Merge pull request #66 from emmanuel-marty/spke/new_z80_small

New small decompressor for LZSA2 (-5 bytes, +1% speed)
This commit is contained in:
Emmanuel Marty 2022-04-05 18:22:18 +02:00 committed by GitHub
commit 86b7fe0738
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -1,5 +1,5 @@
;
; Size-optimized LZSA2 decompressor by spke & uniabis (139 bytes)
; Size-optimized LZSA2 decompressor by spke & uniabis (134 bytes)
;
; ver.00 by spke for LZSA 1.0.0 (02-09/06/2019, 145 bytes);
; ver.01 by spke for LZSA 1.0.5 (24/07/2019, added support for backward decompression);
@ -7,6 +7,8 @@
; ver.03 by spke for LZSA 1.0.7 (01/08/2019, 140(-4) bytes, -1.4% speed and small re-organization of macros);
; ver.04 by spke for LZSA 1.1.0 (26/09/2019, removed usage of IY, added full revision history)
; ver.05 by spke for LZSA 1.1.1 (11/10/2019, 139(-1) bytes, +0.1% speed)
; ver.06 by spke (11-12/04/2021, added some comments)
; ver.07 by spke (04-05/04/2022, 134(-5) bytes, +1% speed, using self-modifying code by default)
;
; The data must be compressed using the command line compressor by Emmanuel Marty
; The compression is done as follows:
@ -55,8 +57,8 @@
; 3. This notice may not be removed or altered from any source distribution.
;
; DEFINE BACKWARD_DECOMPRESS ; uncomment for data compressed with option -b
; DEFINE HD64180 ; uncomment for systems using Hitachi HD64180
; DEFINE BACKWARD_DECOMPRESS ; uncomment for data compressed with option -b (+5 bytes, -3% speed)
; DEFINE AVOID_SELFMODIFYING_CODE ; uncomment to disallow self-modifying code (-1 byte, -4% speed)
IFNDEF BACKWARD_DECOMPRESS
@ -65,7 +67,7 @@
ENDM
MACRO ADD_OFFSET
ex de,hl : add hl,de
add hl,de
ENDM
MACRO BLOCKCOPY
@ -79,7 +81,10 @@
ENDM
MACRO ADD_OFFSET
push hl : or a : sbc hl,de : pop de ; 11+4+15+10 = 40t / 5 bytes
;push hl : or a : sbc hl,de : pop de ; 11+4+15+10 = 40t / 5 bytes
; HL = DE - HL
ld a,e : sub l : ld l,a
ld a,d : sbc h : ld h,a ; 6*4 = 24t / 6 bytes
ENDM
MACRO BLOCKCOPY
@ -88,85 +93,90 @@
ENDIF
IFNDEF HD64180
@DecompressLZSA2:
; in many places we assume that B = 0
; flag P in A' signals the need to re-load the nibble store
xor a : ld b,a : exa : jr .ReadToken
MACRO LD_IX_DE
ld ixl,e : ld ixh,d
ENDM
.CASE00x: ; token "00Z" stands for 5-bit offsets
; (read a nibble for offset bits 1-4 and use the inverted bit Z
; of the token as bit 0 of the offset; set bits 5-15 of the offset to 1)
push af
call ReadNibble.skipLDCA : ld c,a
pop af
cp %00100000 : rl c : jr .SaveOffset
MACRO LD_DE_IX
ld e,ixl : ld d,ixh
ENDM
.CASE0xx dec b : cp %01000000 : jr c,.CASE00x
.CASE01x: ; token "01Z" stands for 9-bit offsets
; (read a byte for offset bits 0-7 and use the inverted bit Z
; for bit 8 of the offset; set bits 9-15 of the offset to 1)
cp %01100000
.doRLB rl b
.OffsetReadC: ld c,(hl) : NEXT_HL
IFNDEF AVOID_SELFMODIFYING_CODE
.SaveOffset: ld (.PrevOffset),bc : ld b,0
ELSE
MACRO LD_IX_DE
push de : pop ix
ENDM
MACRO LD_DE_IX
push ix : pop de
ENDM
.SaveOffset: push bc : pop ix : ld b,0
ENDIF
@DecompressLZSA2:
xor a : ld b,a : exa : jr ReadToken
.MatchLen: and %00000111 : add 2 : cp 9
call z,ExtendedCode
CASE00x: call ReadNibble
ld e,a : ld a,c
cp %00100000 : rl e : jr SaveOffset
.CopyMatch: ld c,a
push hl ; BC = len, DE = dest, HL = -offset, SP -> [src]
CASE0xx ld d,#FF : cp %01000000 : jr c,CASE00x
IFNDEF AVOID_SELFMODIFYING_CODE
.PrevOffset EQU $+1 : ld hl,0
ELSE
push ix : pop hl
ENDIF
ADD_OFFSET
BLOCKCOPY ; BC = 0, DE = dest
pop hl ; HL = src
CASE01x: cp %01100000 : rl d
.ReadToken: ld a,(hl) : NEXT_HL : push af
and %00011000 : jr z,.NoLiterals
OffsetReadE: ld e,(hl) : NEXT_HL
SaveOffset: LD_IX_DE
rrca : rrca : rrca
call pe,ExtendedCode
MatchLen: and %00000111 : add 2 : cp 9 : call z,ExtendedCode
ld c,a
BLOCKCOPY
CopyMatch: ld c,a
ex (sp),hl ; BC = len, DE = -offset, HL = dest, SP -> [src]
ADD_OFFSET ; BC = len, DE = dest, HL = dest+(-offset), SP -> [src]
BLOCKCOPY ; BC = 0, DE = dest
pop hl ; HL = src
.NoLiterals: pop af : or a : jp p,.CASE0xx
ReadToken: ld a,(hl) : NEXT_HL : push af
and %00011000 : jr z,NoLiterals
.CASE1xx cp %11000000 : jr c,.CASE10x
; token "111" stands for repeat offsets
; (reuse the offset value of the previous match command)
cp %11100000 : jr nc,.MatchLen
rrca : rrca : rrca
call pe,ExtendedCode
.CASE110: ; token "110" stands for 16-bit offset
; (read a byte for offset bits 8-15, then another byte for offset bits 0-7)
ld b,(hl) : NEXT_HL : jr .OffsetReadC
ld c,a
BLOCKCOPY
.CASE10x: ; token "10Z" stands for 13-bit offsets
; (read a nibble for offset bits 9-12 and use the inverted bit Z
; for bit 8 of the offset, then read a byte for offset bits 0-7.
; set bits 13-15 of the offset to 1. substract 512 from the offset to get the final value)
call ReadNibble : ld b,a
ld a,c : cp %10100000
dec b : jr .doRLB
NoLiterals: pop af : push de
or a : jp p,CASE0xx
CASE1xx cp %11000000 : jr nc,CASE11x
CASE10x: call ReadNibble
ld d,a : ld a,c
cp %10100000 ;: rl d
dec d : rl d : DB #CA ; jr OffsetReadE ; #CA is JP Z,.. to skip all commands in CASE110 before jr OffsetReadE
CASE110: ld d,(hl) : NEXT_HL : jr OffsetReadE
CASE11x cp %11100000 : jr c,CASE110
CASE111: LD_DE_IX : jr MatchLen
ExtendedCode: call ReadNibble : inc a : jr z,ExtraByte
sub #F0+1 : add c : ret
ExtraByte ld a,15 : add c : add (hl) : NEXT_HL : ret nc
ld a,(hl) : NEXT_HL
ld b,(hl) : NEXT_HL : ret nz
pop de : pop de ; RET is not needed, because RET from ReadNibble is sufficient
pop bc ; RET is not needed, because RET from ReadNibble is sufficient
ReadNibble: ld c,a : xor a : exa : ret m
UpdateNibble ld a,(hl) : or #F0 : exa
ReadNibble: ld c,a
.skipLDCA xor a : exa : ret m
ld a,(hl) : or #F0 : exa
ld a,(hl) : NEXT_HL : or #0F
rrca : rrca : rrca : rrca : ret