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