; ppm2AGA utility functions written by Günther Röhrich ; the version for 68020+ processors is slightly faster and smaller ; the function is called from C: ; EncodeHAM(char *yorig, char *yham, char *ColorTable, ; short NumColors, short xsize); ; yorig = original row in rgbrgb... format ; yham = row in HAM chunky format ; ColorTable = color table in brgbrg... format ; NumColors = (number of valid colors in ColorTable)*3 ; xsize = size of row in pixels ; ConvertMode should contain: 0 for HAM6 encoding ; 1 for HAM8 encoding ; int MapColor(colorhist_vector colormap, pixval r1, pixval g1, pixval b1, ; int NumColors) ; see later IFD MC68020 MACHINE MC68020 ENDC IFD MC68000 MACHINE MC68000 ENDC ;NOTE: GCC stores all data as 32 bit at subroutine calls! yorig EQU 44 ;pointer to actual row (original) yham EQU 48 ;pointer to actual row (HAM) ColorTable EQU 52 ;address of color table (3 bytes per entry) IFD GCC NumColors EQU 58 xsize EQU 62 ELSE NumColors EQU 56 ;number of colors allocated so far (*3) xsize EQU 58 ;number of pixels (x) ENDC colormap EQU 44 ;pointer to colorhist_vector IFD GCC r1 EQU 51 g1 EQU 55 b1 EQU 59 NColors EQU 60 ELSE r1 EQU 48 ;red component (byte format) g1 EQU 50 ;green component b1 EQU 52 ;blue component NColors EQU 54 ;number of colors in colormap ENDC XREF _Mult_Table ;address of multiplication table XREF _Mult_Table32 XREF _ConvertMode ;HAM8 or HAM6 XDEF _EncodeHAM ;entry point for function XDEF _MapColorASM XREF _ColorCache ;an 256K array * XDEF _blue_left ;only for debugging purposes * XDEF _search_finish ;only for debugging XDEF _MaxError XDEF _MaxErrorPos XDEF _hit ;only for debugging XDEF _t1 ;only for debugging dseg cnop 0,2 _blue_left: dc.b 0 ;this order is better for HAM-encoding _red_left: dc.b 0 _green_left: dc.b 0,0 ;additional dummy-value for faster longword access cnop 0,2 _offset_orig: dc.w 0 _offset_ham: dc.w 0 _MaxError: dc.w 0 _MaxErrorPos dc.w 0 _CacheOffset dc.l 0 cseg ;register usage: D0 = general purpose register ; D1 = contains afterwards the error ; D2 = orig_blue ; D3 = orig_red ; D4 = orig_green ; D5 = color that should be set ; D6 = offset to actual pixel (HAM) ; D7 = offset for color table / (0,1,2) at HAM ; A0 = best color so far (color+1)*3 ; A1 = pointer to multiplication table ; A2 = pointer to color table / to _blue_left ; A3 = smallest error that has been reached so far ; A4 = used by the Aztec assembler (small data model) ; A5 = pointer to the actual row (original) ; A6 = pointer to the actual row (HAM) ;WARNING: This will not work with color values higher than 63 ;(HAM8 has a maximum of 63, HAM6 has a maximum of 15) ;(values higher than 63 may result in memory corruption) ;computing the difference of color values is done with signed 8 bit ;arithmetic ;the maximum error value is 63^2+63^2+63^2=11907 ;the summation has to be done therefore with 16 bits ;Initializing _EncodeHAM: movem.l D2-D7/A2-A3/A5/A6,-(A7) ;store registers lea _blue_left,A0 ;load start of left colors move.l ColorTable(sp),A2 IFD MC68020 move.l (A2),(A0) ELSE move.b (A2)+,(A0)+ ;initialize left colors move.b (A2)+,(A0)+ ;A2 may not be word aligned move.b (A2)+,(A0)+ lea -3(A0),A0 ;correct register lea -3(A2),A2 ;correct register ENDC moveq.l #0,D6 ;initialize ham offset move.l D6,_MaxError ;initialize absolute max error ;and max error pos move.l yham(sp),A6 move.l yorig(sp),A5 lea _Mult_Table,A1 lea 255*2(A1),A1 search_begin: move.b (A5)+,D3 ;load registers with original colors move.b (A5)+,D4 ;for faster access move.b (A5)+,D2 IFD GCC lea 1(A5),A5 ;adjust A5 for correct alignment ENDC ;needed for GNU C move.w #15000,A3 ;dummy-value for minimum error so far move.l ColorTable(sp),A2 moveq.l #0,D7 ;initialize offset for color table ;First take a look if we have the value already in the cache moveq.l #0,D0 moveq.l #0,D1 move.b D3,D0 ;compute the cache offset lsl.l #6,D0 or.b D4,D0 lsl.l #6,D0 or.b D2,D0 ;now we have the offset in D0 movea.l _ColorCache,A0 ;load start address of the cache move.b 0(A0,D0.l),D1 ;take the value from the cache bne _hit ;jump if we have a cache hit move.l D0,_CacheOffset ;store the offset to avoid recomputing it move.l #3,A0 ;dummy-value for best colornumber so far ; (3 means color 0) search_start: bsr _compute_error cmp.w D1,A3 ;A3 <= D1 ? bls.s search4 move.w D1,A3 ;D1 is smaller than A3, store it move.w D7,A0 ;store color number tst.w D1 ;do we have the correct color ? beq search5 ;then finish immediately search4: cmp.w NumColors(sp),D7 ;have we reached highest colornum ? bne.s search_start ;no, then once again ;A0 contains now (colornumber+1)*3 ;A3 contains the error for that colornumber _t1: move.w A0,D0 movea.l _ColorCache,A2 move.l _CacheOffset,D1 move.b D0,0(A2,D1.l) ;store the value in the cache lea _blue_left,A2 ;restore A2 ;compute the error when using modify mode start_ham6: move.b D2,D0 ;load orig_blue moveq.l #0,D7 ;assume blue should be changed move.b D2,D5 ;store value to change sub.b _blue_left,D0 ;D0=D0-_blue_left bpl.s ham6_1 neg.b D0 ;make result positive ham6_1: move.b D0,D1 ;store maximum error so far move.b D3,D0 ;load orig_red sub.b _red_left,D0 ;D0=D0-_red_left bpl.s ham6_2 neg.b D0 ham6_2: cmp.b D0,D1 ;check D1-D0 bge.s ham6_3 ;jump if D1>=D0 move.b D0,D1 ;store new maximum error moveq.l #1,D7 ;assume red should be changed move.b D3,D5 ;store value to change ham6_3: move.b D4,D0 ;load orig_green sub.b _green_left,D0 ;D0=D0-_green_left bpl.s ham6_4 neg.b D0 ham6_4: cmp.b D0,D1 bge.s ham6_5 move.b D0,D1 ;store maximum error moveq.l #2,D7 ;green should be changed move.b D4,D5 ;store value to change ham6_5: lea _blue_left,A2 move.b D5,0(A2,D7.W) ;perform change ham_error: moveq.l #0,D1 moveq.l #0,D0 move.b D2,D0 ;load blue_origin sub.b (A2)+,D0 ;D0 = blue_color - blue_origin ext.w D0 IFD MC68020 add.w 0(A1,D0.W*2),D1 ELSE add.w D0,D0 add.w 0(A1,D0.W),D1 ;D1 = D1 + D0*D0 ENDC move.b D3,D0 ;load red_origin sub.b (A2)+,D0 ;D0 = red_color - red_origin ext.w D0 IFD MC68020 add.w 0(A1,D0.W*2),D1 ELSE add.w D0,D0 add.w 0(A1,D0.W),D1 ENDC move.b D4,D0 ;load green_origin sub.b (A2)+,D0 ;D0 = green_color - green_origin ext.w D0 IFD MC68020 add.w 0(A1,D0.W*2),D1 ELSE add.w D0,D0 add.w 0(A1,D0.W),D1 ;D1 contains error from HAM encoding ENDC cmpa.w D1,A3 ;check what error is smaller bls.s _search_finish ;jump if colortable is better cmp.w _MaxError,D1 ;compare with absolute max error ble.s ham6_9 ;jump if _MaxError is greater cmp.w #2,D6 ;less then 2 pixels from left ? bpl.s ham6_8a lsr.w #1,D1 ;half error ham6_8a: move.w D1,_MaxError move.w D6,_MaxErrorPos ham6_9: add.b #1,D7 ;needed to get correct code tst.w _ConvertMode ;ConvertMode = 0 -> HAM6 beq.s ham6_10 ;ConvertMode != 0 -> HAM8 IFD MC68020 lsl.b #2,D7 ;HAM8-adjust ELSE add.b D7,D7 ;this is faster than shifting add.b D7,D7 ;on the 68000 ENDC ham6_10: lsl.b #4,D7 ;HAM6-adjust or.b D5,D7 move.b D7,(A6,D6.W) ;store code in bitmap addq.w #1,D6 ;increase _offset_ham cmp.w xsize(sp),D6 ;end of column ? bne search_begin bra.s search_end _search_finish: move.w A3,D0 ;move error to D0 cmp.w #2,D6 ;less then 2 pixels from left ? bpl.s _search_finish1 lsr.w #1,D0 ;half error _search_finish1: cmp.w _MaxError,D0 ;compare with absolute max error ble.s _search_finish2 move.w D0,_MaxError move.w D6,_MaxErrorPos _search_finish2: move.w A0,D0 move.l ColorTable(sp),A2 lea -3(A2,D0.W),A3 ;load A3 with pointer to colors divu #3,D0 subq.w #1,D0 move.b D0,0(A6,D6.W) ;store colornumber in bitmap lea _blue_left,A2 IFD MC68020 move.l (A3),(A2) ELSE move.b (A3)+,(A2)+ ;store new left colors move.b (A3)+,(A2)+ ;A3 may not be word aligned move.b (A3)+,(A2)+ ; ENDC addq.w #1,D6 ;increase _offset_ham cmp.w xsize(sp),D6 ;have we reached the end ? bne search_begin search_end: movem.l (A7)+,D2-D7/A2-A3/A5/A6 ;restore registers rts ;jump back to caller ;we jump here if we have reached error 0 by colortable only ;and there was not a cache hit search5: move.w A0,D0 movea.l _ColorCache,A2 move.l _CacheOffset,D1 move.b D0,0(A2,D1.l) ;store the value in the cache bra.s _search_finish ;we jump here if we have a cache hit ;D1 contains the best color number, but we have to compute the error _hit: subq.w #3,D1 ;correct color number move.w D1,D7 bsr.s _compute_error move.w D1,A3 ;D1 is smaller than A3, store it move.w D7,A0 ;store color number bra start_ham6 ;continue with HAM encoding ;compute the error _compute_error: moveq.l #0,D1 moveq.l #0,D0 move.b D2,D0 ;load blue_origin sub.b 0(A2,D7.W),D0 ;D0 = blue_color - blue_origin ext.w D0 ;extend result to word IFD MC68020 add.w 0(A1,D0.W*2),D1 ELSE add.w D0,D0 add.w 0(A1,D0.W),D1 ;D1 = D1 + D0*D0 ENDC addq.w #1,D7 ;advance color table offset move.b D3,D0 ;load red_origin sub.b 0(A2,D7.W),D0 ;D0 = red_color - red_origin ext.w D0 IFD MC68020 add.w 0(A1,D0.W*2),D1 ELSE add.w D0,D0 add.w 0(A1,D0.W),D1 ENDC addq.w #1,D7 ;advance color table offset move.b D4,D0 ;load green_origin sub.b 0(A2,D7.W),D0 ;D0 = green_color - green_origin ext.w D0 IFD MC68020 add.w 0(A1,D0.W*2),D1 ;D1 = D1 + D0*D0 ELSE add.w D0,D0 add.w 0(A1,D0.W),D1 ENDC addq.w #1,D7 ;advance color table offset rts ;now comes the brute-force colormap search code ;if you know how to make it faster/smaller please let me know ;the 68020 main loop is only 56 bytes long, it should ;fit completely into the 256 bytes instruction cache of the 68020 ;or 68030 processor ;NOTE: this is a bit different compared to the HAM-code because ;signed 8 bit math gives wrong results with values higher than 127 ;it computes the differences therefore with 16 bit ;the maximum error value is 3*255^2=195075, we will therefore need ;32 bit arithmetic for the summation ;register usage: ; D0 = general purpose register, return value ; D1 = contains error ; D2 = r1 ; D3 = g1 ; D4 = b1 ; D5 = holds color from colormap ; D6 = counter ; D7 = color table offset ; A1 = pointer to multiplication table ; A2 = pointer to colormap ; A3 = smallest error ; A5 = actual error ; A6 = NColors _MapColorASM: movem.l D2-D7/A2-A3/A5/A6,-(A7) ;store registers move.l colormap(sp),A2 move.l NColors(sp),A6 lea _Mult_Table32,A1 lea 255*4(A1),A1 moveq.l #0,D2 ;ensure that the higher bytes moveq.l #0,D3 ;of these registers are cleared moveq.l #0,D4 move.b r1(sp),D2 move.b g1(sp),D3 move.b b1(sp),D4 move.l #2000000000,A3 ;dummy value for min. error moveq.l #0,D6 ;initialize counter moveq.l #0,D7 moveq.l #0,D5 ;now comes the main loop Map_start: suba.l A5,A5 ;clear A5 (actual error) move.l D2,D0 ;load r1 move.b 0(A2,D7.W),D5 ;load colormap.red sub.w D5,D0 ;D0 = colormap.red - r1 IFD MC68020 add.l 0(A1,D0.W*4),A5 ELSE lsl.w #2,D0 add.l 0(A1,D0.W),A5 ;A5 = A5 + D0*D0 ENDC move.l D3,D0 ;load g1 move.b 1(A2,D7.W),D5 ;load colormap.green sub.w D5,D0 ;D5 = colormap.green - g1 IFD MC68020 add.l 0(A1,D0.W*4),A5 ELSE lsl.w #2,D0 add.l 0(A1,D0.W),A5 ENDC move.l D4,D0 ;load b1 move.b 2(A2,D7.W),D5 ;load colormap.blue sub.w D5,D0 ;D0 = colormap.blue - b1 IFD MC68020 add.l 0(A1,D0.W*4),A5 ;A5 = A5 + D0*D0 ELSE lsl.w #2,D0 add.l 0(A1,D0.W),A5 ENDC addq.w #8,D7 ;advance color table offset cmp.l A5,A3 ;A3 - A5 ? bls.s Map1 move.l A5,A3 ;D1 is smaller than A3, store it move.w D6,A0 ;store color number move.l A5,D0 ;do we have the correct color ? beq.s Map_finish ;then finish immediately Map1: addq.w #1,D6 ;advance color number cmp.l A6,D6 ;have we reached highest colornum ? bne.s Map_start ;no, then once again Map_finish: move.l A0,D0 ;store return value movem.l (A7)+,D2-D7/A2-A3/A5/A6 ;restore registers rts ;jump back to caller end