Merge branch '7.8'
[mesa.git] / src / gallium / drivers / llvmpipe / lp_setup_tri.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 * Binning code for triangles
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
38 #define NUM_CHANNELS 4
39
40
41 /**
42 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
43 */
44 static void constant_coef( struct lp_setup_context *setup,
45 struct lp_rast_triangle *tri,
46 unsigned slot,
47 const float value,
48 unsigned i )
49 {
50 tri->inputs.a0[slot][i] = value;
51 tri->inputs.dadx[slot][i] = 0.0f;
52 tri->inputs.dady[slot][i] = 0.0f;
53 }
54
55
56 /**
57 * Compute a0, dadx and dady for a linearly interpolated coefficient,
58 * for a triangle.
59 */
60 static void linear_coef( struct lp_setup_context *setup,
61 struct lp_rast_triangle *tri,
62 float oneoverarea,
63 unsigned slot,
64 const float (*v1)[4],
65 const float (*v2)[4],
66 const float (*v3)[4],
67 unsigned vert_attr,
68 unsigned i)
69 {
70 float a1 = v1[vert_attr][i];
71 float a2 = v2[vert_attr][i];
72 float a3 = v3[vert_attr][i];
73
74 float da12 = a1 - a2;
75 float da31 = a3 - a1;
76 float dadx = (da12 * tri->dy31 - tri->dy12 * da31) * oneoverarea;
77 float dady = (da31 * tri->dx12 - tri->dx31 * da12) * oneoverarea;
78
79 tri->inputs.dadx[slot][i] = dadx;
80 tri->inputs.dady[slot][i] = dady;
81
82 /* calculate a0 as the value which would be sampled for the
83 * fragment at (0,0), taking into account that we want to sample at
84 * pixel centers, in other words (0.5, 0.5).
85 *
86 * this is neat but unfortunately not a good way to do things for
87 * triangles with very large values of dadx or dady as it will
88 * result in the subtraction and re-addition from a0 of a very
89 * large number, which means we'll end up loosing a lot of the
90 * fractional bits and precision from a0. the way to fix this is
91 * to define a0 as the sample at a pixel center somewhere near vmin
92 * instead - i'll switch to this later.
93 */
94 tri->inputs.a0[slot][i] = (a1 -
95 (dadx * (v1[0][0] - setup->pixel_offset) +
96 dady * (v1[0][1] - setup->pixel_offset)));
97 }
98
99
100 /**
101 * Compute a0, dadx and dady for a perspective-corrected interpolant,
102 * for a triangle.
103 * We basically multiply the vertex value by 1/w before computing
104 * the plane coefficients (a0, dadx, dady).
105 * Later, when we compute the value at a particular fragment position we'll
106 * divide the interpolated value by the interpolated W at that fragment.
107 */
108 static void perspective_coef( struct lp_setup_context *setup,
109 struct lp_rast_triangle *tri,
110 float oneoverarea,
111 unsigned slot,
112 const float (*v1)[4],
113 const float (*v2)[4],
114 const float (*v3)[4],
115 unsigned vert_attr,
116 unsigned i)
117 {
118 /* premultiply by 1/w (v[0][3] is always 1/w):
119 */
120 float a1 = v1[vert_attr][i] * v1[0][3];
121 float a2 = v2[vert_attr][i] * v2[0][3];
122 float a3 = v3[vert_attr][i] * v3[0][3];
123 float da12 = a1 - a2;
124 float da31 = a3 - a1;
125 float dadx = (da12 * tri->dy31 - tri->dy12 * da31) * oneoverarea;
126 float dady = (da31 * tri->dx12 - tri->dx31 * da12) * oneoverarea;
127
128 tri->inputs.dadx[slot][i] = dadx;
129 tri->inputs.dady[slot][i] = dady;
130 tri->inputs.a0[slot][i] = (a1 -
131 (dadx * (v1[0][0] - setup->pixel_offset) +
132 dady * (v1[0][1] - setup->pixel_offset)));
133 }
134
135
136 /**
137 * Special coefficient setup for gl_FragCoord.
138 * X and Y are trivial
139 * Z and W are copied from position_coef which should have already been computed.
140 * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
141 */
142 static void
143 setup_fragcoord_coef(struct lp_setup_context *setup,
144 struct lp_rast_triangle *tri,
145 float oneoverarea,
146 unsigned slot,
147 const float (*v1)[4],
148 const float (*v2)[4],
149 const float (*v3)[4])
150 {
151 /*X*/
152 tri->inputs.a0[slot][0] = 0.0;
153 tri->inputs.dadx[slot][0] = 1.0;
154 tri->inputs.dady[slot][0] = 0.0;
155 /*Y*/
156 tri->inputs.a0[slot][1] = 0.0;
157 tri->inputs.dadx[slot][1] = 0.0;
158 tri->inputs.dady[slot][1] = 1.0;
159 /*Z*/
160 linear_coef(setup, tri, oneoverarea, slot, v1, v2, v3, 0, 2);
161 /*W*/
162 linear_coef(setup, tri, oneoverarea, slot, v1, v2, v3, 0, 3);
163 }
164
165
166 static void setup_facing_coef( struct lp_setup_context *setup,
167 struct lp_rast_triangle *tri,
168 unsigned slot,
169 boolean frontface )
170 {
171 constant_coef( setup, tri, slot, 1.0f - frontface, 0 );
172 constant_coef( setup, tri, slot, 0.0f, 1 ); /* wasted */
173 constant_coef( setup, tri, slot, 0.0f, 2 ); /* wasted */
174 constant_coef( setup, tri, slot, 0.0f, 3 ); /* wasted */
175 }
176
177
178 /**
179 * Compute the tri->coef[] array dadx, dady, a0 values.
180 */
181 static void setup_tri_coefficients( struct lp_setup_context *setup,
182 struct lp_rast_triangle *tri,
183 float oneoverarea,
184 const float (*v1)[4],
185 const float (*v2)[4],
186 const float (*v3)[4],
187 boolean frontface)
188 {
189 unsigned slot;
190
191 /* The internal position input is in slot zero:
192 */
193 setup_fragcoord_coef(setup, tri, oneoverarea, 0, v1, v2, v3);
194
195 /* setup interpolation for all the remaining attributes:
196 */
197 for (slot = 0; slot < setup->fs.nr_inputs; slot++) {
198 unsigned vert_attr = setup->fs.input[slot].src_index;
199 unsigned i;
200
201 switch (setup->fs.input[slot].interp) {
202 case LP_INTERP_CONSTANT:
203 for (i = 0; i < NUM_CHANNELS; i++)
204 constant_coef(setup, tri, slot+1, v3[vert_attr][i], i);
205 break;
206
207 case LP_INTERP_LINEAR:
208 for (i = 0; i < NUM_CHANNELS; i++)
209 linear_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
210 break;
211
212 case LP_INTERP_PERSPECTIVE:
213 for (i = 0; i < NUM_CHANNELS; i++)
214 perspective_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
215 break;
216
217 case LP_INTERP_POSITION:
218 /* XXX: fix me - duplicates the values in slot zero.
219 */
220 setup_fragcoord_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3);
221 break;
222
223 case LP_INTERP_FACING:
224 setup_facing_coef(setup, tri, slot+1, frontface);
225 break;
226
227 default:
228 assert(0);
229 }
230 }
231 }
232
233
234
235 static INLINE int subpixel_snap( float a )
236 {
237 return util_iround(FIXED_ONE * a - (FIXED_ONE / 2));
238 }
239
240
241
242 /**
243 * Alloc space for a new triangle plus the input.a0/dadx/dady arrays
244 * immediately after it.
245 * The memory is allocated from the per-scene pool, not per-tile.
246 * \param tri_size returns number of bytes allocated
247 * \param nr_inputs number of fragment shader inputs
248 * \return pointer to triangle space
249 */
250 static INLINE struct lp_rast_triangle *
251 alloc_triangle(struct lp_scene *scene, unsigned nr_inputs, unsigned *tri_size)
252 {
253 unsigned input_array_sz = NUM_CHANNELS * (nr_inputs + 1) * sizeof(float);
254 struct lp_rast_triangle *tri;
255 unsigned bytes;
256 char *inputs;
257
258 assert(sizeof(*tri) % 16 == 0);
259
260 bytes = sizeof(*tri) + (3 * input_array_sz);
261
262 tri = lp_scene_alloc_aligned( scene, bytes, 16 );
263
264 inputs = (char *) (tri + 1);
265 tri->inputs.a0 = (float (*)[4]) inputs;
266 tri->inputs.dadx = (float (*)[4]) (inputs + input_array_sz);
267 tri->inputs.dady = (float (*)[4]) (inputs + 2 * input_array_sz);
268
269 *tri_size = bytes;
270
271 return tri;
272 }
273
274
275
276 /**
277 * Do basic setup for triangle rasterization and determine which
278 * framebuffer tiles are touched. Put the triangle in the scene's
279 * bins for the tiles which we overlap.
280 */
281 static void
282 do_triangle_ccw(struct lp_setup_context *setup,
283 const float (*v1)[4],
284 const float (*v2)[4],
285 const float (*v3)[4],
286 boolean frontfacing )
287 {
288 /* x/y positions in fixed point */
289 const int x1 = subpixel_snap(v1[0][0] + 0.5 - setup->pixel_offset);
290 const int x2 = subpixel_snap(v2[0][0] + 0.5 - setup->pixel_offset);
291 const int x3 = subpixel_snap(v3[0][0] + 0.5 - setup->pixel_offset);
292 const int y1 = subpixel_snap(v1[0][1] + 0.5 - setup->pixel_offset);
293 const int y2 = subpixel_snap(v2[0][1] + 0.5 - setup->pixel_offset);
294 const int y3 = subpixel_snap(v3[0][1] + 0.5 - setup->pixel_offset);
295
296 struct lp_scene *scene = lp_setup_get_current_scene(setup);
297 struct lp_rast_triangle *tri;
298 int area;
299 float oneoverarea;
300 int minx, maxx, miny, maxy;
301 unsigned tri_bytes;
302
303 tri = alloc_triangle(scene, setup->fs.nr_inputs, &tri_bytes);
304
305 #ifdef DEBUG
306 tri->v[0][0] = v1[0][0];
307 tri->v[1][0] = v2[0][0];
308 tri->v[2][0] = v3[0][0];
309 tri->v[0][1] = v1[0][1];
310 tri->v[1][1] = v2[0][1];
311 tri->v[2][1] = v3[0][1];
312 #endif
313
314 tri->dx12 = x1 - x2;
315 tri->dx23 = x2 - x3;
316 tri->dx31 = x3 - x1;
317
318 tri->dy12 = y1 - y2;
319 tri->dy23 = y2 - y3;
320 tri->dy31 = y3 - y1;
321
322 area = (tri->dx12 * tri->dy31 - tri->dx31 * tri->dy12);
323
324 LP_COUNT(nr_tris);
325
326 /* Cull non-ccw and zero-sized triangles.
327 *
328 * XXX: subject to overflow??
329 */
330 if (area <= 0) {
331 lp_scene_putback_data( scene, tri_bytes );
332 LP_COUNT(nr_culled_tris);
333 return;
334 }
335
336 /* Bounding rectangle (in pixels) */
337 minx = (MIN3(x1, x2, x3) + (FIXED_ONE-1)) >> FIXED_ORDER;
338 maxx = (MAX3(x1, x2, x3) + (FIXED_ONE-1)) >> FIXED_ORDER;
339 miny = (MIN3(y1, y2, y3) + (FIXED_ONE-1)) >> FIXED_ORDER;
340 maxy = (MAX3(y1, y2, y3) + (FIXED_ONE-1)) >> FIXED_ORDER;
341
342 if (setup->scissor_test) {
343 minx = MAX2(minx, setup->scissor.current.minx);
344 maxx = MIN2(maxx, setup->scissor.current.maxx);
345 miny = MAX2(miny, setup->scissor.current.miny);
346 maxy = MIN2(maxy, setup->scissor.current.maxy);
347 }
348
349 if (miny == maxy ||
350 minx == maxx) {
351 lp_scene_putback_data( scene, tri_bytes );
352 LP_COUNT(nr_culled_tris);
353 return;
354 }
355
356 /*
357 */
358 oneoverarea = ((float)FIXED_ONE) / (float)area;
359
360 /* Setup parameter interpolants:
361 */
362 setup_tri_coefficients( setup, tri, oneoverarea, v1, v2, v3, frontfacing );
363
364 /* half-edge constants, will be interated over the whole render target.
365 */
366 tri->c1 = tri->dy12 * x1 - tri->dx12 * y1;
367 tri->c2 = tri->dy23 * x2 - tri->dx23 * y2;
368 tri->c3 = tri->dy31 * x3 - tri->dx31 * y3;
369
370 /* correct for top-left fill convention:
371 */
372 if (tri->dy12 < 0 || (tri->dy12 == 0 && tri->dx12 > 0)) tri->c1++;
373 if (tri->dy23 < 0 || (tri->dy23 == 0 && tri->dx23 > 0)) tri->c2++;
374 if (tri->dy31 < 0 || (tri->dy31 == 0 && tri->dx31 > 0)) tri->c3++;
375
376 tri->dy12 *= FIXED_ONE;
377 tri->dy23 *= FIXED_ONE;
378 tri->dy31 *= FIXED_ONE;
379
380 tri->dx12 *= FIXED_ONE;
381 tri->dx23 *= FIXED_ONE;
382 tri->dx31 *= FIXED_ONE;
383
384 /* find trivial reject offsets for each edge for a single-pixel
385 * sized block. These will be scaled up at each recursive level to
386 * match the active blocksize. Scaling in this way works best if
387 * the blocks are square.
388 */
389 tri->eo1 = 0;
390 if (tri->dy12 < 0) tri->eo1 -= tri->dy12;
391 if (tri->dx12 > 0) tri->eo1 += tri->dx12;
392
393 tri->eo2 = 0;
394 if (tri->dy23 < 0) tri->eo2 -= tri->dy23;
395 if (tri->dx23 > 0) tri->eo2 += tri->dx23;
396
397 tri->eo3 = 0;
398 if (tri->dy31 < 0) tri->eo3 -= tri->dy31;
399 if (tri->dx31 > 0) tri->eo3 += tri->dx31;
400
401 /* Calculate trivial accept offsets from the above.
402 */
403 tri->ei1 = tri->dx12 - tri->dy12 - tri->eo1;
404 tri->ei2 = tri->dx23 - tri->dy23 - tri->eo2;
405 tri->ei3 = tri->dx31 - tri->dy31 - tri->eo3;
406
407 /* Fill in the inputs.step[][] arrays.
408 * We've manually unrolled some loops here.
409 */
410 {
411 const int xstep1 = -tri->dy12;
412 const int xstep2 = -tri->dy23;
413 const int xstep3 = -tri->dy31;
414 const int ystep1 = tri->dx12;
415 const int ystep2 = tri->dx23;
416 const int ystep3 = tri->dx31;
417
418 #define SETUP_STEP(i, x, y) \
419 do { \
420 tri->inputs.step[0][i] = x * xstep1 + y * ystep1; \
421 tri->inputs.step[1][i] = x * xstep2 + y * ystep2; \
422 tri->inputs.step[2][i] = x * xstep3 + y * ystep3; \
423 } while (0)
424
425 SETUP_STEP(0, 0, 0);
426 SETUP_STEP(1, 1, 0);
427 SETUP_STEP(2, 0, 1);
428 SETUP_STEP(3, 1, 1);
429
430 SETUP_STEP(4, 2, 0);
431 SETUP_STEP(5, 3, 0);
432 SETUP_STEP(6, 2, 1);
433 SETUP_STEP(7, 3, 1);
434
435 SETUP_STEP(8, 0, 2);
436 SETUP_STEP(9, 1, 2);
437 SETUP_STEP(10, 0, 3);
438 SETUP_STEP(11, 1, 3);
439
440 SETUP_STEP(12, 2, 2);
441 SETUP_STEP(13, 3, 2);
442 SETUP_STEP(14, 2, 3);
443 SETUP_STEP(15, 3, 3);
444 #undef STEP
445 }
446
447 /*
448 * All fields of 'tri' are now set. The remaining code here is
449 * concerned with binning.
450 */
451
452 /* Convert to tile coordinates:
453 */
454 minx = minx / TILE_SIZE;
455 miny = miny / TILE_SIZE;
456 maxx = maxx / TILE_SIZE;
457 maxy = maxy / TILE_SIZE;
458
459 /*
460 * Clamp to framebuffer size
461 */
462 minx = MAX2(minx, 0);
463 miny = MAX2(miny, 0);
464 maxx = MIN2(maxx, scene->tiles_x - 1);
465 maxy = MIN2(maxy, scene->tiles_y - 1);
466
467 /* Determine which tile(s) intersect the triangle's bounding box
468 */
469 if (miny == maxy && minx == maxx)
470 {
471 /* Triangle is contained in a single tile:
472 */
473 lp_scene_bin_command( scene, minx, miny, lp_rast_triangle,
474 lp_rast_arg_triangle(tri) );
475 }
476 else
477 {
478 int c1 = (tri->c1 +
479 tri->dx12 * miny * TILE_SIZE -
480 tri->dy12 * minx * TILE_SIZE);
481 int c2 = (tri->c2 +
482 tri->dx23 * miny * TILE_SIZE -
483 tri->dy23 * minx * TILE_SIZE);
484 int c3 = (tri->c3 +
485 tri->dx31 * miny * TILE_SIZE -
486 tri->dy31 * minx * TILE_SIZE);
487
488 int ei1 = tri->ei1 << TILE_ORDER;
489 int ei2 = tri->ei2 << TILE_ORDER;
490 int ei3 = tri->ei3 << TILE_ORDER;
491
492 int eo1 = tri->eo1 << TILE_ORDER;
493 int eo2 = tri->eo2 << TILE_ORDER;
494 int eo3 = tri->eo3 << TILE_ORDER;
495
496 int xstep1 = -(tri->dy12 << TILE_ORDER);
497 int xstep2 = -(tri->dy23 << TILE_ORDER);
498 int xstep3 = -(tri->dy31 << TILE_ORDER);
499
500 int ystep1 = tri->dx12 << TILE_ORDER;
501 int ystep2 = tri->dx23 << TILE_ORDER;
502 int ystep3 = tri->dx31 << TILE_ORDER;
503 int x, y;
504
505
506 /* Test tile-sized blocks against the triangle.
507 * Discard blocks fully outside the tri. If the block is fully
508 * contained inside the tri, bin an lp_rast_shade_tile command.
509 * Else, bin a lp_rast_triangle command.
510 */
511 for (y = miny; y <= maxy; y++)
512 {
513 int cx1 = c1;
514 int cx2 = c2;
515 int cx3 = c3;
516 boolean in = FALSE; /* are we inside the triangle? */
517
518 for (x = minx; x <= maxx; x++)
519 {
520 if (cx1 + eo1 < 0 ||
521 cx2 + eo2 < 0 ||
522 cx3 + eo3 < 0)
523 {
524 /* do nothing */
525 LP_COUNT(nr_empty_64);
526 if (in)
527 break; /* exiting triangle, all done with this row */
528 }
529 else if (cx1 + ei1 > 0 &&
530 cx2 + ei2 > 0 &&
531 cx3 + ei3 > 0)
532 {
533 /* triangle covers the whole tile- shade whole tile */
534 LP_COUNT(nr_fully_covered_64);
535 in = TRUE;
536 if(setup->fs.current.opaque) {
537 lp_scene_bin_reset( scene, x, y );
538 lp_scene_bin_command( scene, x, y,
539 lp_rast_set_state,
540 lp_rast_arg_state(setup->fs.stored) );
541 }
542 lp_scene_bin_command( scene, x, y,
543 lp_rast_shade_tile,
544 lp_rast_arg_inputs(&tri->inputs) );
545 }
546 else
547 {
548 /* rasterizer/shade partial tile */
549 LP_COUNT(nr_partially_covered_64);
550 in = TRUE;
551 lp_scene_bin_command( scene, x, y,
552 lp_rast_triangle,
553 lp_rast_arg_triangle(tri) );
554 }
555
556 /* Iterate cx values across the region:
557 */
558 cx1 += xstep1;
559 cx2 += xstep2;
560 cx3 += xstep3;
561 }
562
563 /* Iterate c values down the region:
564 */
565 c1 += ystep1;
566 c2 += ystep2;
567 c3 += ystep3;
568 }
569 }
570 }
571
572
573 static void triangle_cw( struct lp_setup_context *setup,
574 const float (*v0)[4],
575 const float (*v1)[4],
576 const float (*v2)[4] )
577 {
578 do_triangle_ccw( setup, v1, v0, v2, !setup->ccw_is_frontface );
579 }
580
581
582 static void triangle_ccw( struct lp_setup_context *setup,
583 const float (*v0)[4],
584 const float (*v1)[4],
585 const float (*v2)[4] )
586 {
587 do_triangle_ccw( setup, v0, v1, v2, setup->ccw_is_frontface );
588 }
589
590
591 static void triangle_both( struct lp_setup_context *setup,
592 const float (*v0)[4],
593 const float (*v1)[4],
594 const float (*v2)[4] )
595 {
596 /* edge vectors e = v0 - v2, f = v1 - v2 */
597 const float ex = v0[0][0] - v2[0][0];
598 const float ey = v0[0][1] - v2[0][1];
599 const float fx = v1[0][0] - v2[0][0];
600 const float fy = v1[0][1] - v2[0][1];
601
602 /* det = cross(e,f).z */
603 if (ex * fy - ey * fx < 0.0f)
604 triangle_ccw( setup, v0, v1, v2 );
605 else
606 triangle_cw( setup, v0, v1, v2 );
607 }
608
609
610 static void triangle_nop( struct lp_setup_context *setup,
611 const float (*v0)[4],
612 const float (*v1)[4],
613 const float (*v2)[4] )
614 {
615 }
616
617
618 void
619 lp_setup_choose_triangle( struct lp_setup_context *setup )
620 {
621 switch (setup->cullmode) {
622 case PIPE_WINDING_NONE:
623 setup->triangle = triangle_both;
624 break;
625 case PIPE_WINDING_CCW:
626 setup->triangle = triangle_cw;
627 break;
628 case PIPE_WINDING_CW:
629 setup->triangle = triangle_ccw;
630 break;
631 default:
632 setup->triangle = triangle_nop;
633 break;
634 }
635 }