Issue 75 2 September, 1999
_______________________________________________________________________________
Core Warrior is a newsletter promoting the game of corewar. Emphasis is placed
on the most active hills - currently the '94 draft hill, the beginner hill and
the '94 no-pspace hill. Coverage will follow where ever the action is. If you
haven't a clue what I'm talking about then check out these five-star internet
locals for more information:
FAQs are available from:
http://www.koth.org/corewar-faq.html
http://homepages.paradise.net.nz/~anton/cw/corewar-faq.html
The ftp site and it's mirrors are at:
ftp://ftp.csua.berkeley.edu/pub/corewar
ftp://ftp.inria.fr/INRIA/Projects/para/doligez/cw/mirror
ftp://www.koth.org/corewar
pMARS itself is also available from:
http://www.koth.org/pmars ;pMARS home page
ftp://members.aol.com/ofechner/corewar ;Fechner ftp site
Web pages are at:
http://www.koth.org/ ;KOTH
http://www.ecst.csuchico.edu/~pizza/koth ;Pizza
http://para.inria.fr/~doligez/corewar ;Planar
Newbies should check the above pages for the FAQ, language specification,
guides, and tutorials. Post questions to rec.games.corewar. All new players
are infinitely welcome!
_______________________________________________________________________________
Greetings...
This issue contains the results of the IWT and the code for Kline's
SnooPy. The hint this issue shows how Khuong's nPaper became a warrior
which shot to KotH on the '94nop hill. Enjoy...
-- John Metcalf
_______________________________________________________________________________
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: 200
instruction set: ICWS '94 Draft
# %W / %L / %T Name Author Score Age
1 43.3/ 34.4/ 22.2 Recycled Bits David Moore 152.2 97
2 35.2/ 18.6/ 46.3 Jade Ben Ford 151.8 20
3 41.4/ 34.8/ 23.8 Self-Modifying Code Ben Ford 148.1 43
4 45.1/ 42.4/ 12.5 Qshot Christian Schmidt 147.8 27
5 41.4/ 38.0/ 20.5 Snowman John Metcalf 144.8 41
6 41.9/ 39.8/ 18.3 SnooPy P.Kline 144.0 23
7 29.2/ 15.6/ 55.2 Return of the Fugitive David Moore 142.9 24
8 34.8/ 26.9/ 38.3 Stonewashed Christian Schmidt 142.7 37
9 34.3/ 27.2/ 38.6 One Step Beyond John Metcalf 141.4 58
10 41.6/ 42.0/ 16.5 goonie David Moore 141.2 18
11 32.4/ 24.4/ 43.1 nPaper II Paul-V Khuong 140.4 0
12 36.8/ 33.6/ 29.7 The Endless Knot John Metcalf 139.9 4
13 35.5/ 31.0/ 33.5 The Stormbringer Christian Schmidt 139.9 97
14 43.4/ 47.5/ 9.1 QHSA Ken Espiritu 139.3 8
15 39.3/ 39.5/ 21.2 Stranger John Metcalf 139.2 54
16 38.2/ 37.5/ 24.3 Draken Fire Ben Ford 138.9 55
17 38.7/ 39.3/ 22.0 death by redcode Simon Wainwright 138.0 76
18 31.2/ 24.8/ 44.0 Slippery Eels Ben Ford 137.6 39
19 27.7/ 18.5/ 53.7 No More Innocuous Leonardo H. Liporati 136.9 29
20 32.0/ 27.3/ 40.6 Icen Ben Ford 136.7 45
21 32.7/ 29.5/ 37.8 Sword'n'shield Christian Schmidt 135.9 8
22 30.6/ 26.0/ 43.4 EvoP Ken Espiritu 135.2 3
23 30.7/ 26.5/ 42.8 Exor Ken Espiritu 134.9 12
24 38.6/ 43.2/ 18.2 myBlur2 Paulsson 134.1 8
25 37.7/ 45.4/ 16.8 TheFlyOnTheWall Christian Schmidt 130.0 0
Age since last issue: 25 ( 43 last issue, 16 the issue before )
Days since last issue: 37 ( 61 last issue, 44 the issue before )
Average age: 33 ( 24 last issue, 19 the issue before )
Average score: 141 ( 141 last issue, 145 the issue before )
The top 25 warriors are represented by just 10 independent authors: Ford
and Schmidt with 5 each, Metcalf with 4, Moore and Espiritu with 3 and the
remaining 5 authors with just one warrior each. ( 9 authors last issue, 11
the issue before )
The scanners are returning to the hill in new unexpected forms... MyBlur's
scan loop contains several processes to implement an airBag technique.
QHSA uses a 'quasi-1c scan' to great effect against Return of the Fugitive,
against which Moore's own 3-shot scanner (Goonie) also scores well.
_______________________________________________________________________________
94 - What's New (Sorted by rank and score)
# %W / %L / %T Name Author Score Age
1 37.8/ 32.0/ 30.2 Am I alive? Christian Schmidt 143.6 1
1 35.0/ 28.1/ 36.8 The Pendragon Christian Schmidt 141.9 1
1 28.3/ 17.6/ 54.1 Jade Ben Ford 139.0 1
2 41.9/ 41.5/ 16.6 SnooPy P.Kline 142.3 1
2 26.0/ 13.0/ 61.0 Return of the Fugitive David Moore 139.1 1
3 40.6/ 40.6/ 18.7 goonie David Moore 140.6 0
7 38.2/ 39.8/ 22.0 The Endless Knot John Metcalf 136.5 0
8 43.2/ 48.0/ 8.8 QHSA Ken Espiritu 138.4 0
8 30.3/ 24.4/ 45.2 nPaper II Paul-V Khuong 136.3 1
12 29.4/ 26.3/ 44.3 Exor Ken Espiritu 132.5 1
14 38.3/ 44.2/ 17.5 myBlur2 Paulsson 132.4 0
16 35.8/ 41.5/ 22.7 Men-An-Tol-4 John Metcalf 130.0 0
17 33.5/ 35.4/ 31.2 Electrum John Metcalf 131.5 1
18 28.8/ 28.4/ 42.8 Sword'n'shield Christian Schmidt 129.2 1
20 28.8/ 26.7/ 44.5 EvoP Ken Espiritu 130.9 1
23 31.7/ 35.3/ 33.1 Baby One Moore Time Phooey 128.0 1
25 34.3/ 42.7/ 23.0 F B I Ken Espiritu 125.8 1
25 34.7/ 47.6/ 17.7 TheFlyOnTheWall Christian Schmidt 121.8 0
Players entering hill since last issue: 9 ( 8 last issue, 6 the issue before )
Average rank of new entries: 11 ( 9 last issue, 14 the issue before )
Schmidt and Ford both claim 1st place with their new entries. The
Pendragon is a curious variation on the stone/imp. Jade contains a
Q^4 -> Stone/imp which seems to be the only imp warrior immune to
Khuong's nPaper... Moore's Fugitive returns stronger than ever.
_______________________________________________________________________________
94 - What's No More (Sorted by age)
# %W / %L / %T Name Author Score Age
26 30.6/ 37.8/ 31.6 The Outsider Simon Wainwright 123.5 51
26 34.1/ 43.0/ 22.9 Circle of Fire John Metcalf 125.1 48
26 35.9/ 41.6/ 22.5 WingShot Ben Ford 130.1 41
26 34.2/ 41.6/ 24.2 Digitalis 5 Christian Schmidt 126.9 40
26 25.4/ 25.6/ 49.0 Alive and K(qu)icking Zet Leonardo H. Liporati 125.2 40
26 29.2/ 34.7/ 36.1 Vain Ian Oversby 123.7 36
26 25.2/ 25.4/ 49.4 Glass Ben Ford 124.9 32
26 31.3/ 37.7/ 30.9 Pitch Shifter Christian Schmidt 124.9 30
26 27.2/ 33.0/ 39.8 Icepick Ben Ford 121.3 21
26 32.4/ 36.1/ 31.5 Am I alive? Christian Schmidt 128.7 16
26 33.4/ 42.1/ 24.6 Rubble P.Kline 124.7 9
26 31.3/ 38.4/ 30.3 Serenade to a Cuckoo WFB 124.2 8
26 31.1/ 39.3/ 29.6 PC WFB 122.9 8
26 2.0/ 0.9/ 1.2 The Pendragon Christian Schmidt 7.1 6
26 30.7/ 35.9/ 33.5 Baby One Moore Time Phooey 125.5 5
26 31.8/ 35.5/ 32.8 Electrum John Metcalf 128.1 3
26 32.6/ 42.1/ 25.4 Men-An-Tol-4 John Metcalf 123.1 3
26 34.3/ 43.3/ 22.4 F B I Ken Espiritu 125.4 2
Am I Alive? drops off the hill from KotH in 15 challenges - though not
quite quickly enough to beat Toki's 12 last issue.
_______________________________________________________________________________
94 - What's Old
# %W / %L / %T Name Author Score Age
1 43.3/ 34.4/ 22.2 Recycled Bits David Moore 152.2 97
13 35.5/ 31.0/ 33.5 The Stormbringer Christian Schmidt 139.9 97
17 38.7/ 39.3/ 22.0 death by redcode Simon Wainwright 138.0 76
9 34.3/ 27.2/ 38.6 One Step Beyond John Metcalf 141.4 58
16 38.2/ 37.5/ 24.3 Draken Fire Ben Ford 138.9 55
15 39.3/ 39.5/ 21.2 Stranger John Metcalf 139.2 54
Recycled Bits continues to perform well.
_______________________________________________________________________________
The Extended New Hall of Fame:
* indicates the warrior is still active.
Pos Name Author Age Strategy
1 Probe Anton Marsden 403 Q^2 -> Bomber
2 Blur 2 Anton Marsden 396 Scanner
3 Damage Incorporated Anton Marsden 373 Q^2 -> Bomber
4 Return Of The Jedimp John K W 357 Q^2 -> Stone/imp
5 unrequited love kafka 346 Q^2 -> Paper
6 Impish v0.2 Ian Oversby 345 Stone/imp
7 Gigolo Core Warrior staff 332 Q^2 -> Stone/imp
8 Falcon v0.3 Ian Oversby 275 P-warrior
9 Nine Seven Six M R Bremer 232 Q^2 -> Stone/imp
10 Rosebud Beppe 218 Stone/imp
11 Newt Ian Oversby 216 Q^2 -> Stone/imp
12 Q^2 Miro Anders Ivner 214 Q^2 -> Scanner/bomber
13 Instant Wolf 3.4 Edgar 205 P-warrior
14 Goldfinch P.Kline 201 P-warrior
15 Simple v0.4b Ian Oversby 197 QScan -> Stone/imp
16 Trident^2 John K W 195 Q^2 -> Stone/imp
17 ompega Steven Morrell 189 Stone/imp
18 Frogz Franz 172 Q^2 -> Paper
19 The Machine Anton Marsden 164 Scanner
20 Memories Beppe 152 Scanner
21 Vain Ian Oversby 147 Q^2 -> Stone/imp
22 Head or Tail Christian Schmidt 142 Q^2 -> Paper
23 Electric Head Anton Marsden 140 P-warrior
24 Vigor Ken Espiritu 138 Q^2 -> Paper
25 Fixed Ken Espiritu 135 Q^2 -> Paper
26 Tiberius 3.1 Franz 130 Q^2 -> Paper
27 Ultraviolet-B Ken Espiritu 120 Q^2 -> Paper
28 obvious to those who k Robert Macrae 117 Q^2 -> Paper
29 Solomon v0.8 Ian Oversby 116 Stone and scanner
30 CC Paper 3.3 Franz 107 Q^2 -> Paper
31 mrb-test M R Bremer 106 *Unknown*
32 T.N.T. pro Maurizio Vittuari 105 Bomber
= Pulp v0.5 Ian Oversby 105 Q^2 -> Paper
34 Fugitive David Moore 102 Q^2 -> Paper/imp
35 Vengeance Robert Hale 101 Q^2 -> Stone/imp
36 Jack in the box II Beppe Bezzi 100 P-warrior
= Fire and Ice David Moore 100 P-warrior
38 Oblivion Ian Sutton 99 P-warrior
39 Recycled Bits David Moore 97* P-warrior
= The Stormbringer Christian Schmidt 97* Q^2 -> Stone/imp
41 Silver Talon 1.2 Edgar 93 Scanner
42 Bodge 1 Robert Macrae 85 Q^2 -> Scanner
43 Inferno 2.4 Philip Kendall 84 Qscan -> Bomber
44 Test Anton Marsden 83 *Unknown*
= NCC-1701-A Philip Kendall 83 P-warrior
46 RetroQ Paul Kline 82 Q^2 -> Paper
47 Tornado 4 Beppe Bezzi 78 Bomber
48 death by redcode Simon Wainwright 76* Q^2 -> Bomber
= He Scans Again Paul Kline 76 Scanner
50 Digitalis 4 Christian Schmidt 73 Q^2 -> Clear/imp
Congratulations to Wainwright whose Death By Redcode gains him access to
the Hall of Fame. Recycled Bits and The Stormbringer climb 9 ranks.
_______________________________________________________________________________
Current Status of the Internet Pizza Server Beginner 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: 200
instruction set: ICWS '94 Draft
# %W / %L / %T Name Author Score Age
1 54.3/ 34.4/ 11.3 Nuke it! Silvio Sampietro 174.3 46
2 44.3/ 25.3/ 30.5 PC WFB 163.3 18
3 39.1/ 19.3/ 41.6 Quickest Zeta Leonardo Humberto 158.9 49
4 41.9/ 29.0/ 29.1 Silken Half Life v4.0 Dale Neal 154.7 54
5 44.6/ 36.3/ 19.1 da mutant Paul-V Khuong 152.9 27
6 36.3/ 20.1/ 43.6 Quicker Zeta Leonardo Humberto 152.6 71
7 41.0/ 30.3/ 28.6 Silken Half Life Dale Neal 151.7 57
8 45.8/ 40.4/ 13.8 Nuke it! (V 0.2) Silvio Sampietro 151.2 53
9 43.5/ 39.2/ 17.3 B-52 #001 A.S. Mehlos 147.8 1
10 32.4/ 22.8/ 44.8 Quick Zeta Leonardo Humberto 142.0 76
11 37.9/ 35.7/ 26.4 Arsonic C P._V._K. 140.1 25
12 41.3/ 43.5/ 15.2 Neverland II John Metcalf 139.2 17
13 37.0/ 37.5/ 25.5 FireMaster 2 P._V._K. 136.6 23
14 39.1/ 41.9/ 19.0 Wild-Fire P_.V_.K./John Metcal 136.2 28
15 37.8/ 40.6/ 21.7 Mel the Great Ken Hubbard 135.0 29
16 28.3/ 21.8/ 49.9 HardCore v0.2b Simon Duff 134.7 77
17 35.8/ 38.3/ 25.9 Red Sand Ken Hubbard 133.2 60
18 36.7/ 40.2/ 23.2 1 shot bomber 0.1 P._V._K. 133.2 26
19 36.9/ 40.8/ 22.3 Simpsons4Ever v0.4 Maurice Fernández 133.1 3
20 36.3/ 39.6/ 24.1 Fire Master P_.V_.K. 133.0 61
21 32.4/ 32.6/ 35.0 Radio Flyer Ken Hubbard 132.2 32
22 24.9/ 18.9/ 56.1 Caladan Ingo S Kacza 131.0 2
23 40.2/ 50.1/ 9.7 UltiBomber P._V._K. 130.4 24
24 31.4/ 32.9/ 35.8 quatrefoil 133 Steve Gunnell 129.9 47
25 35.1/ 43.0/ 21.9 Simpsons4Ever v0.3 Maurice Fernández 127.2 5
Redemption retired, and Death Kiss almost made it, dropping off aged 96 :-(
Welcome to the new players - the two currently on the hill and the three or
four who haven't quite made it yet.
_______________________________________________________________________________
Current Status of the KOTH.ORG '94 No Pspace 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, excluding ldp and stp
# %W/ %L/ %T Name Author Score Age
1 39/ 21/ 41 nPaper II Paul-V Khuong 157 2
2 46/ 40/ 14 Zooom... John Metcalf 151 138
3 46/ 42/ 12 aTest 4 P.Kline 151 1
4 35/ 21/ 44 Jade Ben Ford 149 71
5 36/ 25/ 38 Omnibus John Metcalf 148 11
6 45/ 42/ 13 Boys are Back in Town 1.1 Philip Kendall 147 264
7 46/ 45/ 10 Win! David Moore 146 237
8 36/ 26/ 38 Icen Ben Ford 146 96
9 45/ 45/ 10 Qshot Christian Schmidt 145 8
10 37/ 28/ 35 Ant Factory Christian Schmidt 145 97
11 38/ 32/ 30 The Pendragon Christian Schmidt 144 75
12 38/ 31/ 31 Blacken Ian Oversby 144 249
13 34/ 26/ 40 EvoP Ken Espiritu 142 4
14 43/ 44/ 14 goonie David Moore 142 64
15 39/ 37/ 25 PC WFB 140 88
16 40/ 39/ 22 Napalm Ken Espiritu 140 51
17 41/ 41/ 18 Men-An-Tol-7 John Metcalf 140 20
18 35/ 31/ 33 Recovery Ian Oversby 139 262
19 39/ 41/ 19 K5000R Christian Schmidt 138 6
20 41/ 45/ 14 myBlur2(3) Paulsson 137 18
In the 87 successful challenges since last issue a total of 12 warriors
have left this hill - including the warriors which held 1st, 2nd and 3rd
ranks. Vain (age 330), The Fugitive (274), The Stormbringer (262) and
Romanian Killah (218) all perished... Khuong's nPaper dominates the hill,
see this issue's hint for more details.
_______________________________________________________________________________
Current Status of the Koenigstuhl Recursive ICWS '94 Draft Hill:
Koenigstuhl is an infinite hill similar to Planar's Mount Olympus and can
be found on Christoph C. Birk's Core War page:
http://andromeda.ociw.edu/COREWAR/
Below we show the top 20 of a total 1066 warriors:
1 Zooom... John Metcalf 174.69
2 The Machine Anton Marsden 174.29
3 Silver Talon 1.2 Edgar 173.50
4 Phantasm 50 Robert Macrae 172.91
5 Newt Ian Oversby 172.86
6 Scan Man David van Dam 172.67
7 Fixed Ken Espiritu 172.26
8 The Question David Moore 172.19
9 Recovery Ian Oversby 171.92
10 The Bloodhound Ian Sutton 171.67
11 The Stormbringer Christian Schmidt 171.57
12 Baseline Plus Ken Espiritu 171.44
13 Boys are Back in Town 1.1 Philip Kendall 171.23
14 Alien Christian Schmidt 170.25
15 Recycled Paper Planar 169.99
16 Forty Anton Marsden 169.87
17 Head or Tail Christian Schmidt 168.92
18 Digitalis 4 Christian Schmidt 168.89
19 Gigolo Core Warrior staff 168.76
20 myBlur2 Paulsson 167.55
Many recently published warriors hold positions in the top 20 ranks,
though Newt is still the top rating stone/imp. Other recent entries in
the top 100 are Puddleglum (29th), WingShot (30th), Shadow (31st),
Sharkrage (47th), Icepick (49th), Digital Dragon (52nd), Burning Water
(62nd) and Z-Shot (80th).
Also to be found on the same page are the Koenigstuhl '88 Hill of which
Freight Train is the King with a lead of 13 points over 2nd place, and the
Open Hill ('94 Draft + P-Space) where Pattel's Virus recently became KotH.
_______________________________________________________________________________
The Intelligent Warrior Tournament (1999) RESULTS!
In what would later be called an amazing display of corewar programming
power, eight programmers battled it out for control of the RedOS. In the
end only one would be crowned King.
In a tournament filled with both luck and skill, the eight Warriors would
fight over RedOS registers, attempting to enter a higher score in their
own register than their opponent could. The fighting would be fierce and
started out with a round robin to determine who would move on to the Knock
Out rounds. Only four would survive this grueling test.
The contestants were as follows:
;redcode-94
;name silly & slow
;author Ryan Coleman
;strategy Set my score to 7999!
;strategy set his score to 0
;strategy then spl bomb
;redcode-iwt99
;name Inconspicuous
;author John Metcalf
;strategy .8c scan -> spl carpet -> alter values
;redcode-94
;author Robert Macrae
;name Bride of :-=
; [or, Better Late Than Never...]
;strategy 0.66 Vamp, Decoy and Pit
;redcode
;name Reverse Vampire
;author Josh Yeager
;strategy Find RedOS, then copy self over middle of it.
;strategy Hopefully, opponent won't hit me because he thinks I'm the OS!
;strategy Also, any proccesses the OS has will be running me!
;redcode-94
;name Quick Thinking
;author Ben Ford
;strategy qscan -> HSAish scanner
;strategy
;strategy instead of bombing, qscan sets starting values for scanner
;strategy scanner based on HSA, added part to set my score properly
;strategy also never switches to dat-bombing (stun everything but me!)
;strategy
;strategy beats test warrior (and simple improvements) 80%+
;strategy better than nothing, I suppose someone could try to take
;strategy advantage of 'flaws' in the hsa engine
;redcode-94
;name Mine v0.3
;author John K Wilkinson
;strategy my entry. this is mine. hence the name.
;redcode-94
;name StupidWarrior
;author Planar
;redcode-94
;name Starving Hacker
;author David Moore
;strategy Score 1 point :-) Optimized for single-elimination combat.
The round robin featured single combat between each competitor. Position
was randomly assigned by a coin toss (terribly high tech.) The chart below
shows the results of the round robin portion.
Coleman Metcalf Macrae Yeager Ford Wilkinson Doligez Moore
Coleman X X X 0* -1 0* -7980 1 1 -1
Metcalf 0* X X X -1 0 -7980 0 0 -1
Macrae 1 1 X X X 1 1 5623 1 3
Yeager 0* 0 -1 X X X -7980 -7998 0 -1
Ford 7980 7980 -1 7980 X X X 7980 7980 -1
Wilkinson -1 0 -5623 7998 -7980 X X X 7998 -1
Doligez -1 0 -1 0 -7980 -7998 X X X 0
Moore 1 1 -3 1 1 1 0 X X X
* denotes a fight where the RedOS died.
Total Scores (Round Robin)
39899 Ford -7980 Coleman
5631 Macrae -7982 Metcalf
2391 Wilkinson -15980 Yeager
2 Moore -15980 Doligez
With the round robin finished we came to the four Warrior Knock Out! As
we can see, Ben Ford's entry "Quick Thinking" is clearly the odds on
favorite having scored so highly in the round robin. Macrae and Moore
both entered "dumber" programs that cleared core with sequential clears,
hoping to trade score for speed that might help in the Knock Out portion
of the tourney. Wilkinson's "Mine v0.3" was a jmp stunner/scanner with
bootstrap.
The first battle was between Ben Ford's "Quick Thinking" and David Moore's
"Starving Hacker". Ben had lost in the round robin to "Starving Hacker"
and was looking to take the battle to Moore's quicker program.
Unfortunately for Ben, Moore's attack was too fast and he fell. Having
easily dominated the round robin, "Quick Thinking" was ironically beaten
by a quicker program. (Note: Ben had, in my opinion, on of the best
programs written and might had taken it all had the rules been slightly
different.)
The second battle was going to be one of the most contentious. Macrae's
"Bride of :-=" would face third place "Mine v0.3". The battle took four
rounds to decide as the score was a tie the first three tries. In the
end, "Mine v0.3" was triumphant.
The stage was set for the final round:
Our MAIN EVENT...
;name Starving Hacker ;name Mine v0.3
;author David Moore ;author John K Wilkinson
The first battle lit up the screen with green and yellow pixels as the two
programs battle furiously for control of the registers. Fighting with all
their might, they fought the first round to a tie! The tie breaker was
decided early when Moore's "Starving Hacker" was struck squarely with a jmp
bomb from "Mine v0.3". The rest was clean up and John K. Wilkinson's
program did this brilliantly.
The winner of the 1999 Intelligent Warrior Tournament is:
John K. Wilkinson!
Congratulations to John and thanks to all the contestants for taking the
time to code what was a hard fought battle. I'm glad that this
competitions winner was a program of some size and that it needed a bit of
intelligence to win. The warriors will be displayed on the website soon.
Side Note:
I welcome feedback on this tournament and what could be done to make it
more interesting for the programmers. It was my hope to reward complexity
of strategy and it seems to have worked in most cases. Feel free to send
me an email at <jklewis@umich.edu> or start a thread on r.g.c.
The tournament's homepage can be found at:
http://www.concentric.net/~liminal/IWT/index.html
_______________________________________________________________________________
The Hint - Improving a Beginner's Warrior by Paul-V Khuong and John Metcalf
Once again this hint will follow the development process of a warrior.
This time the warrior is Paul's nPaper and our aim is to improve it
sufficiently to allow it to gain a place on Pizza's '94 draft hill. Here
is the original:
;redcode-94nop
;name nPaper
;author Paul-V Khuong
;strategy Paper
;assert CORESIZE==8000
pStep1 equ 3944
pStep2 equ 2616
pStep3 equ -624
x equ -373
y equ 1915
z equ -287
spl 1
spl 1
spl 1
pap1:spl @0, <pStep1
mov }pap1, >pap1
pap2:spl @0, {pStep2
mov }pap2, >pap2
mov.i #1, }1
mov {x, <y
mov {pap2, {pap3
pap3:djn.f pStep3, <z
As with most papers, nPaper is based on the typical silk replication
concept. However, the methods of attack are particularly interesting:
1) The bomb which nPaper throws is a kind of silk-imp. Mov.i #1,}1
combines with the mov {x,<y line to bomb two strips of memory with
itself.
2) Instead of the commonplace jmp pStep3, nPaper uses a rarely seen djn.f
pStep3,<?. Since there are 8 processes and only 6 lines to copy,
nPaper is still effective even if a couple of proccesses happen to fall
out of the warrior when the djn.f meets it's arch-enemy - dat 1,1.
We agreed to carry out our testing on KOTH's '94nop hill, since it returns
challenge results a little faster than Pizza. The original version of
nPaper scores 100.32 - losing to scanners and tying against everything
else.
What we need to do is take a close look at each section of the paper and
see how it interacts with some typical opponents...
The mov.i #1,}1 silk-imp bomb which Paul has created tends to act
similarly to a standard imp while in empty core, carrying any enemy
processes executing it forward to the next section of code. The mov.i
#1,{1 bomb of Fugitive doesn't have this disadvantage and against a-driven
imps it should help convert a few ties into wins.
As a consequence of changing to Fugitive's bomb, we need to change the
line following the bomb to mov x,<y. Let us test this theory against the
hill...
nPaper now scores 118.61 and as expected shows an improvement against the
Stone/imps on the hill, with the exception of Jade which we suspect may
use standard b-driven imps. Also, a small increase in score was observed
against the majority of scanners.
#W / #L / #T Opponent #W / #L / #T Opponent
86 / 15 / 149 The Stormbringer 0 / 9 / 241 Jade
78 / 12 / 160 The Pendragon 65 / 37 / 148 Vain
60 / 24 / 166 Blacken 53 / 41 / 156 PC
Do the advantages of the djn.f attack outweigh the disadvantages? Simply
replacing the djn with a jmp scores worse, 117.47. However, jmz has been
used successfully in a number of papers (including the classics Mice and
Molerat) to determine if a pointer which should be zero has been somehow
corrupted. Let us then try a couple of variants of this technique:
mov {pap2, {pap3 ; scores 123.62
pap3:jmz.a pStep3, *0
mov {pap2, <pap3 ; scores 122.76
pap3:jmz.a @0, pStep3
A substantial improvement. Of the two variants we tested, the former
appears to score higher by almost a point. However, against most warriors
the two versions scored about the same. The exception is against myBlur -
the former scores 148.4 and the latter 115.2 :-)
Having improved the attack, we now need to concentrate on ways in which
the replication mechanism might be improved... Our first idea was to
reduce the number of processes the paper requires from 8 to 6, by
including an extra copy instruction in the first silk launcher as Macrae
describes in his text accompanying Benj's Revenge (CW64). However, since
this scored a mere 116.25 points; it appears this technique isn't suited
to nPaper. (Can you think why?)
Instead we opted for simply booting a second copy of the paper half way
around core for a score of 127.12. Now let us see how much we can improve
on this score by optimizing the steps.
First the dumb brute force method was tried. A simple program generates
copies of the paper with random constants. These are then battled against
a small selection of warriors to see which scores the highest. After
testing over 200 sets of constants, we had a warrior capable of scoring
127.41 points on '94nop.
An alternative approach is needed... Perhaps we could improve the steps
by making random small tweaks to the steps we currently have. A program
was written to change one or two constants at a time by a small multiple
of 8... After testing over 100 tweaks, we found one which could score
128.08 on the hill.
Our final attempt at optimization proved a little more successful. We
replaced the two bombing instructions with nops and optimized the three
replication constants using the DBF (dumb brute force) method. Next, the
highest scoring set of pSteps were chosen and the nops were replaced with
the bombing instructions. The bombing steps x and y were then optimized
using DBF which left us with a warrior scoring 132.88 :-) NPaper now
scores even more spectacularly against those annoying stone/imps:
#W / #L / #T Opponent #W / #L / #T Opponent
152 / 2 / 96 The Stormbringer 2 / 4 / 244 Jade
131 / 7 / 112 The Pendragon 113 / 24 / 113 Vain
83 / 10 / 170 Blacken 82 / 23 / 145 PC
Since nPaper is now capable of entering the lower end of both the '94nop
and '94 draft hills, we have fulfilled our original criterion. However,
we still haven't added a qscan which is traditionally included in a paper
warriors.
>From previous experience, we know how effective mini-qscans can be.
Therefore we chose to employ a Q^3 trimmed down to scan fewer locations.
A short series of test submissions showed the score seemed to peak at 33
scans, which we can implement in 3 table entries. We also put the unused
b-fields to use decrementing pockets of core.
;redcode-94
;name nPaper II
;author Paul-V Khuong
;strategy MiniQ^3 -> Paper
;assert CORESIZE==8000
org qGo
pStep1 equ 560
pStep2 equ 2680
pStep3 equ -1120
x equ -3781
y equ -2013
pGo: spl 1, >424
spl 1, {761
spl 1, {933
mov {pap1, {pBoo
pBoo:spl pEnd+4000,{2093
pap1:spl @pEnd, <pStep1
mov.i }pap1, >pap1
pap2:spl @0, {pStep2
mov.i }pap2, >pap2
mov.i #1, {1
mov x, <y
mov.i {pap2, <pap3
pap3:jmz.a @0, pStep3
pEnd:
a for 45
dat 0,0
rof
qf equ qKil
qs equ (qd*2)
qd equ 107
qi equ 7
qr equ 11
; -+)>] 0/1 cycles [(<+-
qGo: seq qd+qf+qs, qf+qs ; 1
jmp qSki, {qd+qf+qs+qi
seq qd+qf+6*qs, qf+6*qs ; B
jmp qFas, {qd+qf+6*qs+qi
seq qd+qf+5*qs, qf+5*qs ; B-1
jmp qFas, <qBmb
seq qd+qf+7*qs, qf+7*qs ; B+1
jmp qFas, >qBmb
seq qd+qf+9*qs, qf+9*qs ; A-1
djn qFas, {qFas
seq qd+qf+10*qs, qf+10*qs ; A
jmp qFas, {qFas
; -+>)] 2 cycles [(<+-
seq qd+qf+3*qs, qf+3*qs ; C
jmp >qFas, {qd+qf+3*qs+qi
seq qd+qf+2*qs, qf+2*qs ; C-1
jmp >qFas, {qSlo
seq qd+qf+4*qs, qf+4*qs ; C+1
jmp >qFas, }qSlo
seq qd+qf+12*qs, qf+12*qs ; B*C-B
jmp qSlo, {qSlo
seq qd+qf+15*qs, qf+15*qs ; B*C-C
jmp qSlo, <qBmb
seq qd+qf+21*qs, qf+21*qs ; B*C+C
jmp qSlo, >qBmb
seq qd+qf+24*qs, qf+24*qs ; B*C+B
jmp qSlo, }qSlo
seq qd+qf+27*qs, qf+27*qs ; A*C-C
djn qSlo, {qFas
seq qd+qf+30*qs, qf+30*qs ; A*C
jmp qSlo, {qFas
sne qd+qf+18*qs, qf+18*qs ; B*C
jmz.f pGo, qd+qf+18*qs-10
qSlo:mul.ab #3, qKil ; C=3
qFas:mul.b qBmb, @qSlo
qSki:sne >3456, @qKil
add #qd, qKil
qLoo:mov qBmb, @qKil
qKil:mov qBmb, *qs
sub #qi, qKil
djn qLoo, #qr
jmp pGo, >10 ; A=10
qBmb:dat {qi*qr-10, {6 ; B=6
end
So how does our final version score? King of the '94nop and in the top
ten on the '94 draft :-) On entering the '94nop hill it scored 40%+ wins
against 11 of the warriors - 50%+ against 7. However, it would not
surprise us to see a few more b-driven imps appearing on the hills...
1 36/ 22/ 42 nPaper II Paul-V Khuong 151 1
_______________________________________________________________________________
Extra Extra - SnooPy by Paul Kline
You may remember Alladin's Cave, onetime leader on Pizza's 94 hill. I
thought it might still have some potential and started working on a
replacement, using the same base components and changing over to a p^3
switcher. I really could not get it going however, the one-shot scanner
just is no longer strong enough and AC would end up carouseling the
various parts against stronger opponents. This is one of the problems
with p-programming, if you do not have a good 'stopper' in there, you end
up rotating weaker opponents and give up way too many losses.
So I went looking for some good code, especially something that would
perform well against stone-imps and strong silks. The modern stone-imps
seem to be structured similarly, a small stone with core wipe and a small
imp-pump that is spaced far away from the stone. Both are hard to find by
scanning, and a core-clear takes too long to kill them together. My
approach was to try a quick one-pass something that would stun at least
one of the parts with spl's, followed by a dclear. That way the dclear
would have more time to do its job, and the imp-spiral would not grow too
long.
I went through a number of Torch and vamp variations, but finally settled
on a Blur attack, one scanning pass and a dclear. It works pretty well
against stone-imps and not so well against stronger silks. Additional
stunning would be useful but to kill s-i's you have to start dclear-ing
early, or you will not have time to kill all the spl's and the imp.
Other p-spacers can be expected to try a simple stone which is deadly
against scanners, so I added a paper component. After much diddling I
selected a variation on an old theme of mine, which is a non-silk
replicator with a lot of imp-deadly dat-bombing. This version kills s-i's
like Rosebud and Blue Funk about 50% of the time, but is mainly guaranteed
to at least get a tie. It also effectively kills a class of non-looping
silk replicators. Because they do not loop, they act like simple hopper
generators, once all the generators are dead, the hoppers are quite
vulnerable. CC Paper is a good example.
To protect the replicator against HSA and f-scan attacks I added a Wimp,
which is just a JMP #0,#0 program that is invisible to them.
For a little more punch against certain opponents I included the one-shot
dclear component from Alladin's Cave.
SnooPy is a good all-round performer, so far no programs on the hill are
killing it badly and it is not killing anything else badly either (except
one new one :-). Some challengers are giving up nearly 90% losses
however, which is interesting. I think what is happening is that while
SnooPy does a 16-byte forward scan in 3 steps (actually a 10 and a 6), the
opponent is doing a backward scan with a smaller step or at a slower rate.
I described this situation in an old Push Off article and it is fatal to
the slower program. Other than that I can not explain such a high ratio.
The p-code drives the trio on successive losses from scan to paper to
one-shot and round again. The only variation is that if the paper gets a
tie it returns to the scanner, which works quite well when the opponent is
a replicator or imp.
;redcode-94
;name SnooPy
;author P.Kline
;assert CORESIZE == 8000
;strategy blurish scanner, paper, oneshot-clear
;strategy did i mention a wimp?
w0 EQU vBoot
w1 EQU slPaper
w2 EQU cBoot
PSTATE EQU 278
wimp jmp #0 ,#0 ; frustrate HSA-types, used with paper
for 15
dat 0,0
rof
clGate jmp OsClear,OsClear-6 ; one-shot to a dat-clear
clIncr dat 13 ,13
OsClear add clIncr ,clGate
sne.x *clGate,@clGate
djn.f OsClear,<-1500
clSpl spl #0 ,<clGate
mov clBomb ,>clGate
djn.f -1 ,>clGate
clBomb dat <5535 ,1-clGate
dat 0,0
btClear jmp 1000 ,clGate+10-btV
for 5
dat 0,0
rof
think ldp.a #0 ,in
ldp.a #PSTATE ,table
mod.ba *in ,table
stp.b *table ,#PSTATE
; S L W T
table jmp }0 ,0441 ; 00 1 0 0
dat #w0 ,0154 ; 01 2 0 1 ; w0 = scanner
dat #w0 ,0163 ; 02 3 2 1 ; w1 = paper
dat #w0 ,0092 ; 03 4 1 2 ; w2 = one-shot/clear
dat #w1 ,0102 ; 04 6 4 3
dat #w2 ,0446 ; 05 6 5 5
dat #w2 ,0362 ; 06 2 5 2
dat #w2 ,0362 ; 07 2 5 2
dat #w2 ,0362 ; 08 2 5 2
in dat 0,8 ; must have non-zero b-field in the previous cell
dat 0,7
dat 0,9
cBoot mov btClear ,btV ; boot one-shot or scanner
vBoot
a for 10
mov <btV ,{btV
rof
btV jmp vPtr+10+16*150 ,vPtr+10
for 10
dat 0,0
rof
vStep equ 16
vGate equ (vPtr-3)
Snoop
vPtr mov vScan ,6000 ; Blur-type scanner
vNext mov vIncr ,>vPtr
vScan seq vStep-6 ,vStep
mov.ab vScan ,vPtr
add vIncr ,vScan
djn vNext ,#7900/16
vIncr spl #vStep ,#vStep
mov vClr ,>vGate
djn.f -1 ,>vGate
vClr dat <2667 ,2-vGate
for 8
dat 0,0
rof
slPaper spl wimp
spl 1 ,{1000
spl 1 ,{2000
spl 1 ,{3000
p1s mov p1b,>8 ; imp- and stone-killing paper
p1c mov <p1s,<p1n ; pretty good at killing silks too
p1n spl @p1n,>-1356
mov p1b,<p1n
mov p1b,<-1000
mov p1b,>200
jmn.f @0,>p1s
p1b dat <2667,<5334
end think
"History has pretty well established that no program is going to last
forever, publishing accelerates the demise yet advances the game."
_______________________________________________________________________________
Questions? Concerns? Comments? Complaints? Mail them to people who care.
Beppe Bezzi <giuseppe.bezzi@galactica.it>, Philip Kendall <pak21@cam.ac.uk>,
Anton Marsden <anton@paradise.net.nz>, John Metcalf <grumpy@digitald.uk.com>
and Christian Schmidt <schmidt@chiral.apchem.nagoya-u.ac.jp>
|
