Combatra is a p-switcher with specialist components which attempt to calculate and exploit the opponent's boot distance. Combatra entered the Pizza 94 draft hill on 5th October 2000 and reached age 40 before the hill closed in 2001. Combatra also entered the KOTH 94 hill on 1st December 2002 and survived 131 challenges before being pushed off on 29th March 2003. The code was published in Core Warrior issue 78.
;redcode-94
;name Combatra
;author David Moore
;strategy Start each battle with a few quick scans. If nothing
;strategy is found, then run either a core clear or a bomber.
;strategy Otherwise, assume that Combatra found the opponent's
;strategy original 100 instructions. Probe to discover where its
;strategy processes run. Remember where everything is. Compute
;strategy the bootstrap distance. Later, a scanner finds the enemy
;strategy simply by locating its enormous boot code.
;assert CORESIZE==8000 && MAXLENGTH==100
org start
;----------------------------------------------------------------------
; the core clear is used either alone or with the scanner
cgate dat -1, sFirst
dat <5334, <2667
dat -1, 12
clear spl #3891, sScan ; spl #3891, 18
mov -1, >cgate
mov *-1, >cgate
djn.a -2, cgate
djn.a -3, *-1
dat 0,0
dat 0,0
dat 0,0
dat 0,0 ; scanned
sDiff equ 65 ; difference between sScan and cgate after boot
sStep equ 98
sFirst equ 1283
; Boot the scanner to take advantage of the magic number
sBoot mov <away, <where
ldp.a #pBoot, sStore
mov <away, <where
mov >clear, >sJump
djn -2, #7
sJump djn.f start+dist-3+sDiff+1, #start+dist-3+sDiff
sScan add.ab #sStep+1, @2
jmz.f sScan, <sScan-sDiff
jmz.f sScan, <sScan-sDiff ; ignore the loners
spl (-3-sDiff)+3, <(-3-sDiff) ; write over what I just found
djn.a #6, #4 ; after about 12 SPLs, change
sStore add.ab #0, -5-sDiff ; attack to hit booted opponent
dat 0,0
;----------------------------------------------------------------------
; bomber taken from Kline's Floody River and hacked
bstep equ 7829
btime equ 863
btime2 equ 420
bomb spl #bstep*2, -bstep
mov {(bomb-2)+bstep*btime2, @3-(bstep*2)*btime
mov bomb, @-1
add.x bomb, -2
djn.f -3, {-(bstep*2)*(btime-1) ; hit by SPL to start clear
mov 2, >bomb-3
djn.f -1, >bomb-3
dat <2667, 6-bomb
;----------------------------------------------------------------------
; Did I find something last time?
check ldp.a #pTroll, tab
add.a #5, tab ; ignore small numbers
tab slt.ab #0, #11
jmp compute, 9 ; (yes, go compute the boot distance)
jmp bBoot, 10 ; (no, keep trolling)
; Compile all of my information into one nice number
compute ldp.a #pQuick, diff
diff sub.a #0, tab
add.a #((start-qPtr)+tDist+5-6)-11-196, tab
stp.ab tab, #pBoot
stp.ab #6, #pMode ; run scanner next time
jmp boot, 0 ; I'm too slow to risk booting everything now
dat 0,0
dat 0,0
dat 0,0
dat 0,0 ; scanned
pMode equ 271 ; overall strategy (code for one of the 11 states)
pBoot equ 292 ; opponent's bootstrap distance
; intermediate results used to compute pBoot
pQuick equ 433 ; what cell was quickscanned
pTroll equ 454 ; what cell was occupied by the opponent
start ldp.a #0, tab
ldp.a #pMode, state
mod.ba *tab,state; current; next loss win tie
stp.b *1, #pMode ; state ; state ---- --- ---
state jmp }0, 472 ; 0 check; 472= 42*11+ 10 = 52*9+ 4 = 47*10+ 2
spl #check, 472 ; 1 check; 472= 42*11+ 10 = 52*9+ 4 = 47*10+ 2
spl #bBoot, 472 ; 2 bomb ; 472= 42*11+ 10 = 52*9+ 4 = 47*10+ 2
spl #sBoot, 943 ; 3 scan ; 943= 85*11+ 8 =104*9+ 7 = 94*10+ 3
spl #bBoot, qA ; 4 bomb ;3874=352*11+ 2 =430*9+ 4 =387*10+ 4
qTable spl #cBoot, qB ; 5 clear;2705=245*11+ 10 =300*9+ 5 =270*10+ 5
spl #sBoot, qC ; 6 scan ;1536=139*11+ 7 =170*9+ 6 =153*10+ 6
spl #sBoot, 267 ; 7 scan ; 267= 24*11+ 3 = 29*9+ 6 = 26*10+ 7
spl #sBoot, 768 ; 8 scan ; 768= 69*11+ 9 = 85*9+ 3 = 76*10+ 8
spl #sBoot, 849 ; 9 scan ; 849= 77*11+ 2 = 94*9+ 3 = 84*10+ 9
jmp #cBoot, 455 ;10 clear; 455= 41*11+ 4 = 50*9+ 5 = 45*10+ 5
;----------------------------------------------------------------------
; program intended to "troll" the enemy into revealing its location
tStep equ 98
tTime equ 2448
troll mov @0, *5
add.f 3, 4
jmz.a -2, 1
stp.ab #0, #pTroll
jmp #tStep, <-tStep
; this bomb causes my foe to report its location to me
mov.a #(-2-tStep*tTime), -2-(-2-tStep*tTime)
; quickscan got something! (presumably boot code)
tDist equ 2100 ; difference between start and troll after boot
tBoot mov <tJump, {tJump
mov <tJump, {tJump
djn -2, #3
qCalc mul.ba qTable, qPtr ; compute the scanned location
qPtr sne qX, datZero
add.a #qDiff, qPtr
stp.ab qPtr, #pQuick ; remember where I found it
stp.ab #1, #pMode ; next time, check the results
stp.ab #0, #pTroll ; forget previous troll now
tJump djn.f start+tDist+6, #troll+6 ; start trolling
bAway dat -3, bomb-(clear-3)
dat 0,0
datZero dat 0,0
dat 0,0
dat 0,0 ; scanned
bBoot add.f bAway, away ; converts clear boot into bomber boot
; note: changing the distance between qTable and qPtr affects
; both the quickscan and the pspace!
qX equ 5072
qInv equ 6831 ; inverse of qX-1 (such that (qX-1)*qInv==1)
qDiff equ 5983
qA equ (((qX+(qTable-qPtr)-2)*qInv)%CORESIZE)
qB equ (((qX+(qTable-qPtr)-1)*qInv)%CORESIZE)
qC equ (((qX+(qTable-qPtr)-0)*qInv)%CORESIZE)
cBoot sne qPtr+(qA-0)*qX, qDiff+qPtr+(qA-0)*qX
seq <qTable-1, qDiff+qPtr+(qA-1)*qX
jmp tBoot, {qCalc
sne qPtr+(qC-0)*qX, qDiff+qPtr+(qC-0)*qX
seq <qTable+1, qDiff+qPtr+(qC-1)*qX
jmp tBoot, }qCalc
sne qPtr+(qB-0)*qX, qDiff+qPtr+(qB-0)*qX
seq <qTable+0, qDiff+qPtr+(qB-1)*qX
where jmp tBoot, start+dist+5
dist equ 6942 ; distance between start and clear after boot
boot mov <away, <where
mov <away, <where
mov <away, <where
mov <away, <where
djn -4, #2
away djn.f start+dist, #clear+5
end