Date sent: Wed, 29 Oct 1997 22:37:51 +0200 (EET) From: Harry Sintonen To: amycoders@ibm.net Subject: speedopt entry to #amycoders triangledrawing compo (fwd) ********************************************************************* * * Your init code comes here. * ********************************************************************* _initcode a b rts ********************************************************************* * * Your line routine. * Every time you are given: A0.L : Chunky Screen (256x256x8) * * D0.l color * A1 {x1,y1,x2,y2,x3,y3}.w * * other registers are in undefined state !! * you may trash all regs except a7 * ********************************************************************* CNOP 0,8 _triangle c ;**************************************************************************** ;** ;* Speed-optimized chunky triangledraw routine ;* ;* This is one of my contributions to #amycoders triangledrawing-compo. This ;* routine will participate speed-optimizing competition. This code works ;* with MC68020+ processors with any cache setup. ;* ;* This code benefits from copyback cache so turn it on. ;* ;* Code size without rts: max. 582 bytes. ;* Average time (dec): damn-it's-fast ;* ;* Coded by Harry "Piru" Sintonen . ;* This code is copyright © 1997 by Harry Sintonen, and was written 26th- ;* 28th october 1997. ;* ;**************************************************************************** ;* ; variables T_SWIDTH EQU 256 Screen width ; flags T_ALLOWLINE SET 0 Allow triangles that are lines T_SMALLMODE SET 0 Optimize for small triangles T_4PIXWRITE SET 0 Write 4 pixels at a time instead of 8 T_SPECIALBFLAT SET 1 Have sep routine for flat bottom tris T_0YDELTACHECK SET 0 Have check for 1y=3y (ydelta=0) ; following is my quick "nop" longword align macro ;) mcnop4tempb ; this must be longword aligned! mcnop4 MACRO mcnop4temps SET (*-mcnop4tempb)&2 IFGT (mcnop4temps) or.l d0,d0 ENDC ENDM ; IN: d0.l=color ; a0=chunky buffer (256x256x8) ; a1={x1,y1,x2,y2,x3,y3}.w ; OUT: d0-d7/a0-a4 trashed IFEQ T_SMALLMODE move.l d0,d4 000c ENDC movem.l (a1),d1-d3 d1=1x1y, d2=2x2y, d3=3x3y IFEQ T_SMALLMODE lsl.l #8,d0 00c0 ENDC cmp.w d1,d2 bhi.b .ns1 exg d1,d2 .ns1 IFEQ T_SMALLMODE or.b d4,d0 00cc ENDC cmp.w d1,d3 bhi.b .ns2 exg d1,d3 .ns2 IFEQ T_SMALLMODE move.l d0,d4 00cc ENDC cmp.w d2,d3 bhi.b .ns3 exg d2,d3 .ns3 IFEQ T_SMALLMODE swap d0 cc00 move.w d4,d0 cccc ENDC IFNE T_0YDELTACHECK cmp.w d1,d3 1y=3y? (check for 0 ydelta) beq .done1 ENDC ; d1=uppermost corner, d2=middle corner, d3=lowermost corner cmp.w d1,d2 1y=2y -> top flat triangle beq .top_flat ; add position of corner 1 to chunky buffer start. IFEQ T_SWIDTH-320 ; oh dear, this will be slower than mulu.w #T_SWIDTH ; thing on 060. Should be faster on 040 tho... :) move.l d1,d4 swap d1 001x ext.l d4 0000001y move.l d4,d5 lsl.l #8,d4 1<<8=256 lsl.l #6,d5 1<<6=64 add.l d5,d4 or.w d1,d4 1y*320!1x swap d1 ELSE IFEQ T_SWIDTH-256 move.l d1,d4 001x001y swap d4 001y001x lsl.w #8,d4 001y1x00 lsr.l #8,d4 00001y1x ELSE move.w d1,d4 001y swap d1 mulu.w #T_SWIDTH,d4 or.w d1,d4 +001x swap d1 ENDC ENDC add.l d4,a0 a0=chunky+1y*T_SWIDTH+1x IFNE T_SPECIALBFLAT cmp.w d2,d3 2y=3y -> bottom flat triangle beq .bottom_flat ENDC ; determine on which side middle (2) corner is. move.l d3,d5 NOTE: 3y>1y !! move.w d2,d4 sub.l d1,d5 sub.w d1,d4 d4.w=(2y-1y) swap d5 d5.w=(3x-1x) muls.w d5,d4 d4.l=(2y-1y)*(3x-1x) move.w d3,d5 sub.w d1,d5 d5.l=(3y-1y) divs.w d5,d4 d4.l=(2y-1y)*(3x-1x)/(3y-1y) (4x-1x) move.l d1,d5 move.l d2,d6 swap d5 swap d6 add.w d4,d5 d5.w=4x=(2y/1y)*(3x-1x)/(3y-1y)+1x cmp.w d6,d5 4x>2x blo .middle_on_right IFNE T_ALLOWLINE bne .noladd sub.l #$10000,d2 if it's "line" triangle fix not to be one ;) .noladd ENDC ; middle on left ; 1 ; /\ ; / \ ;2·--._\ ; 3 ;r=fill width (dbf) dr=(3x-1x)<<16/(3y-1y)-(2x-1x)<<16/(2y-1y) ;l=offset in chunky dl=(2x-1x)<<16/(2y-1y) move.l d2,d7 NOTE: 2y>1y !! sub.l d1,d7 (2x2y-1x1y) move.l d7,d6 ext.l d7 (2y-1y) clr.w d6 (2x-1x)<<16 divs.l d7,d6 move.l d6,a3 a3=dl move.l d3,d7 NOTE: 3y>1y !! sub.l d1,d7 (3x3y-1x1y) move.l d7,d6 ext.l d7 (3y-1y) clr.w d6 (3x-1x)<<16 divs.l d7,d6 move.l d6,a2 sub.l a3,a2 a2=dr (dr-dl) ; calculate new left delta move.l d3,d4 NOTE: 3y>2y !! sub.l d2,d4 (3x3y-2x2y) move.l d4,d5 clr.w d4 (3x-2x)<<16 ext.l d5 (3y-2y) IFEQ T_SPECIALBFLAT beq.b .nodiv1 ENDC divs.l d5,d4 .nodiv1 sub.w d1,d2 d2=2y-1y move.l d4,a4 a4=dl2 sub.w d1,d3 d3=3y-1y (NOTE: no subq.w #1,d3) move.l #$8000,d6 d6=l move.l d6,d7 d7=r mcnop4 .oloop1 swap d6 ffffiiii move.l d7,d4 iiiiffff lea (a0,d6.w),a1 swap d4 ffffiiii swap d6 iiiiffff IFEQ T_SMALLMODE cmp.w #8,d4 blo.b .bl1 IFNE T_4PIXWRITE subq.w #4,d4 ELSE subq.w #8,d4 ENDC move.w d4,d5 IFNE T_4PIXWRITE and.w #%11,d4 lsr.w #2,d5 ELSE and.w #%111,d4 lsr.w #3,d5 ENDC .iloop1 IFEQ T_4PIXWRITE move.l d0,(a1)+ ENDC move.l d0,(a1)+ dbf d5,.iloop1 ENDC .bl1 move.b d0,(a1)+ dbf d4,.bl1 add.l a3,d6 l=l+dl (iiiiffff) lea (T_SWIDTH,a0),a0 move to next line add.l a2,d7 r=r+dr (iiiiffff) subq.w #1,d2 are we on middle corner y pos? dbeq d3,.oloop1 nope, do more tst.w d3 bmi.b .done1 ; set new left delta (will affect right delta too!) ; (begin to move towards lowermost corner) ;dl=(3x-2x)<<16/(3y-2y) add.l a3,a2 a2=dr (orig!) move.l a4,a3 dl=dl2 sub.l a4,a2 a2=dr (dr-dl2) dbf d3,.oloop1 ok, do rest of triangle .done1 rts CNOP 0,4 .middle_on_right ; 1 ; /\ ; / \ ; /_.--·2 ; 3 ; middle on right ;r=fill width (dbf) dr=(2x-1x)<<16/(2y-1y)-(3x-1x)<<16/(3y-1y) ;l=offset in chunky dl=(3x-1x)<<16/(3y-1y) move.l d3,d7 NOTE: 3y>1y !! sub.l d1,d7 (3x3y-1x1y) move.l d7,d6 ext.l d7 (3y-1y) clr.w d6 (3x-1x)<<16 divs.l d7,d6 move.l d6,a2 a2=dl move.l d2,d7 NOTE: 2y>1y !! sub.l d1,d7 (2x2y-1x1y) move.l d7,d6 ext.l d7 (2y-1y) clr.w d6 (2x-1x)<<16 divs.l d7,d6 move.l d6,a3 sub.l a2,a3 a3=dr (dr-dl) ; calculate new right delta move.l d3,d4 NOTE: 3y>2y !! sub.l d2,d4 (3x3y-2x2y) move.l d4,d5 clr.w d4 (3x-2x)<<16 ext.l d5 (3y-2y) IFEQ T_SPECIALBFLAT beq.b .nodiv2 ENDC divs.l d5,d4 .nodiv2 move.l d4,a4 sub.w d1,d2 d2=2y-1y sub.l a2,a4 a4=dr2 (dr2-dl) sub.w d1,d3 d3=3y-1y (NOTE: no subq.w #1,d3) move.l #$8000,d6 d6=l move.l d6,d7 d7=r mcnop4 .oloop2 swap d6 ffffiiii move.l d7,d4 iiiiffff lea (a0,d6.w),a1 swap d4 ffffiiii swap d6 iiiiffff IFEQ T_SMALLMODE cmp.w #8,d4 blo.b .bl2 IFNE T_4PIXWRITE subq.w #4,d4 ELSE subq.w #8,d4 ENDC move.w d4,d5 IFNE T_4PIXWRITE and.w #%11,d4 lsr.w #2,d5 ELSE and.w #%111,d4 lsr.w #3,d5 ENDC .iloop2 IFEQ T_4PIXWRITE move.l d0,(a1)+ ENDC move.l d0,(a1)+ dbf d5,.iloop2 ENDC .bl2 move.b d0,(a1)+ dbf d4,.bl2 add.l a2,d6 l=l+dl (iiiiffff) lea (T_SWIDTH,a0),a0 move to next line add.l a3,d7 r=r+dr (iiiiffff) subq.w #1,d2 are we on middle corner y pos? dbeq d3,.oloop2 nope, do more tst.w d3 bmi.b .done2 ; set new right delta ; (begin to move towards lowermost corner) ;dr=(3x-2x)<<16/(3y-2y) move.l a4,a3 dr=dr2 dbf d3,.oloop2 ok, do rest of triangle .done2 rts CNOP 0,4 .top_flat ;1·----·2 (or 2 1) ; \ / ; \/ ; 3 cmp.l d1,d2 bhi.b .nxs1 exg d1,d2 .nxs1 ; add position of corner 1 to chunky buffer start. IFEQ T_SWIDTH-320 ; oh dear, this will be slower than mulu.w #T_SWIDTH ; thing on 060. Should be faster on 040 tho... :) move.l d1,d4 swap d1 001x ext.l d4 0000001y move.l d4,d5 lsl.l #8,d4 1<<8=256 lsl.l #6,d5 1<<6=64 add.l d5,d4 or.w d1,d4 1y*320!1x swap d1 ELSE IFEQ T_SWIDTH-256 move.l d1,d4 001x001y swap d4 001y001x lsl.w #8,d4 001y1x00 lsr.l #8,d4 00001y1x ELSE move.w d1,d4 001y swap d1 mulu.w #T_SWIDTH,d4 or.w d1,d4 +001x swap d1 ENDC ENDC add.l d4,a0 a0=chunky+1y*T_SWIDTH+1x ;r=fill width (dbf) dr=(3x-2x)<<16/(3y-2y)-(2x-1x)<<16/(3y-1y) ;l=offset in chunky dl=(3x-1x)<<16/(3y-1y) move.l d3,d6 NOTE: 3y>1y !! sub.l d1,d6 (3x3y-1x1y) move.l d6,d7 ext.l d6 (3y-1y) clr.w d7 (3x-1x)<<16 divs.l d6,d7 move.l d7,a2 a2=dl move.l d2,d7 sub.l d1,d7 sub.w d3,d1 d3=3y-1y neg.w d1 NOTE: no subq.w #1,d1 sub.l d2,d3 NOTE: 3y>2y !! clr.w d3 (3x-2x)<<16 neg.l d3 divs.l d6,d3 add.l a2,d3 move.l #$8000,d6 d6=l move.w d6,d7 d7=r (2x-1x)<<16!$8000 mcnop4 .oloop3 swap d6 ffffiiii move.l d7,d4 iiiiffff lea (a0,d6.w),a1 swap d4 ffffiiii swap d6 iiiiffff IFEQ T_SMALLMODE cmp.w #8,d4 blo.b .bl3 IFNE T_4PIXWRITE subq.w #4,d4 ELSE subq.w #8,d4 ENDC move.w d4,d5 IFNE T_4PIXWRITE and.w #%11,d4 lsr.w #2,d5 ELSE and.w #%111,d4 lsr.w #3,d5 ENDC .iloop3 IFEQ T_4PIXWRITE move.l d0,(a1)+ ENDC move.l d0,(a1)+ dbf d5,.iloop3 ENDC .bl3 move.b d0,(a1)+ dbf d4,.bl3 add.l a2,d6 l=l+dl (iiiiffff) lea (T_SWIDTH,a0),a0 move to next line sub.l d3,d7 r=r+dr (iiiiffff) dbf d1,.oloop3 IFNE T_SPECIALBFLAT rts CNOP 0,4 .bottom_flat ; 1 ; /\ ; / \ ;2·----· 3 (or 3 2) cmp.l d2,d3 bhi.b .nxs2 exg d2,d3 .nxs2 ;r=fill width (dbf) dr=(3x-1x)<<16/(2y-1y)-(2x-1x)<<16/(2y-1y) ;l=offset in chunky dl=(2x-1x)<<16/(2y-1y) move.l d2,d7 NOTE: 2y>1y !! sub.l d1,d7 (2x2y-1x1y) move.l d7,d6 ext.l d7 (2y-1y) clr.w d6 (2x-1x)<<16 divs.l d7,d6 d6=dl sub.l d1,d3 NOTE: 3y>1y !! clr.w d3 (3x-1x)<<16 divs.l d7,d3 sub.w d1,d2 d2=2y-1y (NOTE: no subq.w #1,d2) sub.l d6,d3 d3=dr (dr-dl) move.l #$8000,d1 d1=l move.l d1,d7 d7=r mcnop4 .oloop4 swap d1 ffffiiii move.l d7,d4 iiiiffff lea (a0,d1.w),a1 swap d4 ffffiiii swap d1 iiiiffff IFEQ T_SMALLMODE cmp.w #8,d4 blo.b .bl4 IFNE T_4PIXWRITE subq.w #4,d4 ELSE subq.w #8,d4 ENDC move.w d4,d5 IFNE T_4PIXWRITE and.w #%11,d4 lsr.w #2,d5 ELSE and.w #%111,d4 lsr.w #3,d5 ENDC .iloop4 IFEQ T_4PIXWRITE move.l d0,(a1)+ ENDC move.l d0,(a1)+ dbf d5,.iloop4 ENDC .bl4 move.b d0,(a1)+ dbf d4,.bl4 add.l d6,d1 l=l+dl (iiiiffff) lea (T_SWIDTH,a0),a0 move to next line add.l d3,d7 r=r+dr (iiiiffff) dbf d2,.oloop4 ENDC d rts ;len dc.l ((b-a)+(d-c))