Advertisement
Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- ; ------------------------------------------------------------------------------
- ; Nemesis decompression routine
- ; ------------------------------------------------------------------------------
- ; Optimized by vladikcomper, further optimizations & comments by carljr17
- ; ------------------------------------------------------------------------------
- NemDec_RAM:
- ; INPUT:
- ; a0 -> number of patterns + mode, followed by compressed data
- ; a4 -> RAM to output patterns to
- movem.l d0-d6/a0-a1/a3-a5,-(sp) ; 8 + 8*12 = 104
- lea NemDec_WriteRowToRAM(pc),a3 ; 8
- bra.s NemDec_Main ; 10
- ; ------------------------------------------------------------------------------
- NemDec:
- ; INPUT:
- ; a0 -> number of patterns + mode, followed by compressed data
- movem.l d0-d6/a0-a1/a3-a5,-(sp)
- lea NemDec_WriteRowToVDP(pc),a3 ; 8 a3 = jump address for pixel row writes
- lea $C00000,a4 ; 12 a4 -> [vdp_data_port]
- NemDec_Main:
- ; INPUT:
- ; a3 = NemDec_WriteRowToRAM/VDP
- ; a4 -> [$C00000] (vdp_data_port) (or anywhere in RAM)
- lea $FFFFAA00.w,a1 ; 8 a1 -> [$FFAA00] (Nemesis decompression buffer)
- move.w (a0)+,d3 ; 8 d3 <- #/patterns (MSB set for Mode 1)
- bpl.s @0 ; 10/8 d3.15 == 1 ? no, so in Mode 0 (not based on changes between rows)
- lea NemDec_WriteRowToVDP_XOR-NemDec_WriteRowToVDP(a3),a3 ; 8 yes, so in Mode 1 (each row XOR'd with last, only changes recorded)
- @0:
- lsl.w #3,d3 ; 6 + 2*3 = 12 -.
- subq.w #1,d3 ; 4 -' d3 = #/patterns * 8 (= number of rows to plot), minus 1
- bsr.w NemDec4 ; 18 d0-d2/d5/a5 build opcode-to-count/color lookup table
- moveq #0,d2 ; for use with Mode 1 only (XOR with first row)
- moveq #1,d4 ; set stop bit (nybble counter) -- 8 pixels per row
- move.b (a0)+,d5 ; -. get first two bytes of compressed data
- asl.w #8,d5 ; : (can't we replace this with move.w (a0)+,d5 !?)
- move.b (a0)+,d5 ; -'
- moveq #16,d6 ; 16 bits ready, set initial shift value
- bsr.s NemDec2 ; read in from bit stream of opcodes/in-line data, output rows
- movem.l (sp)+,d0-d6/a0-a1/a3-a5
- rts
- ; ---------------------------------------------------------------------------
- ; Part of the Nemesis decompressor, processes the actual compressed data
- ; ---------------------------------------------------------------------------
- NemDec3: ; this is an entry point for compatibility with code in sonic.asm
- move.w a5,d3 ; 4 -. number of rows to plot from a5 (for compatibility)
- subq.w #1,d3 ; 4 -'
- moveq #1,d4 ; 4 reset 8-pixel row, set stop bit/nybble counter
- eor.b d1,d4 ; 4 cheaper than a branch around next instruction
- NemDec_WritePixelLoop:
- eor.b d1,d4 ; 4 d4.l = ........ ........ .......! 00001111 etc.
- NemDec3_1:
- dbra d0,NemDec_WritePixel ; 10/12 --repeat count == -1 ? no, so plot another pixel
- NemDec2:
- ; INPUT:
- ; a0 -> stream of bits -- opcodes and/or in-line data
- ; a1 -> [$FFAA000] (LUT: opcode to palette, repeat count, & opcode width, 512 bytes)
- ; a3 = NemDec_WriteRowToRAM/VDP/_XOR
- ; a4 -> [$C00000] (vdp_data_port) (or anywhere in RAM)
- ; d2.l = previous row of 8 pixels, or 0 if first row; XOR'd with new row (mode 1 only)
- ; d3.w = number of patterns * 8 (= number of rows to plot), minus 1
- ; d4.l = holds 8 pixel nibbles, initially 1 (used as a stop bit or nybble counter)
- ; d5.w = [a0 - 2] (first 16 bits of stream, read left to right)
- ; d6.b = #/bits remaining to process, initially 16
- ; TRASHES: d0-d1
- sub.b #9,d6 ; 8 get right shift value to peek ahead at next 9 bits
- move.w d5,d0 ; 4 -. left-justify opcode (high bit in bit #8),
- lsr.w d6,d0 ; 6+2n -' followed by 1 or more unrelated, unprocessed bits
- andi.w #$01FE,d0 ; 8 isolate 8 bits for lookup: opcode and 1 or more unrelated bits
- sub.b 0(a1,d0.w),d6 ; 14 subtract opcode width, minus 9, from #/bits remaining
- move.b 1(a1,d0.w),d0 ; 14 d0.w <- .......? irrrpppp
- bpl.s NemDec_NotInline ; if i == 1, inline: next 7 bits are palette + count (rrrpppp)
- cmpi.b #9,d6 ; 9 or more bits still available ?
- bcc.s @0 ; no, so not enough room to read next byte
- addq.b #8,d6 ; 8 new bits, about to be read in below
- asl.w #8,d5 ; shift all remaining bits into high byte
- move.b (a0)+,d5 ; get next 8 bits into low of d5
- @0:
- subq.b #7,d6 ; 7 bits needed for inline data itself (palette + count) * (could be combined in above sub)
- move.w d5,d0
- lsr.w d6,d0 ; shift so that low bit rrrpppp is in bit position 0
- NemDec_NotInline:
- cmpi.b #9,d6 ; 9 or more bits still available ?
- bcc.s @1 ; no, so not enough room for read next byte
- addq.b #8,d6 ; 8 new bits, about to be read in below
- asl.w #8,d5 ; shift all remaining bits into high byte
- move.b (a0)+,d5 ; get next 8 bits into low of d5
- @1:
- move.b d0,d1 ; d1.w = ???????? ?rrrpppp
- andi.b #$0F,d1 ; = ???????? ....pppp (palette index)
- andi.w #$0070,d0 ; d0.w = ........ .rrr....
- lsr.w #4,d0 ; = ........ .....rrr (repeat count, minus 1; clear bits 8-15)
- NemDec_WritePixel:
- lsl.l #4,d4 ; d4.l = ........ ........ .......! 0000....
- bcc.s NemDec_WritePixelLoop ; 10/8
- or.b d1,d4 ; d4.l = 00001111 22223333 44445555 66667777
- jmp (a3) ; 8
- ; ---------------------------------------------------------------------------
- NemDec_WriteRowToVDP:
- loc_1502:
- move.l d4,(a4) ; write 8-pixel row to VDP control port
- moveq #1,d4 ; 4 reset 8-pixel row, set stop bit/nybble counter
- dbra d3,NemDec3_1 ; 10/12 have all 8-pixel rows been written ? if yes, branch
- rts
- NemDec_WriteRowToVDP_XOR:
- eor.l d4,d2 ; XOR the previous row by the current row
- move.l d2,(a4) ; write new current row to VDP control port
- moveq #1,d4 ; 4 reset 8-pixel row, set stop bit/nybble counter
- dbra d3,NemDec3_1 ; 10/12 have all 8-pixel rows been written ? if yes, branch
- rts
- ; ---------------------------------------------------------------------------
- NemDec_WriteRowToRAM:
- move.l d4,(a4)+ ; write 8-pixel row to RAM, with post-increment
- moveq #1,d4 ; 4 reset 8-pixel row, set stop bit/nybble counter
- dbra d3,NemDec3_1 ; 10/12 have all 8-pixel rows been written ? if yes, branch
- rts
- NemDec_WriteRowToRAM_XOR:
- eor.l d4,d2 ; XOR the previous row by the current row
- move.l d2,(a4)+ ; write new current row to RAM with post-increment
- moveq #1,d4 ; 4 reset 8-pixel row, set stop bit/nybble counter
- dbra d3,NemDec3_1 ; 10/12 have all 8-pixel rows been written ? if yes, branch
- rts
- ; ---------------------------------------------------------------------------
- ; Part of the Nemesis decompressor, builds the code table (in RAM)
- ; ---------------------------------------------------------------------------
- NemDec4:
- ; INPUT:
- ; a0 -> !...pppp (palette index -- 1st need not have bit #7 set)
- ; .rrrcccc (repeat count + code length)
- ; cccccccc (opcode, mysteriously right-justified)
- ; .rrrcccc (repeat count + code length)
- ; cccccccc (opcode, mysteriously right-justified)
- ; ...
- ; $FF (end-of-table)
- ; a1 -> [$FFAA000] (nemesis decompression buffer -- actually 512 byte opcode-to-count/color LUT)
- ; TRASHES: d0-2/d5/a5
- lea $01F8(a1),a5 ; 8 point to last four entries of table (used to flag inline data)
- move.l #$FDFFFDFF,d5 ; 12 set cccc (opcode width) = 6 (minus 9), i (inline) = 1, rrrpppp = don't care
- move.l d5,(a5)+ ; 12 [$01F8 + 0/2].w <- !!!!cccc irrrpppp
- move.l d5,(a5)+ ; 12 [$01F8 + 4/6].w <- !!!!cccc irrrpppp
- move.b (a0)+,d0 ; 8 read 1st byte (????pppp or $FF for end-of-table)
- bra.s @ChkEnd ; 10 if d0.7 == 1, end of table ($FF) OR new palette index (!???pppp)
- @ItemLoop:
- move.b d0,d5 ; 4 d5.w = ???????? .rrrcccc (rrr = repeat count, cccc = opcode width)
- andi.w #$000F,d2 ; 8 d2.w = ........ ....pppp
- andi.w #$0070,d0 ; 8 d0.w = ........ .rrr.... (note bits 8-15 cleared for below)
- or.b d0,d2 ; 4 d2.w = ........ .rrrpppp
- eor.b d0,d5 ; 4 d5.w = ???????? ....cccc
- moveq #8,d1 ; 4 -. d1.b = 8 - opcode width = shift left count (for left-justify)
- sub.b d5,d1 ; 4 -' i.e., 12345678 -> 76543210
- sub.b #9,d5 ; 8 opcode width - 9 needed for NemDec2 *
- lsl.w #8,d5 ; 6 + 2*8 = 22 d5.w = !!!!cccc ........
- or.w d5,d2 ; 4 d2.w = !!!!cccc .rrrpppp (table entry)
- move.b (a0)+,d0 ; 8 get actual opcode (mysteriously not shifted into position)
- lsl.b d1,d0 ; 6 + 2*d1 d0.b = opcode << ( 8 - opcode width ) (= left justify opcode; but why isn't this just pre-shifted ?)
- add.w d0,d0 ; 4 double to index into table of words
- moveq #0,d5 ; 4 -.
- bset d1,d5 ; 6 : d5 = 1 << ( 8 - opcode length ) - 1
- subq.b #1,d5 ; 4 -' = #/times to repeat table entry
- lea (a1,d0.w),a5 ; 12 (versus 4 move.w a1, a5 + 8 add.w d0, a5 = 12)
- @ItemShortCodeLoop:
- move.w d2,(a5)+ ; 8 store entry: !!!!cccc .rrrpppp
- dbra d5,@ItemShortCodeLoop ; 10/14 repeat for required number of entries
- @ChkNext:
- move.b (a0)+,d0 ; 8 palette index, repeat count/opcode width, or end-of-table
- bpl.s @ItemLoop ; 10/8 if d0.7 == 0, repeat count/opcode width
- @ChkEnd:
- move.b d0,d2 ; 4 d2.w = ???????? ????pppp (pppp = palette index)
- not.b d0 ; 4 reached end-of-table ?
- bne.s @ChkNext ; 10/8 yes, so return to caller (RTS)
- rts ; return to caller
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
