From news-rocq!jussieu.fr!freenix!sunqbc.risq.qc.ca!wesley.videotron.net!weber.videotron.net.POSTED!not-for-mail Tue Jul 25 11:25:28 2000
Article: 11560 of rec.games.corewar
Path: news-rocq!jussieu.fr!freenix!sunqbc.risq.qc.ca!wesley.videotron.net!weber.videotron.net.POSTED!not-for-mail
Message-ID: <397C497E.13EF725D@videotron.ca>
From: pk <thida@videotron.ca>
X-Sender: "pk" <vlohqbsf@pop.videotron.ca> (Unverified)
X-Mailer: Mozilla 4.05 [en]C-NECCK  (Win95; I)
MIME-Version: 1.0
Newsgroups: rec.games.corewar
Subject: Help! Mayday! SOS! Whatever 8)
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Lines: 98
Date: Mon, 24 Jul 2000 09:49:50 -0400
NNTP-Posting-Host: 24.201.17.77
X-Complaints-To: abuse@videotron.net
X-Trace: weber.videotron.net 964446517 24.201.17.77 (Mon, 24 Jul 2000 09:48:37 EDT)
NNTP-Posting-Date: Mon, 24 Jul 2000 09:48:37 EDT
Xref: news-rocq rec.games.corewar:11560

Well, it seems that after months of work, i've gotten a working
non-one-shot scanner using the:
Add 	inc, 1
Mov.i	<x, {y
Jmz.f	-2, <-1
(it'll call it... FI for now(fine idea or f***ing idiot, as you wish 8)
In this case, it use y=x+25, and a step of 30. For bombing, it uses
active carpetting ala HSA(thanx Macrae 8), and then a 4 line clear(i've
one version that uses a backward spl/dat/spl... clear, and the other one
uses dclear)... It doesn't do too bad against papers, but all others
just kill it 8(

Side comment: A backward clear with the pointer just after the clear
kills CLP: it puts a vamp on the pointer, the clear start clearing
exactly where CLP is, going right through the fences 8) Sorry if this is
common knowledge, but i was just testing for fun, and FI completely beat
CLP 8)

The problems I see:
Big footprint(like 30 non zero instructions 8): this is the only reason
i can see why scanners kill it, since it trashes the core at the same
speed as emerald, or it might just be the step... i'll think about a mod
30 step later 8)
Have to find as better way of switching the scan positions & going to
clear

Here they are:

Ver 1(spl/dat/spl/... clear)
;redcode
;name FI V1
;author PVK, but don't tell anyone
;assert 1

	Org Scan-1
x	equ	(scan-15); neat trick: since the two scan positions are 15 cells
apart, moving the b-field to 			 ;scan-15 points directly to scan+x 8)

clear	spl	#-5,	{-2 ;not sure about the {-2
	mov	@4,	{4
	jmp	-1,	<-4
db	dat.f	-5,	-5
	spl	#-100,	-100
inc	dat.f	31,	30
sb	spl	#1,	0
for 22
	spl	#0,0
rof
	add.x	inc,	1
scan	mov.i	<-2,	{13
	jmz.f	-2,	<-1
chkself	slt.ab	scan,	#-8
	djn.f	scan-1,	scan
	jmp	attack,	-100;i think that this line is hit, if not, i'll just o the
b-field
db2	dat.f	-5,	-5;inc's b-field point to this line
for 9
	dat.f	-100,	-100 ; might not be needed, except for the 1st one, but i'm
not sure
rof
attack	mov.b	scan,	x
	mov.i	sb,	<x
	mov.i	>x,	>x
	jmn.f	-2,	>x
check	jmn.b	scan-1,	-1
	jmp clear

to have the version with a dclear, just change the lines before the "for
22" in:

	org	scan-1
x	equ	scan-15

inc	dat.f	31,	30

clear	spl	#0,0
	mov.i	2,	>-2
	djn.f	-1,	>-3
	dat.f	>5335,	6
sb	spl	#1,	0
	spl	#0,0

There is something in the version with the multipass clear that leaves
some process running in the attack block, i think, and bc of its
position, it survives the first self-dat-wipe, get spl bombed, and only
after like 24000 cycles get dat-bombed... so i have processes running
spl 0 for a long time, which worries me, even though it doesn't seem to
affect the clearing too much(no visible speed change)

Anybody has idea to ameliorate FI?

TIA
 Paul
PS. this seems to be the first warrior listing presented to the list in
months... what's happening?? 8)
--
AKA paul_virak_khuong at yahoo.com, pkhuong at deja.com, pkhuong at
crosswinds.net and pkhuong at technologist.com(list not complete)...