6502/lzsa
1
0
Fork 0
mirror of https://github.com/emmanuel-marty/lzsa.git synced 2024-06-01 06:41:43 +00:00
Code Issues Projects Releases Wiki Activity

slightly faster

Browse Source
This commit is contained in:
Peter Ferrie 2023-11-22 10:56:27 -08:00
parent 15ee2dfe11
commit 6a10fb0087
2 changed files with 157 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

79
asm/6502/decompress_faster_v1.asm
View File

@ -7,7 +7,7 @@
;
; This code is written for the ACME assembler.
;
; The code is 165 bytes for the small version, and 191 bytes for the normal.
; The code is 165 bytes for the small version, and 205 bytes for the normal.
;
; Copyright John Brandwood 2021.
;
@ -81,8 +81,7 @@ lzsa1_unpack: ldy #0 ; Initialize source index.
lda (lzsa_srcptr),y
inc <lzsa_srcptr + 0
bne .cp_skip0
inc <lzsa_srcptr + 1
beq .inc_src1
}
@ -105,11 +104,31 @@ lzsa1_unpack: ldy #0 ; Initialize source index.
.cp_byte: lda (lzsa_srcptr),y ; CC throughout the execution of
sta (lzsa_dstptr),y ; of this .cp_page loop.
inc <lzsa_srcptr + 0
!if LZSA_SMALL_SIZE {
bne .cp_skip1
inc <lzsa_srcptr + 1
} else {
beq .inc_src2
}
.cp_skip1: inc <lzsa_dstptr + 0
!if LZSA_SMALL_SIZE {
bne .cp_skip2
inc <lzsa_dstptr + 1
} else {
beq .inc_dst
}
.cp_skip2: dex
bne .cp_byte
.cp_npages: lda #0 ; Any full pages left to copy?
@ -171,6 +190,21 @@ lzsa1_unpack: ldy #0 ; Initialize source index.
} else {
.inc_src1: inc <lzsa_srcptr + 1
bne .cp_skip0 ; Always taken
.inc_src2: inc <lzsa_srcptr + 1
bne .cp_skip1 ; Always taken
.inc_dst: inc <lzsa_dstptr + 1
bne .cp_skip2 ; Always taken
.inc_src3: inc <lzsa_srcptr + 1
bne .offset_lo ; Always taken
.inc_src4: inc <lzsa_srcptr + 1
bne .offset_hi ; Always taken
;
; Copy bytes from decompressed window.
;
@ -181,8 +215,7 @@ lzsa1_unpack: ldy #0 ; Initialize source index.
.lz_offset: lda (lzsa_srcptr),y ; Get offset-lo.
inc <lzsa_srcptr + 0
bne .offset_lo
inc <lzsa_srcptr + 1
beq .inc_src3
.offset_lo: sta <lzsa_offset + 0
@ -192,8 +225,7 @@ lzsa1_unpack: ldy #0 ; Initialize source index.
lda (lzsa_srcptr),y
inc <lzsa_srcptr + 0
bne .offset_hi
inc <lzsa_srcptr + 1
beq .inc_src4
.offset_hi: sta <lzsa_offset + 1
@ -238,6 +270,9 @@ lzsa1_unpack: ldy #0 ; Initialize source index.
.lz_more: inc <lzsa_winptr + 1 ; Unlikely, so can be slow.
bne .lz_byte ; Always true!
.inc_src5: inc <lzsa_srcptr + 1 ; Unlikely, so can be slow.
bne .skip_inc ; Always true!
}
;
@ -249,9 +284,18 @@ lzsa1_unpack: ldy #0 ; Initialize source index.
.get_length: clc ; Add on the next byte to get
adc (lzsa_srcptr),y ; the length.
inc <lzsa_srcptr + 0
!if LZSA_SMALL_SIZE {
bne .skip_inc
inc <lzsa_srcptr + 1
} else {
beq .inc_src5
}
.skip_inc: bcc .got_length ; No overflow means done.
clc ; MUST return CC!
tax ; Preserve overflow value.
@ -277,6 +321,27 @@ lzsa1_unpack: ldy #0 ; Initialize source index.
.get_byte: lda (lzsa_srcptr),y ; Subroutine version for when
inc <lzsa_srcptr + 0 ; inlining isn't advantageous.
!if LZSA_SMALL_SIZE {
bne .got_byte
inc <lzsa_srcptr + 1 ; Inc & test for bank overflow.
} else {
beq .inc_src6
}
.got_byte: rts
!if LZSA_SMALL_SIZE {
; nothing
} else {
.inc_src6: inc <lzsa_srcptr + 1 ; Unlikely, so can be slow.
bne .got_byte ; Always true!
}

92
asm/6502/decompress_faster_v2.asm
View File

@ -7,7 +7,7 @@
;
; This code is written for the ACME assembler.
;
; The code is 241 bytes for the small version, and 256 bytes for the normal.
; The code is 241 bytes for the small version, and 268 bytes for the normal.
;
; Copyright John Brandwood 2021.
;
@ -87,8 +87,7 @@ lzsa2_unpack: ldx #$00 ; Hi-byte of length or offset.
lda (lzsa_srcptr),y
inc <lzsa_srcptr + 0
bne .cp_skip0
inc <lzsa_srcptr + 1
beq .inc_src1
}
@ -110,11 +109,31 @@ lzsa2_unpack: ldx #$00 ; Hi-byte of length or offset.
.cp_byte: lda (lzsa_srcptr),y ; CC throughout the execution of
sta (lzsa_dstptr),y ; of this .cp_page loop.
inc <lzsa_srcptr + 0
!if LZSA_SMALL_SIZE {
bne .cp_skip1
inc <lzsa_srcptr + 1
} else {
beq .inc_src2
}
.cp_skip1: inc <lzsa_dstptr + 0
!if LZSA_SMALL_SIZE {
bne .cp_skip2
inc <lzsa_dstptr + 1
} else {
beq .inc_dst
}
.cp_skip2: dex
bne .cp_byte
.cp_npages: lda #0 ; Any full pages left to copy?
@ -123,6 +142,23 @@ lzsa2_unpack: ldx #$00 ; Hi-byte of length or offset.
dec .cp_npages + 1 ; Unlikely, so can be slow.
bcc .cp_byte ; Always true!
!if LZSA_SMALL_SIZE {
; nothing
} else {
.inc_src1: inc <lzsa_srcptr + 1 ; Unlikely, so can be slow.
bne .cp_skip0 ; Always true!
.inc_src2: inc <lzsa_srcptr + 1 ; Unlikely, so can be slow.
bne .cp_skip1 ; Always true!
.inc_dst: inc <lzsa_dstptr + 1 ; Unlikely, so can be slow.
bne .cp_skip2 ; Always true!
}
;
; Copy bytes from decompressed window.
;
@ -178,8 +214,7 @@ lzsa2_unpack: ldx #$00 ; Hi-byte of length or offset.
lda (lzsa_srcptr),y ; Get lo-byte of offset.
inc <lzsa_srcptr + 0
bne .set_offset
inc <lzsa_srcptr + 1
beq .inc_src3
}
@ -229,6 +264,17 @@ lzsa2_unpack: ldx #$00 ; Hi-byte of length or offset.
.lz_more: inc <lzsa_winptr + 1 ; Unlikely, so can be slow.
bne .lz_byte ; Always true!
!if LZSA_SMALL_SIZE {
; nothing
} else {
.inc_src3: inc <lzsa_srcptr + 1 ; Unlikely, so can be slow.
bne .set_offset ; Always true!
}
;
; Lookup tables to differentiate literal and match lengths.
;
@ -268,10 +314,31 @@ lzsa2_unpack: ldx #$00 ; Hi-byte of length or offset.
.get_byte: lda (lzsa_srcptr),y ; Subroutine version for when
inc <lzsa_srcptr + 0 ; inlining isn't advantageous.
!if LZSA_SMALL_SIZE {
bne .got_byte
inc <lzsa_srcptr + 1
} else {
beq .inc_src4
}
.got_byte: rts
!if LZSA_SMALL_SIZE {
; nothing
} else {
.inc_src4: inc <lzsa_srcptr + 1 ; Unlikely, so can be slow.
bne .got_byte ; Always true!
}
.finished: pla ; Decompression completed, pop
pla ; return address.
rts
@ -287,14 +354,14 @@ lzsa2_unpack: ldx #$00 ; Hi-byte of length or offset.
inc <lzsa_nibflg ; Reset the flag.
!if LZSA_SMALL_SIZE {
jsr .get_byte
} else {
lda (lzsa_srcptr),y
inc <lzsa_srcptr + 0
bne .set_nibble
inc <lzsa_srcptr + 1
beq .inc_src5
}
@ -306,3 +373,14 @@ lzsa2_unpack: ldx #$00 ; Hi-byte of length or offset.
.got_nibble: and #$0F
rts
!if LZSA_SMALL_SIZE {
; nothing
} else {
.inc_src5: inc <lzsa_srcptr + 1 ; Unlikely, so can be slow.
bne .set_nibble ; Always true!
}