Sections
Home
Hills
Infinite Hills
Tournaments
Software
Evolving
Optimizer
Community
Newsletter
Discussion
History
Sections
 
For Beginners
First Steps
FAQ
Guides
Lexicon
Benchmarks
For Beginners
> Home > The Corewar Newsletters > Core Warrior > Issue #2

Issue 2                                                           Oct 23, 1995
______________________________________________________________________________

Core_Warrior_ is a weekly newsletter promoting the game of corewar.  Emphasis
is placed on the most active hills--currently the '94 draft hill and the 
beginner hill.  Coverage will follow where ever the action is.  If you have 
no clue what I'm talking about then check out these five-star internet locals
for more information:

FAQs are available by anonymous FTP from rtfm.mit.edu as
pub/usenet/news.answers/games/corewar-faq.Z
FTP site is: ftp.csua.berkeley.edu /pub/corewar
Web pages are at:
http://www.stormking.com/~koth
http://www.ecst.csuchico.edu/~pizza/koth
______________________________________________________________________________
Greetings.

I can't speak for Beppe Bezzi, but I'm pretty new at this, so bear with our 
format and content changes until we produce something nice.  Complaints and 
criticism are welcome.  Mail them to Beppe <bezzi@iol.it>.  Praise and
encouragment are also accepted.  Just mail those flowery comments to me--
Myer R Bremer <bremermr@ecn.purdue.edu>.  We would love to have more people
join our staff.  I promise I'll take all the credit for your hard work, too.
How can you resist?

Another note of appreciation to Andersen F. Scholl, for the Core_Warrior_
logo.
______________________________________________________________________________
Tournament Time

NSFCWT round 2 is over, and it's been a tough challenge (so we were told :).
There were lots of interesting and unique solutions to the problem how to
kill the white warrior most effectively. The white warrior was an aggressive
silk style paper which was launched after a brief bombing phase. Most
challengers featured some form of multi-pass core clear (one or more spl
runs followed by a dat clear) teamed up with a scanner. A brief description
of the strategies is below. The winning challenger and the two runners-
up are posted in the next message.

twoOFaKIND
        Is two jmn.F scanners that scan in parallel for the white warrior body,
        avoiding bombs. The two parts then perform a checksum on the attack
        body and commit suicide when overrun by white. Multi-pass core clear
        code is then placed inside the found white warrior body where it is
        safe from attack. This proved to be the most effective strategy.

v15
        Is an unrolled (sne/sne/seq) scanner that bombs with a SPL carpet
        and finishes in a multi-pass core clear. For each bombed location,
        the scanner also checks 1800 addresses away, which is a likely
        place for white to be.

blackTN
        Tandem sne/seq scanners bombing with spl/spl/spl/jmp and
        finishing with a multi-pass clear.

safeclear
        has two redundant quickscan parts that checks for the 80 
        most dangerous white starting positions before a multi-pass
        core clear is executed.

Umbrella
        A multi-pass core clear protected by a bomber that concentrates
        on the most dangerous locations.

WhiteSimple
        Multi-pass core clear

Taking Trains (;-)
        sne/seq scanner bombing with spl carpet

Futility
        Two redundant multi-pass core clears

Mr Speculative
        cmp scan/bomber with spl bomb and single dat core clear

PlinyScan
        cmp scanner with spl bomb and multi-pass core clear

so I won't lose a lotta points
        imp spiral

Welcome to our new player Anders Scholl, it's good to have you. Now
for the scores; Magnus Paulsson won this round with a comfortable
lead over Steven Morrell and Paul Kline.

Rank    Name                    Author                   %W  %L  %T   Score
___________________________________________________________________________
  1     twoOFaKIND              Paulsson                 89   5   6   2738
  2     v15                     Steven Morrell           73  18   8   2283
  3     blackTN                 P.Kline                  64  16  20   2128
  4     safeclear               Anders Ivner             61  16  23   2058
  5     Umbrella                Beppe Bezzi              56  12  32   1996
  6     WhiteSimple             John K. Wilkinson        56  14  30   1986
  7     Taking Trains           Maurizio Vittuari        59  32  10   1853
  8     Futility                M R Bremer               42   8  50   1768
  9     Mr Speculative          Derek Ross               45  39  16   1506
 10     PlinyScan               G. Eadon                 41  35  24   1479
 11     so i won't lose a lotta Anders Scholl             0  12  88   892

After the first two rounds, these are the points:

Name    pts for round   1        2
__________________________________
Beppe Bezzi             7        7
M R Bremer              7        4
G. Eadon                1.5      2
Anders Ivner            5.5      8
P.Kline                 7.5      9
Steven Morrell          5       10
Paulsson                7.5     11
Derek Ross              3.5      3
Anders Scholl            -       1
Maurizio Vittuari       6.5      5
John K. Wilkinson       4        6

Magnus Paulsson is leading with 18.5 points, followed by Paul Kline with
16.5 and Steven Morrell with 15 points.

Good luck for round 3!
        Nandor & Stefan
______________________________________________________________________________
Current Status of the Internet Pizza Server ICWS '94 Draft Hill:

Hill Specs:
         coresize: 8000
   max. processes: 8000
         duration: after 80,000 cycles, a tie is declared.
max. entry length: 100
 minimum distance: 100
    rounds fought: 250
  instruction set: ICWS '94 Draft

Last challenge: Mon Oct 23 11:06:37 PDT 1995 

 #  %W/ %L/ %T                      Name               Author   Score     Age
 1  48/ 40/ 12                Frontwards       Steven Morrell     155     201
 2  46/ 41/ 13       Leprechaun on speed         Anders Ivner     150      62
 3  35/ 28/ 37                    test r          Beppe Bezzi     142       4
 4  37/ 32/ 31                 Torch t18              P.Kline     142     266
 5  35/ 28/ 36           Jack in the box          Beppe Bezzi     142     254
 6  39/ 36/ 24               Armory - A5            Wilkinson     141     405
 7  40/ 40/ 21                 myZizzor8             Paulsson     139       1
 8  34/ 31/ 35              Father & Son    Maurizio Vittuari     137     124
 9  41/ 45/ 14   Anti Die-Hard Bevo (3c)       John Wilkinson     137      71
10  38/ 39/ 22                 session Z           M R Bremer     137      20
11  36/ 35/ 29                Brain Vamp  B.Bezzi, M.Paulsson     136     136
12  38/ 41/ 21               myVamp v3.7             Paulsson     136     234
13  40/ 45/ 15         Leprechaun deluxe         Anders Ivner     134     171
14  34/ 34/ 31                       Phq    Maurizio Vittuari     134     368
15  41/ 48/ 11    Anti-DieHard Bevo (3f)       John Wilkinson     134      29
16  23/ 12/ 65                  Die Hard              P.Kline     133      10
17  28/ 24/ 48                    simple           M R Bremer     132      13
18  38/ 46/ 16                   blackTN              P.Kline     131      27
19  34/ 38/ 28               Tornado 1.8          Beppe Bezzi     130     120
20  40/ 51/ 10              SandBlast 1a    John K. Wilkinson     129      30

It's been a difficult time for replicators this week.  It seems some authors
are testing their tournament entries on the hill.  P.Kline's blackTN appeared
along with myZizzorX from Magnus Paulsson.  Paulsson's earlier test program  
produced huge losses for papers, but was too specialized to score well on the
hill.  simple--a non-pspace replicator/bomber once held a lofty position until
all those scanners came along.  It quickly needed to be recoded to survive.
Phq is also suffering a bit from the increased scanner presence.

The many versions of Die Hard have attempted the hill lately.  Perhaps Kline 
will finally settle on a version he's satisfied with.  Maybe then he'll tell
us what's in that thing!  For a interesting clue, read Extra Extra for 
Paulsson's comments on Die Hard and Kline's response.
______________________________________________________________________________
94 - What's New

 3  35/ 28/ 37                    test r          Beppe Bezzi     142       4
 7  40/ 40/ 21                 myZizzor8             Paulsson     139       1
10  38/ 39/ 22                 session Z           M R Bremer     137      20
15  41/ 48/ 11    Anti-DieHard Bevo (3f)       John Wilkinson     134      29
18  38/ 46/ 16                   blackTN              P.Kline     131      27
______________________________________________________________________________
94 - What's No More

21  34/ 39/ 27                Cthulhu v7    John K. Wilkinson     129      66
21  34/ 41/ 24              Thermite 1.0        Robert Macrae     128     802

Congratulations to Thermite for its long and successful run.  Armory is now
the oldest program on the hill at 405.
______________________________________________________________________________
Current Status of the Internet Pizza Server Beginner's Hill:

Hill Specs:
         coresize: 8000
   max. processes: 8000
         duration: after 80,000 cycles, a tie is declared.
max. entry length: 100
 minimum distance: 100
      maximum age: At age 100, warriors are retired.
    rounds fought: 250
  instruction set: ICWS '94 Draft

Last challenge: Mon Oct 23 08:14:39 PDT 1995 

 #  %W/ %L/ %T                      Name               Author   Score     Age
 1  57/  7/ 36              juliet storm           M R Bremer     208      25
 2  56/ 31/ 13                 Heatseek2        Phil Whineray     182       3
 3  49/ 32/ 20                   Test-Fc             G. Eadon     165      11
 4  49/ 38/ 13                       Web        Gareth Thomas     160      26
 5  34/ 11/ 55                 ShadowImp     Robert J. Street     156      99
 6  42/ 32/ 26                    Weasel          Kurt Franke     153      17
 7  44/ 38/ 18                 Hint Test           M R Bremer     151       5
 8  32/ 19/ 49                   Cyclone         Scott Manley     145      13
 9  42/ 41/ 17                   Obvious        John K. Lewis     143      10
10  42/ 43/ 15            Mythicon v1.1a             G. Eadon     142      84
11  42/ 45/ 13                     Eradu         Scott Manley     139       4
12  29/ 23/ 48           Imp Craze 91.43               Planar     135       1
13  19/  3/ 78              Cotton-DH 8a            Wilkinson     135      20
14  41/ 47/ 12                   Skimmer        Gareth Thomas     134      52
15  38/ 48/ 14                 PlinyScan             G. Eadon     128       7
16  33/ 42/ 25               DeathWalker         Scott Manley     125      36
17  21/ 43/ 35              Mokkori v0.2         David Boeren      99      47
18  25/ 52/ 23          Enlightenment II         Scott Manley      98      38
19  29/ 60/ 12              Apple attack        Gareth Thomas      98      53
20  31/ 67/  2                 Banzai0.2           Calvin Loh      96      18

John Wilkinson has decided to play on the beginner hill even though his
programs on the '94 draft hill are doing well.  I suppose I shouldn't   
talk.  juliet storm was just meant to monitor the hill, but has jumped to
the number one position.  Below is a copy of the code in case it's not in
your collection already.  Simplicity and compactness are the keys to its
success.  Is the actual running code ( code that is executed many times--like
a bomber--as opposed to code that is only executed once--like a quick scan )
of your warrior more than 15 lines long?  If it is, you're probably trying to
do too much.  Keep it simple.  Focus on one strategy and do it well. 

;redcode-94
;name juliet storm 
;author M R Bremer
;strategy bomber--yippee 

ptr     EQU -1333

gate    dat <-445, <-446
s       spl #445, <-445
        spl #0, <-446
        mov {445-1, -445+2
        add -3, -1
        djn.f -2, <-2667-500
        mov 33, <-20
go      dat #0, #ptr
start   mov {-1, <-1            ;boot the bomber into core
        mov {-2, <-2
        mov {-3, <-3
        mov {-4, <-4
        mov {-5, <-5
        mov {-6, <-6
        mov gate, ptr+24 
        mov gate, ptr+24
        spl @go, <-4000
        jmp boot, <-4013
for 73
        dat #0, #0
rof

imp_sz  equ     2667

boot    spl     1      ,#0              ;vector launched imp ala T. Hsu
        spl     1      ,#0
        spl     <0     ,#vector+1
        djn.a   @vector,#0

imp     mov.i   #0,imp_sz

        jmp     imp+imp_sz*7,imp+imp_sz*6   
        jmp     imp+imp_sz*5,imp+imp_sz*4   
        jmp     imp+imp_sz*3,imp+imp_sz*2   
vector  jmp     imp+imp_sz  ,imp

        end     start

______________________________________________________________________________
The hint

Do you often ask yourself:  Why do my warriors do well on the beginner's
and get thrashed on the '94 draft hill?  This hint should help all newcomers
in writing more viable code.

I have asked the 15th place author to send me his warrior so we can dissect
it as a case study.  Thanks to Scott Manley for sharing his code.  If you
are in 15th place around the end of the week, you could be getting mail from
me.  Unfortunately, Mutagen has dropped off the hill, but I suspect that  
Scott will have a brand new version real soon.

;redcode-b
;name Mutagen
;author Scott Manley
;strategy Scan -> SPL/JMP bomb -> split to Imp gate / 2 pass  forward 
;strategy travelling core clear
;assert CORESIZE==8000

plength EQU     35
inc     EQU     6
carpet  EQU     (CORESIZE-MINDISTANCE)/inc


adds    DAT     #inc , #inc
begin   SPL     clear2
start   SEQ.I   *scan, @scan
        JMP     scan1        
cont    ADD.F   adds , scan
        DJN     start , count
        JMP     clear3  
scan1   MOV     sbomb , * scan 
scan2   MOV     sbomb2 , @ scan 
        JMP     cont
scan    DAT     MINDISTANCE-10 , MINDISTANCE-9
target1 DAT     23 , -23
target2 DAT     22 , -22
target3 DAT     -1 , -21
target4 DAT     -2 , -20
        DAT     0 , 0
        DAT     0 , 0
        DAT     0 , 0
        JMP     0 , } cont
clear2  MOV.I   sbomb , } target1
        DJN     clear2 , target1
        JMP     clear4
clear3  MOV.I   target2 , } target2
        DJN     clear3 , target2
clear1  MOV.I   target4 , { target4  
        DJN     clear1 , target4
        JMP     0 , > -10
clear4  MOV.I   sbomb , { target3
        DJN     clear4 , target3
        MOV     0,1
sbomb   SPL     0 , 0
sbomb2  JMP     -1
count   DAT     #carpet-1,#carpet-1
        END     begin

The basic concept is to scan and spl coreclear simultaneously.  Once 
the scan is finished (one pass through core), jump to a dat coreclear to 
follow the spl clear. There are other startegies involved but we'll just 
focus on the scan engine and multi-pass coreclear.

Mutagen's scan is 11 lines long--13 lines if the stun bomb is included in
the count.  We can reduce the length a bit without affecting the behavior 
too much by using the following code:

        add.f split, scan
scan    sne.i loc1, loc+1   
        djn.f -2, <DJNSTREAM1   ;djn.f will decrement both the a and b field.
        mov.i split, *scan
        mov.i jump, @scan
        djn.f -5, <DJNSTREAM2   ;need to have a trigger so this falls thourgh
        jmp clear               ;after scan is complete.
split   spl #step, #step        ;try to find other uses for these
jump    jmp -1                  ;can be anywhere in code

Remember that the bigger your executing code is, the more inviting target
you make to the enemy. The split and jump lines can be moved away from the 
executing code to provide the smallest possible profile.  Also: the scanning 
step is not optimal.  If an enemy instruction is at location N, most likely 
there will be another instruction at N+1.  So there is no need to scan there.
There are two basic ways to spread the scan out.  One is to scan N and N+step
then bomb every location in between.  Another is to bomb N and then check
N+step in a separate step.  Check out Agony and Irongate for good examples
of these two methods.

Next let's talk multi-pass coreclears.  Muatagen's clear is lengthy and easy  
prey for scanners.  In a multi-pass coreclear, only a few things change--the
bomb being swept through core and the pointer that does it.  We can reuse
the movement instructions and just change the bomb and pointer with this
code:

org a1
ptr1    dat a1, end+100
a4      dat 0,  end+2+1
a3      dat 1,  end+2+2
a2      spl #2, end+2+3         ; spl #X, <-Y acts like a split 0.
a1      spl #3, end+2+4         ; you can use x and Y as step values
        mov *ptr1, >ptr1        ; and use the decrement in the b-field 
        mov *ptr1, >ptr1        ; as part of an imp gate.
        mov *ptr1, >ptr1        ; > (post-increment) keeps adding 1 to
end     djn.f -3, <4000         ; the b-field of ptr1 to move the bomb
                                ; through core.

The clear starts with a1 being swept forward through core.  Notice that the 
move instructions use a pointer (ptr1) to determine which bomb to sweep with.  
ptr1's b-field is also used to do the actual clearing.  When the clear wraps
around the core, it will eventually overwrite ptr1 with a1.  Now the move
instructions will look where the old ptr1 used to be to find what they should
move through core.  The new ptr1 (a1) points towards a2, so the move 
instructions will move a2 through core.  The new pointer also has a new
b-field.  The value of this b-field ensures that the clear skips over the
core clear code.  Then a2 overwrites the pointer.  The pointer is now   
points to a3.  And so on.  This code will continue to clear until time run  
out.

 7  44/ 38/ 18                 Hint Test           M R Bremer     151       5
A simple version of these improvements has been submitted to the beginner 
hill.  The code does reasonably well against bombers and replicators, but  
its large size makes it vulnerable to scanners.  A good project would be
to pspace it with a small fast bomber.
______________________________________________________________________________
Extra Extra:

P.Kline usually publishes his warriors fairly quickly.  However Die Hard has
been a mystery for quite some time.  Magnus Paulsson is one of the few 
authors who has had some success against the many interations of Die Hard.
Paulsson has been kind enough to share his thoughts ( along with his newest
program--theMystery ) with us:

How does Die Hard work?
That thought came to mind when some version of myVamp didn't score more than 
1% wins!  It was double disturbing as myVamp has a coreclear spl/spl/dat and 
a djn stream at the same time, no normal imp could survive it (nothing else 
either if lucky).

Why is that?
If you get the other program in such a clear it will fast spl it self to 8000 
processes.  No you have about 8000 cycles and then the imp spiral will 
execute. In my coreclear I change a location in core every cycle which means 
that when the imp-spiral execute it is (should be) overwritten/djned already.

What then?
Now if you place two spirals on top of each other, and plan in which order 
they and the rest of the code will be in the execution. In the coreclear the 
spl processes will be like 4000 processes, spiral, 4000 processes, spiral. 
That means the clear has to kill spiral in 4000 cycles which isn't possible 
in a clear.
 
Is that how Die Hard works?
(Actually I don't know :-) But I'll keep on guessing for a while.)
So, because there is a thing called gate which kills spirals. In order to tie 
you have to have something like 100 processes in a spiral to slow it down as 
much that it doesn't reach the gate. Now you can't launch such a monster 
without getting killed before the launch is complete.

How then?
Have a look at theMystery!
It is three papers working together to launch imp-rings in such a way that 
there will never be more that say 1500 cycles between a ring is executed.

What is the diffrence between Die Hard and theMystery?
Die Hard is a lot better :-(, I don't know how Paul did it but theMystery 
doesn't kill anything! (I'm totaly wortless on papers, I can't get the 
constants right and it's to bloody complicated to get a paper effective.)

Actually I don't belive that theMystery could ever beat anything 20%,
maybee Paul has taken the idea one step further?  (stone?)

(Now I resubmitted theMystery1.5 to the hill and it wins 17% :-)
15  25/ 19/ 57                  Die Hard              P.Kline     131       3
..
20  17/  7/ 76             theMystery1.5             Paulsson     127       1


/Magnus Paulsson
(More fun to write this than to think about Peierls substitution)
(I should heve mentioned CottonDH (J.Wilkinson) but I wont :-) )
______________________________________________________________________________
Paul responds:

Magnus' approach is intriguing.  Lots of ties but not many wins, which
is about where Die Hard was until I worked in some bombing.  Not much but
enough to pump up the score and move him up the Hill.  Die Hard doesn't
look much like theMystery, but his kernel is based on something I published
a looong time ago.

Another tidbit about Die Hard.  He starts with a very brief quick-scan
and vamp.  The pit does a one-pass suicidal clear incorporating a brainwash
which, when it works forces programs like Jack in the Box to revert from
paper to something which Die Hard can kill.  Instead of going 1/0
he now goes about 12/0 against JitB :-)

theMystery sure does a lot better against myZizzor than Die Hard does :-(
(I have a pretty good idea what Magnus' missing line in myZizzor might be!)

Paul
______________________________________________________________________________
;redcode-94
;name theMystery1.5
;author Paulsson
;strategy How does Die Hard work? (this way maybee?)
;strategy Looking gooood, could it be somthing that begins
;strategy with an i and ends with p?
;assert CORESIZE > 1
;kill theMystery

org start

step1   equ             1800
step2   equ             -1922

start   spl 1           ;\
        mov.i -1,0      ;- make 7 processes 
        mov.i -1,0      ;/
        mov {ptr2,<ptr2 ; move out second paper
        mov {ptr1,<ptr1 ; move out first paper
        spl 3
        spl 4
        jmp @ptr1       ; jump to 1 
        mov {ptr3,<ptr3 ; move out third paper
        jmp     @ptr3   ; jump to 3
        jmp @ptr2       ; jump to 2

pap     spl step1,0             ;\
        mov.i >-1,}-1           ;\\ Normal paper, with bad constants
        mov.i <-2,<1            ;// (I think I riped it from timescape :-) )
        spl     @0,}step2       ;/
        mov.i #0,2667           ;Impy!

ptr1 dat 5+pap,pap+5+500
ptr2 dat pap+5,pap+5+2667+500
ptr3 dat pap+5,pap+5+2667*2+500
______________________________________________________________________________
Most losses:
Program "myZizzor" (length 58) by "Paulsson"
(contact address "mpn@ifm.liu.se"):
;strategy Cissors (or whatever way you spell it)
;strategy Let's se how hard Die Hard is this time :-)
theMystery1.5 wins: 27
myZizzor wins: 55
Ties: 168

Program "Anti Die-Hard Bevo (3c)" (length 76) by "John Wilkinson"
(contact address "jwilkinson@mail.utexas.edu"):
;strategy I didn't have a single program on my HD that could beat
;strategy my Cotton-DH, so I made this.  Let's see how it fares
;strategy against the real Die Hard. :)
theMystery1.5 wins: 60
Anti Die-Hard Bevo (3c) wins: 54
Ties: 136
______________________________________________________________________________
Questions?  Concerns?  Comments?   Complaints?  Mail them to:
Beppe Bezzi <bezzi@iol.it> or Myer R Bremer <bremermr@ecn.purdue.edu>

2002-2005 corewar.info. Logo C. Schmidt