1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
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:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
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.
26 **************************************************************************/
29 * Binning code for lines
32 #include "util/u_math.h"
33 #include "util/u_memory.h"
35 #include "lp_setup_context.h"
37 #include "lp_state_fs.h"
38 #include "lp_state_setup.h"
40 #define NUM_CHANNELS 4
58 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
60 static void constant_coef( struct lp_setup_context *setup,
61 struct lp_line_info *info,
66 info->a0[slot][i] = value;
67 info->dadx[slot][i] = 0.0f;
68 info->dady[slot][i] = 0.0f;
73 * Compute a0, dadx and dady for a linearly interpolated coefficient,
76 static void linear_coef( struct lp_setup_context *setup,
77 struct lp_line_info *info,
82 float a1 = info->v1[vert_attr][i];
83 float a2 = info->v2[vert_attr][i];
86 float dadx = da21 * info->dx * info->oneoverarea;
87 float dady = da21 * info->dy * info->oneoverarea;
89 info->dadx[slot][i] = dadx;
90 info->dady[slot][i] = dady;
92 info->a0[slot][i] = (a1 -
93 (dadx * (info->v1[0][0] - setup->pixel_offset) +
94 dady * (info->v1[0][1] - setup->pixel_offset)));
99 * Compute a0, dadx and dady for a perspective-corrected interpolant,
101 * We basically multiply the vertex value by 1/w before computing
102 * the plane coefficients (a0, dadx, dady).
103 * Later, when we compute the value at a particular fragment position we'll
104 * divide the interpolated value by the interpolated W at that fragment.
106 static void perspective_coef( struct lp_setup_context *setup,
107 struct lp_line_info *info,
112 /* premultiply by 1/w (v[0][3] is always 1/w):
114 float a1 = info->v1[vert_attr][i] * info->v1[0][3];
115 float a2 = info->v2[vert_attr][i] * info->v2[0][3];
117 float da21 = a1 - a2;
118 float dadx = da21 * info->dx * info->oneoverarea;
119 float dady = da21 * info->dy * info->oneoverarea;
121 info->dadx[slot][i] = dadx;
122 info->dady[slot][i] = dady;
124 info->a0[slot][i] = (a1 -
125 (dadx * (info->v1[0][0] - setup->pixel_offset) +
126 dady * (info->v1[0][1] - setup->pixel_offset)));
130 setup_fragcoord_coef( struct lp_setup_context *setup,
131 struct lp_line_info *info,
136 if (usage_mask & TGSI_WRITEMASK_X) {
137 info->a0[slot][0] = 0.0;
138 info->dadx[slot][0] = 1.0;
139 info->dady[slot][0] = 0.0;
143 if (usage_mask & TGSI_WRITEMASK_Y) {
144 info->a0[slot][1] = 0.0;
145 info->dadx[slot][1] = 0.0;
146 info->dady[slot][1] = 1.0;
150 if (usage_mask & TGSI_WRITEMASK_Z) {
151 linear_coef(setup, info, slot, 0, 2);
155 if (usage_mask & TGSI_WRITEMASK_W) {
156 linear_coef(setup, info, slot, 0, 3);
161 * Compute the tri->coef[] array dadx, dady, a0 values.
163 static void setup_line_coefficients( struct lp_setup_context *setup,
164 struct lp_line_info *info)
166 const struct lp_setup_variant_key *key = &setup->setup.variant->key;
167 unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ;
170 /* setup interpolation for all the remaining attributes:
172 for (slot = 0; slot < key->num_inputs; slot++) {
173 unsigned vert_attr = key->inputs[slot].src_index;
174 unsigned usage_mask = key->inputs[slot].usage_mask;
177 switch (key->inputs[slot].interp) {
178 case LP_INTERP_CONSTANT:
179 if (key->flatshade_first) {
180 for (i = 0; i < NUM_CHANNELS; i++)
181 if (usage_mask & (1 << i))
182 constant_coef(setup, info, slot+1, info->v1[vert_attr][i], i);
185 for (i = 0; i < NUM_CHANNELS; i++)
186 if (usage_mask & (1 << i))
187 constant_coef(setup, info, slot+1, info->v2[vert_attr][i], i);
191 case LP_INTERP_LINEAR:
192 for (i = 0; i < NUM_CHANNELS; i++)
193 if (usage_mask & (1 << i))
194 linear_coef(setup, info, slot+1, vert_attr, i);
197 case LP_INTERP_PERSPECTIVE:
198 for (i = 0; i < NUM_CHANNELS; i++)
199 if (usage_mask & (1 << i))
200 perspective_coef(setup, info, slot+1, vert_attr, i);
201 fragcoord_usage_mask |= TGSI_WRITEMASK_W;
204 case LP_INTERP_POSITION:
206 * The generated pixel interpolators will pick up the coeffs from
207 * slot 0, so all need to ensure that the usage mask is covers all
210 fragcoord_usage_mask |= usage_mask;
213 case LP_INTERP_FACING:
214 for (i = 0; i < NUM_CHANNELS; i++)
215 if (usage_mask & (1 << i))
216 constant_coef(setup, info, slot+1, 1.0, i);
224 /* The internal position input is in slot zero:
226 setup_fragcoord_coef(setup, info, 0,
227 fragcoord_usage_mask);
232 static INLINE int subpixel_snap( float a )
234 return util_iround(FIXED_ONE * a);
239 * Print line vertex attribs (for debug).
242 print_line(struct lp_setup_context *setup,
243 const float (*v1)[4],
244 const float (*v2)[4])
246 const struct lp_setup_variant_key *key = &setup->setup.variant->key;
249 debug_printf("llvmpipe line\n");
250 for (i = 0; i < 1 + key->num_inputs; i++) {
251 debug_printf(" v1[%d]: %f %f %f %f\n", i,
252 v1[i][0], v1[i][1], v1[i][2], v1[i][3]);
254 for (i = 0; i < 1 + key->num_inputs; i++) {
255 debug_printf(" v2[%d]: %f %f %f %f\n", i,
256 v2[i][0], v2[i][1], v2[i][2], v2[i][3]);
261 static INLINE boolean sign(float x){
266 /* Used on positive floats only:
268 static INLINE float fracf(float f)
270 return f - floorf(f);
276 try_setup_line( struct lp_setup_context *setup,
277 const float (*v1)[4],
278 const float (*v2)[4])
280 struct lp_scene *scene = setup->scene;
281 const struct lp_setup_variant_key *key = &setup->setup.variant->key;
282 struct lp_rast_triangle *line;
283 struct lp_rast_plane *plane;
284 struct lp_line_info info;
285 float width = MAX2(1.0, setup->line_width);
292 unsigned scissor_index = 0;
294 /* linewidth should be interpreted as integer */
295 int fixed_width = util_iround(width) * FIXED_ONE;
299 float x_offset_end=0;
300 float y_offset_end=0;
311 boolean will_draw_start;
312 boolean will_draw_end;
315 print_line(setup, v1, v2);
317 if (setup->scissor_test) {
319 if (setup->viewport_index_slot > 0) {
320 unsigned *udata = (unsigned*)v1[setup->viewport_index_slot];
321 scissor_index = lp_clamp_scissor_idx(*udata);
329 dx = v1[0][0] - v2[0][0];
330 dy = v1[0][1] - v2[0][1];
331 area = (dx * dx + dy * dy);
333 LP_COUNT(nr_culled_tris);
337 info.oneoverarea = 1.0f / area;
345 if (fabsf(dx) >= fabsf(dy)) {
346 float dydx = dy / dx;
348 x1diff = v1[0][0] - (float) floor(v1[0][0]) - 0.5;
349 y1diff = v1[0][1] - (float) floor(v1[0][1]) - 0.5;
350 x2diff = v2[0][0] - (float) floor(v2[0][0]) - 0.5;
351 y2diff = v2[0][1] - (float) floor(v2[0][1]) - 0.5;
353 if (y2diff==-0.5 && dy<0){
358 * Diamond exit rule test for starting point
360 if (fabsf(x1diff) + fabsf(y1diff) < 0.5) {
363 else if (sign(x1diff) == sign(-dx)) {
366 else if (sign(-y1diff) != sign(dy)) {
370 /* do intersection test */
371 float yintersect = fracf(v1[0][1]) + x1diff * dydx;
372 draw_start = (yintersect < 1.0 && yintersect > 0.0);
377 * Diamond exit rule test for ending point
379 if (fabsf(x2diff) + fabsf(y2diff) < 0.5) {
382 else if (sign(x2diff) != sign(-dx)) {
385 else if (sign(-y2diff) == sign(dy)) {
389 /* do intersection test */
390 float yintersect = fracf(v2[0][1]) + x2diff * dydx;
391 draw_end = (yintersect < 1.0 && yintersect > 0.0);
394 /* Are we already drawing start/end?
396 will_draw_start = sign(-x1diff) != sign(dx);
397 will_draw_end = (sign(x2diff) == sign(-dx)) || x2diff==0;
400 /* if v2 is to the right of v1, swap pointers */
401 const float (*temp)[4] = v1;
406 /* Otherwise shift planes appropriately */
407 if (will_draw_start != draw_start) {
408 x_offset_end = - x1diff - 0.5;
409 y_offset_end = x_offset_end * dydx;
412 if (will_draw_end != draw_end) {
413 x_offset = - x2diff - 0.5;
414 y_offset = x_offset * dydx;
419 /* Otherwise shift planes appropriately */
420 if (will_draw_start != draw_start) {
421 x_offset = - x1diff + 0.5;
422 y_offset = x_offset * dydx;
424 if (will_draw_end != draw_end) {
425 x_offset_end = - x2diff + 0.5;
426 y_offset_end = x_offset_end * dydx;
430 /* x/y positions in fixed point */
431 x[0] = subpixel_snap(v1[0][0] + x_offset - setup->pixel_offset);
432 x[1] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset);
433 x[2] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset);
434 x[3] = subpixel_snap(v1[0][0] + x_offset - setup->pixel_offset);
436 y[0] = subpixel_snap(v1[0][1] + y_offset - setup->pixel_offset) - fixed_width/2;
437 y[1] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset) - fixed_width/2;
438 y[2] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset) + fixed_width/2;
439 y[3] = subpixel_snap(v1[0][1] + y_offset - setup->pixel_offset) + fixed_width/2;
443 const float dxdy = dx / dy;
446 x1diff = v1[0][0] - (float) floor(v1[0][0]) - 0.5;
447 y1diff = v1[0][1] - (float) floor(v1[0][1]) - 0.5;
448 x2diff = v2[0][0] - (float) floor(v2[0][0]) - 0.5;
449 y2diff = v2[0][1] - (float) floor(v2[0][1]) - 0.5;
451 if (x2diff==-0.5 && dx<0) {
456 * Diamond exit rule test for starting point
458 if (fabsf(x1diff) + fabsf(y1diff) < 0.5) {
461 else if (sign(-y1diff) == sign(dy)) {
464 else if (sign(x1diff) != sign(-dx)) {
468 /* do intersection test */
469 float xintersect = fracf(v1[0][0]) + y1diff * dxdy;
470 draw_start = (xintersect < 1.0 && xintersect > 0.0);
474 * Diamond exit rule test for ending point
476 if (fabsf(x2diff) + fabsf(y2diff) < 0.5) {
479 else if (sign(-y2diff) != sign(dy) ) {
482 else if (sign(x2diff) == sign(-dx) ) {
486 /* do intersection test */
487 float xintersect = fracf(v2[0][0]) + y2diff * dxdy;
488 draw_end = (xintersect < 1.0 && xintersect >= 0.0);
491 /* Are we already drawing start/end?
493 will_draw_start = sign(y1diff) == sign(dy);
494 will_draw_end = (sign(-y2diff) == sign(dy)) || y2diff==0;
497 /* if v2 is on top of v1, swap pointers */
498 const float (*temp)[4] = v1;
504 /* Otherwise shift planes appropriately */
505 if (will_draw_start != draw_start) {
506 y_offset_end = - y1diff + 0.5;
507 x_offset_end = y_offset_end * dxdy;
509 if (will_draw_end != draw_end) {
510 y_offset = - y2diff + 0.5;
511 x_offset = y_offset * dxdy;
515 /* Otherwise shift planes appropriately */
516 if (will_draw_start != draw_start) {
517 y_offset = - y1diff - 0.5;
518 x_offset = y_offset * dxdy;
521 if (will_draw_end != draw_end) {
522 y_offset_end = - y2diff - 0.5;
523 x_offset_end = y_offset_end * dxdy;
527 /* x/y positions in fixed point */
528 x[0] = subpixel_snap(v1[0][0] + x_offset - setup->pixel_offset) - fixed_width/2;
529 x[1] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset) - fixed_width/2;
530 x[2] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset) + fixed_width/2;
531 x[3] = subpixel_snap(v1[0][0] + x_offset - setup->pixel_offset) + fixed_width/2;
533 y[0] = subpixel_snap(v1[0][1] + y_offset - setup->pixel_offset);
534 y[1] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset);
535 y[2] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset);
536 y[3] = subpixel_snap(v1[0][1] + y_offset - setup->pixel_offset);
544 /* Bounding rectangle (in pixels) */
546 /* Yes this is necessary to accurately calculate bounding boxes
547 * with the two fill-conventions we support. GL (normally) ends
548 * up needing a bottom-left fill convention, which requires
549 * slightly different rounding.
551 int adj = (setup->pixel_offset != 0) ? 1 : 0;
553 bbox.x0 = (MIN4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER;
554 bbox.x1 = (MAX4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER;
555 bbox.y0 = (MIN4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
556 bbox.y1 = (MAX4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
558 /* Inclusive coordinates:
564 if (bbox.x1 < bbox.x0 ||
566 if (0) debug_printf("empty bounding box\n");
567 LP_COUNT(nr_culled_tris);
571 if (!u_rect_test_intersection(&setup->draw_regions[scissor_index], &bbox)) {
572 if (0) debug_printf("offscreen\n");
573 LP_COUNT(nr_culled_tris);
577 /* Can safely discard negative regions:
579 bbox.x0 = MAX2(bbox.x0, 0);
580 bbox.y0 = MAX2(bbox.y0, 0);
582 line = lp_setup_alloc_triangle(scene,
590 line->v[0][0] = v1[0][0];
591 line->v[1][0] = v2[0][0];
592 line->v[0][1] = v1[0][1];
593 line->v[1][1] = v2[0][1];
596 /* calculate the deltas */
597 plane = GET_PLANES(line);
598 plane[0].dcdy = x[0] - x[1];
599 plane[1].dcdy = x[1] - x[2];
600 plane[2].dcdy = x[2] - x[3];
601 plane[3].dcdy = x[3] - x[0];
603 plane[0].dcdx = y[0] - y[1];
604 plane[1].dcdx = y[1] - y[2];
605 plane[2].dcdx = y[2] - y[3];
606 plane[3].dcdx = y[3] - y[0];
609 /* Setup parameter interpolants:
611 info.a0 = GET_A0(&line->inputs);
612 info.dadx = GET_DADX(&line->inputs);
613 info.dady = GET_DADY(&line->inputs);
614 setup_line_coefficients(setup, &info);
616 line->inputs.frontfacing = TRUE;
617 line->inputs.disable = FALSE;
618 line->inputs.opaque = FALSE;
620 for (i = 0; i < 4; i++) {
622 /* half-edge constants, will be interated over the whole render
625 plane[i].c = plane[i].dcdx * x[i] - plane[i].dcdy * y[i];
628 /* correct for top-left vs. bottom-left fill convention.
630 if (plane[i].dcdx < 0) {
631 /* both fill conventions want this - adjust for left edges */
634 else if (plane[i].dcdx == 0) {
635 if (setup->pixel_offset == 0) {
636 /* correct for top-left fill convention:
638 if (plane[i].dcdy > 0) plane[i].c++;
641 /* correct for bottom-left fill convention:
643 if (plane[i].dcdy < 0) plane[i].c++;
647 plane[i].dcdx *= FIXED_ONE;
648 plane[i].dcdy *= FIXED_ONE;
650 /* find trivial reject offsets for each edge for a single-pixel
651 * sized block. These will be scaled up at each recursive level to
652 * match the active blocksize. Scaling in this way works best if
653 * the blocks are square.
656 if (plane[i].dcdx < 0) plane[i].eo -= plane[i].dcdx;
657 if (plane[i].dcdy > 0) plane[i].eo += plane[i].dcdy;
662 * When rasterizing scissored tris, use the intersection of the
663 * triangle bounding box and the scissor rect to generate the
666 * This permits us to cut off the triangle "tails" that are present
667 * in the intermediate recursive levels caused when two of the
668 * triangles edges don't diverge quickly enough to trivially reject
669 * exterior blocks from the triangle.
671 * It's not really clear if it's worth worrying about these tails,
672 * but since we generate the planes for each scissored tri, it's
673 * free to trim them in this case.
675 * Note that otherwise, the scissor planes only vary in 'C' value,
676 * and even then only on state-changes. Could alternatively store
677 * these planes elsewhere.
679 if (nr_planes == 8) {
680 const struct u_rect *scissor =
681 &setup->scissors[scissor_index];
685 plane[4].c = 1-scissor->x0;
690 plane[5].c = scissor->x1+1;
695 plane[6].c = 1-scissor->y0;
700 plane[7].c = scissor->y1+1;
704 return lp_setup_bin_triangle(setup, line, &bbox, nr_planes, scissor_index);
708 static void lp_setup_line( struct lp_setup_context *setup,
709 const float (*v0)[4],
710 const float (*v1)[4] )
712 if (!try_setup_line( setup, v0, v1 ))
714 if (!lp_setup_flush_and_restart(setup))
717 if (!try_setup_line( setup, v0, v1 ))
723 void lp_setup_choose_line( struct lp_setup_context *setup )
725 setup->line = lp_setup_line;