From: nowhere@never.at
Kerr Mudd-John wrote:
>> now it's more than obvious that a direct 4:3 SMC gain at least
>> 25 bytes against all 2:1 variants:
>> all modifications can be done within the first two lines and
>> the string itself is halfed and now fits the third line !
> This was the realization for me; that the (eventual) B64 decoder was less
> than a line's width.
> I've been assuming that it must be the first line (bar a jmp past it, but
> even that can be fixed up!) with the fixup code afterwards;
> how do you otherwise deal with the crlf/cr/lf/nul possible line endings
> hitting the wrong fixup locations?
I put my 56byte decoder string at 0184 starting with four'36' nops,
so it wont contain any CRLF.
> the move code itself needs to be fixedup or written on the fly, so I leave it
> 'til after fixing in place.
>> we had ~016B as stringstart for 2:1.
>> now I made it 0184 (4 slide nops) to not have CRLF inbetween.
>> even this are 25 byte more it allow REP MOVS after start-scan.
>> further size reduction seems possible, I'll check.
>> btw: you posted something to the other 'too long' thread,
>> I see the header but can't open it.
> Not that exciting: saved a jmp, cost a lodsb:
> On Fri, 26 May 2017 08:24:38 +0100, Kerr Mudd-John
> wrote:
> ll63 bytes 252
> Saved a few more! (OK just 1 in the 2:1 stage2 decoder)
> ll64 bytes 249
> ZRPRhF0XPR5F1PRRjF4gPa5LU1GI1GKRX+G=1GdSVG8+r6238wat083t+RUZ=rY=
> 662202Zffffff0222F6====MJ=20=17===eeee+++slOBI9UnoAlpinMIupasbaA
> NE9E7jelperHQE9sAWKsuH8lpssMIG7asA9babaRKi8A99Wh8G7w7d8v7h8WKh8f
> 7I8sCoAlpadITw7YoH9f7=tAm6CAHNIcNIZWxsbyB3b3JsZCEk=
thanks.
> Here's the 47 byte b64 decoder (that I'm trying to fixup)
let me count:
3 Singles 100,114,118,(12e,+ 12f RET)
1 8x single 110, (80.. are only pairable if low byte with my way)
12 Pairs 108,10e,111,11b,11d,11f,121,123,125,127,129,12e
16 modifiers needed, shouldn't be a problem..
> ->u100 l2E
> 1622:0100 AC lodsb
> 1622:0101 3C 2B cmp al,2B
> 1622:0103 72 25 jb 012A
> 1622:0105 2C 41 sub al,41
> 1622:0107 73 0A jnb 0113
> 1622:0109 04 45 add al,45
> 1622:010B 3C 33 cmp al,33
> 1622:010D 77 0A ja 0119
> 1622:010F B0 86 mov al,86
> 1622:0111 1C 00 sbb al,00
> 1622:0113 3C 1A cmp al,1A
> 1622:0115 72 02 jb 0119
> 1622:0117 2C 06 sub al,06
> 1622:0119 41 inc cx
> 1622:011A 41 inc cx
> 1622:011B 83 E1 06 and cx,0006
> 1622:011E D2 EC shr ah,cl
> 1622:0120 C0 E0 02 shl al,02
> 1622:0123 D3 E0 shl ax,cl
> 1622:0125 86 E0 xchg ah,al
> 1622:0127 E3 01 jcxzw 012A
> 1622:0129 AA stosb
> 1622:012A AC lodsb
> 1622:012B 3C 3D cmp al,3D
> 1622:012D 75 D2 jnz 0101
why mov al,86 ?
I hope to not have inserted too many typos here:
my decoder is a bit larger (56 byte):
it needs: 5 singles| 1 paired 8x | 9 pairs (=15)
044 41 41 41 41 inc cx dup4 ;cx=0 from previous
048 55 55 push bp dup2
04a 66 8f 05 pop dword [DI] ;clr four (p)
04d AC lodsb ;(s)
04e 3c 2b cmp al,2b
050 72 fb jc 04d ;(s)
052 3c 3d cmp al,3d
054 75 01 jnz 057 ;
056 C3 ret ;c301 (p)
057 2c 47 sub al,47
059 73 10 jnc 06b ;0410 (p),06(s)
05b 04 06 add al,6 ;no gain with sub al,fa
05d 3c 1a cmp al,1a ;(s)
05f 72 0a jc 06b ;040a (p)
061 04 45 add al,45
063 3c 33 cmp al,33
065 77 04 ja 06b ;b004 (p)
067 b0 3f mov al,3f
069 1c 00 sbb al,00 ;(p)
06b 08 05 or [di],al ;(p)
06d 66 c1 05 46 ROR dword [di],6;(p)
071 e2 da loop 04d ;(p)
073 66 c1 2d 48 SHR dword [di],8;(s)
077 47 47 47 inc di dup 3
07a eb c8 jmp 044 ;(p)
but you might be more interested in how I modify:
100 pop cx
101 push cx
102 push cx ;ax=0
103 push byte 56 ;cx=0056
105 push 4130 ;dx=4130 to find my first byte with DH
108 pop ax
109 push ax
10a xor ax,4030
10d push ax ;bx=0100
10e push cx dup2 ;sp,bp=0
110 sub al,7e ;
112 push ax ;si=0182
113 push byte 044 ;di=0044
115 popa ;
116 push BX ;for ret
117 push DI ;for ret
118 sub ax,5c73 ;-> a38d
11b xor [bx+5b],ax ;make a back-branch to 0120
11e jnc 0142 ;=jmp always after xor (73 22)
120 cmp [si],dh ;search my first byte (41)
122 jnz 0158 ;skip
124 push bx
125 3636 REP MOVSB ;
127 pop di
128 sub [bp+7a],ax ;ax= 7761 [07a] was 404c
12b sub [bp+74],al ; [074] was 22
12e sub [bp+71],ax ; and so on...
131 sub [bb+6e],ax
134 sub [bp+6b],ax
137 sub [bp+69],ax
13a sub [bp+66],ax
13d jnz 0160 ;=jmp always 75 21
13f 3d ;unused can be any
140 0d 0a
142 36 36
144 push bp
145 pop ax
146 sub ax,6d3b ;92c5 xor 3636 = A4F3
149 xor [bx+25],ax ;create REP MOVS
14c sub al,21 ;c5-21=a4
14d inc ax dup2 ;a6 xor 65 = c3
14f xor [bx+7e],al ;create a RET in a field of 65 nops
152 36 ;just a forgotten byte
153 push 7761
156 pop ax
157 dec si
158 36 36
15a inc si
15b 6666 JMP 120
15d 6666
15f 36
160 sub [bp+5e],al
163 sub [bp+60],ax
166 sub [bp+5c],al
169 sub [bp+5a],ax
16c sub [bp+55],ax
16f sub [bb+51],al
172 sub [bp+4d],al
175 sub al,7a
178 sub [bp+4b],ax
17b 363636 ;free for one more
17e 65656565 ;the return goes there
182 0d0a
184 36363636
188 41414141 ... my raw string resides here
__
wolfgang
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
