mirror of
https://github.com/emmanuel-marty/lzsa.git
synced 2025-02-16 17:30:44 +00:00
Merge pull request #14 from MobyGamer/decompressor/8086_speed
Additional minor speedups
This commit is contained in:
Emmanuel Marty
GitHub
commit
4d7d90b893
@ -27,12 +27,22 @@
|
||||
|
||||
IDEAL
|
||||
P8086
|
||||
MODEL SMALL
|
||||
|
||||
CODESEG
|
||||
SEGMENT CODE para public
|
||||
|
||||
ASSUME cs:CODE, ds:CODE
|
||||
|
||||
PUBLIC lzsa1_decompress_speed
|
||||
|
||||
; ---------------------------------------------------------------------------
|
||||
; Decompress raw LZSA1 block
|
||||
; inputs:
|
||||
; * ds:si: raw LZSA1 block
|
||||
; * es:di: output buffer
|
||||
; output:
|
||||
; * ax: decompressed size
|
||||
; ---------------------------------------------------------------------------
|
||||
|
||||
; Must declare this in the code segment:
|
||||
SHR4table:
|
||||
DB 00h,00h,00h,00h,00h,00h,00h,00h,00h,00h,00h,00h,00h,00h,00h,00h
|
||||
@ -44,27 +54,18 @@ SHR4table:
|
||||
DB 06h,06h,06h,06h,06h,06h,06h,06h,06h,06h,06h,06h,06h,06h,06h,06h
|
||||
DB 07h,07h,07h,07h,07h,07h,07h,07h,07h,07h,07h,07h,07h,07h,07h,07h
|
||||
|
||||
; ---------------------------------------------------------------------------
|
||||
; Decompress raw LZSA1 block
|
||||
; inputs:
|
||||
; * ds:si: raw LZSA1 block
|
||||
; * es:di: output buffer
|
||||
; output:
|
||||
; * ax: decompressed size
|
||||
; ---------------------------------------------------------------------------
|
||||
|
||||
PROC lzsa1_decompress_speed NEAR
|
||||
|
||||
lzsa1_start:
|
||||
push di ;remember decompression offset
|
||||
cld ;ensure string ops move forward
|
||||
mov bx,offset SHR4table ;table lookup faster than bitops if op > 3
|
||||
mov bx,offset SHR4table
|
||||
xor cx,cx
|
||||
|
||||
@@decode_token:
|
||||
mov ax,cx ;clear ah (cx = 0 from match copy's rep movsb)
|
||||
xchg cx,ax ;clear ah (cx = 0 from match copy's rep movsb)
|
||||
lodsb ;read token byte: O|LLL|MMMM
|
||||
mov dx,ax ;keep token in dl
|
||||
mov dx,ax ;copy our token to dl for later MMMM handling
|
||||
|
||||
and al,070H ;isolate literals length in token (LLL)
|
||||
jz @@check_offset_size ;if LLL=0, we have no literals; goto match
|
||||
@ -74,16 +75,7 @@ lzsa1_start:
|
||||
lodsb ;grab extra length byte
|
||||
add al,07H ;add LITERALS_RUN_LEN
|
||||
jnc @@got_literals_exact ;if no overflow, we have full count
|
||||
jne @@mid_literals
|
||||
|
||||
@@big_literals:
|
||||
lodsw ;grab 16-bit extra length
|
||||
xchg cx,ax ;with longer counts, we can save some time
|
||||
shr cx,1 ;by doing a word copy instead of a byte copy.
|
||||
rep movsw
|
||||
adc cx,0
|
||||
rep movsb
|
||||
jmp @@check_offset_size
|
||||
je @@big_literals
|
||||
|
||||
@@mid_literals:
|
||||
lodsb ;grab single extra length byte
|
||||
@ -95,11 +87,21 @@ lzsa1_start:
|
||||
rep movsb
|
||||
jmp @@check_offset_size
|
||||
|
||||
@@big_literals:
|
||||
lodsw ;grab 16-bit extra length
|
||||
xchg cx,ax ;with longer counts, we can save some time
|
||||
shr cx,1 ;by doing a word copy instead of a byte copy.
|
||||
rep movsw
|
||||
adc cx,0
|
||||
rep movsb
|
||||
jmp @@check_offset_size
|
||||
|
||||
@@got_literals:
|
||||
segcs xlat ;shift literals length into place
|
||||
@@got_literals_exact:
|
||||
xchg cx,ax
|
||||
rep movsb ;copy cx literals from ds:si to es:di
|
||||
|
||||
@@check_offset_size:
|
||||
test dl,dl ;check match offset size in token (O bit)
|
||||
js @@get_long_offset ;load absolute 16-bit match offset
|
||||
@ -110,9 +112,7 @@ lzsa1_start:
|
||||
@@get_match_length:
|
||||
xchg dx,ax ;dx: match offset ax: original token
|
||||
and al,0FH ;isolate match length in token (MMMM)
|
||||
add al,3 ;add MIN_MATCH_SIZE
|
||||
|
||||
cmp al,012H ;MATCH_RUN_LEN?
|
||||
cmp al,0FH ;MATCH_RUN_LEN?
|
||||
jne @@got_matchlen_short ;no, we have the full match length from the token, go copy
|
||||
|
||||
lodsb ;grab extra length byte
|
||||
@ -126,6 +126,7 @@ lzsa1_start:
|
||||
jmp @@copy_len_preset ;otherwise, do the copy
|
||||
|
||||
@@got_matchlen_short:
|
||||
add al,3 ;add MIN_MATCH_SIZE
|
||||
xchg cx,ax ;copy match length into cx
|
||||
mov bp,ds ;save ds
|
||||
mov ax,es
|
||||
@ -207,7 +208,7 @@ lzsa1_start:
|
||||
|
||||
ENDP lzsa1_decompress_speed
|
||||
|
||||
ENDS
|
||||
ENDS CODE
|
||||
|
||||
END
|
||||
|
||||
@ -234,5 +235,17 @@ END
|
||||
; reverse 16-bit len compar shuttle 102000 alice 59263 robotron 364460 --- rb
|
||||
; jcxz for EOD detection no change to speed, but is 1 byte shorter +++
|
||||
; force movsw for literals shuttle 107183 alice 62555 robotron 379524 --- rb
|
||||
; defer shr4 until necessry shuttle 102069 alice 60236 robotron 364096 --- rb
|
||||
; skip literals if LLL=0 shuttle 98655 alice 57849 robotron 363358 --- rb
|
||||
; defer shr4 until necessry shuttle 102069 alice 60236 robotron 364096 ---
|
||||
; skip literals if LLL=0 shuttle 98655 alice 57849 robotron 363358 ---
|
||||
; fall through to mid_liter shuttle 98595 alice 57789 robotron 361998 +++
|
||||
; == jumptable experiments begin ==
|
||||
; jumptable for small copys shuttle 101594 alice 61078 robotron 386018 ---
|
||||
; start:xchg instead of mov shuttle 100948 alice 60467 robotron 381112 +++
|
||||
; use table for LLL=0 check shuttle 106972 alice 63333 robotron 388304 --- rb
|
||||
; jmptbl to fallthrough mov shuttle 102532 alice 60760 robotron 383070 ---
|
||||
; cpy fallthrough check_ofs shuttle 98939 alice 58917 robotron 371019 +**
|
||||
; single jumptable jump shuttle 97528 alice 57264 robotron 362194 ++*
|
||||
; conditional check for L=7 shuttle 98610 alice 58521 robotron 368153 --- rb
|
||||
; rip out the jumptable :-/ shuttle 97616 alice 57128 robotron 360697 +++
|
||||
; defer add MIN_MATCH_SIZE shuttle 97250 alice 57004 robotron 361191 ++?
|
||||
; cache constants in regs shuttle 104681 alice 59939 robotron 380125 --- rb
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user