projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
softpipe: First attempts at multithreaded softpipe.
[mesa.git] / src / gallium / drivers / softpipe / sp_setup.c
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 /**
29 * \brief Primitive rasterization/rendering (points, lines, triangles)
30 *
31 * \author Keith Whitwell <keith@tungstengraphics.com>
32 * \author Brian Paul
33 */
34
35 #include "sp_setup.h"
36
37 #include "sp_context.h"
38 #include "sp_headers.h"
39 #include "sp_quad.h"
40 #include "sp_state.h"
41 #include "sp_prim_setup.h"
42 #include "draw/draw_context.h"
43 #include "draw/draw_private.h"
44 #include "draw/draw_vertex.h"
45 #include "pipe/p_shader_tokens.h"
46 #include "pipe/p_thread.h"
47 #include "util/u_math.h"
48 #include "util/u_memory.h"
49
50
51 #define DEBUG_VERTS 0
52 #define DEBUG_FRAGS 0
53
54 /**
55 * Triangle edge info
56 */
57 struct edge {
58 float dx; /**< X(v1) - X(v0), used only during setup */
59 float dy; /**< Y(v1) - Y(v0), used only during setup */
60 float dxdy; /**< dx/dy */
61 float sx, sy; /**< first sample point coord */
62 int lines; /**< number of lines on this edge */
63 };
64
65 #if SP_NUM_QUAD_THREADS > 1
66
67 struct thread_info
68 {
69 struct setup_context *setup;
70 uint id;
71 pipe_thread handle;
72 };
73
74 struct quad_job;
75
76 typedef void (* quad_job_routine)( struct setup_context *setup, uint thread, struct quad_job *job );
77
78 struct quad_job
79 {
80 int x, y;
81 unsigned mask;
82 struct quad_header quad;
83 quad_job_routine routine;
84 };
85
86 #define NUM_QUAD_JOBS 64
87
88 struct quad_job_que
89 {
90 struct quad_job jobs[NUM_QUAD_JOBS];
91 uint first;
92 uint last;
93 pipe_mutex mutex;
94 };
95
96 static void
97 add_quad_job( struct quad_job_que *que, int x, int y, unsigned mask, struct quad_header *quad, quad_job_routine routine )
98 {
99 while ((que->last + 1) % NUM_QUAD_JOBS == que->first)
100 usleep( 10 );
101 que->jobs[que->last].x = x;
102 que->jobs[que->last].y = y;
103 que->jobs[que->last].mask = mask;
104 que->jobs[que->last].quad = *quad;
105 que->jobs[que->last].routine = routine;
106 que->last = (que->last + 1) % NUM_QUAD_JOBS;
107 }
108
109 #endif
110
111 /**
112 * Triangle setup info (derived from draw_stage).
113 * Also used for line drawing (taking some liberties).
114 */
115 struct setup_context {
116 struct softpipe_context *softpipe;
117
118 /* Vertices are just an array of floats making up each attribute in
119 * turn. Currently fixed at 4 floats, but should change in time.
120 * Codegen will help cope with this.
121 */
122 const float (*vmax)[4];
123 const float (*vmid)[4];
124 const float (*vmin)[4];
125 const float (*vprovoke)[4];
126
127 struct edge ebot;
128 struct edge etop;
129 struct edge emaj;
130
131 float oneoverarea;
132
133 struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS];
134 struct tgsi_interp_coef posCoef; /* For Z, W */
135 struct quad_header quad;
136
137 #if SP_NUM_QUAD_THREADS > 1
138 struct quad_job_que que;
139 struct thread_info threads[SP_NUM_QUAD_THREADS];
140
141 #endif
142
143 struct {
144 int left[2]; /**< [0] = row0, [1] = row1 */
145 int right[2];
146 int y;
147 unsigned y_flags;
148 unsigned mask; /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */
149 } span;
150
151 #if DEBUG_FRAGS
152 uint numFragsEmitted; /**< per primitive */
153 uint numFragsWritten; /**< per primitive */
154 #endif
155
156 unsigned winding; /* which winding to cull */
157 };
158
159 #if SP_NUM_QUAD_THREADS > 1
160
161 static PIPE_THREAD_ROUTINE( quad_thread, param )
162 {
163 struct thread_info *info = (struct thread_info *) param;
164
165 for (;;) {
166 struct quad_job *job;
167
168 while (info->setup->que.last == info->setup->que.first)
169 usleep( 10 );
170 pipe_mutex_lock( info->setup->que.mutex );
171 job = &info->setup->que.jobs[info->setup->que.first];
172 info->setup->que.first = (info->setup->que.first + 1) % NUM_QUAD_JOBS;
173 job->routine( info->setup, info->id, job );
174 pipe_mutex_unlock( info->setup->que.mutex );
175 }
176 }
177
178 #define WAIT_FOR_COMPLETION(setup) \
179 do {\
180 while (setup->que.last != setup->que.first)\
181 usleep( 10 );\
182 } while (0)
183
184 #else
185
186 #define WAIT_FOR_COMPLETION(setup) ((void) 0)
187
188 #endif
189
190 /**
191 * Test if x is NaN or +/- infinity.
192 */
193 static INLINE boolean
194 is_inf_or_nan(float x)
195 {
196 union fi tmp;
197 tmp.f = x;
198 return !(int)((unsigned int)((tmp.i & 0x7fffffff)-0x7f800000) >> 31);
199 }
200
201
202 static boolean cull_tri( struct setup_context *setup,
203 float det )
204 {
205 if (det != 0)
206 {
207 /* if (det < 0 then Z points toward camera and triangle is
208 * counter-clockwise winding.
209 */
210 unsigned winding = (det < 0) ? PIPE_WINDING_CCW : PIPE_WINDING_CW;
211
212 if ((winding & setup->winding) == 0)
213 return FALSE;
214 }
215
216 /* Culled:
217 */
218 return TRUE;
219 }
220
221
222
223 /**
224 * Clip setup->quad against the scissor/surface bounds.
225 */
226 static INLINE void
227 quad_clip( struct setup_context *setup, struct quad_header *quad )
228 {
229 const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
230 const int minx = (int) cliprect->minx;
231 const int maxx = (int) cliprect->maxx;
232 const int miny = (int) cliprect->miny;
233 const int maxy = (int) cliprect->maxy;
234
235 if (quad->x0 >= maxx ||
236 quad->y0 >= maxy ||
237 quad->x0 + 1 < minx ||
238 quad->y0 + 1 < miny) {
239 /* totally clipped */
240 quad->mask = 0x0;
241 return;
242 }
243 if (quad->x0 < minx)
244 quad->mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
245 if (quad->y0 < miny)
246 quad->mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
247 if (quad->x0 == maxx - 1)
248 quad->mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
249 if (quad->y0 == maxy - 1)
250 quad->mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
251 }
252
253
254 /**
255 * Emit a quad (pass to next stage) with clipping.
256 */
257 static INLINE void
258 clip_emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread )
259 {
260 quad_clip( setup, quad );
261 if (quad->mask) {
262 struct softpipe_context *sp = setup->softpipe;
263
264 sp->quad[thread].first->run( sp->quad[thread].first, quad );
265 }
266 }
267
268 #if SP_NUM_QUAD_THREADS > 1
269
270 static void
271 clip_emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
272 {
273 clip_emit_quad( setup, &job->quad, thread );
274 }
275
276 #define CLIP_EMIT_QUAD(setup) add_quad_job( &setup->que, 0, 0, 0, &setup->quad, clip_emit_quad_job )
277
278 #else
279
280 #define CLIP_EMIT_QUAD(setup) clip_emit_quad( setup, &setup->quad, 0 )
281
282 #endif
283
284 /**
285 * Emit a quad (pass to next stage). No clipping is done.
286 */
287 static INLINE void
288 emit_quad( struct setup_context *setup, int x, int y, unsigned mask, struct quad_header *quad, uint thread )
289 {
290 struct softpipe_context *sp = setup->softpipe;
291
292 quad->x0 = x;
293 quad->y0 = y;
294 quad->mask = mask;
295 #if DEBUG_FRAGS
296 if (mask & 1) setup->numFragsEmitted++;
297 if (mask & 2) setup->numFragsEmitted++;
298 if (mask & 4) setup->numFragsEmitted++;
299 if (mask & 8) setup->numFragsEmitted++;
300 #endif
301 sp->quad[thread].first->run( sp->quad[thread].first, quad );
302 #if DEBUG_FRAGS
303 mask = quad->mask;
304 if (mask & 1) setup->numFragsWritten++;
305 if (mask & 2) setup->numFragsWritten++;
306 if (mask & 4) setup->numFragsWritten++;
307 if (mask & 8) setup->numFragsWritten++;
308 #endif
309 }
310
311 #if SP_NUM_QUAD_THREADS > 1
312
313 static void
314 emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
315 {
316 emit_quad( setup, job->x, job->y, job->mask, &job->quad, thread );
317 }
318
319 #define EMIT_QUAD(setup,x,y,mask) add_quad_job( &setup->que, x, y, mask, &setup->quad, emit_quad_job )
320
321 #else
322
323 #define EMIT_QUAD(setup,x,y,mask) emit_quad( setup, x, y, mask, &setup->quad, 0 )
324
325 #endif
326
327 /**
328 * Given an X or Y coordinate, return the block/quad coordinate that it
329 * belongs to.
330 */
331 static INLINE int block( int x )
332 {
333 return x & ~1;
334 }
335
336
337 /**
338 * Render a horizontal span of quads
339 */
340 static void flush_spans( struct setup_context *setup )
341 {
342 const int xleft0 = setup->span.left[0];
343 const int xleft1 = setup->span.left[1];
344 const int xright0 = setup->span.right[0];
345 const int xright1 = setup->span.right[1];
346 int minleft, maxright;
347 int x;
348
349 switch (setup->span.y_flags) {
350 case 0x3:
351 /* both odd and even lines written (both quad rows) */
352 minleft = block(MIN2(xleft0, xleft1));
353 maxright = block(MAX2(xright0, xright1));
354 for (x = minleft; x <= maxright; x += 2) {
355 /* determine which of the four pixels is inside the span bounds */
356 uint mask = 0x0;
357 if (x >= xleft0 && x < xright0)
358 mask |= MASK_TOP_LEFT;
359 if (x >= xleft1 && x < xright1)
360 mask |= MASK_BOTTOM_LEFT;
361 if (x+1 >= xleft0 && x+1 < xright0)
362 mask |= MASK_TOP_RIGHT;
363 if (x+1 >= xleft1 && x+1 < xright1)
364 mask |= MASK_BOTTOM_RIGHT;
365 EMIT_QUAD( setup, x, setup->span.y, mask );
366 }
367 break;
368
369 case 0x1:
370 /* only even line written (quad top row) */
371 minleft = block(xleft0);
372 maxright = block(xright0);
373 for (x = minleft; x <= maxright; x += 2) {
374 uint mask = 0x0;
375 if (x >= xleft0 && x < xright0)
376 mask |= MASK_TOP_LEFT;
377 if (x+1 >= xleft0 && x+1 < xright0)
378 mask |= MASK_TOP_RIGHT;
379 EMIT_QUAD( setup, x, setup->span.y, mask );
380 }
381 break;
382
383 case 0x2:
384 /* only odd line written (quad bottom row) */
385 minleft = block(xleft1);
386 maxright = block(xright1);
387 for (x = minleft; x <= maxright; x += 2) {
388 uint mask = 0x0;
389 if (x >= xleft1 && x < xright1)
390 mask |= MASK_BOTTOM_LEFT;
391 if (x+1 >= xleft1 && x+1 < xright1)
392 mask |= MASK_BOTTOM_RIGHT;
393 EMIT_QUAD( setup, x, setup->span.y, mask );
394 }
395 break;
396
397 default:
398 return;
399 }
400
401 setup->span.y = 0;
402 setup->span.y_flags = 0;
403 setup->span.right[0] = 0;
404 setup->span.right[1] = 0;
405 }
406
407
408 #if DEBUG_VERTS
409 static void print_vertex(const struct setup_context *setup,
410 const float (*v)[4])
411 {
412 int i;
413 debug_printf(" Vertex: (%p)\n", v);
414 for (i = 0; i < setup->quad.nr_attrs; i++) {
415 debug_printf(" %d: %f %f %f %f\n", i,
416 v[i][0], v[i][1], v[i][2], v[i][3]);
417 }
418 }
419 #endif
420
421 /**
422 * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise
423 */
424 static boolean setup_sort_vertices( struct setup_context *setup,
425 float det,
426 const float (*v0)[4],
427 const float (*v1)[4],
428 const float (*v2)[4] )
429 {
430 setup->vprovoke = v2;
431
432 /* determine bottom to top order of vertices */
433 {
434 float y0 = v0[0][1];
435 float y1 = v1[0][1];
436 float y2 = v2[0][1];
437 if (y0 <= y1) {
438 if (y1 <= y2) {
439 /* y0<=y1<=y2 */
440 setup->vmin = v0;
441 setup->vmid = v1;
442 setup->vmax = v2;
443 }
444 else if (y2 <= y0) {
445 /* y2<=y0<=y1 */
446 setup->vmin = v2;
447 setup->vmid = v0;
448 setup->vmax = v1;
449 }
450 else {
451 /* y0<=y2<=y1 */
452 setup->vmin = v0;
453 setup->vmid = v2;
454 setup->vmax = v1;
455 }
456 }
457 else {
458 if (y0 <= y2) {
459 /* y1<=y0<=y2 */
460 setup->vmin = v1;
461 setup->vmid = v0;
462 setup->vmax = v2;
463 }
464 else if (y2 <= y1) {
465 /* y2<=y1<=y0 */
466 setup->vmin = v2;
467 setup->vmid = v1;
468 setup->vmax = v0;
469 }
470 else {
471 /* y1<=y2<=y0 */
472 setup->vmin = v1;
473 setup->vmid = v2;
474 setup->vmax = v0;
475 }
476 }
477 }
478
479 setup->ebot.dx = setup->vmid[0][0] - setup->vmin[0][0];
480 setup->ebot.dy = setup->vmid[0][1] - setup->vmin[0][1];
481 setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0];
482 setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1];
483 setup->etop.dx = setup->vmax[0][0] - setup->vmid[0][0];
484 setup->etop.dy = setup->vmax[0][1] - setup->vmid[0][1];
485
486 /*
487 * Compute triangle's area. Use 1/area to compute partial
488 * derivatives of attributes later.
489 *
490 * The area will be the same as prim->det, but the sign may be
491 * different depending on how the vertices get sorted above.
492 *
493 * To determine whether the primitive is front or back facing we
494 * use the prim->det value because its sign is correct.
495 */
496 {
497 const float area = (setup->emaj.dx * setup->ebot.dy -
498 setup->ebot.dx * setup->emaj.dy);
499
500 setup->oneoverarea = 1.0f / area;
501
502 /*
503 debug_printf("%s one-over-area %f area %f det %f\n",
504 __FUNCTION__, setup->oneoverarea, area, det );
505 */
506 if (is_inf_or_nan(setup->oneoverarea))
507 return FALSE;
508 }
509
510 /* We need to know if this is a front or back-facing triangle for:
511 * - the GLSL gl_FrontFacing fragment attribute (bool)
512 * - two-sided stencil test
513 */
514 setup->quad.facing = (det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
515
516 return TRUE;
517 }
518
519
520 /**
521 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
522 * The value value comes from vertex[slot][i].
523 * The result will be put into setup->coef[slot].a0[i].
524 * \param slot which attribute slot
525 * \param i which component of the slot (0..3)
526 */
527 static void const_coeff( struct setup_context *setup,
528 struct tgsi_interp_coef *coef,
529 uint vertSlot, uint i)
530 {
531 assert(i <= 3);
532
533 coef->dadx[i] = 0;
534 coef->dady[i] = 0;
535
536 /* need provoking vertex info!
537 */
538 coef->a0[i] = setup->vprovoke[vertSlot][i];
539 }
540
541
542 /**
543 * Compute a0, dadx and dady for a linearly interpolated coefficient,
544 * for a triangle.
545 */
546 static void tri_linear_coeff( struct setup_context *setup,
547 struct tgsi_interp_coef *coef,
548 uint vertSlot, uint i)
549 {
550 float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i];
551 float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
552 float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
553 float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
554 float dadx = a * setup->oneoverarea;
555 float dady = b * setup->oneoverarea;
556
557 assert(i <= 3);
558
559 coef->dadx[i] = dadx;
560 coef->dady[i] = dady;
561
562 /* calculate a0 as the value which would be sampled for the
563 * fragment at (0,0), taking into account that we want to sample at
564 * pixel centers, in other words (0.5, 0.5).
565 *
566 * this is neat but unfortunately not a good way to do things for
567 * triangles with very large values of dadx or dady as it will
568 * result in the subtraction and re-addition from a0 of a very
569 * large number, which means we'll end up loosing a lot of the
570 * fractional bits and precision from a0. the way to fix this is
571 * to define a0 as the sample at a pixel center somewhere near vmin
572 * instead - i'll switch to this later.
573 */
574 coef->a0[i] = (setup->vmin[vertSlot][i] -
575 (dadx * (setup->vmin[0][0] - 0.5f) +
576 dady * (setup->vmin[0][1] - 0.5f)));
577
578 /*
579 debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
580 slot, "xyzw"[i],
581 setup->coef[slot].a0[i],
582 setup->coef[slot].dadx[i],
583 setup->coef[slot].dady[i]);
584 */
585 }
586
587
588 /**
589 * Compute a0, dadx and dady for a perspective-corrected interpolant,
590 * for a triangle.
591 * We basically multiply the vertex value by 1/w before computing
592 * the plane coefficients (a0, dadx, dady).
593 * Later, when we compute the value at a particular fragment position we'll
594 * divide the interpolated value by the interpolated W at that fragment.
595 */
596 static void tri_persp_coeff( struct setup_context *setup,
597 struct tgsi_interp_coef *coef,
598 uint vertSlot, uint i)
599 {
600 /* premultiply by 1/w (v[0][3] is always W):
601 */
602 float mina = setup->vmin[vertSlot][i] * setup->vmin[0][3];
603 float mida = setup->vmid[vertSlot][i] * setup->vmid[0][3];
604 float maxa = setup->vmax[vertSlot][i] * setup->vmax[0][3];
605 float botda = mida - mina;
606 float majda = maxa - mina;
607 float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
608 float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
609 float dadx = a * setup->oneoverarea;
610 float dady = b * setup->oneoverarea;
611
612 /*
613 debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i,
614 setup->vmin[vertSlot][i],
615 setup->vmid[vertSlot][i],
616 setup->vmax[vertSlot][i]
617 );
618 */
619 assert(i <= 3);
620
621 coef->dadx[i] = dadx;
622 coef->dady[i] = dady;
623 coef->a0[i] = (mina -
624 (dadx * (setup->vmin[0][0] - 0.5f) +
625 dady * (setup->vmin[0][1] - 0.5f)));
626 }
627
628
629 /**
630 * Special coefficient setup for gl_FragCoord.
631 * X and Y are trivial, though Y has to be inverted for OpenGL.
632 * Z and W are copied from posCoef which should have already been computed.
633 * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
634 */
635 static void
636 setup_fragcoord_coeff(struct setup_context *setup, uint slot)
637 {
638 /*X*/
639 setup->coef[slot].a0[0] = 0;
640 setup->coef[slot].dadx[0] = 1.0;
641 setup->coef[slot].dady[0] = 0.0;
642 /*Y*/
643 if (setup->softpipe->rasterizer->origin_lower_left) {
644 /* y=0=bottom */
645 const int winHeight = setup->softpipe->framebuffer.height;
646 setup->coef[slot].a0[1] = (float) (winHeight - 1);
647 setup->coef[slot].dady[1] = -1.0;
648 }
649 else {
650 /* y=0=top */
651 setup->coef[slot].a0[1] = 0.0;
652 setup->coef[slot].dady[1] = 1.0;
653 }
654 setup->coef[slot].dadx[1] = 0.0;
655 /*Z*/
656 setup->coef[slot].a0[2] = setup->posCoef.a0[2];
657 setup->coef[slot].dadx[2] = setup->posCoef.dadx[2];
658 setup->coef[slot].dady[2] = setup->posCoef.dady[2];
659 /*W*/
660 setup->coef[slot].a0[3] = setup->posCoef.a0[3];
661 setup->coef[slot].dadx[3] = setup->posCoef.dadx[3];
662 setup->coef[slot].dady[3] = setup->posCoef.dady[3];
663 }
664
665
666
667 /**
668 * Compute the setup->coef[] array dadx, dady, a0 values.
669 * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized.
670 */
671 static void setup_tri_coefficients( struct setup_context *setup )
672 {
673 struct softpipe_context *softpipe = setup->softpipe;
674 const struct sp_fragment_shader *spfs = softpipe->fs;
675 const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
676 uint fragSlot;
677
678 /* z and w are done by linear interpolation:
679 */
680 tri_linear_coeff(setup, &setup->posCoef, 0, 2);
681 tri_linear_coeff(setup, &setup->posCoef, 0, 3);
682
683 /* setup interpolation for all the remaining attributes:
684 */
685 for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
686 const uint vertSlot = vinfo->src_index[fragSlot];
687 uint j;
688
689 switch (vinfo->interp_mode[fragSlot]) {
690 case INTERP_CONSTANT:
691 for (j = 0; j < NUM_CHANNELS; j++)
692 const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
693 break;
694 case INTERP_LINEAR:
695 for (j = 0; j < NUM_CHANNELS; j++)
696 tri_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
697 break;
698 case INTERP_PERSPECTIVE:
699 for (j = 0; j < NUM_CHANNELS; j++)
700 tri_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
701 break;
702 case INTERP_POS:
703 setup_fragcoord_coeff(setup, fragSlot);
704 break;
705 default:
706 assert(0);
707 }
708
709 if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
710 /* FOG.y = front/back facing XXX fix this */
711 setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
712 setup->coef[fragSlot].dadx[1] = 0.0;
713 setup->coef[fragSlot].dady[1] = 0.0;
714 }
715 }
716 }
717
718
719
720 static void setup_tri_edges( struct setup_context *setup )
721 {
722 float vmin_x = setup->vmin[0][0] + 0.5f;
723 float vmid_x = setup->vmid[0][0] + 0.5f;
724
725 float vmin_y = setup->vmin[0][1] - 0.5f;
726 float vmid_y = setup->vmid[0][1] - 0.5f;
727 float vmax_y = setup->vmax[0][1] - 0.5f;
728
729 setup->emaj.sy = ceilf(vmin_y);
730 setup->emaj.lines = (int) ceilf(vmax_y - setup->emaj.sy);
731 setup->emaj.dxdy = setup->emaj.dx / setup->emaj.dy;
732 setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy;
733
734 setup->etop.sy = ceilf(vmid_y);
735 setup->etop.lines = (int) ceilf(vmax_y - setup->etop.sy);
736 setup->etop.dxdy = setup->etop.dx / setup->etop.dy;
737 setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy;
738
739 setup->ebot.sy = ceilf(vmin_y);
740 setup->ebot.lines = (int) ceilf(vmid_y - setup->ebot.sy);
741 setup->ebot.dxdy = setup->ebot.dx / setup->ebot.dy;
742 setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy;
743 }
744
745
746 /**
747 * Render the upper or lower half of a triangle.
748 * Scissoring/cliprect is applied here too.
749 */
750 static void subtriangle( struct setup_context *setup,
751 struct edge *eleft,
752 struct edge *eright,
753 unsigned lines )
754 {
755 const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
756 const int minx = (int) cliprect->minx;
757 const int maxx = (int) cliprect->maxx;
758 const int miny = (int) cliprect->miny;
759 const int maxy = (int) cliprect->maxy;
760 int y, start_y, finish_y;
761 int sy = (int)eleft->sy;
762
763 assert((int)eleft->sy == (int) eright->sy);
764
765 /* clip top/bottom */
766 start_y = sy;
767 finish_y = sy + lines;
768
769 if (start_y < miny)
770 start_y = miny;
771
772 if (finish_y > maxy)
773 finish_y = maxy;
774
775 start_y -= sy;
776 finish_y -= sy;
777
778 /*
779 debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);
780 */
781
782 for (y = start_y; y < finish_y; y++) {
783
784 /* avoid accumulating adds as floats don't have the precision to
785 * accurately iterate large triangle edges that way. luckily we
786 * can just multiply these days.
787 *
788 * this is all drowned out by the attribute interpolation anyway.
789 */
790 int left = (int)(eleft->sx + y * eleft->dxdy);
791 int right = (int)(eright->sx + y * eright->dxdy);
792
793 /* clip left/right */
794 if (left < minx)
795 left = minx;
796 if (right > maxx)
797 right = maxx;
798
799 if (left < right) {
800 int _y = sy + y;
801 if (block(_y) != setup->span.y) {
802 flush_spans(setup);
803 setup->span.y = block(_y);
804 }
805
806 setup->span.left[_y&1] = left;
807 setup->span.right[_y&1] = right;
808 setup->span.y_flags |= 1<<(_y&1);
809 }
810 }
811
812
813 /* save the values so that emaj can be restarted:
814 */
815 eleft->sx += lines * eleft->dxdy;
816 eright->sx += lines * eright->dxdy;
817 eleft->sy += lines;
818 eright->sy += lines;
819 }
820
821
822 /**
823 * Recalculate prim's determinant. This is needed as we don't have
824 * get this information through the vbuf_render interface & we must
825 * calculate it here.
826 */
827 static float
828 calc_det( const float (*v0)[4],
829 const float (*v1)[4],
830 const float (*v2)[4] )
831 {
832 /* edge vectors e = v0 - v2, f = v1 - v2 */
833 const float ex = v0[0][0] - v2[0][0];
834 const float ey = v0[0][1] - v2[0][1];
835 const float fx = v1[0][0] - v2[0][0];
836 const float fy = v1[0][1] - v2[0][1];
837
838 /* det = cross(e,f).z */
839 return ex * fy - ey * fx;
840 }
841
842
843 /**
844 * Do setup for triangle rasterization, then render the triangle.
845 */
846 void setup_tri( struct setup_context *setup,
847 const float (*v0)[4],
848 const float (*v1)[4],
849 const float (*v2)[4] )
850 {
851 float det;
852
853 #if DEBUG_VERTS
854 debug_printf("Setup triangle:\n");
855 print_vertex(setup, v0);
856 print_vertex(setup, v1);
857 print_vertex(setup, v2);
858 #endif
859
860 if (setup->softpipe->no_rast)
861 return;
862
863 det = calc_det(v0, v1, v2);
864 /*
865 debug_printf("%s\n", __FUNCTION__ );
866 */
867
868 #if DEBUG_FRAGS
869 setup->numFragsEmitted = 0;
870 setup->numFragsWritten = 0;
871 #endif
872
873 if (cull_tri( setup, det ))
874 return;
875
876 if (!setup_sort_vertices( setup, det, v0, v1, v2 ))
877 return;
878 setup_tri_coefficients( setup );
879 setup_tri_edges( setup );
880
881 setup->quad.prim = PRIM_TRI;
882
883 setup->span.y = 0;
884 setup->span.y_flags = 0;
885 setup->span.right[0] = 0;
886 setup->span.right[1] = 0;
887 /* setup->span.z_mode = tri_z_mode( setup->ctx ); */
888
889 /* init_constant_attribs( setup ); */
890
891 if (setup->oneoverarea < 0.0) {
892 /* emaj on left:
893 */
894 subtriangle( setup, &setup->emaj, &setup->ebot, setup->ebot.lines );
895 subtriangle( setup, &setup->emaj, &setup->etop, setup->etop.lines );
896 }
897 else {
898 /* emaj on right:
899 */
900 subtriangle( setup, &setup->ebot, &setup->emaj, setup->ebot.lines );
901 subtriangle( setup, &setup->etop, &setup->emaj, setup->etop.lines );
902 }
903
904 flush_spans( setup );
905
906 WAIT_FOR_COMPLETION(setup);
907
908 #if DEBUG_FRAGS
909 printf("Tri: %u frags emitted, %u written\n",
910 setup->numFragsEmitted,
911 setup->numFragsWritten);
912 #endif
913 }
914
915
916
917 /**
918 * Compute a0, dadx and dady for a linearly interpolated coefficient,
919 * for a line.
920 */
921 static void
922 line_linear_coeff(struct setup_context *setup,
923 struct tgsi_interp_coef *coef,
924 uint vertSlot, uint i)
925 {
926 const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
927 const float dadx = da * setup->emaj.dx * setup->oneoverarea;
928 const float dady = da * setup->emaj.dy * setup->oneoverarea;
929 coef->dadx[i] = dadx;
930 coef->dady[i] = dady;
931 coef->a0[i] = (setup->vmin[vertSlot][i] -
932 (dadx * (setup->vmin[0][0] - 0.5f) +
933 dady * (setup->vmin[0][1] - 0.5f)));
934 }
935
936
937 /**
938 * Compute a0, dadx and dady for a perspective-corrected interpolant,
939 * for a line.
940 */
941 static void
942 line_persp_coeff(struct setup_context *setup,
943 struct tgsi_interp_coef *coef,
944 uint vertSlot, uint i)
945 {
946 /* XXX double-check/verify this arithmetic */
947 const float a0 = setup->vmin[vertSlot][i] * setup->vmin[0][3];
948 const float a1 = setup->vmax[vertSlot][i] * setup->vmax[0][3];
949 const float da = a1 - a0;
950 const float dadx = da * setup->emaj.dx * setup->oneoverarea;
951 const float dady = da * setup->emaj.dy * setup->oneoverarea;
952 coef->dadx[i] = dadx;
953 coef->dady[i] = dady;
954 coef->a0[i] = (setup->vmin[vertSlot][i] -
955 (dadx * (setup->vmin[0][0] - 0.5f) +
956 dady * (setup->vmin[0][1] - 0.5f)));
957 }
958
959
960 /**
961 * Compute the setup->coef[] array dadx, dady, a0 values.
962 * Must be called after setup->vmin,vmax are initialized.
963 */
964 static INLINE boolean
965 setup_line_coefficients(struct setup_context *setup,
966 const float (*v0)[4],
967 const float (*v1)[4])
968 {
969 struct softpipe_context *softpipe = setup->softpipe;
970 const struct sp_fragment_shader *spfs = softpipe->fs;
971 const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
972 uint fragSlot;
973 float area;
974
975 /* use setup->vmin, vmax to point to vertices */
976 setup->vprovoke = v1;
977 setup->vmin = v0;
978 setup->vmax = v1;
979
980 setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0];
981 setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1];
982
983 /* NOTE: this is not really area but something proportional to it */
984 area = setup->emaj.dx * setup->emaj.dx + setup->emaj.dy * setup->emaj.dy;
985 if (area == 0.0f || is_inf_or_nan(area))
986 return FALSE;
987 setup->oneoverarea = 1.0f / area;
988
989 /* z and w are done by linear interpolation:
990 */
991 line_linear_coeff(setup, &setup->posCoef, 0, 2);
992 line_linear_coeff(setup, &setup->posCoef, 0, 3);
993
994 /* setup interpolation for all the remaining attributes:
995 */
996 for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
997 const uint vertSlot = vinfo->src_index[fragSlot];
998 uint j;
999
1000 switch (vinfo->interp_mode[fragSlot]) {
1001 case INTERP_CONSTANT:
1002 for (j = 0; j < NUM_CHANNELS; j++)
1003 const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
1004 break;
1005 case INTERP_LINEAR:
1006 for (j = 0; j < NUM_CHANNELS; j++)
1007 line_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
1008 break;
1009 case INTERP_PERSPECTIVE:
1010 for (j = 0; j < NUM_CHANNELS; j++)
1011 line_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
1012 break;
1013 case INTERP_POS:
1014 setup_fragcoord_coeff(setup, fragSlot);
1015 break;
1016 default:
1017 assert(0);
1018 }
1019
1020 if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
1021 /* FOG.y = front/back facing XXX fix this */
1022 setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
1023 setup->coef[fragSlot].dadx[1] = 0.0;
1024 setup->coef[fragSlot].dady[1] = 0.0;
1025 }
1026 }
1027 return TRUE;
1028 }
1029
1030
1031 /**
1032 * Plot a pixel in a line segment.
1033 */
1034 static INLINE void
1035 plot(struct setup_context *setup, int x, int y)
1036 {
1037 const int iy = y & 1;
1038 const int ix = x & 1;
1039 const int quadX = x - ix;
1040 const int quadY = y - iy;
1041 const int mask = (1 << ix) << (2 * iy);
1042
1043 if (quadX != setup->quad.x0 ||
1044 quadY != setup->quad.y0)
1045 {
1046 /* flush prev quad, start new quad */
1047
1048 if (setup->quad.x0 != -1)
1049 CLIP_EMIT_QUAD(setup);
1050
1051 setup->quad.x0 = quadX;
1052 setup->quad.y0 = quadY;
1053 setup->quad.mask = 0x0;
1054 }
1055
1056 setup->quad.mask |= mask;
1057 }
1058
1059
1060 /**
1061 * Do setup for line rasterization, then render the line.
1062 * Single-pixel width, no stipple, etc. We rely on the 'draw' module
1063 * to handle stippling and wide lines.
1064 */
1065 void
1066 setup_line(struct setup_context *setup,
1067 const float (*v0)[4],
1068 const float (*v1)[4])
1069 {
1070 int x0 = (int) v0[0][0];
1071 int x1 = (int) v1[0][0];
1072 int y0 = (int) v0[0][1];
1073 int y1 = (int) v1[0][1];
1074 int dx = x1 - x0;
1075 int dy = y1 - y0;
1076 int xstep, ystep;
1077
1078 #if DEBUG_VERTS
1079 debug_printf("Setup line:\n");
1080 print_vertex(setup, v0);
1081 print_vertex(setup, v1);
1082 #endif
1083
1084 if (setup->softpipe->no_rast)
1085 return;
1086
1087 if (dx == 0 && dy == 0)
1088 return;
1089
1090 if (!setup_line_coefficients(setup, v0, v1))
1091 return;
1092
1093 assert(v0[0][0] < 1.0e9);
1094 assert(v0[0][1] < 1.0e9);
1095 assert(v1[0][0] < 1.0e9);
1096 assert(v1[0][1] < 1.0e9);
1097
1098 if (dx < 0) {
1099 dx = -dx; /* make positive */
1100 xstep = -1;
1101 }
1102 else {
1103 xstep = 1;
1104 }
1105
1106 if (dy < 0) {
1107 dy = -dy; /* make positive */
1108 ystep = -1;
1109 }
1110 else {
1111 ystep = 1;
1112 }
1113
1114 assert(dx >= 0);
1115 assert(dy >= 0);
1116
1117 setup->quad.x0 = setup->quad.y0 = -1;
1118 setup->quad.mask = 0x0;
1119 setup->quad.prim = PRIM_LINE;
1120 /* XXX temporary: set coverage to 1.0 so the line appears
1121 * if AA mode happens to be enabled.
1122 */
1123 setup->quad.coverage[0] =
1124 setup->quad.coverage[1] =
1125 setup->quad.coverage[2] =
1126 setup->quad.coverage[3] = 1.0;
1127
1128 if (dx > dy) {
1129 /*** X-major line ***/
1130 int i;
1131 const int errorInc = dy + dy;
1132 int error = errorInc - dx;
1133 const int errorDec = error - dx;
1134
1135 for (i = 0; i < dx; i++) {
1136 plot(setup, x0, y0);
1137
1138 x0 += xstep;
1139 if (error < 0) {
1140 error += errorInc;
1141 }
1142 else {
1143 error += errorDec;
1144 y0 += ystep;
1145 }
1146 }
1147 }
1148 else {
1149 /*** Y-major line ***/
1150 int i;
1151 const int errorInc = dx + dx;
1152 int error = errorInc - dy;
1153 const int errorDec = error - dy;
1154
1155 for (i = 0; i < dy; i++) {
1156 plot(setup, x0, y0);
1157
1158 y0 += ystep;
1159 if (error < 0) {
1160 error += errorInc;
1161 }
1162 else {
1163 error += errorDec;
1164 x0 += xstep;
1165 }
1166 }
1167 }
1168
1169 /* draw final quad */
1170 if (setup->quad.mask) {
1171 CLIP_EMIT_QUAD(setup);
1172 }
1173
1174 WAIT_FOR_COMPLETION(setup);
1175 }
1176
1177
1178 static void
1179 point_persp_coeff(struct setup_context *setup,
1180 const float (*vert)[4],
1181 struct tgsi_interp_coef *coef,
1182 uint vertSlot, uint i)
1183 {
1184 assert(i <= 3);
1185 coef->dadx[i] = 0.0F;
1186 coef->dady[i] = 0.0F;
1187 coef->a0[i] = vert[vertSlot][i] * vert[0][3];
1188 }
1189
1190
1191 /**
1192 * Do setup for point rasterization, then render the point.
1193 * Round or square points...
1194 * XXX could optimize a lot for 1-pixel points.
1195 */
1196 void
1197 setup_point( struct setup_context *setup,
1198 const float (*v0)[4] )
1199 {
1200 struct softpipe_context *softpipe = setup->softpipe;
1201 const struct sp_fragment_shader *spfs = softpipe->fs;
1202 const int sizeAttr = setup->softpipe->psize_slot;
1203 const float size
1204 = sizeAttr > 0 ? v0[sizeAttr][0]
1205 : setup->softpipe->rasterizer->point_size;
1206 const float halfSize = 0.5F * size;
1207 const boolean round = (boolean) setup->softpipe->rasterizer->point_smooth;
1208 const float x = v0[0][0]; /* Note: data[0] is always position */
1209 const float y = v0[0][1];
1210 const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
1211 uint fragSlot;
1212
1213 #if DEBUG_VERTS
1214 debug_printf("Setup point:\n");
1215 print_vertex(setup, v0);
1216 #endif
1217
1218 if (softpipe->no_rast)
1219 return;
1220
1221 /* For points, all interpolants are constant-valued.
1222 * However, for point sprites, we'll need to setup texcoords appropriately.
1223 * XXX: which coefficients are the texcoords???
1224 * We may do point sprites as textured quads...
1225 *
1226 * KW: We don't know which coefficients are texcoords - ultimately
1227 * the choice of what interpolation mode to use for each attribute
1228 * should be determined by the fragment program, using
1229 * per-attribute declaration statements that include interpolation
1230 * mode as a parameter. So either the fragment program will have
1231 * to be adjusted for pointsprite vs normal point behaviour, or
1232 * otherwise a special interpolation mode will have to be defined
1233 * which matches the required behaviour for point sprites. But -
1234 * the latter is not a feature of normal hardware, and as such
1235 * probably should be ruled out on that basis.
1236 */
1237 setup->vprovoke = v0;
1238
1239 /* setup Z, W */
1240 const_coeff(setup, &setup->posCoef, 0, 2);
1241 const_coeff(setup, &setup->posCoef, 0, 3);
1242
1243 for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
1244 const uint vertSlot = vinfo->src_index[fragSlot];
1245 uint j;
1246
1247 switch (vinfo->interp_mode[fragSlot]) {
1248 case INTERP_CONSTANT:
1249 /* fall-through */
1250 case INTERP_LINEAR:
1251 for (j = 0; j < NUM_CHANNELS; j++)
1252 const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
1253 break;
1254 case INTERP_PERSPECTIVE:
1255 for (j = 0; j < NUM_CHANNELS; j++)
1256 point_persp_coeff(setup, setup->vprovoke,
1257 &setup->coef[fragSlot], vertSlot, j);
1258 break;
1259 case INTERP_POS:
1260 setup_fragcoord_coeff(setup, fragSlot);
1261 break;
1262 default:
1263 assert(0);
1264 }
1265
1266 if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
1267 /* FOG.y = front/back facing XXX fix this */
1268 setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
1269 setup->coef[fragSlot].dadx[1] = 0.0;
1270 setup->coef[fragSlot].dady[1] = 0.0;
1271 }
1272 }
1273
1274 setup->quad.prim = PRIM_POINT;
1275
1276 if (halfSize <= 0.5 && !round) {
1277 /* special case for 1-pixel points */
1278 const int ix = ((int) x) & 1;
1279 const int iy = ((int) y) & 1;
1280 setup->quad.x0 = (int) x - ix;
1281 setup->quad.y0 = (int) y - iy;
1282 setup->quad.mask = (1 << ix) << (2 * iy);
1283 CLIP_EMIT_QUAD(setup);
1284 }
1285 else {
1286 if (round) {
1287 /* rounded points */
1288 const int ixmin = block((int) (x - halfSize));
1289 const int ixmax = block((int) (x + halfSize));
1290 const int iymin = block((int) (y - halfSize));
1291 const int iymax = block((int) (y + halfSize));
1292 const float rmin = halfSize - 0.7071F; /* 0.7071 = sqrt(2)/2 */
1293 const float rmax = halfSize + 0.7071F;
1294 const float rmin2 = MAX2(0.0F, rmin * rmin);
1295 const float rmax2 = rmax * rmax;
1296 const float cscale = 1.0F / (rmax2 - rmin2);
1297 int ix, iy;
1298
1299 for (iy = iymin; iy <= iymax; iy += 2) {
1300 for (ix = ixmin; ix <= ixmax; ix += 2) {
1301 float dx, dy, dist2, cover;
1302
1303 setup->quad.mask = 0x0;
1304
1305 dx = (ix + 0.5f) - x;
1306 dy = (iy + 0.5f) - y;
1307 dist2 = dx * dx + dy * dy;
1308 if (dist2 <= rmax2) {
1309 cover = 1.0F - (dist2 - rmin2) * cscale;
1310 setup->quad.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
1311 setup->quad.mask |= MASK_TOP_LEFT;
1312 }
1313
1314 dx = (ix + 1.5f) - x;
1315 dy = (iy + 0.5f) - y;
1316 dist2 = dx * dx + dy * dy;
1317 if (dist2 <= rmax2) {
1318 cover = 1.0F - (dist2 - rmin2) * cscale;
1319 setup->quad.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
1320 setup->quad.mask |= MASK_TOP_RIGHT;
1321 }
1322
1323 dx = (ix + 0.5f) - x;
1324 dy = (iy + 1.5f) - y;
1325 dist2 = dx * dx + dy * dy;
1326 if (dist2 <= rmax2) {
1327 cover = 1.0F - (dist2 - rmin2) * cscale;
1328 setup->quad.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
1329 setup->quad.mask |= MASK_BOTTOM_LEFT;
1330 }
1331
1332 dx = (ix + 1.5f) - x;
1333 dy = (iy + 1.5f) - y;
1334 dist2 = dx * dx + dy * dy;
1335 if (dist2 <= rmax2) {
1336 cover = 1.0F - (dist2 - rmin2) * cscale;
1337 setup->quad.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
1338 setup->quad.mask |= MASK_BOTTOM_RIGHT;
1339 }
1340
1341 if (setup->quad.mask) {
1342 setup->quad.x0 = ix;
1343 setup->quad.y0 = iy;
1344 CLIP_EMIT_QUAD(setup);
1345 }
1346 }
1347 }
1348 }
1349 else {
1350 /* square points */
1351 const int xmin = (int) (x + 0.75 - halfSize);
1352 const int ymin = (int) (y + 0.25 - halfSize);
1353 const int xmax = xmin + (int) size;
1354 const int ymax = ymin + (int) size;
1355 /* XXX could apply scissor to xmin,ymin,xmax,ymax now */
1356 const int ixmin = block(xmin);
1357 const int ixmax = block(xmax - 1);
1358 const int iymin = block(ymin);
1359 const int iymax = block(ymax - 1);
1360 int ix, iy;
1361
1362 /*
1363 debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax);
1364 */
1365 for (iy = iymin; iy <= iymax; iy += 2) {
1366 uint rowMask = 0xf;
1367 if (iy < ymin) {
1368 /* above the top edge */
1369 rowMask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
1370 }
1371 if (iy + 1 >= ymax) {
1372 /* below the bottom edge */
1373 rowMask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
1374 }
1375
1376 for (ix = ixmin; ix <= ixmax; ix += 2) {
1377 uint mask = rowMask;
1378
1379 if (ix < xmin) {
1380 /* fragment is past left edge of point, turn off left bits */
1381 mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
1382 }
1383 if (ix + 1 >= xmax) {
1384 /* past the right edge */
1385 mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
1386 }
1387
1388 setup->quad.mask = mask;
1389 setup->quad.x0 = ix;
1390 setup->quad.y0 = iy;
1391 CLIP_EMIT_QUAD(setup);
1392 }
1393 }
1394 }
1395 }
1396
1397 WAIT_FOR_COMPLETION(setup);
1398 }
1399
1400 void setup_prepare( struct setup_context *setup )
1401 {
1402 struct softpipe_context *sp = setup->softpipe;
1403 unsigned i;
1404
1405 if (sp->dirty) {
1406 softpipe_update_derived(sp);
1407 }
1408
1409 /* Mark surfaces as defined now */
1410 for (i = 0; i < sp->framebuffer.num_cbufs; i++){
1411 if (sp->framebuffer.cbufs[i]) {
1412 sp->framebuffer.cbufs[i]->status = PIPE_SURFACE_STATUS_DEFINED;
1413 }
1414 }
1415 if (sp->framebuffer.zsbuf) {
1416 sp->framebuffer.zsbuf->status = PIPE_SURFACE_STATUS_DEFINED;
1417 }
1418
1419 /* Note: nr_attrs is only used for debugging (vertex printing) */
1420 setup->quad.nr_attrs = draw_num_vs_outputs(sp->draw);
1421
1422 for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
1423 sp->quad[i].first->begin( sp->quad[i].first );
1424 }
1425
1426 if (sp->reduced_api_prim == PIPE_PRIM_TRIANGLES &&
1427 sp->rasterizer->fill_cw == PIPE_POLYGON_MODE_FILL &&
1428 sp->rasterizer->fill_ccw == PIPE_POLYGON_MODE_FILL) {
1429 /* we'll do culling */
1430 setup->winding = sp->rasterizer->cull_mode;
1431 }
1432 else {
1433 /* 'draw' will do culling */
1434 setup->winding = PIPE_WINDING_NONE;
1435 }
1436 }
1437
1438
1439
1440 void setup_destroy_context( struct setup_context *setup )
1441 {
1442 FREE( setup );
1443 }
1444
1445
1446 /**
1447 * Create a new primitive setup/render stage.
1448 */
1449 struct setup_context *setup_create_context( struct softpipe_context *softpipe )
1450 {
1451 struct setup_context *setup = CALLOC_STRUCT(setup_context);
1452 uint i;
1453
1454 setup->softpipe = softpipe;
1455
1456 setup->quad.coef = setup->coef;
1457 setup->quad.posCoef = &setup->posCoef;
1458
1459 #if SP_NUM_QUAD_THREADS > 1
1460 setup->que.first = 0;
1461 setup->que.last = 0;
1462 pipe_mutex_init( setup->que.mutex );
1463 for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
1464 setup->threads[i].setup = setup;
1465 setup->threads[i].id = i;
1466 setup->threads[i].handle = pipe_thread_create( quad_thread, &setup->threads[i] );
1467 }
1468 #endif
1469
1470 return setup;
1471 }
1472