1 #include <u.h>
2 #include <libc.h>
3 #include <draw.h>
4 #include <memdraw.h>
5 
6 int drawdebug;
7 static int	tablesbuilt;
8 
9 /* perfect approximation to NTSC = .299r+.587g+.114b when 0 ≤ r,g,b < 256 */
10 #define RGB2K(r,g,b)	((156763*(r)+307758*(g)+59769*(b))>>19)
11 
12 /*
13  * for 0 ≤ x ≤ 255*255, (x*0x0101+0x100)>>16 is a perfect approximation.
14  * for 0 ≤ x < (1<<16), x/255 = ((x+1)*0x0101)>>16 is a perfect approximation.
15  * the last one is perfect for all up to 1<<16, avoids a multiply, but requires a rathole.
16  */
17 /* #define DIV255(x) (((x)*257+256)>>16)  */
18 #define DIV255(x) ((((x)+1)*257)>>16)
19 /* #define DIV255(x) (tmp=(x)+1, (tmp+(tmp>>8))>>8) */
20 
21 #define MUL(x, y, t)	(t = (x)*(y)+128, (t+(t>>8))>>8)
22 #define MASK13	0xFF00FF00
23 #define MASK02	0x00FF00FF
24 #define MUL13(a, x, t)		(t = (a)*(((x)&MASK13)>>8)+128, ((t+((t>>8)&MASK02))>>8)&MASK02)
25 #define MUL02(a, x, t)		(t = (a)*(((x)&MASK02)>>0)+128, ((t+((t>>8)&MASK02))>>8)&MASK02)
26 #define MUL0123(a, x, s, t)	((MUL13(a, x, s)<<8)|MUL02(a, x, t))
27 
28 #define MUL2(u, v, x, y)	(t = (u)*(v)+(x)*(y)+256, (t+(t>>8))>>8)
29 
30 static void mktables(void);
31 typedef int Subdraw(Memdrawparam*);
32 static Subdraw chardraw, alphadraw, memoptdraw;
33 
34 static Memimage*	memones;
35 static Memimage*	memzeros;
36 Memimage *memwhite;
37 Memimage *memblack;
38 Memimage *memtransparent;
39 Memimage *memopaque;
40 
41 int	__ifmt(Fmt*);
42 
43 void
44 memimageinit(void)
45 {
46 	static int didinit = 0;
47 
48 	if(didinit)
49 		return;
50 
51 	didinit = 1;
52 
53 	mktables();
54 	_memmkcmap();
55 
56 	fmtinstall('R', Rfmt); 
57 	fmtinstall('P', Pfmt);
58 	fmtinstall('b', __ifmt);
59 
60 	memones = allocmemimage(Rect(0,0,1,1), GREY1);
61 	memones->flags |= Frepl;
62 	memones->clipr = Rect(-0x3FFFFFF, -0x3FFFFFF, 0x3FFFFFF, 0x3FFFFFF);
63 	*byteaddr(memones, ZP) = ~0;
64 
65 	memzeros = allocmemimage(Rect(0,0,1,1), GREY1);
66 	memzeros->flags |= Frepl;
67 	memzeros->clipr = Rect(-0x3FFFFFF, -0x3FFFFFF, 0x3FFFFFF, 0x3FFFFFF);
68 	*byteaddr(memzeros, ZP) = 0;
69 
70 	if(memones == nil || memzeros == nil)
71 		assert(0 /*cannot initialize memimage library */);	/* RSC BUG */
72 
73 	memwhite = memones;
74 	memblack = memzeros;
75 	memopaque = memones;
76 	memtransparent = memzeros;
77 }
78 
79 u32int _imgtorgba(Memimage*, u32int);
80 u32int _rgbatoimg(Memimage*, u32int);
81 u32int _pixelbits(Memimage*, Point);
82 
83 #define DBG if(drawdebug)
84 static Memdrawparam par;
85 
86 Memdrawparam*
87 _memimagedrawsetup(Memimage *dst, Rectangle r, Memimage *src, Point p0, Memimage *mask, Point p1, int op)
88 {
89 	if(mask == nil)
90 		mask = memopaque;
91 
92 DBG	print("memimagedraw %p/%luX %R @ %p %p/%luX %P %p/%luX %P... ", dst, dst->chan, r, dst->data->bdata, src, src->chan, p0, mask, mask->chan, p1);
93 
94 	if(drawclip(dst, &r, src, &p0, mask, &p1, &par.sr, &par.mr) == 0){
95 /*		if(drawdebug) */
96 /*			iprint("empty clipped rectangle\n"); */
97 		return nil;
98 	}
99 
100 	if(op < Clear || op > SoverD){
101 /*		if(drawdebug) */
102 /*			iprint("op out of range: %d\n", op); */
103 		return nil;
104 	}
105 
106 	par.op = op;
107 	par.dst = dst;
108 	par.r = r;
109 	par.src = src;
110 	/* par.sr set by drawclip */
111 	par.mask = mask;
112 	/* par.mr set by drawclip */
113 
114 	par.state = 0;
115 	if(src->flags&Frepl){
116 		par.state |= Replsrc;
117 		if(Dx(src->r)==1 && Dy(src->r)==1){
118 			par.sval = pixelbits(src, src->r.min);
119 			par.state |= Simplesrc;
120 			par.srgba = _imgtorgba(src, par.sval);
121 			par.sdval = _rgbatoimg(dst, par.srgba);
122 			if((par.srgba&0xFF) == 0 && (op&DoutS)){
123 /*				if (drawdebug) iprint("fill with transparent source\n"); */
124 				return nil;	/* no-op successfully handled */
125 			}
126 			if((par.srgba&0xFF) == 0xFF)
127 				par.state |= Fullsrc;
128 		}
129 	}
130 
131 	if(mask->flags & Frepl){
132 		par.state |= Replmask;
133 		if(Dx(mask->r)==1 && Dy(mask->r)==1){
134 			par.mval = pixelbits(mask, mask->r.min);
135 			if(par.mval == 0 && (op&DoutS)){
136 /*				if(drawdebug) iprint("fill with zero mask\n"); */
137 				return nil;	/* no-op successfully handled */
138 			}
139 			par.state |= Simplemask;
140 			if(par.mval == ~0)
141 				par.state |= Fullmask;
142 			par.mrgba = _imgtorgba(mask, par.mval);
143 		}
144 	}
145 
146 /*	if(drawdebug) */
147 /*		iprint("dr %R sr %R mr %R...", r, par.sr, par.mr); */
148 DBG print("draw dr %R sr %R mr %R %lux\n", r, par.sr, par.mr, par.state);
149 
150 	return &par;
151 }
152 
153 void
154 _memimagedraw(Memdrawparam *par)
155 {
156 	/*
157 	 * Now that we've clipped the parameters down to be consistent, we 
158 	 * simply try sub-drawing routines in order until we find one that was able
159 	 * to handle us.  If the sub-drawing routine returns zero, it means it was
160 	 * unable to satisfy the request, so we do not return.
161 	 */
162 
163 	/*
164 	 * Hardware support.  Each video driver provides this function,
165 	 * which checks to see if there is anything it can help with.
166 	 * There could be an if around this checking to see if dst is in video memory.
167 	 */
168 DBG print("test hwdraw\n");
169 	if(hwdraw(par)){
170 /*if(drawdebug) iprint("hw handled\n"); */
171 DBG print("hwdraw handled\n");
172 		return;
173 	}
174 	/*
175 	 * Optimizations using memmove and memset.
176 	 */
177 DBG print("test memoptdraw\n");
178 	if(memoptdraw(par)){
179 /*if(drawdebug) iprint("memopt handled\n"); */
180 DBG print("memopt handled\n");
181 		return;
182 	}
183 
184 	/*
185 	 * Character drawing.
186 	 * Solid source color being painted through a boolean mask onto a high res image.
187 	 */
188 DBG print("test chardraw\n");
189 	if(chardraw(par)){
190 /*if(drawdebug) iprint("chardraw handled\n"); */
191 DBG print("chardraw handled\n");
192 		return;
193 	}
194 
195 	/*
196 	 * General calculation-laden case that does alpha for each pixel.
197 	 */
198 DBG print("do alphadraw\n");
199 	alphadraw(par);
200 /*if(drawdebug) iprint("alphadraw handled\n"); */
201 DBG print("alphadraw handled\n");
202 }
203 #undef DBG
204 
205 /*
206  * Clip the destination rectangle further based on the properties of the 
207  * source and mask rectangles.  Once the destination rectangle is properly
208  * clipped, adjust the source and mask rectangles to be the same size.
209  * Then if source or mask is replicated, move its clipped rectangle
210  * so that its minimum point falls within the repl rectangle.
211  *
212  * Return zero if the final rectangle is null.
213  */
214 int
215 drawclip(Memimage *dst, Rectangle *r, Memimage *src, Point *p0, Memimage *mask, Point *p1, Rectangle *sr, Rectangle *mr)
216 {
217 	Point rmin, delta;
218 	int splitcoords;
219 	Rectangle omr;
220 
221 	if(r->min.x>=r->max.x || r->min.y>=r->max.y)
222 		return 0;
223 	splitcoords = (p0->x!=p1->x) || (p0->y!=p1->y);
224 	/* clip to destination */
225 	rmin = r->min;
226 	if(!rectclip(r, dst->r) || !rectclip(r, dst->clipr))
227 		return 0;
228 	/* move mask point */
229 	p1->x += r->min.x-rmin.x;
230 	p1->y += r->min.y-rmin.y;
231 	/* move source point */
232 	p0->x += r->min.x-rmin.x;
233 	p0->y += r->min.y-rmin.y;
234 	/* map destination rectangle into source */
235 	sr->min = *p0;
236 	sr->max.x = p0->x+Dx(*r);
237 	sr->max.y = p0->y+Dy(*r);
238 	/* sr is r in source coordinates; clip to source */
239 	if(!(src->flags&Frepl) && !rectclip(sr, src->r))
240 		return 0;
241 	if(!rectclip(sr, src->clipr))
242 		return 0;
243 	/* compute and clip rectangle in mask */
244 	if(splitcoords){
245 		/* move mask point with source */
246 		p1->x += sr->min.x-p0->x;
247 		p1->y += sr->min.y-p0->y;
248 		mr->min = *p1;
249 		mr->max.x = p1->x+Dx(*sr);
250 		mr->max.y = p1->y+Dy(*sr);
251 		omr = *mr;
252 		/* mr is now rectangle in mask; clip it */
253 		if(!(mask->flags&Frepl) && !rectclip(mr, mask->r))
254 			return 0;
255 		if(!rectclip(mr, mask->clipr))
256 			return 0;
257 		/* reflect any clips back to source */
258 		sr->min.x += mr->min.x-omr.min.x;
259 		sr->min.y += mr->min.y-omr.min.y;
260 		sr->max.x += mr->max.x-omr.max.x;
261 		sr->max.y += mr->max.y-omr.max.y;
262 		*p1 = mr->min;
263 	}else{
264 		if(!(mask->flags&Frepl) && !rectclip(sr, mask->r))
265 			return 0;
266 		if(!rectclip(sr, mask->clipr))
267 			return 0;
268 		*p1 = sr->min;
269 	}
270 
271 	/* move source clipping back to destination */
272 	delta.x = r->min.x - p0->x;
273 	delta.y = r->min.y - p0->y;
274 	r->min.x = sr->min.x + delta.x;
275 	r->min.y = sr->min.y + delta.y;
276 	r->max.x = sr->max.x + delta.x;
277 	r->max.y = sr->max.y + delta.y;
278 
279 	/* move source rectangle so sr->min is in src->r */
280 	if(src->flags&Frepl) {
281 		delta.x = drawreplxy(src->r.min.x, src->r.max.x, sr->min.x) - sr->min.x;
282 		delta.y = drawreplxy(src->r.min.y, src->r.max.y, sr->min.y) - sr->min.y;
283 		sr->min.x += delta.x;
284 		sr->min.y += delta.y;
285 		sr->max.x += delta.x;
286 		sr->max.y += delta.y;
287 	}
288 	*p0 = sr->min;
289 
290 	/* move mask point so it is in mask->r */
291 	*p1 = drawrepl(mask->r, *p1);
292 	mr->min = *p1;
293 	mr->max.x = p1->x+Dx(*sr);
294 	mr->max.y = p1->y+Dy(*sr);
295 
296 	assert(Dx(*sr) == Dx(*mr) && Dx(*mr) == Dx(*r));
297 	assert(Dy(*sr) == Dy(*mr) && Dy(*mr) == Dy(*r));
298 	assert(ptinrect(*p0, src->r));
299 	assert(ptinrect(*p1, mask->r));
300 	assert(ptinrect(r->min, dst->r));
301 
302 	return 1;
303 }
304 
305 /*
306  * Conversion tables.
307  */
308 static uchar replbit[1+8][256];		/* replbit[x][y] is the replication of the x-bit quantity y to 8-bit depth */
309 static uchar conv18[256][8];		/* conv18[x][y] is the yth pixel in the depth-1 pixel x */
310 static uchar conv28[256][4];		/* ... */
311 static uchar conv48[256][2];
312 
313 /*
314  * bitmap of how to replicate n bits to fill 8, for 1 ≤ n ≤ 8.
315  * the X's are where to put the bottom (ones) bit of the n-bit pattern.
316  * only the top 8 bits of the result are actually used.
317  * (the lower 8 bits are needed to get bits in the right place
318  * when n is not a divisor of 8.)
319  *
320  * Should check to see if its easier to just refer to replmul than
321  * use the precomputed values in replbit.  On PCs it may well
322  * be; on machines with slow multiply instructions it probably isn't.
323  */
324 #define a ((((((((((((((((0
325 #define X *2+1)
326 #define _ *2)
327 static int replmul[1+8] = {
328 	0,
329 	a X X X X X X X X X X X X X X X X,
330 	a _ X _ X _ X _ X _ X _ X _ X _ X,
331 	a _ _ X _ _ X _ _ X _ _ X _ _ X _,
332 	a _ _ _ X _ _ _ X _ _ _ X _ _ _ X,
333 	a _ _ _ _ X _ _ _ _ X _ _ _ _ X _,
334 	a _ _ _ _ _ X _ _ _ _ _ X _ _ _ _, 
335 	a _ _ _ _ _ _ X _ _ _ _ _ _ X _ _,
336 	a _ _ _ _ _ _ _ X _ _ _ _ _ _ _ X,
337 };
338 #undef a
339 #undef X
340 #undef _
341 
342 static void
343 mktables(void)
344 {
345 	int i, j, mask, sh, small;
346 		
347 	if(tablesbuilt)
348 		return;
349 
350 	fmtinstall('R', Rfmt);
351 	fmtinstall('P', Pfmt);
352 	tablesbuilt = 1;
353 
354 	/* bit replication up to 8 bits */
355 	for(i=0; i<256; i++){
356 		for(j=0; j<=8; j++){	/* j <= 8 [sic] */
357 			small = i & ((1<<j)-1);
358 			replbit[j][i] = (small*replmul[j])>>8;
359 		}
360 	}
361 
362 	/* bit unpacking up to 8 bits, only powers of 2 */
363 	for(i=0; i<256; i++){
364 		for(j=0, sh=7, mask=1; j<8; j++, sh--)
365 			conv18[i][j] = replbit[1][(i>>sh)&mask];
366 
367 		for(j=0, sh=6, mask=3; j<4; j++, sh-=2)
368 			conv28[i][j] = replbit[2][(i>>sh)&mask];
369 
370 		for(j=0, sh=4, mask=15; j<2; j++, sh-=4)
371 			conv48[i][j] = replbit[4][(i>>sh)&mask];
372 	}
373 }
374 
375 static uchar ones = 0xff;
376 
377 /*
378  * General alpha drawing case.  Can handle anything.
379  */
380 typedef struct	Buffer	Buffer;
381 struct Buffer {
382 	/* used by most routines */
383 	uchar	*red;
384 	uchar	*grn;
385 	uchar	*blu;
386 	uchar	*alpha;
387 	uchar	*grey;
388 	u32int	*rgba;
389 	int	delta;	/* number of bytes to add to pointer to get next pixel to the right */
390 
391 	/* used by boolcalc* for mask data */
392 	uchar	*m;		/* ptr to mask data r.min byte; like p->bytermin */
393 	int		mskip;	/* no. of left bits to skip in *m */
394 	uchar	*bm;		/* ptr to mask data img->r.min byte; like p->bytey0s */
395 	int		bmskip;	/* no. of left bits to skip in *bm */
396 	uchar	*em;		/* ptr to mask data img->r.max.x byte; like p->bytey0e */
397 	int		emskip;	/* no. of right bits to skip in *em */
398 };
399 
400 typedef struct	Param	Param;
401 typedef Buffer	Readfn(Param*, uchar*, int);
402 typedef void	Writefn(Param*, uchar*, Buffer);
403 typedef Buffer	Calcfn(Buffer, Buffer, Buffer, int, int, int);
404 
405 enum {
406 	MAXBCACHE = 16
407 };
408 
409 /* giant rathole to customize functions with */
410 struct Param {
411 	Readfn	*replcall;
412 	Readfn	*greymaskcall;	
413 	Readfn	*convreadcall;
414 	Writefn	*convwritecall;
415 
416 	Memimage *img;
417 	Rectangle	r;
418 	int	dx;	/* of r */
419 	int	needbuf;
420 	int	convgrey;
421 	int	alphaonly;
422 
423 	uchar	*bytey0s;		/* byteaddr(Pt(img->r.min.x, img->r.min.y)) */
424 	uchar	*bytermin;	/* byteaddr(Pt(r.min.x, img->r.min.y)) */
425 	uchar	*bytey0e;		/* byteaddr(Pt(img->r.max.x, img->r.min.y)) */
426 	int		bwidth;
427 
428 	int	replcache;	/* if set, cache buffers */
429 	Buffer	bcache[MAXBCACHE];
430 	u32int	bfilled;
431 	uchar	*bufbase;
432 	int	bufoff;
433 	int	bufdelta;
434 
435 	int	dir;
436 
437 	int	convbufoff;
438 	uchar	*convbuf;
439 	Param	*convdpar;
440 	int	convdx;
441 };
442 
443 static uchar *drawbuf;
444 static int	ndrawbuf;
445 static int	mdrawbuf;
446 static Param spar, mpar, dpar;	/* easier on the stacks */
447 static Readfn	greymaskread, replread, readptr;
448 static Writefn	nullwrite;
449 static Calcfn	alphacalc0, alphacalc14, alphacalc2810, alphacalc3679, alphacalc5, alphacalc11, alphacalcS;
450 static Calcfn	boolcalc14, boolcalc236789, boolcalc1011;
451 
452 static Readfn*	readfn(Memimage*);
453 static Readfn*	readalphafn(Memimage*);
454 static Writefn*	writefn(Memimage*);
455 
456 static Calcfn*	boolcopyfn(Memimage*, Memimage*);
457 static Readfn*	convfn(Memimage*, Param*, Memimage*, Param*);
458 
459 static Calcfn *alphacalc[Ncomp] = 
460 {
461 	alphacalc0,		/* Clear */
462 	alphacalc14,		/* DoutS */
463 	alphacalc2810,		/* SoutD */
464 	alphacalc3679,		/* DxorS */
465 	alphacalc14,		/* DinS */
466 	alphacalc5,		/* D */
467 	alphacalc3679,		/* DatopS */
468 	alphacalc3679,		/* DoverS */
469 	alphacalc2810,		/* SinD */
470 	alphacalc3679,		/* SatopD */
471 	alphacalc2810,		/* S */
472 	alphacalc11,		/* SoverD */
473 };
474 
475 static Calcfn *boolcalc[Ncomp] =
476 {
477 	alphacalc0,		/* Clear */
478 	boolcalc14,		/* DoutS */
479 	boolcalc236789,		/* SoutD */
480 	boolcalc236789,		/* DxorS */
481 	boolcalc14,		/* DinS */
482 	alphacalc5,		/* D */
483 	boolcalc236789,		/* DatopS */
484 	boolcalc236789,		/* DoverS */
485 	boolcalc236789,		/* SinD */
486 	boolcalc236789,		/* SatopD */
487 	boolcalc1011,		/* S */
488 	boolcalc1011,		/* SoverD */
489 };
490 
491 static int
492 allocdrawbuf(void)
493 {
494 	uchar *p;
495 
496 	if(ndrawbuf > mdrawbuf){
497 		p = realloc(drawbuf, ndrawbuf);
498 		if(p == nil){
499 			werrstr("memimagedraw out of memory");
500 			return -1;
501 		}
502 		drawbuf = p;
503 		mdrawbuf = ndrawbuf;
504 	}
505 	return 0;
506 }
507 
508 static void
509 getparam(Param *p, Memimage *img, Rectangle r, int convgrey, int needbuf)
510 {
511 	int nbuf;
512 
513 	memset(p, 0, sizeof *p);
514 
515 	p->img = img;
516 	p->r = r;
517 	p->dx = Dx(r);
518 	p->needbuf = needbuf;
519 	p->convgrey = convgrey;
520 
521 	assert(img->r.min.x <= r.min.x && r.min.x < img->r.max.x);
522 
523 	p->bytey0s = byteaddr(img, Pt(img->r.min.x, img->r.min.y));
524 	p->bytermin = byteaddr(img, Pt(r.min.x, img->r.min.y));
525 	p->bytey0e = byteaddr(img, Pt(img->r.max.x, img->r.min.y));
526 	p->bwidth = sizeof(u32int)*img->width;
527 
528 	assert(p->bytey0s <= p->bytermin && p->bytermin <= p->bytey0e);
529 
530 	if(p->r.min.x == p->img->r.min.x)
531 		assert(p->bytermin == p->bytey0s);
532 
533 	nbuf = 1;
534 	if((img->flags&Frepl) && Dy(img->r) <= MAXBCACHE && Dy(img->r) < Dy(r)){
535 		p->replcache = 1;
536 		nbuf = Dy(img->r);
537 	}
538 	p->bufdelta = 4*p->dx;
539 	p->bufoff = ndrawbuf;
540 	ndrawbuf += p->bufdelta*nbuf;
541 }
542 
543 static void
544 clipy(Memimage *img, int *y)
545 {
546 	int dy;
547 
548 	dy = Dy(img->r);
549 	if(*y == dy)
550 		*y = 0;
551 	else if(*y == -1)
552 		*y = dy-1;
553 	assert(0 <= *y && *y < dy);
554 }
555 
556 static void
557 dumpbuf(char *s, Buffer b, int n)
558 {
559 	int i;
560 	uchar *p;
561 	
562 	print("%s", s);
563 	for(i=0; i<n; i++){
564 		print(" ");
565 		if(p=b.grey){
566 			print(" k%.2uX", *p);
567 			b.grey += b.delta;
568 		}else{	
569 			if(p=b.red){
570 				print(" r%.2uX", *p);
571 				b.red += b.delta;
572 			}
573 			if(p=b.grn){
574 				print(" g%.2uX", *p);
575 				b.grn += b.delta;
576 			}
577 			if(p=b.blu){
578 				print(" b%.2uX", *p);
579 				b.blu += b.delta;
580 			}
581 		}
582 		if((p=b.alpha) != &ones){
583 			print(" α%.2uX", *p);
584 			b.alpha += b.delta;
585 		}
586 	}
587 	print("\n");
588 }
589 
590 /*
591  * For each scan line, we expand the pixels from source, mask, and destination
592  * into byte-aligned red, green, blue, alpha, and grey channels.  If buffering is not
593  * needed and the channels were already byte-aligned (grey8, rgb24, rgba32, rgb32),
594  * the readers need not copy the data: they can simply return pointers to the data.
595  * If the destination image is grey and the source is not, it is converted using the NTSC
596  * formula.
597  *
598  * Once we have all the channels, we call either rgbcalc or greycalc, depending on 
599  * whether the destination image is color.  This is allowed to overwrite the dst buffer (perhaps
600  * the actual data, perhaps a copy) with its result.  It should only overwrite the dst buffer
601  * with the same format (i.e. red bytes with red bytes, etc.)  A new buffer is returned from
602  * the calculator, and that buffer is passed to a function to write it to the destination.
603  * If the buffer is already pointing at the destination, the writing function is a no-op.
604  */
605 #define DBG if(drawdebug)
606 static int
607 alphadraw(Memdrawparam *par)
608 {
609 	int isgrey, starty, endy, op;
610 	int needbuf, dsty, srcy, masky;
611 	int y, dir, dx, dy;
612 	Buffer bsrc, bdst, bmask;
613 	Readfn *rdsrc, *rdmask, *rddst;
614 	Calcfn *calc;
615 	Writefn *wrdst;
616 	Memimage *src, *mask, *dst;
617 	Rectangle r, sr, mr;
618 
619 	if(drawdebug)
620 		print("alphadraw %R\n", par->r);
621 	r = par->r;
622 	dx = Dx(r);
623 	dy = Dy(r);
624 
625 	ndrawbuf = 0;
626 
627 	src = par->src;
628 	mask = par->mask;	
629 	dst = par->dst;
630 	sr = par->sr;
631 	mr = par->mr;
632 	op = par->op;
633 
634 	isgrey = dst->flags&Fgrey;
635 
636 	/*
637 	 * Buffering when src and dst are the same bitmap is sufficient but not 
638 	 * necessary.  There are stronger conditions we could use.  We could
639 	 * check to see if the rectangles intersect, and if simply moving in the
640 	 * correct y direction can avoid the need to buffer.
641 	 */
642 	needbuf = (src->data == dst->data);
643 
644 	getparam(&spar, src, sr, isgrey, needbuf);
645 	getparam(&dpar, dst, r, isgrey, needbuf);
646 	getparam(&mpar, mask, mr, 0, needbuf);
647 
648 	dir = (needbuf && byteaddr(dst, r.min) > byteaddr(src, sr.min)) ? -1 : 1;
649 	spar.dir = mpar.dir = dpar.dir = dir;
650 
651 	/*
652 	 * If the mask is purely boolean, we can convert from src to dst format
653 	 * when we read src, and then just copy it to dst where the mask tells us to.
654 	 * This requires a boolean (1-bit grey) mask and lack of a source alpha channel.
655 	 *
656 	 * The computation is accomplished by assigning the function pointers as follows:
657 	 *	rdsrc - read and convert source into dst format in a buffer
658 	 * 	rdmask - convert mask to bytes, set pointer to it
659 	 * 	rddst - fill with pointer to real dst data, but do no reads
660 	 *	calc - copy src onto dst when mask says to.
661 	 *	wrdst - do nothing
662 	 * This is slightly sleazy, since things aren't doing exactly what their names say,
663 	 * but it avoids a fair amount of code duplication to make this a case here
664 	 * rather than have a separate booldraw.
665 	 */
666 /*if(drawdebug) iprint("flag %lud mchan %lux=?%x dd %d\n", src->flags&Falpha, mask->chan, GREY1, dst->depth); */
667 	if(!(src->flags&Falpha) && mask->chan == GREY1 && dst->depth >= 8 && op == SoverD){
668 /*if(drawdebug) iprint("boolcopy..."); */
669 		rdsrc = convfn(dst, &dpar, src, &spar);
670 		rddst = readptr;
671 		rdmask = readfn(mask);
672 		calc = boolcopyfn(dst, mask);
673 		wrdst = nullwrite;
674 	}else{
675 		/* usual alphadraw parameter fetching */
676 		rdsrc = readfn(src);
677 		rddst = readfn(dst);
678 		wrdst = writefn(dst);
679 		calc = alphacalc[op];
680 
681 		/*
682 		 * If there is no alpha channel, we'll ask for a grey channel
683 		 * and pretend it is the alpha.
684 		 */
685 		if(mask->flags&Falpha){
686 			rdmask = readalphafn(mask);
687 			mpar.alphaonly = 1;
688 		}else{
689 			mpar.greymaskcall = readfn(mask);
690 			mpar.convgrey = 1;
691 			rdmask = greymaskread;
692 
693 			/*
694 			 * Should really be above, but then boolcopyfns would have
695 			 * to deal with bit alignment, and I haven't written that.
696 			 *
697 			 * This is a common case for things like ellipse drawing.
698 			 * When there's no alpha involved and the mask is boolean,
699 			 * we can avoid all the division and multiplication.
700 			 */
701 			if(mask->chan == GREY1 && !(src->flags&Falpha))
702 				calc = boolcalc[op];
703 			else if(op == SoverD && !(src->flags&Falpha))
704 				calc = alphacalcS;
705 		}
706 	}
707 
708 	/*
709 	 * If the image has a small enough repl rectangle,
710 	 * we can just read each line once and cache them.
711 	 */
712 	if(spar.replcache){
713 		spar.replcall = rdsrc;
714 		rdsrc = replread;
715 	}
716 	if(mpar.replcache){
717 		mpar.replcall = rdmask;
718 		rdmask = replread;
719 	}
720 
721 	if(allocdrawbuf() < 0)
722 		return 0;
723 
724 	/*
725 	 * Before we were saving only offsets from drawbuf in the parameter
726 	 * structures; now that drawbuf has been grown to accomodate us,
727 	 * we can fill in the pointers.
728 	 */
729 	spar.bufbase = drawbuf+spar.bufoff;
730 	mpar.bufbase = drawbuf+mpar.bufoff;
731 	dpar.bufbase = drawbuf+dpar.bufoff;
732 	spar.convbuf = drawbuf+spar.convbufoff;
733 
734 	if(dir == 1){
735 		starty = 0;
736 		endy = dy;
737 	}else{
738 		starty = dy-1;
739 		endy = -1;
740 	}
741 
742 	/*
743 	 * srcy, masky, and dsty are offsets from the top of their
744 	 * respective Rectangles.  they need to be contained within
745 	 * the rectangles, so clipy can keep them there without division.
746  	 */
747 	srcy = (starty + sr.min.y - src->r.min.y)%Dy(src->r);
748 	masky = (starty + mr.min.y - mask->r.min.y)%Dy(mask->r);
749 	dsty = starty + r.min.y - dst->r.min.y;
750 
751 	assert(0 <= srcy && srcy < Dy(src->r));
752 	assert(0 <= masky && masky < Dy(mask->r));
753 	assert(0 <= dsty && dsty < Dy(dst->r));
754 
755 	if(drawdebug)
756 		print("alphadraw: rdsrc=%p rdmask=%p rddst=%p calc=%p wrdst=%p\n",
757 			rdsrc, rdmask, rddst, calc, wrdst);
758 	for(y=starty; y!=endy; y+=dir, srcy+=dir, masky+=dir, dsty+=dir){
759 		clipy(src, &srcy);
760 		clipy(dst, &dsty);
761 		clipy(mask, &masky);
762 
763 		bsrc = rdsrc(&spar, spar.bufbase, srcy);
764 DBG print("[");
765 		bmask = rdmask(&mpar, mpar.bufbase, masky);
766 DBG print("]\n");
767 		bdst = rddst(&dpar, dpar.bufbase, dsty);
768 DBG		dumpbuf("src", bsrc, dx);
769 DBG		dumpbuf("mask", bmask, dx);
770 DBG		dumpbuf("dst", bdst, dx);
771 		bdst = calc(bdst, bsrc, bmask, dx, isgrey, op);
772 DBG		dumpbuf("bdst", bdst, dx);
773 		wrdst(&dpar, dpar.bytermin+dsty*dpar.bwidth, bdst);
774 	}
775 
776 	return 1;
777 }
778 #undef DBG
779 
780 static Buffer
781 alphacalc0(Buffer bdst, Buffer b1, Buffer b2, int dx, int grey, int op)
782 {
783 	USED(grey);
784 	USED(op);
785 	memset(bdst.rgba, 0, dx*bdst.delta);
786 	return bdst;
787 }
788 
789 static Buffer
790 alphacalc14(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int grey, int op)
791 {
792 	Buffer obdst;
793 	int fd, sadelta;
794 	int i, sa, ma, q;
795 	u32int s, t;
796 
797 	obdst = bdst;
798 	sadelta = bsrc.alpha == &ones ? 0 : bsrc.delta;
799 	q = bsrc.delta == 4 && bdst.delta == 4;
800 
801 	for(i=0; i<dx; i++){
802 		sa = *bsrc.alpha;
803 		ma = *bmask.alpha;
804 		fd = MUL(sa, ma, t);
805 		if(op == DoutS)
806 			fd = 255-fd;
807 
808 		if(grey){
809 			*bdst.grey = MUL(fd, *bdst.grey, t);
810 			bsrc.grey += bsrc.delta;
811 			bdst.grey += bdst.delta;
812 		}else{
813 			if(q){
814 				*bdst.rgba = MUL0123(fd, *bdst.rgba, s, t);
815 				bsrc.rgba++;
816 				bdst.rgba++;
817 				bsrc.alpha += sadelta;
818 				bmask.alpha += bmask.delta;
819 				continue;
820 			}
821 			*bdst.red = MUL(fd, *bdst.red, t);
822 			*bdst.grn = MUL(fd, *bdst.grn, t);
823 			*bdst.blu = MUL(fd, *bdst.blu, t);
824 			bsrc.red += bsrc.delta;
825 			bsrc.blu += bsrc.delta;
826 			bsrc.grn += bsrc.delta;
827 			bdst.red += bdst.delta;
828 			bdst.blu += bdst.delta;
829 			bdst.grn += bdst.delta;
830 		}
831 		if(bdst.alpha != &ones){
832 			*bdst.alpha = MUL(fd, *bdst.alpha, t);
833 			bdst.alpha += bdst.delta;
834 		}
835 		bmask.alpha += bmask.delta;
836 		bsrc.alpha += sadelta;
837 	}
838 	return obdst;
839 }
840 
841 static Buffer
842 alphacalc2810(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int grey, int op)
843 {
844 	Buffer obdst;
845 	int fs, sadelta;
846 	int i, ma, da, q;
847 	u32int s, t;
848 
849 	obdst = bdst;
850 	sadelta = bsrc.alpha == &ones ? 0 : bsrc.delta;
851 	q = bsrc.delta == 4 && bdst.delta == 4;
852 
853 	for(i=0; i<dx; i++){
854 		ma = *bmask.alpha;
855 		da = *bdst.alpha;
856 		if(op == SoutD)
857 			da = 255-da;
858 		fs = ma;
859 		if(op != S)
860 			fs = MUL(fs, da, t);
861 
862 		if(grey){
863 			*bdst.grey = MUL(fs, *bsrc.grey, t);
864 			bsrc.grey += bsrc.delta;
865 			bdst.grey += bdst.delta;
866 		}else{
867 			if(q){
868 				*bdst.rgba = MUL0123(fs, *bsrc.rgba, s, t);
869 				bsrc.rgba++;
870 				bdst.rgba++;
871 				bmask.alpha += bmask.delta;
872 				bdst.alpha += bdst.delta;
873 				continue;
874 			}
875 			*bdst.red = MUL(fs, *bsrc.red, t);
876 			*bdst.grn = MUL(fs, *bsrc.grn, t);
877 			*bdst.blu = MUL(fs, *bsrc.blu, t);
878 			bsrc.red += bsrc.delta;
879 			bsrc.blu += bsrc.delta;
880 			bsrc.grn += bsrc.delta;
881 			bdst.red += bdst.delta;
882 			bdst.blu += bdst.delta;
883 			bdst.grn += bdst.delta;
884 		}
885 		if(bdst.alpha != &ones){
886 			*bdst.alpha = MUL(fs, *bsrc.alpha, t);
887 			bdst.alpha += bdst.delta;
888 		}
889 		bmask.alpha += bmask.delta;
890 		bsrc.alpha += sadelta;
891 	}
892 	return obdst;
893 }
894 
895 static Buffer
896 alphacalc3679(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int grey, int op)
897 {
898 	Buffer obdst;
899 	int fs, fd, sadelta;
900 	int i, sa, ma, da, q;
901 	u32int s, t, u, v;
902 
903 	obdst = bdst;
904 	sadelta = bsrc.alpha == &ones ? 0 : bsrc.delta;
905 	q = bsrc.delta == 4 && bdst.delta == 4;
906 
907 	for(i=0; i<dx; i++){
908 		sa = *bsrc.alpha;
909 		ma = *bmask.alpha;
910 		da = *bdst.alpha;
911 		if(op == SatopD)
912 			fs = MUL(ma, da, t);
913 		else
914 			fs = MUL(ma, 255-da, t);
915 		if(op == DoverS)
916 			fd = 255;
917 		else{
918 			fd = MUL(sa, ma, t);
919 			if(op != DatopS)
920 				fd = 255-fd;
921 		}
922 
923 		if(grey){
924 			*bdst.grey = MUL(fs, *bsrc.grey, s)+MUL(fd, *bdst.grey, t);
925 			bsrc.grey += bsrc.delta;
926 			bdst.grey += bdst.delta;
927 		}else{
928 			if(q){
929 				*bdst.rgba = MUL0123(fs, *bsrc.rgba, s, t)+MUL0123(fd, *bdst.rgba, u, v);
930 				bsrc.rgba++;
931 				bdst.rgba++;
932 				bsrc.alpha += sadelta;
933 				bmask.alpha += bmask.delta;
934 				bdst.alpha += bdst.delta;
935 				continue;
936 			}
937 			*bdst.red = MUL(fs, *bsrc.red, s)+MUL(fd, *bdst.red, t);
938 			*bdst.grn = MUL(fs, *bsrc.grn, s)+MUL(fd, *bdst.grn, t);
939 			*bdst.blu = MUL(fs, *bsrc.blu, s)+MUL(fd, *bdst.blu, t);
940 			bsrc.red += bsrc.delta;
941 			bsrc.blu += bsrc.delta;
942 			bsrc.grn += bsrc.delta;
943 			bdst.red += bdst.delta;
944 			bdst.blu += bdst.delta;
945 			bdst.grn += bdst.delta;
946 		}
947 		if(bdst.alpha != &ones){
948 			*bdst.alpha = MUL(fs, sa, s)+MUL(fd, da, t);
949 			bdst.alpha += bdst.delta;
950 		}
951 		bmask.alpha += bmask.delta;
952 		bsrc.alpha += sadelta;
953 	}
954 	return obdst;
955 }
956 
957 static Buffer
958 alphacalc5(Buffer bdst, Buffer b1, Buffer b2, int dx, int grey, int op)
959 {
960 	USED(dx);
961 	USED(grey);
962 	USED(op);
963 	return bdst;
964 }
965 
966 static Buffer
967 alphacalc11(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int grey, int op)
968 {
969 	Buffer obdst;
970 	int fd, sadelta;
971 	int i, sa, ma, q;
972 	u32int s, t, u, v;
973 
974 	USED(op);
975 	obdst = bdst;
976 	sadelta = bsrc.alpha == &ones ? 0 : bsrc.delta;
977 	q = bsrc.delta == 4 && bdst.delta == 4;
978 
979 	for(i=0; i<dx; i++){
980 		sa = *bsrc.alpha;
981 		ma = *bmask.alpha;
982 		fd = 255-MUL(sa, ma, t);
983 
984 		if(grey){
985 			*bdst.grey = MUL(ma, *bsrc.grey, s)+MUL(fd, *bdst.grey, t);
986 			bsrc.grey += bsrc.delta;
987 			bdst.grey += bdst.delta;
988 		}else{
989 			if(q){
990 				*bdst.rgba = MUL0123(ma, *bsrc.rgba, s, t)+MUL0123(fd, *bdst.rgba, u, v);
991 				bsrc.rgba++;
992 				bdst.rgba++;
993 				bsrc.alpha += sadelta;
994 				bmask.alpha += bmask.delta;
995 				continue;
996 			}
997 			*bdst.red = MUL(ma, *bsrc.red, s)+MUL(fd, *bdst.red, t);
998 			*bdst.grn = MUL(ma, *bsrc.grn, s)+MUL(fd, *bdst.grn, t);
999 			*bdst.blu = MUL(ma, *bsrc.blu, s)+MUL(fd, *bdst.blu, t);
1000 			bsrc.red += bsrc.delta;
1001 			bsrc.blu += bsrc.delta;
1002 			bsrc.grn += bsrc.delta;
1003 			bdst.red += bdst.delta;
1004 			bdst.blu += bdst.delta;
1005 			bdst.grn += bdst.delta;
1006 		}
1007 		if(bdst.alpha != &ones){
1008 			*bdst.alpha = MUL(ma, sa, s)+MUL(fd, *bdst.alpha, t);
1009 			bdst.alpha += bdst.delta;
1010 		}
1011 		bmask.alpha += bmask.delta;
1012 		bsrc.alpha += sadelta;
1013 	}
1014 	return obdst;
1015 }
1016 
1017 /*
1018 not used yet
1019 source and mask alpha 1
1020 static Buffer
1021 alphacalcS0(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int grey, int op)
1022 {
1023 	Buffer obdst;
1024 	int i;
1025 
1026 	USED(op);
1027 	obdst = bdst;
1028 	if(bsrc.delta == bdst.delta){
1029 		memmove(bdst.rgba, bsrc.rgba, dx*bdst.delta);
1030 		return obdst;
1031 	}
1032 	for(i=0; i<dx; i++){
1033 		if(grey){
1034 			*bdst.grey = *bsrc.grey;
1035 			bsrc.grey += bsrc.delta;
1036 			bdst.grey += bdst.delta;
1037 		}else{
1038 			*bdst.red = *bsrc.red;
1039 			*bdst.grn = *bsrc.grn;
1040 			*bdst.blu = *bsrc.blu;
1041 			bsrc.red += bsrc.delta;
1042 			bsrc.blu += bsrc.delta;
1043 			bsrc.grn += bsrc.delta;
1044 			bdst.red += bdst.delta;
1045 			bdst.blu += bdst.delta;
1046 			bdst.grn += bdst.delta;
1047 		}
1048 		if(bdst.alpha != &ones){
1049 			*bdst.alpha = 255;
1050 			bdst.alpha += bdst.delta;
1051 		}
1052 	}
1053 	return obdst;
1054 }
1055 */
1056 
1057 /* source alpha 1 */
1058 static Buffer
1059 alphacalcS(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int grey, int op)
1060 {
1061 	Buffer obdst;
1062 	int fd;
1063 	int i, ma;
1064 	u32int s, t;
1065 
1066 	USED(op);
1067 	obdst = bdst;
1068 
1069 	for(i=0; i<dx; i++){
1070 		ma = *bmask.alpha;
1071 		fd = 255-ma;
1072 
1073 		if(grey){
1074 			*bdst.grey = MUL(ma, *bsrc.grey, s)+MUL(fd, *bdst.grey, t);
1075 			bsrc.grey += bsrc.delta;
1076 			bdst.grey += bdst.delta;
1077 		}else{
1078 			*bdst.red = MUL(ma, *bsrc.red, s)+MUL(fd, *bdst.red, t);
1079 			*bdst.grn = MUL(ma, *bsrc.grn, s)+MUL(fd, *bdst.grn, t);
1080 			*bdst.blu = MUL(ma, *bsrc.blu, s)+MUL(fd, *bdst.blu, t);
1081 			bsrc.red += bsrc.delta;
1082 			bsrc.blu += bsrc.delta;
1083 			bsrc.grn += bsrc.delta;
1084 			bdst.red += bdst.delta;
1085 			bdst.blu += bdst.delta;
1086 			bdst.grn += bdst.delta;
1087 		}
1088 		if(bdst.alpha != &ones){
1089 			*bdst.alpha = ma+MUL(fd, *bdst.alpha, t);
1090 			bdst.alpha += bdst.delta;
1091 		}
1092 		bmask.alpha += bmask.delta;
1093 	}
1094 	return obdst;
1095 }
1096 
1097 static Buffer
1098 boolcalc14(Buffer bdst, Buffer b1, Buffer bmask, int dx, int grey, int op)
1099 {
1100 	Buffer obdst;
1101 	int i, ma, zero;
1102 
1103 	obdst = bdst;
1104 
1105 	for(i=0; i<dx; i++){
1106 		ma = *bmask.alpha;
1107 		zero = ma ? op == DoutS : op == DinS;
1108 
1109 		if(grey){
1110 			if(zero)
1111 				*bdst.grey = 0;
1112 			bdst.grey += bdst.delta;
1113 		}else{
1114 			if(zero)
1115 				*bdst.red = *bdst.grn = *bdst.blu = 0;
1116 			bdst.red += bdst.delta;
1117 			bdst.blu += bdst.delta;
1118 			bdst.grn += bdst.delta;
1119 		}
1120 		bmask.alpha += bmask.delta;
1121 		if(bdst.alpha != &ones){
1122 			if(zero)
1123 				*bdst.alpha = 0;
1124 			bdst.alpha += bdst.delta;
1125 		}
1126 	}
1127 	return obdst;
1128 }
1129 
1130 static Buffer
1131 boolcalc236789(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int grey, int op)
1132 {
1133 	Buffer obdst;
1134 	int fs, fd;
1135 	int i, ma, da, zero;
1136 	u32int s, t;
1137 
1138 	obdst = bdst;
1139 	zero = !(op&1);
1140 
1141 	for(i=0; i<dx; i++){
1142 		ma = *bmask.alpha;
1143 		da = *bdst.alpha;
1144 		fs = da;
1145 		if(op&2)
1146 			fs = 255-da;
1147 		fd = 0;
1148 		if(op&4)
1149 			fd = 255;
1150 
1151 		if(grey){
1152 			if(ma)
1153 				*bdst.grey = MUL(fs, *bsrc.grey, s)+MUL(fd, *bdst.grey, t);
1154 			else if(zero)
1155 				*bdst.grey = 0;
1156 			bsrc.grey += bsrc.delta;
1157 			bdst.grey += bdst.delta;
1158 		}else{
1159 			if(ma){
1160 				*bdst.red = MUL(fs, *bsrc.red, s)+MUL(fd, *bdst.red, t);
1161 				*bdst.grn = MUL(fs, *bsrc.grn, s)+MUL(fd, *bdst.grn, t);
1162 				*bdst.blu = MUL(fs, *bsrc.blu, s)+MUL(fd, *bdst.blu, t);
1163 			}
1164 			else if(zero)
1165 				*bdst.red = *bdst.grn = *bdst.blu = 0;
1166 			bsrc.red += bsrc.delta;
1167 			bsrc.blu += bsrc.delta;
1168 			bsrc.grn += bsrc.delta;
1169 			bdst.red += bdst.delta;
1170 			bdst.blu += bdst.delta;
1171 			bdst.grn += bdst.delta;
1172 		}
1173 		bmask.alpha += bmask.delta;
1174 		if(bdst.alpha != &ones){
1175 			if(ma)
1176 				*bdst.alpha = fs+MUL(fd, da, t);
1177 			else if(zero)
1178 				*bdst.alpha = 0;
1179 			bdst.alpha += bdst.delta;
1180 		}
1181 	}
1182 	return obdst;
1183 }
1184 
1185 static Buffer
1186 boolcalc1011(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int grey, int op)
1187 {
1188 	Buffer obdst;
1189 	int i, ma, zero;
1190 
1191 	obdst = bdst;
1192 	zero = !(op&1);
1193 
1194 	for(i=0; i<dx; i++){
1195 		ma = *bmask.alpha;
1196 
1197 		if(grey){
1198 			if(ma)
1199 				*bdst.grey = *bsrc.grey;
1200 			else if(zero)
1201 				*bdst.grey = 0;
1202 			bsrc.grey += bsrc.delta;
1203 			bdst.grey += bdst.delta;
1204 		}else{
1205 			if(ma){
1206 				*bdst.red = *bsrc.red;
1207 				*bdst.grn = *bsrc.grn;
1208 				*bdst.blu = *bsrc.blu;
1209 			}
1210 			else if(zero)
1211 				*bdst.red = *bdst.grn = *bdst.blu = 0;
1212 			bsrc.red += bsrc.delta;
1213 			bsrc.blu += bsrc.delta;
1214 			bsrc.grn += bsrc.delta;
1215 			bdst.red += bdst.delta;
1216 			bdst.blu += bdst.delta;
1217 			bdst.grn += bdst.delta;
1218 		}
1219 		bmask.alpha += bmask.delta;
1220 		if(bdst.alpha != &ones){
1221 			if(ma)
1222 				*bdst.alpha = 255;
1223 			else if(zero)
1224 				*bdst.alpha = 0;
1225 			bdst.alpha += bdst.delta;
1226 		}
1227 	}
1228 	return obdst;
1229 }
1230 /*
1231  * Replicated cached scan line read.  Call the function listed in the Param,
1232  * but cache the result so that for replicated images we only do the work once.
1233  */
1234 static Buffer
1235 replread(Param *p, uchar *s, int y)
1236 {
1237 	Buffer *b;
1238 
1239 	USED(s);
1240 	b = &p->bcache[y];
1241 	if((p->bfilled & (1<<y)) == 0){
1242 		p->bfilled |= 1<<y;
1243 		*b = p->replcall(p, p->bufbase+y*p->bufdelta, y);
1244 	}
1245 	return *b;
1246 }
1247 
1248 /*
1249  * Alpha reading function that simply relabels the grey pointer.
1250  */
1251 static Buffer
1252 greymaskread(Param *p, uchar *buf, int y)
1253 {
1254 	Buffer b;
1255 
1256 	b = p->greymaskcall(p, buf, y);
1257 	b.alpha = b.grey;
1258 	return b;
1259 }
1260 
1261 #define DBG if(0)
1262 static Buffer
1263 readnbit(Param *p, uchar *buf, int y)
1264 {
1265 	Buffer b;
1266 	Memimage *img;
1267 	uchar *repl, *r, *w, *ow, bits;
1268 	int i, n, sh, depth, x, dx, npack, nbits;
1269 
1270 	b.rgba = (u32int*)buf;
1271 	b.grey = w = buf;
1272 	b.red = b.blu = b.grn = w;
1273 	b.alpha = &ones;
1274 	b.delta = 1;
1275 
1276 	dx = p->dx;
1277 	img = p->img;
1278 	depth = img->depth;
1279 	repl = &replbit[depth][0];
1280 	npack = 8/depth;
1281 	sh = 8-depth;
1282 
1283 	/* copy from p->r.min.x until end of repl rectangle */
1284 	x = p->r.min.x;
1285 	n = dx;
1286 	if(n > p->img->r.max.x - x)
1287 		n = p->img->r.max.x - x;
1288 
1289 	r = p->bytermin + y*p->bwidth;
1290 DBG print("readnbit dx %d %p=%p+%d*%d, *r=%d fetch %d ", dx, r, p->bytermin, y, p->bwidth, *r, n);
1291 	bits = *r++;
1292 	nbits = 8;
1293 	if(i=x&(npack-1)){
1294 DBG print("throwaway %d...", i);
1295 		bits <<= depth*i;
1296 		nbits -= depth*i;
1297 	}
1298 	for(i=0; i<n; i++){
1299 		if(nbits == 0){
1300 DBG print("(%.2ux)...", *r);
1301 			bits = *r++;
1302 			nbits = 8;
1303 		}
1304 		*w++ = repl[bits>>sh];
1305 DBG print("bit %x...", repl[bits>>sh]);
1306 		bits <<= depth;
1307 		nbits -= depth;
1308 	}
1309 	dx -= n;
1310 	if(dx == 0)
1311 		return b;
1312 
1313 	assert(x+i == p->img->r.max.x);
1314 
1315 	/* copy from beginning of repl rectangle until where we were before. */
1316 	x = p->img->r.min.x;
1317 	n = dx;
1318 	if(n > p->r.min.x - x)
1319 		n = p->r.min.x - x;
1320 
1321 	r = p->bytey0s + y*p->bwidth;
1322 DBG print("x=%d r=%p...", x, r);
1323 	bits = *r++;
1324 	nbits = 8;
1325 	if(i=x&(npack-1)){
1326 		bits <<= depth*i;
1327 		nbits -= depth*i;
1328 	}
1329 DBG print("nbits=%d...", nbits);
1330 	for(i=0; i<n; i++){
1331 		if(nbits == 0){
1332 			bits = *r++;
1333 			nbits = 8;
1334 		}
1335 		*w++ = repl[bits>>sh];
1336 DBG print("bit %x...", repl[bits>>sh]);
1337 		bits <<= depth;
1338 		nbits -= depth;
1339 DBG print("bits %x nbits %d...", bits, nbits);
1340 	}
1341 	dx -= n;
1342 	if(dx == 0)
1343 		return b;
1344 
1345 	assert(dx > 0);
1346 	/* now we have exactly one full scan line: just replicate the buffer itself until we are done */
1347 	ow = buf;
1348 	while(dx--)
1349 		*w++ = *ow++;
1350 
1351 	return b;
1352 }
1353 #undef DBG
1354 
1355 #define DBG if(0)
1356 static void
1357 writenbit(Param *p, uchar *w, Buffer src)
1358 {
1359 	uchar *r;
1360 	u32int bits;
1361 	int i, sh, depth, npack, nbits, x, ex;
1362 
1363 	assert(src.grey != nil && src.delta == 1);
1364 
1365 	x = p->r.min.x;
1366 	ex = x+p->dx;
1367 	depth = p->img->depth;
1368 	npack = 8/depth;
1369 
1370 	i=x&(npack-1);
1371 	bits = i ? (*w >> (8-depth*i)) : 0;
1372 	nbits = depth*i;
1373 	sh = 8-depth;
1374 	r = src.grey;
1375 
1376 	for(; x<ex; x++){
1377 		bits <<= depth;
1378 DBG print(" %x", *r);
1379 		bits |= (*r++ >> sh);
1380 		nbits += depth;
1381 		if(nbits == 8){
1382 			*w++ = bits;
1383 			nbits = 0;
1384 		}
1385 	}
1386 
1387 	if(nbits){
1388 		sh = 8-nbits;
1389 		bits <<= sh;
1390 		bits |= *w & ((1<<sh)-1);
1391 		*w = bits;
1392 	}
1393 DBG print("\n");
1394 	return;
1395 }
1396 #undef DBG
1397 
1398 static Buffer
1399 readcmap(Param *p, uchar *buf, int y)
1400 {
1401 	Buffer b;
1402 	int a, convgrey, copyalpha, dx, i, m;
1403 	uchar *q, *cmap, *begin, *end, *r, *w;
1404 
1405 	begin = p->bytey0s + y*p->bwidth;
1406 	r = p->bytermin + y*p->bwidth;
1407 	end = p->bytey0e + y*p->bwidth;
1408 	cmap = p->img->cmap->cmap2rgb;
1409 	convgrey = p->convgrey;
1410 	copyalpha = (p->img->flags&Falpha) ? 1 : 0;
1411 
1412 	w = buf;
1413 	dx = p->dx;
1414 	if(copyalpha){
1415 		b.alpha = buf++;
1416 		a = p->img->shift[CAlpha]/8;
1417 		m = p->img->shift[CMap]/8;
1418 		for(i=0; i<dx; i++){
1419 			*w++ = r[a];
1420 			q = cmap+r[m]*3;
1421 			r += 2;
1422 			if(r == end)
1423 				r = begin;
1424 			if(convgrey){
1425 				*w++ = RGB2K(q[0], q[1], q[2]);
1426 			}else{
1427 				*w++ = q[2];	/* blue */
1428 				*w++ = q[1];	/* green */
1429 				*w++ = q[0];	/* red */
1430 			}
1431 		}
1432 	}else{
1433 		b.alpha = &ones;
1434 		for(i=0; i<dx; i++){
1435 			q = cmap+*r++*3;
1436 			if(r == end)
1437 				r = begin;
1438 			if(convgrey){
1439 				*w++ = RGB2K(q[0], q[1], q[2]);
1440 			}else{
1441 				*w++ = q[2];	/* blue */
1442 				*w++ = q[1];	/* green */
1443 				*w++ = q[0];	/* red */
1444 			}
1445 		}
1446 	}
1447 
1448 	b.rgba = (u32int*)(buf-copyalpha);
1449 
1450 	if(convgrey){
1451 		b.grey = buf;
1452 		b.red = b.blu = b.grn = buf;
1453 		b.delta = 1+copyalpha;
1454 	}else{
1455 		b.blu = buf;
1456 		b.grn = buf+1;
1457 		b.red = buf+2;
1458 		b.grey = nil;
1459 		b.delta = 3+copyalpha;
1460 	}
1461 	return b;
1462 }
1463 
1464 static void
1465 writecmap(Param *p, uchar *w, Buffer src)
1466 {
1467 	uchar *cmap, *red, *grn, *blu;
1468 	int i, dx, delta;
1469 
1470 	cmap = p->img->cmap->rgb2cmap;
1471 	
1472 	delta = src.delta;
1473 	red= src.red;
1474 	grn = src.grn;
1475 	blu = src.blu;
1476 
1477 	dx = p->dx;
1478 	for(i=0; i<dx; i++, red+=delta, grn+=delta, blu+=delta)
1479 		*w++ = cmap[(*red>>4)*256+(*grn>>4)*16+(*blu>>4)];
1480 }
1481 
1482 #define DBG if(drawdebug)
1483 static Buffer
1484 readbyte(Param *p, uchar *buf, int y)
1485 {
1486 	Buffer b;
1487 	Memimage *img;
1488 	int dx, isgrey, convgrey, alphaonly, copyalpha, i, nb;
1489 	uchar *begin, *end, *r, *w, *rrepl, *grepl, *brepl, *arepl, *krepl;
1490 	uchar ured, ugrn, ublu;
1491 	u32int u;
1492 
1493 	img = p->img;
1494 	begin = p->bytey0s + y*p->bwidth;
1495 	r = p->bytermin + y*p->bwidth;
1496 	end = p->bytey0e + y*p->bwidth;
1497 
1498 	w = buf;
1499 	dx = p->dx;
1500 	nb = img->depth/8;
1501 
1502 	convgrey = p->convgrey;	/* convert rgb to grey */
1503 	isgrey = img->flags&Fgrey;
1504 	alphaonly = p->alphaonly;
1505 	copyalpha = (img->flags&Falpha) ? 1 : 0;
1506 
1507 	/* if we can, avoid processing everything */
1508 	if(!(img->flags&Frepl) && !convgrey && (img->flags&Fbytes)){
1509 		memset(&b, 0, sizeof b);
1510 		if(p->needbuf){
1511 			memmove(buf, r, dx*nb);
1512 			r = buf;
1513 		}
1514 		b.rgba = (u32int*)r;
1515 		if(copyalpha)
1516 			b.alpha = r+img->shift[CAlpha]/8;
1517 		else
1518 			b.alpha = &ones;
1519 		if(isgrey){
1520 			b.grey = r+img->shift[CGrey]/8;
1521 			b.red = b.grn = b.blu = b.grey;
1522 		}else{
1523 			b.red = r+img->shift[CRed]/8;
1524 			b.grn = r+img->shift[CGreen]/8;
1525 			b.blu = r+img->shift[CBlue]/8;
1526 		}
1527 		b.delta = nb;
1528 		return b;
1529 	}
1530 
1531 	rrepl = replbit[img->nbits[CRed]];
1532 	grepl = replbit[img->nbits[CGreen]];
1533 	brepl = replbit[img->nbits[CBlue]];
1534 	arepl = replbit[img->nbits[CAlpha]];
1535 	krepl = replbit[img->nbits[CGrey]];
1536 
1537 	for(i=0; i<dx; i++){
1538 		u = r[0] | (r[1]<<8) | (r[2]<<16) | (r[3]<<24);
1539 		if(copyalpha)
1540 			*w++ = arepl[(u>>img->shift[CAlpha]) & img->mask[CAlpha]];
1541 
1542 		if(isgrey)
1543 			*w++ = krepl[(u >> img->shift[CGrey]) & img->mask[CGrey]];
1544 		else if(!alphaonly){
1545 			ured = rrepl[(u >> img->shift[CRed]) & img->mask[CRed]];
1546 			ugrn = grepl[(u >> img->shift[CGreen]) & img->mask[CGreen]];
1547 			ublu = brepl[(u >> img->shift[CBlue]) & img->mask[CBlue]];
1548 			if(convgrey){
1549 				*w++ = RGB2K(ured, ugrn, ublu);
1550 			}else{
1551 				*w++ = brepl[(u >> img->shift[CBlue]) & img->mask[CBlue]];
1552 				*w++ = grepl[(u >> img->shift[CGreen]) & img->mask[CGreen]];
1553 				*w++ = rrepl[(u >> img->shift[CRed]) & img->mask[CRed]];
1554 			}
1555 		}
1556 		r += nb;
1557 		if(r == end)
1558 			r = begin;
1559 	}
1560 	
1561 	b.alpha = copyalpha ? buf : &ones;
1562 	b.rgba = (u32int*)buf;
1563 	if(alphaonly){
1564 		b.red = b.grn = b.blu = b.grey = nil;
1565 		if(!copyalpha)
1566 			b.rgba = nil;
1567 		b.delta = 1;
1568 	}else if(isgrey || convgrey){
1569 		b.grey = buf+copyalpha;
1570 		b.red = b.grn = b.blu = buf+copyalpha;
1571 		b.delta = copyalpha+1;
1572 	}else{
1573 		b.blu = buf+copyalpha;
1574 		b.grn = buf+copyalpha+1;
1575 		b.grey = nil;
1576 		b.red = buf+copyalpha+2;
1577 		b.delta = copyalpha+3;
1578 	}
1579 	return b;
1580 }
1581 #undef DBG
1582 
1583 #define DBG if(drawdebug)
1584 static void
1585 writebyte(Param *p, uchar *w, Buffer src)
1586 {
1587 	Memimage *img;
1588 	int i, isalpha, isgrey, nb, delta, dx, adelta;
1589 	uchar ff, *red, *grn, *blu, *grey, *alpha;
1590 	u32int u, mask;
1591 
1592 	img = p->img;
1593 
1594 	red = src.red;
1595 	grn = src.grn;
1596 	blu = src.blu;
1597 	alpha = src.alpha;
1598 	delta = src.delta;
1599 	grey = src.grey;
1600 	dx = p->dx;
1601 
1602 	nb = img->depth/8;
1603 	mask = (nb==4) ? 0 : ~((1<<img->depth)-1);
1604 
1605 	isalpha = img->flags&Falpha;
1606 	isgrey = img->flags&Fgrey;
1607 	adelta = src.delta;
1608 
1609 	if(isalpha && (alpha == nil || alpha == &ones)){
1610 		ff = 0xFF;
1611 		alpha = &ff;
1612 		adelta = 0;
1613 	}
1614 
1615 	for(i=0; i<dx; i++){
1616 		u = w[0] | (w[1]<<8) | (w[2]<<16) | (w[3]<<24);
1617 DBG print("u %.8lux...", u);
1618 		u &= mask;
1619 DBG print("&mask %.8lux...", u);
1620 		if(isgrey){
1621 			u |= ((*grey >> (8-img->nbits[CGrey])) & img->mask[CGrey]) << img->shift[CGrey];
1622 DBG print("|grey %.8lux...", u);
1623 			grey += delta;
1624 		}else{
1625 			u |= ((*red >> (8-img->nbits[CRed])) & img->mask[CRed]) << img->shift[CRed];
1626 			u |= ((*grn >> (8-img->nbits[CGreen])) & img->mask[CGreen]) << img->shift[CGreen];
1627 			u |= ((*blu >> (8-img->nbits[CBlue])) & img->mask[CBlue]) << img->shift[CBlue];
1628 			red += delta;
1629 			grn += delta;
1630 			blu += delta;
1631 DBG print("|rgb %.8lux...", u);
1632 		}
1633 
1634 		if(isalpha){
1635 			u |= ((*alpha >> (8-img->nbits[CAlpha])) & img->mask[CAlpha]) << img->shift[CAlpha];
1636 			alpha += adelta;
1637 DBG print("|alpha %.8lux...", u);
1638 		}
1639 
1640 		w[0] = u;
1641 		w[1] = u>>8;
1642 		w[2] = u>>16;
1643 		w[3] = u>>24;
1644 DBG print("write back %.8lux...", u);
1645 		w += nb;
1646 	}
1647 }
1648 #undef DBG
1649 
1650 static Readfn*
1651 readfn(Memimage *img)
1652 {
1653 	if(img->depth < 8)
1654 		return readnbit;
1655 	if(img->nbits[CMap] == 8)
1656 		return readcmap;
1657 	return readbyte;
1658 }
1659 
1660 static Readfn*
1661 readalphafn(Memimage *m)
1662 {
1663 	USED(m);
1664 	return readbyte;
1665 }
1666 
1667 static Writefn*
1668 writefn(Memimage *img)
1669 {
1670 	if(img->depth < 8)
1671 		return writenbit;
1672 	if(img->chan == CMAP8)
1673 		return writecmap;
1674 	return writebyte;
1675 }
1676 
1677 static void
1678 nullwrite(Param *p, uchar *s, Buffer b)
1679 {
1680 	USED(p);
1681 	USED(s);
1682 }
1683 
1684 static Buffer
1685 readptr(Param *p, uchar *s, int y)
1686 {
1687 	Buffer b;
1688 	uchar *q;
1689 
1690 	USED(s);
1691 	q = p->bytermin + y*p->bwidth;
1692 	b.red = q;	/* ptr to data */
1693 	b.grn = b.blu = b.grey = b.alpha = nil;
1694 	b.rgba = (u32int*)q;
1695 	b.delta = p->img->depth/8;
1696 	return b;
1697 }
1698 
1699 static Buffer
1700 boolmemmove(Buffer bdst, Buffer bsrc, Buffer b1, int dx, int i, int o)
1701 {
1702 	USED(i);
1703 	USED(o);
1704 	memmove(bdst.red, bsrc.red, dx*bdst.delta);
1705 	return bdst;
1706 }
1707 
1708 static Buffer
1709 boolcopy8(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int i, int o)
1710 {
1711 	uchar *m, *r, *w, *ew;
1712 
1713 	USED(i);
1714 	USED(o);
1715 	m = bmask.grey;
1716 	w = bdst.red;
1717 	r = bsrc.red;
1718 	ew = w+dx;
1719 	for(; w < ew; w++,r++)
1720 		if(*m++)
1721 			*w = *r;
1722 	return bdst;	/* not used */
1723 }
1724 
1725 static Buffer
1726 boolcopy16(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int i, int o)
1727 {
1728 	uchar *m;
1729 	ushort *r, *w, *ew;
1730 
1731 	USED(i);
1732 	USED(o);
1733 	m = bmask.grey;
1734 	w = (ushort*)bdst.red;
1735 	r = (ushort*)bsrc.red;
1736 	ew = w+dx;
1737 	for(; w < ew; w++,r++)
1738 		if(*m++)
1739 			*w = *r;
1740 	return bdst;	/* not used */
1741 }
1742 
1743 static Buffer
1744 boolcopy24(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int i, int o)
1745 {
1746 	uchar *m;
1747 	uchar *r, *w, *ew;
1748 
1749 	USED(i);
1750 	USED(o);
1751 	m = bmask.grey;
1752 	w = bdst.red;
1753 	r = bsrc.red;
1754 	ew = w+dx*3;
1755 	while(w < ew){
1756 		if(*m++){
1757 			*w++ = *r++;
1758 			*w++ = *r++;
1759 			*w++ = *r++;
1760 		}else{
1761 			w += 3;
1762 			r += 3;
1763 		}
1764 	}
1765 	return bdst;	/* not used */
1766 }
1767 
1768 static Buffer
1769 boolcopy32(Buffer bdst, Buffer bsrc, Buffer bmask, int dx, int i, int o)
1770 {
1771 	uchar *m;
1772 	u32int *r, *w, *ew;
1773 
1774 	USED(i);
1775 	USED(o);
1776 	m = bmask.grey;
1777 	w = (u32int*)bdst.red;
1778 	r = (u32int*)bsrc.red;
1779 	ew = w+dx;
1780 	for(; w < ew; w++,r++)
1781 		if(*m++)
1782 			*w = *r;
1783 	return bdst;	/* not used */
1784 }
1785 
1786 static Buffer
1787 genconv(Param *p, uchar *buf, int y)
1788 {
1789 	Buffer b;
1790 	int nb;
1791 	uchar *r, *w, *ew;
1792 
1793 	/* read from source into RGB format in convbuf */
1794 	b = p->convreadcall(p, p->convbuf, y);
1795 
1796 	/* write RGB format into dst format in buf */
1797 	p->convwritecall(p->convdpar, buf, b);
1798 
1799 	if(p->convdx){
1800 		nb = p->convdpar->img->depth/8;
1801 		r = buf;
1802 		w = buf+nb*p->dx;
1803 		ew = buf+nb*p->convdx;
1804 		while(w<ew)
1805 			*w++ = *r++;
1806 	}
1807 
1808 	b.red = buf;
1809 	b.blu = b.grn = b.grey = b.alpha = nil;
1810 	b.rgba = (u32int*)buf;
1811 	b.delta = 0;
1812 	
1813 	return b;
1814 }
1815 
1816 static Readfn*
1817 convfn(Memimage *dst, Param *dpar, Memimage *src, Param *spar)
1818 {
1819 	if(dst->chan == src->chan && !(src->flags&Frepl)){
1820 /*if(drawdebug) iprint("readptr..."); */
1821 		return readptr;
1822 	}
1823 
1824 	if(dst->chan==CMAP8 && (src->chan==GREY1||src->chan==GREY2||src->chan==GREY4)){
1825 		/* cheat because we know the replicated value is exactly the color map entry. */
1826 /*if(drawdebug) iprint("Readnbit..."); */
1827 		return readnbit;
1828 	}
1829 
1830 	spar->convreadcall = readfn(src);
1831 	spar->convwritecall = writefn(dst);
1832 	spar->convdpar = dpar;
1833 
1834 	/* allocate a conversion buffer */
1835 	spar->convbufoff = ndrawbuf;
1836 	ndrawbuf += spar->dx*4;
1837 
1838 	if(spar->dx > Dx(spar->img->r)){
1839 		spar->convdx = spar->dx;
1840 		spar->dx = Dx(spar->img->r);
1841 	}
1842 
1843 /*if(drawdebug) iprint("genconv..."); */
1844 	return genconv;
1845 }
1846 
1847 /*
1848  * Do NOT call this directly.  pixelbits is a wrapper
1849  * around this that fetches the bits from the X server
1850  * when necessary.
1851  */
1852 u32int
1853 _pixelbits(Memimage *i, Point pt)
1854 {
1855 	uchar *p;
1856 	u32int val;
1857 	int off, bpp, npack;
1858 
1859 	val = 0;
1860 	p = byteaddr(i, pt);
1861 	switch(bpp=i->depth){
1862 	case 1:
1863 	case 2:
1864 	case 4:
1865 		npack = 8/bpp;
1866 		off = pt.x%npack;
1867 		val = p[0] >> bpp*(npack-1-off);
1868 		val &= (1<<bpp)-1;
1869 		break;
1870 	case 8:
1871 		val = p[0];
1872 		break;
1873 	case 16:
1874 		val = p[0]|(p[1]<<8);
1875 		break;
1876 	case 24:
1877 		val = p[0]|(p[1]<<8)|(p[2]<<16);
1878 		break;
1879 	case 32:
1880 		val = p[0]|(p[1]<<8)|(p[2]<<16)|(p[3]<<24);
1881 		break;
1882 	}
1883 	while(bpp<32){
1884 		val |= val<<bpp;
1885 		bpp *= 2;
1886 	}
1887 	return val;
1888 }
1889 
1890 static Calcfn*
1891 boolcopyfn(Memimage *img, Memimage *mask)
1892 {
1893 	if(mask->flags&Frepl && Dx(mask->r)==1 && Dy(mask->r)==1 && pixelbits(mask, mask->r.min)==~0)
1894 		return boolmemmove;
1895 
1896 	switch(img->depth){
1897 	case 8:
1898 		return boolcopy8;
1899 	case 16:
1900 		return boolcopy16;
1901 	case 24:
1902 		return boolcopy24;
1903 	case 32:
1904 		return boolcopy32;
1905 	default:
1906 		assert(0 /* boolcopyfn */);
1907 	}
1908 	return 0;
1909 }
1910 
1911 /*
1912  * Optimized draw for filling and scrolling; uses memset and memmove.
1913  */
1914 static void
1915 memsets(void *vp, ushort val, int n)
1916 {
1917 	ushort *p, *ep;
1918 
1919 	p = vp;
1920 	ep = p+n;
1921 	while(p<ep)
1922 		*p++ = val;
1923 }
1924 
1925 static void
1926 memsetl(void *vp, u32int val, int n)
1927 {
1928 	u32int *p, *ep;
1929 
1930 	p = vp;
1931 	ep = p+n;
1932 	while(p<ep)
1933 		*p++ = val;
1934 }
1935 
1936 static void
1937 memset24(void *vp, u32int val, int n)
1938 {
1939 	uchar *p, *ep;
1940 	uchar a,b,c;
1941 
1942 	p = vp;
1943 	ep = p+3*n;
1944 	a = val;
1945 	b = val>>8;
1946 	c = val>>16;
1947 	while(p<ep){
1948 		*p++ = a;
1949 		*p++ = b;
1950 		*p++ = c;
1951 	}
1952 }
1953 
1954 u32int
1955 _imgtorgba(Memimage *img, u32int val)
1956 {
1957 	uchar r, g, b, a;
1958 	int nb, ov, v;
1959 	u32int chan;
1960 	uchar *p;
1961 
1962 	a = 0xFF;
1963 	r = g = b = 0xAA;	/* garbage */
1964 	for(chan=img->chan; chan; chan>>=8){
1965 		nb = NBITS(chan);
1966 		ov = v = val&((1<<nb)-1);
1967 		val >>= nb;
1968 
1969 		while(nb < 8){
1970 			v |= v<<nb;
1971 			nb *= 2;
1972 		}
1973 		v >>= (nb-8);
1974 
1975 		switch(TYPE(chan)){
1976 		case CRed:
1977 			r = v;
1978 			break;
1979 		case CGreen:
1980 			g = v;
1981 			break;
1982 		case CBlue:
1983 			b = v;
1984 			break;
1985 		case CAlpha:
1986 			a = v;
1987 			break;
1988 		case CGrey:
1989 			r = g = b = v;
1990 			break;
1991 		case CMap:
1992 			p = img->cmap->cmap2rgb+3*ov;
1993 			r = *p++;
1994 			g = *p++;	
1995 			b = *p;
1996 			break;
1997 		}
1998 	}
1999 	return (r<<24)|(g<<16)|(b<<8)|a;	
2000 }
2001 
2002 u32int
2003 _rgbatoimg(Memimage *img, u32int rgba)
2004 {
2005 	u32int chan;
2006 	int d, nb;
2007 	u32int v;
2008 	uchar *p, r, g, b, a, m;
2009 
2010 	v = 0;
2011 	r = rgba>>24;
2012 	g = rgba>>16;
2013 	b = rgba>>8;
2014 	a = rgba;
2015 	d = 0;
2016 	for(chan=img->chan; chan; chan>>=8){
2017 		nb = NBITS(chan);
2018 		switch(TYPE(chan)){
2019 		case CRed:
2020 			v |= (r>>(8-nb))<<d;
2021 			break;
2022 		case CGreen:
2023 			v |= (g>>(8-nb))<<d;
2024 			break;
2025 		case CBlue:
2026 			v |= (b>>(8-nb))<<d;
2027 			break;
2028 		case CAlpha:
2029 			v |= (a>>(8-nb))<<d;
2030 			break;
2031 		case CMap:
2032 			p = img->cmap->rgb2cmap;
2033 			m = p[(r>>4)*256+(g>>4)*16+(b>>4)];
2034 			v |= (m>>(8-nb))<<d;
2035 			break;
2036 		case CGrey:
2037 			m = RGB2K(r,g,b);
2038 			v |= (m>>(8-nb))<<d;
2039 			break;
2040 		}
2041 		d += nb;
2042 	}
2043 /*	print("rgba2img %.8lux = %.*lux\n", rgba, 2*d/8, v); */
2044 	return v;
2045 }
2046 
2047 #define DBG if(0)
2048 static int
2049 memoptdraw(Memdrawparam *par)
2050 {
2051 	int m, y, dy, dx, op;
2052 	u32int v;
2053 	Memimage *src;
2054 	Memimage *dst;
2055 
2056 	dx = Dx(par->r);
2057 	dy = Dy(par->r);
2058 	src = par->src;
2059 	dst = par->dst;
2060 	op = par->op;
2061 
2062 DBG print("state %lux mval %lux dd %d\n", par->state, par->mval, dst->depth);
2063 	/*
2064 	 * If we have an opaque mask and source is one opaque pixel we can convert to the
2065 	 * destination format and just replicate with memset.
2066 	 */
2067 	m = Simplesrc|Simplemask|Fullmask;
2068 	if((par->state&m)==m && (par->srgba&0xFF) == 0xFF && (op ==S || op == SoverD)){
2069 		uchar *dp, p[4];
2070 		int d, dwid, ppb, np, nb;
2071 		uchar lm, rm;
2072 
2073 DBG print("memopt, dst %p, dst->data->bdata %p\n", dst, dst->data->bdata);
2074 		dwid = dst->width*sizeof(u32int);
2075 		dp = byteaddr(dst, par->r.min);
2076 		v = par->sdval;
2077 DBG print("sdval %lud, depth %d\n", v, dst->depth);
2078 		switch(dst->depth){
2079 		case 1:
2080 		case 2:
2081 		case 4:
2082 			for(d=dst->depth; d<8; d*=2)
2083 				v |= (v<<d);
2084 			ppb = 8/dst->depth;	/* pixels per byte */
2085 			m = ppb-1;
2086 			/* left edge */
2087 			np = par->r.min.x&m;		/* no. pixels unused on left side of word */
2088 			dx -= (ppb-np);
2089 			nb = 8 - np * dst->depth;		/* no. bits used on right side of word */
2090 			lm = (1<<nb)-1;
2091 DBG print("np %d x %d nb %d lm %ux ppb %d m %ux\n", np, par->r.min.x, nb, lm, ppb, m);	
2092 
2093 			/* right edge */
2094 			np = par->r.max.x&m;	/* no. pixels used on left side of word */
2095 			dx -= np;
2096 			nb = 8 - np * dst->depth;		/* no. bits unused on right side of word */
2097 			rm = ~((1<<nb)-1);
2098 DBG print("np %d x %d nb %d rm %ux ppb %d m %ux\n", np, par->r.max.x, nb, rm, ppb, m);	
2099 
2100 DBG print("dx %d Dx %d\n", dx, Dx(par->r));
2101 			/* lm, rm are masks that are 1 where we should touch the bits */
2102 			if(dx < 0){	/* just one byte */
2103 				lm &= rm;
2104 				for(y=0; y<dy; y++, dp+=dwid)
2105 					*dp ^= (v ^ *dp) & lm;
2106 			}else if(dx == 0){	/* no full bytes */
2107 				if(lm)
2108 					dwid--;
2109 
2110 				for(y=0; y<dy; y++, dp+=dwid){
2111 					if(lm){
2112 DBG print("dp %p v %lux lm %ux (v ^ *dp) & lm %lux\n", dp, v, lm, (v^*dp)&lm);
2113 						*dp ^= (v ^ *dp) & lm;
2114 						dp++;
2115 					}
2116 					*dp ^= (v ^ *dp) & rm;
2117 				}
2118 			}else{		/* full bytes in middle */
2119 				dx /= ppb;
2120 				if(lm)
2121 					dwid--;
2122 				dwid -= dx;
2123 
2124 				for(y=0; y<dy; y++, dp+=dwid){
2125 					if(lm){
2126 						*dp ^= (v ^ *dp) & lm;
2127 						dp++;
2128 					}
2129 					memset(dp, v, dx);
2130 					dp += dx;
2131 					*dp ^= (v ^ *dp) & rm;
2132 				}
2133 			}
2134 			return 1;
2135 		case 8:
2136 			for(y=0; y<dy; y++, dp+=dwid)
2137 				memset(dp, v, dx);
2138 			return 1;
2139 		case 16:
2140 			p[0] = v;		/* make little endian */
2141 			p[1] = v>>8;
2142 			v = *(ushort*)p;
2143 DBG print("dp=%p; dx=%d; for(y=0; y<%d; y++, dp+=%d)\nmemsets(dp, v, dx);\n",
2144 	dp, dx, dy, dwid);
2145 			for(y=0; y<dy; y++, dp+=dwid)
2146 				memsets(dp, v, dx);
2147 			return 1;
2148 		case 24:
2149 			for(y=0; y<dy; y++, dp+=dwid)
2150 				memset24(dp, v, dx);
2151 			return 1;
2152 		case 32:
2153 			p[0] = v;		/* make little endian */
2154 			p[1] = v>>8;
2155 			p[2] = v>>16;
2156 			p[3] = v>>24;
2157 			v = *(u32int*)p;
2158 			for(y=0; y<dy; y++, dp+=dwid)
2159 				memsetl(dp, v, dx);
2160 			return 1;
2161 		default:
2162 			assert(0 /* bad dest depth in memoptdraw */);
2163 		}
2164 	}
2165 
2166 	/*
2167 	 * If no source alpha, an opaque mask, we can just copy the
2168 	 * source onto the destination.  If the channels are the same and
2169 	 * the source is not replicated, memmove suffices.
2170 	 */
2171 	m = Simplemask|Fullmask;
2172 	if((par->state&(m|Replsrc))==m && src->depth >= 8 
2173 	&& src->chan == dst->chan && !(src->flags&Falpha) && (op == S || op == SoverD)){
2174 		uchar *sp, *dp;
2175 		long swid, dwid, nb;
2176 		int dir;
2177 
2178 		if(src->data == dst->data && byteaddr(dst, par->r.min) > byteaddr(src, par->sr.min))
2179 			dir = -1;
2180 		else
2181 			dir = 1;
2182 
2183 		swid = src->width*sizeof(u32int);
2184 		dwid = dst->width*sizeof(u32int);
2185 		sp = byteaddr(src, par->sr.min);
2186 		dp = byteaddr(dst, par->r.min);
2187 		if(dir == -1){
2188 			sp += (dy-1)*swid;
2189 			dp += (dy-1)*dwid;
2190 			swid = -swid;
2191 			dwid = -dwid;
2192 		}
2193 		nb = (dx*src->depth)/8;
2194 		for(y=0; y<dy; y++, sp+=swid, dp+=dwid)
2195 			memmove(dp, sp, nb);
2196 		return 1;
2197 	}
2198 
2199 	/*
2200 	 * If we have a 1-bit mask, 1-bit source, and 1-bit destination, and
2201 	 * they're all bit aligned, we can just use bit operators.  This happens
2202 	 * when we're manipulating boolean masks, e.g. in the arc code.
2203 	 */
2204 	if((par->state&(Simplemask|Simplesrc|Replmask|Replsrc))==0 
2205 	&& dst->chan==GREY1 && src->chan==GREY1 && par->mask->chan==GREY1 
2206 	&& (par->r.min.x&7)==(par->sr.min.x&7) && (par->r.min.x&7)==(par->mr.min.x&7)){
2207 		uchar *sp, *dp, *mp;
2208 		uchar lm, rm;
2209 		long swid, dwid, mwid;
2210 		int i, x, dir;
2211 
2212 		sp = byteaddr(src, par->sr.min);
2213 		dp = byteaddr(dst, par->r.min);
2214 		mp = byteaddr(par->mask, par->mr.min);
2215 		swid = src->width*sizeof(u32int);
2216 		dwid = dst->width*sizeof(u32int);
2217 		mwid = par->mask->width*sizeof(u32int);
2218 
2219 		if(src->data == dst->data && byteaddr(dst, par->r.min) > byteaddr(src, par->sr.min)){
2220 			dir = -1;
2221 		}else
2222 			dir = 1;
2223 
2224 		lm = 0xFF>>(par->r.min.x&7);
2225 		rm = 0xFF<<(8-(par->r.max.x&7));
2226 		dx -= (8-(par->r.min.x&7)) + (par->r.max.x&7);
2227 
2228 		if(dx < 0){	/* one byte wide */
2229 			lm &= rm;
2230 			if(dir == -1){
2231 				dp += dwid*(dy-1);
2232 				sp += swid*(dy-1);
2233 				mp += mwid*(dy-1);
2234 				dwid = -dwid;
2235 				swid = -swid;
2236 				mwid = -mwid;
2237 			}
2238 			for(y=0; y<dy; y++){
2239 				*dp ^= (*dp ^ *sp) & *mp & lm;
2240 				dp += dwid;
2241 				sp += swid;
2242 				mp += mwid;
2243 			}
2244 			return 1;
2245 		}
2246 
2247 		dx /= 8;
2248 		if(dir == 1){
2249 			i = (lm!=0)+dx+(rm!=0);
2250 			mwid -= i;
2251 			swid -= i;
2252 			dwid -= i;
2253 			for(y=0; y<dy; y++, dp+=dwid, sp+=swid, mp+=mwid){
2254 				if(lm){
2255 					*dp ^= (*dp ^ *sp++) & *mp++ & lm;
2256 					dp++;
2257 				}
2258 				for(x=0; x<dx; x++){
2259 					*dp ^= (*dp ^ *sp++) & *mp++;
2260 					dp++;
2261 				}
2262 				if(rm){
2263 					*dp ^= (*dp ^ *sp++) & *mp++ & rm;
2264 					dp++;
2265 				}
2266 			}
2267 			return 1;
2268 		}else{
2269 		/* dir == -1 */
2270 			i = (lm!=0)+dx+(rm!=0);
2271 			dp += dwid*(dy-1)+i-1;
2272 			sp += swid*(dy-1)+i-1;
2273 			mp += mwid*(dy-1)+i-1;
2274 			dwid = -dwid+i;
2275 			swid = -swid+i;
2276 			mwid = -mwid+i;
2277 			for(y=0; y<dy; y++, dp+=dwid, sp+=swid, mp+=mwid){
2278 				if(rm){
2279 					*dp ^= (*dp ^ *sp--) & *mp-- & rm;
2280 					dp--;
2281 				}
2282 				for(x=0; x<dx; x++){
2283 					*dp ^= (*dp ^ *sp--) & *mp--;
2284 					dp--;
2285 				}
2286 				if(lm){
2287 					*dp ^= (*dp ^ *sp--) & *mp-- & lm;
2288 					dp--;
2289 				}
2290 			}
2291 		}
2292 		return 1;
2293 	}
2294 	return 0;	
2295 }
2296 #undef DBG
2297 
2298 /*
2299  * Boolean character drawing.
2300  * Solid opaque color through a 1-bit greyscale mask.
2301  */
2302 #define DBG if(0)
2303 static int
2304 chardraw(Memdrawparam *par)
2305 {
2306 	u32int bits;
2307 	int i, ddepth, dy, dx, x, bx, ex, y, npack, bsh, depth, op;
2308 	u32int v, maskwid, dstwid;
2309 	uchar *wp, *rp, *q, *wc;
2310 	ushort *ws;
2311 	u32int *wl;
2312 	uchar sp[4];
2313 	Rectangle r, mr;
2314 	Memimage *mask, *src, *dst;
2315 
2316 if(0) if(drawdebug) iprint("chardraw? mf %lux md %d sf %lux dxs %d dys %d dd %d ddat %p sdat %p\n",
2317 		par->mask->flags, par->mask->depth, par->src->flags, 
2318 		Dx(par->src->r), Dy(par->src->r), par->dst->depth, par->dst->data, par->src->data);
2319 
2320 	mask = par->mask;
2321 	src = par->src;
2322 	dst = par->dst;
2323 	r = par->r;
2324 	mr = par->mr;
2325 	op = par->op;
2326 
2327 	if((par->state&(Replsrc|Simplesrc|Fullsrc|Replmask)) != (Replsrc|Simplesrc|Fullsrc)
2328 	|| mask->depth != 1 || dst->depth<8 || dst->data==src->data
2329 	|| op != SoverD)
2330 		return 0;
2331 
2332 /*if(drawdebug) iprint("chardraw..."); */
2333 
2334 	depth = mask->depth;
2335 	maskwid = mask->width*sizeof(u32int);
2336 	rp = byteaddr(mask, mr.min);
2337 	npack = 8/depth;
2338 	bsh = (mr.min.x % npack) * depth;
2339 
2340 	wp = byteaddr(dst, r.min);
2341 	dstwid = dst->width*sizeof(u32int);
2342 DBG print("bsh %d\n", bsh);
2343 	dy = Dy(r);
2344 	dx = Dx(r);
2345 
2346 	ddepth = dst->depth;
2347 
2348 	/*
2349 	 * for loop counts from bsh to bsh+dx
2350 	 *
2351 	 * we want the bottom bits to be the amount
2352 	 * to shift the pixels down, so for n≡0 (mod 8) we want 
2353 	 * bottom bits 7.  for n≡1, 6, etc.
2354 	 * the bits come from -n-1.
2355 	 */
2356 
2357 	bx = -bsh-1;
2358 	ex = -bsh-1-dx;
2359 	SET(bits);
2360 	v = par->sdval;
2361 
2362 	/* make little endian */
2363 	sp[0] = v;
2364 	sp[1] = v>>8;
2365 	sp[2] = v>>16;
2366 	sp[3] = v>>24;
2367 
2368 /*print("sp %x %x %x %x\n", sp[0], sp[1], sp[2], sp[3]); */
2369 	for(y=0; y<dy; y++, rp+=maskwid, wp+=dstwid){
2370 		q = rp;
2371 		if(bsh)
2372 			bits = *q++;
2373 		switch(ddepth){
2374 		case 8:
2375 /*if(drawdebug) iprint("8loop..."); */
2376 			wc = wp;
2377 			for(x=bx; x>ex; x--, wc++){
2378 				i = x&7;
2379 				if(i == 8-1)
2380 					bits = *q++;
2381 DBG print("bits %lux sh %d...", bits, i);
2382 				if((bits>>i)&1)
2383 					*wc = v;
2384 			}
2385 			break;
2386 		case 16:
2387 			ws = (ushort*)wp;
2388 			v = *(ushort*)sp;
2389 			for(x=bx; x>ex; x--, ws++){
2390 				i = x&7;
2391 				if(i == 8-1)
2392 					bits = *q++;
2393 DBG print("bits %lux sh %d...", bits, i);
2394 				if((bits>>i)&1)
2395 					*ws = v;
2396 			}
2397 			break;
2398 		case 24:
2399 			wc = wp;
2400 			for(x=bx; x>ex; x--, wc+=3){
2401 				i = x&7;
2402 				if(i == 8-1)
2403 					bits = *q++;
2404 DBG print("bits %lux sh %d...", bits, i);
2405 				if((bits>>i)&1){
2406 					wc[0] = sp[0];
2407 					wc[1] = sp[1];
2408 					wc[2] = sp[2];
2409 				}
2410 			}
2411 			break;
2412 		case 32:
2413 			wl = (u32int*)wp;
2414 			v = *(u32int*)sp;
2415 			for(x=bx; x>ex; x--, wl++){
2416 				i = x&7;
2417 				if(i == 8-1)
2418 					bits = *q++;
2419 DBG iprint("bits %lux sh %d...", bits, i);
2420 				if((bits>>i)&1)
2421 					*wl = v;
2422 			}
2423 			break;
2424 		}
2425 	}
2426 
2427 DBG print("\n");	
2428 	return 1;	
2429 }
2430 #undef DBG
2431 
2432 
2433 /*
2434  * Fill entire byte with replicated (if necessary) copy of source pixel,
2435  * assuming destination ldepth is >= source ldepth.
2436  *
2437  * This code is just plain wrong for >8bpp.
2438  *
2439 u32int
2440 membyteval(Memimage *src)
2441 {
2442 	int i, val, bpp;
2443 	uchar uc;
2444 
2445 	unloadmemimage(src, src->r, &uc, 1);
2446 	bpp = src->depth;
2447 	uc <<= (src->r.min.x&(7/src->depth))*src->depth;
2448 	uc &= ~(0xFF>>bpp);
2449 	* pixel value is now in high part of byte. repeat throughout byte 
2450 	val = uc;
2451 	for(i=bpp; i<8; i<<=1)
2452 		val |= val>>i;
2453 	return val;
2454 }
2455  * 
2456  */
2457 
2458 void
2459 _memfillcolor(Memimage *i, u32int val)
2460 {
2461 	u32int bits;
2462 	int d, y;
2463 	uchar p[4];
2464 
2465 	if(val == DNofill)
2466 		return;
2467 
2468 	bits = _rgbatoimg(i, val);
2469 	switch(i->depth){
2470 	case 24:	/* 24-bit images suck */
2471 		for(y=i->r.min.y; y<i->r.max.y; y++)
2472 			memset24(byteaddr(i, Pt(i->r.min.x, y)), bits, Dx(i->r));
2473 		break;
2474 	default:	/* 1, 2, 4, 8, 16, 32 */
2475 		for(d=i->depth; d<32; d*=2)
2476 			bits = (bits << d) | bits;
2477 		p[0] = bits;		/* make little endian */
2478 		p[1] = bits>>8;
2479 		p[2] = bits>>16;
2480 		p[3] = bits>>24;
2481 		bits = *(u32int*)p;
2482 		memsetl(wordaddr(i, i->r.min), bits, i->width*Dy(i->r));
2483 		break;
2484 	}
2485 }
2486