lzsa/asm/z80/unlzsa2_fast_v1.asm

168 lines
3.8 KiB
NASM
Raw Normal View History

2019-06-10 11:24:15 +00:00
;
; Speed-optimized LZSA2 decompressor by spke (v.1 02-07/06/2019, 218 bytes)
;
; The data must be compressed using the command line compressor by Emmanuel Marty
; The compression is done as follows:
;
; lzsa.exe -f2 -r <sourcefile> <outfile>
;
; where option -r asks for the generation of raw (frame-less) data.
;
; The decompression is done in the standard way:
;
; ld hl,CompressedData
; ld de,WhereToDecompress
; call DecompressLZSA2
;
; Of course, LZSA2 compression algorithm is (c) 2019 Emmanuel Marty,
; see https://github.com/emmanuel-marty/lzsa for more information
;
; Drop me an email if you have any comments/ideas/suggestions: zxintrospec@gmail.com
;
@DecompressLZSA2:
; A' stores next nibble as %1111.... or assumed to contain trash
; B is assumed to be 0
xor a : ld b,a : exa : jr ReadToken
LongerMatch: exa : jp m,.noUpdate
ld a,(hl) : or #F0 : exa
ld a,(hl) : inc hl : or #0F
rrca : rrca : rrca : rrca
.noUpdate sub #F0-9 : cp 15+9 : jr c,CopyMatch
;inc a : jr z,LongMatch : sub #F0-9+1 : jp CopyMatch
LongMatch: ;ld a,24 :
add (hl) : inc hl : jr nc,CopyMatch
ld c,(hl) : inc hl
ld b,(hl) : inc hl
jr nz,CopyMatch.useC
pop de : ret
ManyLiterals: ld a,18 :
add (hl) : inc hl : jr nc,CopyLiterals
ld c,(hl) : inc hl
ld a,b : ld b,(hl) : inc hl
jr CopyLiterals.useBC
MoreLiterals: ld b,(hl) : inc hl
exa : jp m,.noUpdate
ld a,(hl) : or #F0 : exa
ld a,(hl) : inc hl : or #0F
rrca : rrca : rrca : rrca
.noUpdate ;sub #F0-3 : cp 15+3 : jr z,ManyLiterals
inc a : jr z,ManyLiterals : sub #F0-3+1
CopyLiterals: ld c,a
.useC ld a,b : ld b,0
.useBC ldir
push de : or a : jp p,CASE0xx : jr CASE1xx
; if "LL" of the byte token is equal to 0,
; there are no literals to copy
NoLiterals: xor (hl) : inc hl
push de : jp m,CASE1xx
; short (5 or 9 bit long) offsets
CASE0xx ld d,#FF : cp %01000000 : jr c,CASE00x
; "01x": the case of the 9-bit offset
CASE01x: cp %01100000 : rl d
ReadOffsetE: ld e,(hl) : inc hl
SaveOffset: ld ixl,e : ld ixh,d
MatchLen: inc a : and %00000111 : jr z,LongerMatch : inc a
CopyMatch: ld c,a
.useC ex (sp),hl : push hl ; BC = len, DE = offset, HL = dest, SP ->[dest,src]
add hl,de : pop de ; BC = len, DE = dest, HL = dest-offset, SP->[src]
ldir : pop hl
; compressed data stream contains records
; each record begins with the byte token "XYZ|LL|MMM"
ReadToken: ld a,(hl) : and %00011000 : jr z,NoLiterals
jp pe,MoreLiterals ; 00 has already been processed; this identifies the case of 11
rrca : rrca : rrca
ld c,a : ld a,(hl) : inc hl ; token is re-read for further processing
ldir
; the token and literals are followed by the offset
push de : or a : jp p,CASE0xx
CASE1xx cp %11000000 : jr nc,CASE11x
; "10x": the case of the 5-bit offset
CASE10x: ld c,a : xor a
exa : jp m,.noUpdate
ld a,(hl) : or #F0 : exa
ld a,(hl) : inc hl : or #0F
rrca : rrca : rrca : rrca
.noUpdate ld d,a : ld a,c
cp %10100000 : rl d
dec d : dec d : jr ReadOffsetE
; "00x": the case of the 5-bit offset
CASE00x: ld c,a : xor a
exa : jp m,.noUpdate
ld a,(hl) : or #F0 : exa
ld a,(hl) : inc hl : or #0F
rrca : rrca : rrca : rrca
.noUpdate ld e,a : ld a,c
cp %00100000 : rl e : jp SaveOffset
; two remaining cases
CASE11x cp %11100000 : jr c,CASE110
; "111": repeated offset
CASE111: ld e,ixl : ld d,ixh : jr MatchLen
; "110": 16-bit offset
CASE110: ld d,(hl) : inc hl : jr ReadOffsetE
;ReadNibble: ; 17 bytes, 44 t-state per nibble
; exa : ret m ; 4+11 = 15t
;UpdateNibble:
; ld a,(hl) : or #F0 : exa
; ld a,(hl) : inc hl : or #0F
; rrca : rrca : rrca : rrca : ret ; 4+5 + 7+7+4+7+6+7+4+4+4+4+10 = 73t