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 triangles
32 #include "util/u_math.h"
33 #include "util/u_memory.h"
34 #include "util/u_rect.h"
36 #include "lp_setup_context.h"
37 #include "lp_setup_coef.h"
39 #include "lp_state_fs.h"
41 #define NUM_CHANNELS 4
46 subpixel_snap(float a
)
48 return util_iround(FIXED_ONE
* a
);
54 return a
* (1.0 / FIXED_ONE
);
64 * Alloc space for a new triangle plus the input.a0/dadx/dady arrays
65 * immediately after it.
66 * The memory is allocated from the per-scene pool, not per-tile.
67 * \param tri_size returns number of bytes allocated
68 * \param nr_inputs number of fragment shader inputs
69 * \return pointer to triangle space
71 struct lp_rast_triangle
*
72 lp_setup_alloc_triangle(struct lp_scene
*scene
,
77 unsigned input_array_sz
= NUM_CHANNELS
* (nr_inputs
+ 1) * sizeof(float);
78 struct lp_rast_triangle
*tri
;
79 unsigned tri_bytes
, bytes
;
82 tri_bytes
= align(Offset(struct lp_rast_triangle
, plane
[nr_planes
]), 16);
83 bytes
= tri_bytes
+ (3 * input_array_sz
);
85 tri
= lp_scene_alloc_aligned( scene
, bytes
, 16 );
88 inputs
= ((char *)tri
) + tri_bytes
;
89 tri
->inputs
.a0
= (float (*)[4]) inputs
;
90 tri
->inputs
.dadx
= (float (*)[4]) (inputs
+ input_array_sz
);
91 tri
->inputs
.dady
= (float (*)[4]) (inputs
+ 2 * input_array_sz
);
100 lp_setup_print_vertex(struct lp_setup_context
*setup
,
106 debug_printf(" wpos (%s[0]) xyzw %f %f %f %f\n",
108 v
[0][0], v
[0][1], v
[0][2], v
[0][3]);
110 for (i
= 0; i
< setup
->fs
.nr_inputs
; i
++) {
111 const float *in
= v
[setup
->fs
.input
[i
].src_index
];
113 debug_printf(" in[%d] (%s[%d]) %s%s%s%s ",
115 name
, setup
->fs
.input
[i
].src_index
,
116 (setup
->fs
.input
[i
].usage_mask
& 0x1) ? "x" : " ",
117 (setup
->fs
.input
[i
].usage_mask
& 0x2) ? "y" : " ",
118 (setup
->fs
.input
[i
].usage_mask
& 0x4) ? "z" : " ",
119 (setup
->fs
.input
[i
].usage_mask
& 0x8) ? "w" : " ");
121 for (j
= 0; j
< 4; j
++)
122 if (setup
->fs
.input
[i
].usage_mask
& (1<<j
))
123 debug_printf("%.5f ", in
[j
]);
131 * Print triangle vertex attribs (for debug).
134 lp_setup_print_triangle(struct lp_setup_context
*setup
,
135 const float (*v0
)[4],
136 const float (*v1
)[4],
137 const float (*v2
)[4])
139 debug_printf("triangle\n");
142 const float ex
= v0
[0][0] - v2
[0][0];
143 const float ey
= v0
[0][1] - v2
[0][1];
144 const float fx
= v1
[0][0] - v2
[0][0];
145 const float fy
= v1
[0][1] - v2
[0][1];
147 /* det = cross(e,f).z */
148 const float det
= ex
* fy
- ey
* fx
;
150 debug_printf(" - ccw\n");
152 debug_printf(" - cw\n");
154 debug_printf(" - zero area\n");
157 lp_setup_print_vertex(setup
, "v0", v0
);
158 lp_setup_print_vertex(setup
, "v1", v1
);
159 lp_setup_print_vertex(setup
, "v2", v2
);
163 lp_rast_cmd lp_rast_tri_tab
[9] = {
164 NULL
, /* should be impossible */
178 * The primitive covers the whole tile- shade whole tile.
180 * \param tx, ty the tile position in tiles, not pixels
183 lp_setup_whole_tile(struct lp_setup_context
*setup
,
184 const struct lp_rast_shader_inputs
*inputs
,
187 struct lp_scene
*scene
= setup
->scene
;
189 LP_COUNT(nr_fully_covered_64
);
191 /* if variant is opaque and scissor doesn't effect the tile */
192 if (inputs
->opaque
) {
193 if (!scene
->fb
.zsbuf
) {
195 * All previous rendering will be overwritten so reset the bin.
197 lp_scene_bin_reset( scene
, tx
, ty
);
200 LP_COUNT(nr_shade_opaque_64
);
201 lp_scene_bin_command( scene
, tx
, ty
,
202 lp_rast_shade_tile_opaque
,
203 lp_rast_arg_inputs(inputs
) );
205 LP_COUNT(nr_shade_64
);
206 lp_scene_bin_command( scene
, tx
, ty
,
208 lp_rast_arg_inputs(inputs
) );
214 * Do basic setup for triangle rasterization and determine which
215 * framebuffer tiles are touched. Put the triangle in the scene's
216 * bins for the tiles which we overlap.
219 do_triangle_ccw(struct lp_setup_context
*setup
,
220 const float (*v0
)[4],
221 const float (*v1
)[4],
222 const float (*v2
)[4],
223 boolean frontfacing
)
225 struct lp_scene
*scene
= lp_setup_get_current_scene(setup
);
226 struct lp_fragment_shader_variant
*variant
= setup
->fs
.current
.variant
;
227 struct lp_rast_triangle
*tri
;
233 struct lp_tri_info info
;
241 lp_setup_print_triangle(setup
, v0
, v1
, v2
);
243 if (setup
->scissor_test
) {
250 /* x/y positions in fixed point */
251 x
[0] = subpixel_snap(v0
[0][0] - setup
->pixel_offset
);
252 x
[1] = subpixel_snap(v1
[0][0] - setup
->pixel_offset
);
253 x
[2] = subpixel_snap(v2
[0][0] - setup
->pixel_offset
);
254 y
[0] = subpixel_snap(v0
[0][1] - setup
->pixel_offset
);
255 y
[1] = subpixel_snap(v1
[0][1] - setup
->pixel_offset
);
256 y
[2] = subpixel_snap(v2
[0][1] - setup
->pixel_offset
);
259 /* Bounding rectangle (in pixels) */
261 /* Yes this is necessary to accurately calculate bounding boxes
262 * with the two fill-conventions we support. GL (normally) ends
263 * up needing a bottom-left fill convention, which requires
264 * slightly different rounding.
266 int adj
= (setup
->pixel_offset
!= 0) ? 1 : 0;
268 bbox
.x0
= (MIN3(x
[0], x
[1], x
[2]) + (FIXED_ONE
-1)) >> FIXED_ORDER
;
269 bbox
.x1
= (MAX3(x
[0], x
[1], x
[2]) + (FIXED_ONE
-1)) >> FIXED_ORDER
;
270 bbox
.y0
= (MIN3(y
[0], y
[1], y
[2]) + (FIXED_ONE
-1) + adj
) >> FIXED_ORDER
;
271 bbox
.y1
= (MAX3(y
[0], y
[1], y
[2]) + (FIXED_ONE
-1) + adj
) >> FIXED_ORDER
;
273 /* Inclusive coordinates:
279 if (bbox
.x1
< bbox
.x0
||
281 if (0) debug_printf("empty bounding box\n");
282 LP_COUNT(nr_culled_tris
);
286 if (!u_rect_test_intersection(&setup
->draw_region
, &bbox
)) {
287 if (0) debug_printf("offscreen\n");
288 LP_COUNT(nr_culled_tris
);
292 u_rect_find_intersection(&setup
->draw_region
, &bbox
);
294 tri
= lp_setup_alloc_triangle(scene
,
302 tri
->v
[0][0] = v0
[0][0];
303 tri
->v
[1][0] = v1
[0][0];
304 tri
->v
[2][0] = v2
[0][0];
305 tri
->v
[0][1] = v0
[0][1];
306 tri
->v
[1][1] = v1
[0][1];
307 tri
->v
[2][1] = v2
[0][1];
310 tri
->plane
[0].dcdy
= x
[0] - x
[1];
311 tri
->plane
[1].dcdy
= x
[1] - x
[2];
312 tri
->plane
[2].dcdy
= x
[2] - x
[0];
314 tri
->plane
[0].dcdx
= y
[0] - y
[1];
315 tri
->plane
[1].dcdx
= y
[1] - y
[2];
316 tri
->plane
[2].dcdx
= y
[2] - y
[0];
318 area
= (tri
->plane
[0].dcdy
* tri
->plane
[2].dcdx
-
319 tri
->plane
[2].dcdy
* tri
->plane
[0].dcdx
);
323 /* Cull non-ccw and zero-sized triangles.
325 * XXX: subject to overflow??
328 lp_scene_putback_data( scene
, tri_bytes
);
329 LP_COUNT(nr_culled_tris
);
336 dx01
= v0
[0][0] - v1
[0][0];
337 dy01
= v0
[0][1] - v1
[0][1];
338 dx20
= v2
[0][0] - v0
[0][0];
339 dy20
= v2
[0][1] - v0
[0][1];
340 oneoverarea
= 1.0f
/ (dx01
* dy20
- dx20
* dy01
);
345 info
.frontfacing
= frontfacing
;
346 info
.x0_center
= v0
[0][0] - setup
->pixel_offset
;
347 info
.y0_center
= v0
[0][1] - setup
->pixel_offset
;
348 info
.dx01_ooa
= dx01
* oneoverarea
;
349 info
.dx20_ooa
= dx20
* oneoverarea
;
350 info
.dy01_ooa
= dy01
* oneoverarea
;
351 info
.dy20_ooa
= dy20
* oneoverarea
;
353 /* Setup parameter interpolants:
355 lp_setup_tri_coef( setup
, &tri
->inputs
, &info
);
357 tri
->inputs
.facing
= frontfacing
? 1.0F
: -1.0F
;
358 tri
->inputs
.opaque
= variant
->opaque
;
359 tri
->inputs
.state
= setup
->fs
.stored
;
363 for (i
= 0; i
< 3; i
++) {
364 struct lp_rast_plane
*plane
= &tri
->plane
[i
];
366 /* half-edge constants, will be interated over the whole render
369 plane
->c
= plane
->dcdx
* x
[i
] - plane
->dcdy
* y
[i
];
371 /* correct for top-left vs. bottom-left fill convention.
373 * note that we're overloading gl_rasterization_rules to mean
374 * both (0.5,0.5) pixel centers *and* bottom-left filling
377 * GL actually has a top-left filling convention, but GL's
378 * notion of "top" differs from gallium's...
380 * Also, sometimes (in FBO cases) GL will render upside down
381 * to its usual method, in which case it will probably want
382 * to use the opposite, top-left convention.
384 if (plane
->dcdx
< 0) {
385 /* both fill conventions want this - adjust for left edges */
388 else if (plane
->dcdx
== 0) {
389 if (setup
->pixel_offset
== 0) {
390 /* correct for top-left fill convention:
392 if (plane
->dcdy
> 0) plane
->c
++;
395 /* correct for bottom-left fill convention:
397 if (plane
->dcdy
< 0) plane
->c
++;
401 plane
->dcdx
*= FIXED_ONE
;
402 plane
->dcdy
*= FIXED_ONE
;
404 /* find trivial reject offsets for each edge for a single-pixel
405 * sized block. These will be scaled up at each recursive level to
406 * match the active blocksize. Scaling in this way works best if
407 * the blocks are square.
410 if (plane
->dcdx
< 0) plane
->eo
-= plane
->dcdx
;
411 if (plane
->dcdy
> 0) plane
->eo
+= plane
->dcdy
;
413 /* Calculate trivial accept offsets from the above.
415 plane
->ei
= plane
->dcdy
- plane
->dcdx
- plane
->eo
;
420 * When rasterizing scissored tris, use the intersection of the
421 * triangle bounding box and the scissor rect to generate the
424 * This permits us to cut off the triangle "tails" that are present
425 * in the intermediate recursive levels caused when two of the
426 * triangles edges don't diverge quickly enough to trivially reject
427 * exterior blocks from the triangle.
429 * It's not really clear if it's worth worrying about these tails,
430 * but since we generate the planes for each scissored tri, it's
431 * free to trim them in this case.
433 * Note that otherwise, the scissor planes only vary in 'C' value,
434 * and even then only on state-changes. Could alternatively store
435 * these planes elsewhere.
437 if (nr_planes
== 7) {
438 tri
->plane
[3].dcdx
= -1;
439 tri
->plane
[3].dcdy
= 0;
440 tri
->plane
[3].c
= 1-bbox
.x0
;
441 tri
->plane
[3].ei
= 0;
442 tri
->plane
[3].eo
= 1;
444 tri
->plane
[4].dcdx
= 1;
445 tri
->plane
[4].dcdy
= 0;
446 tri
->plane
[4].c
= bbox
.x1
+1;
447 tri
->plane
[4].ei
= -1;
448 tri
->plane
[4].eo
= 0;
450 tri
->plane
[5].dcdx
= 0;
451 tri
->plane
[5].dcdy
= 1;
452 tri
->plane
[5].c
= 1-bbox
.y0
;
453 tri
->plane
[5].ei
= 0;
454 tri
->plane
[5].eo
= 1;
456 tri
->plane
[6].dcdx
= 0;
457 tri
->plane
[6].dcdy
= -1;
458 tri
->plane
[6].c
= bbox
.y1
+1;
459 tri
->plane
[6].ei
= -1;
460 tri
->plane
[6].eo
= 0;
463 lp_setup_bin_triangle( setup
, tri
, &bbox
, nr_planes
);
468 lp_setup_bin_triangle( struct lp_setup_context
*setup
,
469 struct lp_rast_triangle
*tri
,
470 const struct u_rect
*bbox
,
473 struct lp_scene
*scene
= setup
->scene
;
474 int ix0
, ix1
, iy0
, iy1
;
478 * All fields of 'tri' are now set. The remaining code here is
479 * concerned with binning.
482 /* Convert to tile coordinates, and inclusive ranges:
484 if (nr_planes
== 3) {
485 int ix0
= bbox
->x0
/ 16;
486 int iy0
= bbox
->y0
/ 16;
487 int ix1
= bbox
->x1
/ 16;
488 int iy1
= bbox
->y1
/ 16;
490 if (iy0
== iy1
&& ix0
== ix1
)
493 /* Triangle is contained in a single 16x16 block:
495 int mask
= (ix0
& 3) | ((iy0
& 3) << 4);
497 lp_scene_bin_command( scene
, ix0
/4, iy0
/4,
498 lp_rast_triangle_3_16
,
499 lp_rast_arg_triangle(tri
, mask
) );
504 ix0
= bbox
->x0
/ TILE_SIZE
;
505 iy0
= bbox
->y0
/ TILE_SIZE
;
506 ix1
= bbox
->x1
/ TILE_SIZE
;
507 iy1
= bbox
->y1
/ TILE_SIZE
;
510 * Clamp to framebuffer size
512 assert(ix0
== MAX2(ix0
, 0));
513 assert(iy0
== MAX2(iy0
, 0));
514 assert(ix1
== MIN2(ix1
, scene
->tiles_x
- 1));
515 assert(iy1
== MIN2(iy1
, scene
->tiles_y
- 1));
517 /* Determine which tile(s) intersect the triangle's bounding box
519 if (iy0
== iy1
&& ix0
== ix1
)
521 /* Triangle is contained in a single tile:
523 lp_scene_bin_command( scene
, ix0
, iy0
,
524 lp_rast_tri_tab
[nr_planes
],
525 lp_rast_arg_triangle(tri
, (1<<nr_planes
)-1) );
536 for (i
= 0; i
< nr_planes
; i
++) {
537 c
[i
] = (tri
->plane
[i
].c
+
538 tri
->plane
[i
].dcdy
* iy0
* TILE_SIZE
-
539 tri
->plane
[i
].dcdx
* ix0
* TILE_SIZE
);
541 ei
[i
] = tri
->plane
[i
].ei
<< TILE_ORDER
;
542 eo
[i
] = tri
->plane
[i
].eo
<< TILE_ORDER
;
543 xstep
[i
] = -(tri
->plane
[i
].dcdx
<< TILE_ORDER
);
544 ystep
[i
] = tri
->plane
[i
].dcdy
<< TILE_ORDER
;
549 /* Test tile-sized blocks against the triangle.
550 * Discard blocks fully outside the tri. If the block is fully
551 * contained inside the tri, bin an lp_rast_shade_tile command.
552 * Else, bin a lp_rast_triangle command.
554 for (y
= iy0
; y
<= iy1
; y
++)
556 boolean in
= FALSE
; /* are we inside the triangle? */
559 for (i
= 0; i
< nr_planes
; i
++)
562 for (x
= ix0
; x
<= ix1
; x
++)
567 for (i
= 0; i
< nr_planes
; i
++) {
568 int planeout
= cx
[i
] + eo
[i
];
569 int planepartial
= cx
[i
] + ei
[i
] - 1;
570 out
|= (planeout
>> 31);
571 partial
|= (planepartial
>> 31) & (1<<i
);
577 break; /* exiting triangle, all done with this row */
578 LP_COUNT(nr_empty_64
);
581 /* Not trivially accepted by at least one plane -
582 * rasterize/shade partial tile
584 int count
= util_bitcount(partial
);
586 lp_scene_bin_command( scene
, x
, y
,
587 lp_rast_tri_tab
[count
],
588 lp_rast_arg_triangle(tri
, partial
) );
590 LP_COUNT(nr_partially_covered_64
);
593 /* triangle covers the whole tile- shade whole tile */
594 LP_COUNT(nr_fully_covered_64
);
596 lp_setup_whole_tile(setup
, &tri
->inputs
, x
, y
);
599 /* Iterate cx values across the region:
601 for (i
= 0; i
< nr_planes
; i
++)
605 /* Iterate c values down the region:
607 for (i
= 0; i
< nr_planes
; i
++)
615 * Draw triangle if it's CW, cull otherwise.
617 static void triangle_cw( struct lp_setup_context
*setup
,
618 const float (*v0
)[4],
619 const float (*v1
)[4],
620 const float (*v2
)[4] )
622 do_triangle_ccw( setup
, v1
, v0
, v2
, !setup
->ccw_is_frontface
);
627 * Draw triangle if it's CCW, cull otherwise.
629 static void triangle_ccw( struct lp_setup_context
*setup
,
630 const float (*v0
)[4],
631 const float (*v1
)[4],
632 const float (*v2
)[4] )
634 do_triangle_ccw( setup
, v0
, v1
, v2
, setup
->ccw_is_frontface
);
640 * Draw triangle whether it's CW or CCW.
642 static void triangle_both( struct lp_setup_context
*setup
,
643 const float (*v0
)[4],
644 const float (*v1
)[4],
645 const float (*v2
)[4] )
647 /* edge vectors e = v0 - v2, f = v1 - v2 */
648 const float ex
= v0
[0][0] - v2
[0][0];
649 const float ey
= v0
[0][1] - v2
[0][1];
650 const float fx
= v1
[0][0] - v2
[0][0];
651 const float fy
= v1
[0][1] - v2
[0][1];
653 /* det = cross(e,f).z */
654 const float det
= ex
* fy
- ey
* fx
;
656 triangle_ccw( setup
, v0
, v1
, v2
);
658 triangle_cw( setup
, v0
, v1
, v2
);
662 static void triangle_nop( struct lp_setup_context
*setup
,
663 const float (*v0
)[4],
664 const float (*v1
)[4],
665 const float (*v2
)[4] )
671 lp_setup_choose_triangle( struct lp_setup_context
*setup
)
673 switch (setup
->cullmode
) {
675 setup
->triangle
= triangle_both
;
678 setup
->triangle
= setup
->ccw_is_frontface
? triangle_ccw
: triangle_cw
;
680 case PIPE_FACE_FRONT
:
681 setup
->triangle
= setup
->ccw_is_frontface
? triangle_cw
: triangle_ccw
;
684 setup
->triangle
= triangle_nop
;