1 /**************************************************************************
2 *
3 * Copyright 2007 VMware, Inc.
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 VMWARE 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 * Binning code for lines
30 */
31
32 #include "util/u_math.h"
33 #include "util/u_memory.h"
34 #include "lp_perf.h"
35 #include "lp_setup_context.h"
36 #include "lp_rast.h"
37 #include "lp_state_fs.h"
38 #include "lp_state_setup.h"
39 #include "lp_context.h"
40 #include "draw/draw_context.h"
41
42 #define NUM_CHANNELS 4
43
44 struct lp_line_info {
45
46 float dx;
47 float dy;
48 float oneoverarea;
49 boolean frontfacing;
50
51 const float (*v1)[4];
52 const float (*v2)[4];
53
54 float (*a0)[4];
55 float (*dadx)[4];
56 float (*dady)[4];
57 };
58
59
60 /**
61 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
62 */
constant_coef(struct lp_setup_context * setup,struct lp_line_info * info,unsigned slot,const float value,unsigned i)63 static void constant_coef( struct lp_setup_context *setup,
64 struct lp_line_info *info,
65 unsigned slot,
66 const float value,
67 unsigned i )
68 {
69 info->a0[slot][i] = value;
70 info->dadx[slot][i] = 0.0f;
71 info->dady[slot][i] = 0.0f;
72 }
73
74
75 /**
76 * Compute a0, dadx and dady for a linearly interpolated coefficient,
77 * for a triangle.
78 */
linear_coef(struct lp_setup_context * setup,struct lp_line_info * info,unsigned slot,unsigned vert_attr,unsigned i)79 static void linear_coef( struct lp_setup_context *setup,
80 struct lp_line_info *info,
81 unsigned slot,
82 unsigned vert_attr,
83 unsigned i)
84 {
85 float a1 = info->v1[vert_attr][i];
86 float a2 = info->v2[vert_attr][i];
87
88 float da21 = a1 - a2;
89 float dadx = da21 * info->dx * info->oneoverarea;
90 float dady = da21 * info->dy * info->oneoverarea;
91
92 info->dadx[slot][i] = dadx;
93 info->dady[slot][i] = dady;
94
95 info->a0[slot][i] = (a1 -
96 (dadx * (info->v1[0][0] - setup->pixel_offset) +
97 dady * (info->v1[0][1] - setup->pixel_offset)));
98 }
99
100
101 /**
102 * Compute a0, dadx and dady for a perspective-corrected interpolant,
103 * for a triangle.
104 * We basically multiply the vertex value by 1/w before computing
105 * the plane coefficients (a0, dadx, dady).
106 * Later, when we compute the value at a particular fragment position we'll
107 * divide the interpolated value by the interpolated W at that fragment.
108 */
perspective_coef(struct lp_setup_context * setup,struct lp_line_info * info,unsigned slot,unsigned vert_attr,unsigned i)109 static void perspective_coef( struct lp_setup_context *setup,
110 struct lp_line_info *info,
111 unsigned slot,
112 unsigned vert_attr,
113 unsigned i)
114 {
115 /* premultiply by 1/w (v[0][3] is always 1/w):
116 */
117 float a1 = info->v1[vert_attr][i] * info->v1[0][3];
118 float a2 = info->v2[vert_attr][i] * info->v2[0][3];
119
120 float da21 = a1 - a2;
121 float dadx = da21 * info->dx * info->oneoverarea;
122 float dady = da21 * info->dy * info->oneoverarea;
123
124 info->dadx[slot][i] = dadx;
125 info->dady[slot][i] = dady;
126
127 info->a0[slot][i] = (a1 -
128 (dadx * (info->v1[0][0] - setup->pixel_offset) +
129 dady * (info->v1[0][1] - setup->pixel_offset)));
130 }
131
132 static void
setup_fragcoord_coef(struct lp_setup_context * setup,struct lp_line_info * info,unsigned slot,unsigned usage_mask)133 setup_fragcoord_coef( struct lp_setup_context *setup,
134 struct lp_line_info *info,
135 unsigned slot,
136 unsigned usage_mask)
137 {
138 /*X*/
139 if (usage_mask & TGSI_WRITEMASK_X) {
140 info->a0[slot][0] = 0.0;
141 info->dadx[slot][0] = 1.0;
142 info->dady[slot][0] = 0.0;
143 }
144
145 /*Y*/
146 if (usage_mask & TGSI_WRITEMASK_Y) {
147 info->a0[slot][1] = 0.0;
148 info->dadx[slot][1] = 0.0;
149 info->dady[slot][1] = 1.0;
150 }
151
152 /*Z*/
153 if (usage_mask & TGSI_WRITEMASK_Z) {
154 linear_coef(setup, info, slot, 0, 2);
155 }
156
157 /*W*/
158 if (usage_mask & TGSI_WRITEMASK_W) {
159 linear_coef(setup, info, slot, 0, 3);
160 }
161 }
162
163 /**
164 * Compute the tri->coef[] array dadx, dady, a0 values.
165 */
setup_line_coefficients(struct lp_setup_context * setup,struct lp_line_info * info)166 static void setup_line_coefficients( struct lp_setup_context *setup,
167 struct lp_line_info *info)
168 {
169 const struct lp_setup_variant_key *key = &setup->setup.variant->key;
170 unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ;
171 unsigned slot;
172
173 /* setup interpolation for all the remaining attributes:
174 */
175 for (slot = 0; slot < key->num_inputs; slot++) {
176 unsigned vert_attr = key->inputs[slot].src_index;
177 unsigned usage_mask = key->inputs[slot].usage_mask;
178 unsigned i;
179
180 switch (key->inputs[slot].interp) {
181 case LP_INTERP_CONSTANT:
182 if (key->flatshade_first) {
183 for (i = 0; i < NUM_CHANNELS; i++)
184 if (usage_mask & (1 << i))
185 constant_coef(setup, info, slot+1, info->v1[vert_attr][i], i);
186 }
187 else {
188 for (i = 0; i < NUM_CHANNELS; i++)
189 if (usage_mask & (1 << i))
190 constant_coef(setup, info, slot+1, info->v2[vert_attr][i], i);
191 }
192 break;
193
194 case LP_INTERP_LINEAR:
195 for (i = 0; i < NUM_CHANNELS; i++)
196 if (usage_mask & (1 << i))
197 linear_coef(setup, info, slot+1, vert_attr, i);
198 break;
199
200 case LP_INTERP_PERSPECTIVE:
201 for (i = 0; i < NUM_CHANNELS; i++)
202 if (usage_mask & (1 << i))
203 perspective_coef(setup, info, slot+1, vert_attr, i);
204 fragcoord_usage_mask |= TGSI_WRITEMASK_W;
205 break;
206
207 case LP_INTERP_POSITION:
208 /*
209 * The generated pixel interpolators will pick up the coeffs from
210 * slot 0, so all need to ensure that the usage mask is covers all
211 * usages.
212 */
213 fragcoord_usage_mask |= usage_mask;
214 break;
215
216 case LP_INTERP_FACING:
217 for (i = 0; i < NUM_CHANNELS; i++)
218 if (usage_mask & (1 << i))
219 constant_coef(setup, info, slot+1,
220 info->frontfacing ? 1.0f : -1.0f, i);
221 break;
222
223 default:
224 assert(0);
225 }
226 }
227
228 /* The internal position input is in slot zero:
229 */
230 setup_fragcoord_coef(setup, info, 0,
231 fragcoord_usage_mask);
232 }
233
234
235
subpixel_snap(float a)236 static inline int subpixel_snap( float a )
237 {
238 return util_iround(FIXED_ONE * a);
239 }
240
241
242 /**
243 * Print line vertex attribs (for debug).
244 */
245 static void
print_line(struct lp_setup_context * setup,const float (* v1)[4],const float (* v2)[4])246 print_line(struct lp_setup_context *setup,
247 const float (*v1)[4],
248 const float (*v2)[4])
249 {
250 const struct lp_setup_variant_key *key = &setup->setup.variant->key;
251 uint i;
252
253 debug_printf("llvmpipe line\n");
254 for (i = 0; i < 1 + key->num_inputs; i++) {
255 debug_printf(" v1[%d]: %f %f %f %f\n", i,
256 v1[i][0], v1[i][1], v1[i][2], v1[i][3]);
257 }
258 for (i = 0; i < 1 + key->num_inputs; i++) {
259 debug_printf(" v2[%d]: %f %f %f %f\n", i,
260 v2[i][0], v2[i][1], v2[i][2], v2[i][3]);
261 }
262 }
263
264
sign(float x)265 static inline boolean sign(float x){
266 return x >= 0;
267 }
268
269
270 /* Used on positive floats only:
271 */
fracf(float f)272 static inline float fracf(float f)
273 {
274 return f - floorf(f);
275 }
276
277
278
279 static boolean
try_setup_line(struct lp_setup_context * setup,const float (* v1)[4],const float (* v2)[4])280 try_setup_line( struct lp_setup_context *setup,
281 const float (*v1)[4],
282 const float (*v2)[4])
283 {
284 struct llvmpipe_context *lp_context = (struct llvmpipe_context *)setup->pipe;
285 struct lp_scene *scene = setup->scene;
286 const struct lp_setup_variant_key *key = &setup->setup.variant->key;
287 struct lp_rast_triangle *line;
288 struct lp_rast_plane *plane;
289 struct lp_line_info info;
290 float width = MAX2(1.0, setup->line_width);
291 const struct u_rect *scissor;
292 struct u_rect bbox, bboxpos;
293 boolean s_planes[4];
294 unsigned tri_bytes;
295 int x[4];
296 int y[4];
297 int i;
298 int nr_planes = 4;
299 unsigned viewport_index = 0;
300 unsigned layer = 0;
301 float pixel_offset = setup->multisample ? 0.0 : setup->pixel_offset;
302 /* linewidth should be interpreted as integer */
303 int fixed_width = util_iround(width) * FIXED_ONE;
304
305 float x_offset=0;
306 float y_offset=0;
307 float x_offset_end=0;
308 float y_offset_end=0;
309
310 float x1diff;
311 float y1diff;
312 float x2diff;
313 float y2diff;
314 float dx, dy;
315 float area;
316 const float (*pv)[4];
317
318 boolean draw_start;
319 boolean draw_end;
320 boolean will_draw_start;
321 boolean will_draw_end;
322
323 if (lp_context->active_statistics_queries) {
324 lp_context->pipeline_statistics.c_primitives++;
325 }
326
327 if (0)
328 print_line(setup, v1, v2);
329
330 if (setup->flatshade_first) {
331 pv = v1;
332 }
333 else {
334 pv = v2;
335 }
336 if (setup->viewport_index_slot > 0) {
337 unsigned *udata = (unsigned*)pv[setup->viewport_index_slot];
338 viewport_index = lp_clamp_viewport_idx(*udata);
339 }
340 if (setup->layer_slot > 0) {
341 layer = *(unsigned*)pv[setup->layer_slot];
342 layer = MIN2(layer, scene->fb_max_layer);
343 }
344
345 dx = v1[0][0] - v2[0][0];
346 dy = v1[0][1] - v2[0][1];
347 area = (dx * dx + dy * dy);
348 if (area == 0) {
349 LP_COUNT(nr_culled_tris);
350 return TRUE;
351 }
352
353 info.oneoverarea = 1.0f / area;
354 info.dx = dx;
355 info.dy = dy;
356 info.v1 = v1;
357 info.v2 = v2;
358
359
360 if (setup->rectangular_lines) {
361 float scale = (setup->line_width * 0.5f) / sqrtf(area);
362 int tx = subpixel_snap(-dy * scale);
363 int ty = subpixel_snap(+dx * scale);
364
365 x[0] = subpixel_snap(v1[0][0] - pixel_offset) - tx;
366 x[1] = subpixel_snap(v2[0][0] - pixel_offset) - tx;
367 x[2] = subpixel_snap(v2[0][0] - pixel_offset) + tx;
368 x[3] = subpixel_snap(v1[0][0] - pixel_offset) + tx;
369
370 y[0] = subpixel_snap(v1[0][1] - pixel_offset) - ty;
371 y[1] = subpixel_snap(v2[0][1] - pixel_offset) - ty;
372 y[2] = subpixel_snap(v2[0][1] - pixel_offset) + ty;
373 y[3] = subpixel_snap(v1[0][1] - pixel_offset) + ty;
374 } else if (fabsf(dx) >= fabsf(dy)) {
375 float dydx = dy / dx;
376
377 /* X-MAJOR LINE */
378 x1diff = v1[0][0] - floorf(v1[0][0]) - 0.5f;
379 y1diff = v1[0][1] - floorf(v1[0][1]) - 0.5f;
380 x2diff = v2[0][0] - floorf(v2[0][0]) - 0.5f;
381 y2diff = v2[0][1] - floorf(v2[0][1]) - 0.5f;
382
383 if (y2diff==-0.5 && dy<0){
384 y2diff = 0.5;
385 }
386
387 /*
388 * Diamond exit rule test for starting point
389 */
390 if (fabsf(x1diff) + fabsf(y1diff) < 0.5) {
391 draw_start = TRUE;
392 }
393 else if (sign(x1diff) == sign(-dx)) {
394 draw_start = FALSE;
395 }
396 else if (sign(-y1diff) != sign(dy)) {
397 draw_start = TRUE;
398 }
399 else {
400 /* do intersection test */
401 float yintersect = fracf(v1[0][1]) + x1diff * dydx;
402 draw_start = (yintersect < 1.0 && yintersect > 0.0);
403 }
404
405
406 /*
407 * Diamond exit rule test for ending point
408 */
409 if (fabsf(x2diff) + fabsf(y2diff) < 0.5) {
410 draw_end = FALSE;
411 }
412 else if (sign(x2diff) != sign(-dx)) {
413 draw_end = FALSE;
414 }
415 else if (sign(-y2diff) == sign(dy)) {
416 draw_end = TRUE;
417 }
418 else {
419 /* do intersection test */
420 float yintersect = fracf(v2[0][1]) + x2diff * dydx;
421 draw_end = (yintersect < 1.0 && yintersect > 0.0);
422 }
423
424 /* Are we already drawing start/end?
425 */
426 will_draw_start = sign(-x1diff) != sign(dx);
427 will_draw_end = (sign(x2diff) == sign(-dx)) || x2diff==0;
428
429 /* interpolate using the preferred wide-lines formula */
430 info.dx *= 1 + dydx * dydx;
431 info.dy = 0;
432
433 if (dx < 0) {
434 /* if v2 is to the right of v1, swap pointers */
435 const float (*temp)[4] = v1;
436 v1 = v2;
437 v2 = temp;
438 dx = -dx;
439 dy = -dy;
440 /* Otherwise shift planes appropriately */
441 if (will_draw_start != draw_start) {
442 x_offset_end = - x1diff - 0.5;
443 y_offset_end = x_offset_end * dydx;
444
445 }
446 if (will_draw_end != draw_end) {
447 x_offset = - x2diff - 0.5;
448 y_offset = x_offset * dydx;
449 }
450
451 }
452 else{
453 /* Otherwise shift planes appropriately */
454 if (will_draw_start != draw_start) {
455 x_offset = - x1diff + 0.5;
456 y_offset = x_offset * dydx;
457 }
458 if (will_draw_end != draw_end) {
459 x_offset_end = - x2diff + 0.5;
460 y_offset_end = x_offset_end * dydx;
461 }
462 }
463
464 /* x/y positions in fixed point */
465 x[0] = subpixel_snap(v1[0][0] + x_offset - pixel_offset);
466 x[1] = subpixel_snap(v2[0][0] + x_offset_end - pixel_offset);
467 x[2] = subpixel_snap(v2[0][0] + x_offset_end - pixel_offset);
468 x[3] = subpixel_snap(v1[0][0] + x_offset - pixel_offset);
469
470 y[0] = subpixel_snap(v1[0][1] + y_offset - pixel_offset) - fixed_width/2;
471 y[1] = subpixel_snap(v2[0][1] + y_offset_end - pixel_offset) - fixed_width/2;
472 y[2] = subpixel_snap(v2[0][1] + y_offset_end - pixel_offset) + fixed_width/2;
473 y[3] = subpixel_snap(v1[0][1] + y_offset - pixel_offset) + fixed_width/2;
474
475 }
476 else {
477 const float dxdy = dx / dy;
478
479 /* Y-MAJOR LINE */
480 x1diff = v1[0][0] - floorf(v1[0][0]) - 0.5f;
481 y1diff = v1[0][1] - floorf(v1[0][1]) - 0.5f;
482 x2diff = v2[0][0] - floorf(v2[0][0]) - 0.5f;
483 y2diff = v2[0][1] - floorf(v2[0][1]) - 0.5f;
484
485 if (x2diff==-0.5 && dx<0) {
486 x2diff = 0.5;
487 }
488
489 /*
490 * Diamond exit rule test for starting point
491 */
492 if (fabsf(x1diff) + fabsf(y1diff) < 0.5) {
493 draw_start = TRUE;
494 }
495 else if (sign(-y1diff) == sign(dy)) {
496 draw_start = FALSE;
497 }
498 else if (sign(x1diff) != sign(-dx)) {
499 draw_start = TRUE;
500 }
501 else {
502 /* do intersection test */
503 float xintersect = fracf(v1[0][0]) + y1diff * dxdy;
504 draw_start = (xintersect < 1.0 && xintersect > 0.0);
505 }
506
507 /*
508 * Diamond exit rule test for ending point
509 */
510 if (fabsf(x2diff) + fabsf(y2diff) < 0.5) {
511 draw_end = FALSE;
512 }
513 else if (sign(-y2diff) != sign(dy) ) {
514 draw_end = FALSE;
515 }
516 else if (sign(x2diff) == sign(-dx) ) {
517 draw_end = TRUE;
518 }
519 else {
520 /* do intersection test */
521 float xintersect = fracf(v2[0][0]) + y2diff * dxdy;
522 draw_end = (xintersect < 1.0 && xintersect >= 0.0);
523 }
524
525 /* Are we already drawing start/end?
526 */
527 will_draw_start = sign(y1diff) == sign(dy);
528 will_draw_end = (sign(-y2diff) == sign(dy)) || y2diff==0;
529
530 /* interpolate using the preferred wide-lines formula */
531 info.dx = 0;
532 info.dy *= 1 + dxdy * dxdy;
533
534 if (dy > 0) {
535 /* if v2 is on top of v1, swap pointers */
536 const float (*temp)[4] = v1;
537 v1 = v2;
538 v2 = temp;
539 dx = -dx;
540 dy = -dy;
541
542 /* Otherwise shift planes appropriately */
543 if (will_draw_start != draw_start) {
544 y_offset_end = - y1diff + 0.5;
545 x_offset_end = y_offset_end * dxdy;
546 }
547 if (will_draw_end != draw_end) {
548 y_offset = - y2diff + 0.5;
549 x_offset = y_offset * dxdy;
550 }
551 }
552 else {
553 /* Otherwise shift planes appropriately */
554 if (will_draw_start != draw_start) {
555 y_offset = - y1diff - 0.5;
556 x_offset = y_offset * dxdy;
557
558 }
559 if (will_draw_end != draw_end) {
560 y_offset_end = - y2diff - 0.5;
561 x_offset_end = y_offset_end * dxdy;
562 }
563 }
564
565 /* x/y positions in fixed point */
566 x[0] = subpixel_snap(v1[0][0] + x_offset - pixel_offset) - fixed_width/2;
567 x[1] = subpixel_snap(v2[0][0] + x_offset_end - pixel_offset) - fixed_width/2;
568 x[2] = subpixel_snap(v2[0][0] + x_offset_end - pixel_offset) + fixed_width/2;
569 x[3] = subpixel_snap(v1[0][0] + x_offset - pixel_offset) + fixed_width/2;
570
571 y[0] = subpixel_snap(v1[0][1] + y_offset - pixel_offset);
572 y[1] = subpixel_snap(v2[0][1] + y_offset_end - pixel_offset);
573 y[2] = subpixel_snap(v2[0][1] + y_offset_end - pixel_offset);
574 y[3] = subpixel_snap(v1[0][1] + y_offset - pixel_offset);
575 }
576
577 /* Bounding rectangle (in pixels) */
578 {
579 /* Yes this is necessary to accurately calculate bounding boxes
580 * with the two fill-conventions we support. GL (normally) ends
581 * up needing a bottom-left fill convention, which requires
582 * slightly different rounding.
583 */
584 int adj = (setup->bottom_edge_rule != 0) ? 1 : 0;
585
586 bbox.x0 = (MIN4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER;
587 bbox.x1 = (MAX4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER;
588 bbox.y0 = (MIN4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
589 bbox.y1 = (MAX4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
590
591 /* Inclusive coordinates:
592 */
593 bbox.x1--;
594 bbox.y1--;
595 }
596
597 if (!u_rect_test_intersection(&setup->draw_regions[viewport_index], &bbox)) {
598 if (0) debug_printf("no intersection\n");
599 LP_COUNT(nr_culled_tris);
600 return TRUE;
601 }
602
603 bboxpos = bbox;
604
605 /* Can safely discard negative regions:
606 */
607 bboxpos.x0 = MAX2(bboxpos.x0, 0);
608 bboxpos.y0 = MAX2(bboxpos.y0, 0);
609
610 nr_planes = 4;
611 /*
612 * Determine how many scissor planes we need, that is drop scissor
613 * edges if the bounding box of the tri is fully inside that edge.
614 */
615 scissor = &setup->draw_regions[viewport_index];
616 scissor_planes_needed(s_planes, &bboxpos, scissor);
617 nr_planes += s_planes[0] + s_planes[1] + s_planes[2] + s_planes[3];
618
619 line = lp_setup_alloc_triangle(scene,
620 key->num_inputs,
621 nr_planes,
622 &tri_bytes);
623 if (!line)
624 return FALSE;
625
626 #ifdef DEBUG
627 line->v[0][0] = v1[0][0];
628 line->v[1][0] = v2[0][0];
629 line->v[0][1] = v1[0][1];
630 line->v[1][1] = v2[0][1];
631 #endif
632
633 LP_COUNT(nr_tris);
634
635 /* calculate the deltas */
636 plane = GET_PLANES(line);
637 plane[0].dcdy = x[0] - x[1];
638 plane[1].dcdy = x[1] - x[2];
639 plane[2].dcdy = x[2] - x[3];
640 plane[3].dcdy = x[3] - x[0];
641
642 plane[0].dcdx = y[0] - y[1];
643 plane[1].dcdx = y[1] - y[2];
644 plane[2].dcdx = y[2] - y[3];
645 plane[3].dcdx = y[3] - y[0];
646
647 if (draw_will_inject_frontface(lp_context->draw) &&
648 setup->face_slot > 0) {
649 line->inputs.frontfacing = v1[setup->face_slot][0];
650 } else {
651 line->inputs.frontfacing = TRUE;
652 }
653
654 /* Setup parameter interpolants:
655 */
656 info.a0 = GET_A0(&line->inputs);
657 info.dadx = GET_DADX(&line->inputs);
658 info.dady = GET_DADY(&line->inputs);
659 info.frontfacing = line->inputs.frontfacing;
660 setup_line_coefficients(setup, &info);
661
662 line->inputs.disable = FALSE;
663 line->inputs.opaque = FALSE;
664 line->inputs.layer = layer;
665 line->inputs.viewport_index = viewport_index;
666 line->inputs.view_index = setup->view_index;
667
668 /*
669 * XXX: this code is mostly identical to the one in lp_setup_tri, except it
670 * uses 4 planes instead of 3. Could share the code (including the sse
671 * assembly, in fact we'd get the 4th plane for free).
672 * The only difference apart from storing the 4th plane would be some
673 * different shuffle for calculating dcdx/dcdy.
674 */
675 for (i = 0; i < 4; i++) {
676
677 /* half-edge constants, will be iterated over the whole render
678 * target.
679 */
680 plane[i].c = IMUL64(plane[i].dcdx, x[i]) - IMUL64(plane[i].dcdy, y[i]);
681
682 /* correct for top-left vs. bottom-left fill convention.
683 */
684 if (plane[i].dcdx < 0) {
685 /* both fill conventions want this - adjust for left edges */
686 plane[i].c++;
687 }
688 else if (plane[i].dcdx == 0) {
689 if (setup->bottom_edge_rule == 0) {
690 /* correct for top-left fill convention:
691 */
692 if (plane[i].dcdy > 0) plane[i].c++;
693 }
694 else {
695 /* correct for bottom-left fill convention:
696 */
697 if (plane[i].dcdy < 0) plane[i].c++;
698 }
699 }
700
701 plane[i].dcdx *= FIXED_ONE;
702 plane[i].dcdy *= FIXED_ONE;
703
704 /* find trivial reject offsets for each edge for a single-pixel
705 * sized block. These will be scaled up at each recursive level to
706 * match the active blocksize. Scaling in this way works best if
707 * the blocks are square.
708 */
709 plane[i].eo = 0;
710 if (plane[i].dcdx < 0) plane[i].eo -= plane[i].dcdx;
711 if (plane[i].dcdy > 0) plane[i].eo += plane[i].dcdy;
712 }
713
714 if (nr_planes > 4) {
715 lp_setup_add_scissor_planes(scissor, &plane[4], s_planes, setup->multisample);
716 }
717
718 return lp_setup_bin_triangle(setup, line, &bbox, &bboxpos, nr_planes, viewport_index);
719 }
720
721
lp_setup_line_discard(struct lp_setup_context * setup,const float (* v0)[4],const float (* v1)[4])722 static void lp_setup_line_discard(struct lp_setup_context *setup,
723 const float (*v0)[4],
724 const float (*v1)[4])
725 {
726 }
727
lp_setup_line(struct lp_setup_context * setup,const float (* v0)[4],const float (* v1)[4])728 static void lp_setup_line(struct lp_setup_context *setup,
729 const float (*v0)[4],
730 const float (*v1)[4])
731 {
732 if (!try_setup_line(setup, v0, v1)) {
733 if (!lp_setup_flush_and_restart(setup))
734 return;
735
736 if (!try_setup_line(setup, v0, v1))
737 return;
738 }
739 }
740
741
lp_setup_choose_line(struct lp_setup_context * setup)742 void lp_setup_choose_line(struct lp_setup_context *setup)
743 {
744 if (setup->rasterizer_discard) {
745 setup->line = lp_setup_line_discard;
746 } else {
747 setup->line = lp_setup_line;
748 }
749 }
750
751
752