From news-rocq.inria.fr!jussieu.fr!univ-lyon1.fr!newsfeed.nacamar.de!news.maxwell.syr.edu!news.mel.connect.com.au!munnari.OZ.AU!comp.vuw.ac.nz!paramount.mcs.vuw.ac.nz!amarsden Fri Mar  6 11:34:53 1998
Article: 8511 of rec.games.corewar
Path: news-rocq.inria.fr!jussieu.fr!univ-lyon1.fr!newsfeed.nacamar.de!news.maxwell.syr.edu!news.mel.connect.com.au!munnari.OZ.AU!comp.vuw.ac.nz!paramount.mcs.vuw.ac.nz!amarsden
From: Anton Marsden <amarsden@mcs.vuw.ac.nz>
Newsgroups: rec.games.corewar
Subject: P^2b - another p-switcher
Date: Fri, 6 Mar 1998 15:38:07 +1300
Organization: School of Mathematical and Computing Sciences, VUW
Lines: 70
Message-ID: <Pine.GSO.3.96.980306153506.4764A-100000@paramount.mcs.vuw.ac.nz>
Reply-To: Anton Marsden <amarsden@mcs.vuw.ac.nz>
NNTP-Posting-Host: kaukau.mcs.vuw.ac.nz
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Cache-Post-Path: kaukau.mcs.vuw.ac.nz!unknown@paramount.mcs.vuw.ac.nz
X-Cache: nntpcache 2.3.2.1 (see http://www.nntpcache.org/)

P^2b is based upon the ideas presented in the P^2 switcher. P^2 has a major
weakness: it needs a very large table to implement even simple state
transitions. Generally, the state transition rules we use for one component 
can be used with another - but P^2 doesn't take this into account. P^2b
attempts to use this transition similarity property.

P^2b takes 7 instructions to execute (as opposed to 6 for P^2).

In order to keep the code size to a minimum, it is required that state 0 and
state 1 execute the same warrior. Apart from this restriction, it should be
possible to implement any state transition table that was possible in P^2.
However, keep in mind that the initial state is a little bit tricky to set
up properly - this is left as an exercise for the reader.

I'm not going to describe how P^2b works any further - there are some hints
in the code and I don't have much time at the moment. If you don't
understand it, figure out how P^2 works first.

;redcode-94
;name P^2b
;author Anton Marsden
;strategy P-switcher
;assert CORESIZE==8000

PSTATE  EQU 666                 ; pspace location containing current state
STATES  EQU 9                   ; number of states

;T1, T2 and T3 correspond to different states a component can be in.
;In this case, T2 and T3 use the same win/loss/tie table to switch states.

T1 EQU (t1-in)
T2 EQU (t2-in)
T3 EQU (t2-in)

think   ldp.a  #0,in            ; get input value
load    ldp.a  #PSTATE,state    ; load old state
        mod.a  #STATES,state    ; brainwash protection
        add.ba *state,in        ; select correct state table
in      add.ba 0,state
store   stp.a  state,load       ; store new state

state   jmp    }0,T1            ; jump to warrior code
        dat    w1,T2            ; if state.A == 0, the A field of this 
                                ; instruction will be jumped to.
        dat    w1,T3
        dat    w2,T1
        dat    w2,T2
        dat    w2,T3
        dat    w3,T1
        dat    w3,T2
        dat    w3,T3

t2      dat    w1,1     ; loss - add 1 to the state
        dat    0,-1     ; win - subtract 1 from the state
        dat    0,0      ; tie - leave the state as it is
t1      dat    0,1      ; loss - add 1 to the state
        dat    0,0      ; win - leave the state as it is
        dat    0,0      ; tie - leave the state as it is

;Warrior code

w1:     jmp    #1,1
w2:     jmp    #2,2
w3:     jmp    #3,3

END think