I tried
    it without running it as well.  (i only did 5 runs for each, but
    all of the fastest numbers came with Dmouse running...fluke?)
    
    brad
    

re:	.5

->     were the numbers, cpu seconds used or were those the total times??
->     were other people using the VMS vaxes?

USER CPU-TIME. More people on the VMS-8700 (which should have charged me less
in that case). The uVaxen had the same load.

re: .6 GURU

Have you tried recompiling the 68881 code for your system?

re: .6 .7

Thanks! Here is my current table of info (keep those numbers rolling in)!


===============
Results so far:
===============


	Sys	SINGLE	68020		DOUBLE	68020D	68881	68020C
	-------	-------	-------		-------	-------	-------	------
format:		+ff 	+ff +2		+fi 	+fi +2 	+f8	+f8 +2
lib:		-lm32 	-lm32		-lma32  -lma32	-m832	-m832
	-------	-------	-------		-------	-------	-------	------
Amiga:	1	 397			1527		  ?
	2	 397			 760		  ?
	3	 337			 436		  ?
	4	  ?			  ?		  ?
	5	 323	 266		 981	 932	 169	 128
	6	 326	 268		 990	 920	 136	  97
	7	  ?	  ?		  ?	  ?	  ?	  ?
Vax:	8	  88			  69
	9	  57			  46
	10	  18			  15
	11	  10			   9


Sys	Description
-------	--------------------------------------------------------------
1 	Amiga with 68000 (at 7.16 MHZ) and 68881 (at 7.16 MHZ) O/S 1.2
2 	Amiga with 68000 (at 7.16 MHZ) and 68881 (at 7.16 MHZ) O/S 1.3
3	Amiga with 68000 (at 14.3 MHZ) and 68881 (at 12   MHZ) O/S 1.3
4 	Amiga with 68010 (at 7.16 MHZ) and 68881 (at 7.16 MHZ) O/S 1.2
5 	Amiga with 68020 (at 7.16 MHZ) and 68881 (at 7.16 MHZ) O/S 1.2
6 	Amiga with 68020 (at 7.16 MHZ) and 68881 (at 16   MHZ) O/S 1.2
7 	Amiga with 68020 (at 16   MHZ) and 68881 (at 16   MHZ) O/S 1.2
8 	uVAX-II under VMS
9 	uVAX-II under ULTRIX
10 	VAX8700 under VMS
11 	VAX8700 under ULTRIX

    re .1
    
    For some reason, my computer was refusing to download the code.
    I'll try again tonight if it isn't too late.  (I tried for 3 hours).
    It used to work.  ??
    
     - Chris
    

    
    Can someone with an Atari ST or a Mac SE compile and run the benchmark?
    It would be interesting to see which 68000 box is the fastest. 
    I know that the results will probably measure which compiler produces
    the fastest executable, but what the hell, could be fun anyway.
    
    Ed.

    Hardware:
    	Amiga 1000
    	512K  (I'm cheap)
    	2 3.5" drives (1 internal)
    	68010
    
    Software:
    	Kickstart 1.2
    	Workbench 1.2
    
    Results:
    Precision             Run 1       Run 2       Run 3
    ---------             -----       -----       -----
    Single (3.141649)     372 us      372 us      372 us
    Double (3.141604)     1465 us     1450 us     1451 us
    
    
    
    Sorry all of this took so long, but at least I finally got it to
    work.
       

    Hardware:
    	Amiga 500
    	1 MEG
    	68010
    
    Software:
    	Workbench 1.3 GAMMA of some sort
    
	Single 	415 uS   415 uS
	Double 	795 uS   795 uS


It all ran just fine.

Just for comparison here are the numbers during a download of HARRIER.ARC
(BTW - Neat Paul [HARRIER] - I was at that show!)

	Single	841 uS   815 uS
	Double  1579 uS  1173 uS


John