He Scans Alone

[Stephen Gunnell]

Many years ago I was investigating the components of Recycled Bits and I pulled apart the scanner and rebuilt it to be one line shorter. The result was a JMZ scanner with a decoy generator that I called Keshin. Because an ADD.F instruction will change both the A and B fields I tried to develop a dual pointer JMN/JMZ scanner. Aoshi was never quite good enough to make the hills but not so bad that I purged the code. Around the same time I was familiar with Mini HSA from Anton Marsden's Electric Head which was a fairly unprepossessing JMZ scanner so I never investigated Paul Kline's He Scans Alone (which I guess was another fail :). Fast forward to August 2022. I had just developed some code to approximate the contents of the hill from the warriors on hand. This is a really old Idea but it needs modern processor speeds to be effective. And effective it is. One of the old warriors it highlighted was He Scans Alone. When I looked at it there was the Aoshi core but with a brilliant SNE/SUB front end that vastly improved the efficiency. So I tinkered ...

;redcode-94nop
;name He Scans Alone qhoz
;author P.Kline
;strategy 80% f-scanner switches from SPL to DAT carpet
;strategy Tweaked and upcycled by Steve G
;assert CORESIZE == 8000

STEP equ 10
TIME equ 19
OFFSET equ 2813

tPtr dat 100 ,OFFSET ; widely-spaced pointers
dat 0 ,0
for 3
dat 0 ,0
rof

tWipe mov tSpl ,<tPtr ; positive wipe of opponent
mov >tPtr ,>tPtr
jmn.f tWipe ,>tPtr

tScan sub.x #0-STEP ,tPtr ; increment and look
sne *tPtr ,@tPtr
sub.x *pScan ,@tScan ; increment and look
jmn.f tSelf ,@tPtr
jmz.f tScan ,*tPtr
pScan mov.x @tScan ,@tScan ; swap pointers for attack
tSelf slt.b @tScan ,#tEnd+4-tPtr ; self-check
djn tWipe ,@tScan ; go to attack
djn *pScan ,#TIME ; after 19 self-scans
jmp *pScan ,}tWipe ; switch to dat-wiping
dat 0 ,0
tSpl spl #1 ,{1
dat 0 ,0
dat 0 ,0
tEnd dat 0 ,0

for MAXLENGTH-CURLINE-2-1
dat 0 ,0
rof

;tDecoy equ (tWipe+7007)
;tStart mov <tDecoy+0 ,{tDecoy+2 ; make a quick-decoy
; mov <tDecoy+3 ,{tDecoy+5 ; to foil one-shots
; mov <tDecoy+6 ,{tDecoy+8 ; and the occasional q-scan
; djn.f tScan+1 ,<tDecoy+10

B equ (tWipe+7007)

tStart mov.i <B ,{B+3
I for 1
mov.i <B+1+(3*I) ,{(B+(3*I)+3)
rof
djn.f tScan+1 ,<B+5+(3*1)

end tStart

[Stephen Gunnell]

On Sunday, 30 October 2022 at 7:16:48 am UTC+8, Stephen Gunnell wrote:

> ;name He Scans Alone qhoz

> tScan sub.x #0-STEP ,tPtr ; increment and look

It is probably worth noting that I kept STEP as a fairly small value as did Paul in his original code ( the optimiser selected 10/8 in qhoz and the original was 12/9 ). During the development of Aoshi I also noted that using different steps for the A and B fields was very effective. I also note that a mod 10 step seems to also be a value that my Hazy Lazy scanners gravitate towards. If you set STEP to be a large mod 10 value the A field may efficiently scan the core and when it found something it would pass the value to the B field which would scan forwards in small increments.

Steve

[John Metcalf]

Congrats, impressive score on 94nop :-) I'd be interested to compare
which warriors your code selected with those in my hand-picked benchmark.

There are a few tweaks in some other HSA-style scanners which might be
worth trying:

Seven-Eleven (.8c, steps 7/11), DAT 0, {0 for the dat-wipe

Razor (.66c, steps 7/9), perfect gate, SNE.I scanner

Willow (.66c, steps 13/9), DAT >0, 0 for the dat-wipe

Ash (.8c, steps 17/11), use SNE.X and alternates between SUB.F and SUB.X

Trouble (.8c, steps 11/16), start scan at 3rd instruction after self-scan

Here's the code for Trouble:

;redcode-94nop
;name Trouble
;author John Metcalf
;strategy scanner
;assert CORESIZE==8000

x equ 3100
y equ 600
stepx equ 11
stepy equ 16
count equ 16

ptr dat y, x+y-1

for stepx-8
dat 0, 0
rof

dat >0, 0
sb spl #0, {0

wipe mov @bp, <ptr
mov >ptr, >ptr
jmn.f wipe, >ptr

scan sub.x inc, ptr
sne.x @ptr, *ptr
inc sub.x #-stepy, ptr
jmn.f hit, @ptr
jmz.f scan, *ptr

mov.x @scan, @scan
hit slt.b @scan, #last+2-ptr
djn.f wipe, ptr
djn inc, #count
bp djn.f inc, #sb

last end scan+1

[Stephen Gunnell]

On Monday, 7 November 2022 at 2:35:22 am UTC+8, John Metcalf wrote:

> Congrats, impressive score on 94nop :-) I'd be interested to compare
> which warriors your code selected with those in my hand-picked benchmark.
>

The current list is:
Starting WarriorZ ... lore2.red - 0.963639104067866
Starting Her IMPertinent Majesty ... yellownewspaper.red - 0.968492684024264
Starting Visigoth ... sharky.red - 0.97160387599864
Starting biSHO ... k310.red - 0.971778549072255
Starting ShadowWeaver ... borgir.red - 0.969684633279653
Starting dreadful cross ... lastjudgement.red - 0.977390396931096
Starting AshII ... hsaopt-qhoh.red - 0.979644545493337
Starting Resinoid ... lordimprings2.red - 0.963729691567763
Starting King Cobra ... twilight.red - 0.935340724763019
Starting Tolyputes ... blowrag2.red - 0.966682391867335
Starting Despair ... borgir.red - 0.970876224369984
Starting Pocket Rocket ... zplusplus.red - 0.929591489465872
Starting Electra ... lordimprings2.red - 0.974757202343255
Starting Monster Human Grunt ... elvenking2.red - 0.975380831472071
Starting incubus ... hullabaloo3.red - 0.969622420270168

The number at the end is the correlation coefficient between the selected proxy warrior's scores and the scores of the real warrior.

Here is the corresponding list for 94x:
Starting DriedFrogPillsAX9 ... dfpopt-qreo.red - 0.988565232237808
Starting PuddleglumX ... puddleglum-xjcl.red - 0.99928754183971
Starting paperX ... grimoireA-ugbc.red - 0.991370859383051
Starting DriedFrogPillsAX133 ... dfpopt-nxac.red - 0.995470490449125
Starting Morlock ... deadletter-xlwm.red - 0.931948655588622
Starting PendulumX ... memories07x.red - 0.932070614564449
Starting BlackMoods ... dflyoptB-prrt.red - 0.887106456201055
Starting Slayer ... mavinopt1-dtat.red - 0.93102766140238
Starting DespairX ... deadletter-tiru.red - 0.98989167508046

I left PuddleglumX floating as a sanity check and the program has correctly found the correct version. For the two copies of Dried Frog Pills the originals were deleted and I let the program select stand ins.
As you can see with Black Moods I have no good approximations in my repertoire.

I'm happy to publish the code if people are interested but it is in Perl.

Steve

[Stephen Gunnell]

On Monday, 7 November 2022 at 6:50:41 am UTC+8, Stephen Gunnell wrote:

> Starting Resinoid ... lordimprings2.red - 0.963729691567763

> Starting Electra ... lordimprings2.red - 0.974757202343255

I got annoyed at multiple unknown warriors mapping to the same proxy so I added code to allow unknown warriors to differentiate from each other.
94nop is now:

Starting Monster Human Grunt ... elvenking2.red - 0.975380831472071

Starting biSHO ... k310.red - 0.971778549072255

Starting incubus ... spiritual.red - 0.970529104184943

Starting Despair ... borgir.red - 0.970876224369984

Starting WarriorZ ... lore2.red - 0.951270800865769
Starting Resinoid ... unheardof.red - 0.954543201503372
Starting dreadful cross ... lastjudgement.red - 0.977390396931095

Starting Pocket Rocket ... zplusplus.red - 0.929591489465872

Starting King Cobra ... twilight.red - 0.935340724763019

Starting Her IMPertinent Majesty ... yellownewspaper.red - 0.968492684024264

Starting Electra ... lordimprings2.red - 0.980289980569215

Starting Visigoth ... sharky.red - 0.97160387599864

Starting Tolyputes ... blowrag2.red - 0.966682391867335

Starting ShadowWeaver ... borgir.red - 0.969684633279653

Mind you, I'm struggling to keep up with all John's submissions and it isn't helping that messages from the hill are randomly dissapearing.

Steve

[John Metcalf]

I'd be interested :-) It appears to be doing a good job of matching the
warriors to a proxy.

John