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
);
165 lp_rast_tri_tab
[MAX_PLANES
+1] = {
166 0, /* should be impossible */
167 LP_RAST_OP_TRIANGLE_1
,
168 LP_RAST_OP_TRIANGLE_2
,
169 LP_RAST_OP_TRIANGLE_3
,
170 LP_RAST_OP_TRIANGLE_4
,
171 LP_RAST_OP_TRIANGLE_5
,
172 LP_RAST_OP_TRIANGLE_6
,
173 LP_RAST_OP_TRIANGLE_7
,
174 LP_RAST_OP_TRIANGLE_8
180 * The primitive covers the whole tile- shade whole tile.
182 * \param tx, ty the tile position in tiles, not pixels
185 lp_setup_whole_tile(struct lp_setup_context
*setup
,
186 const struct lp_rast_shader_inputs
*inputs
,
189 struct lp_scene
*scene
= setup
->scene
;
191 LP_COUNT(nr_fully_covered_64
);
193 /* if variant is opaque and scissor doesn't effect the tile */
194 if (inputs
->opaque
) {
195 if (!scene
->fb
.zsbuf
) {
197 * All previous rendering will be overwritten so reset the bin.
199 lp_scene_bin_reset( scene
, tx
, ty
);
202 LP_COUNT(nr_shade_opaque_64
);
203 return lp_scene_bin_command( scene
, tx
, ty
,
204 LP_RAST_OP_SHADE_TILE_OPAQUE
,
205 lp_rast_arg_inputs(inputs
) );
207 LP_COUNT(nr_shade_64
);
208 return lp_scene_bin_command( scene
, tx
, ty
,
209 LP_RAST_OP_SHADE_TILE
,
210 lp_rast_arg_inputs(inputs
) );
216 * Do basic setup for triangle rasterization and determine which
217 * framebuffer tiles are touched. Put the triangle in the scene's
218 * bins for the tiles which we overlap.
221 do_triangle_ccw(struct lp_setup_context
*setup
,
222 const float (*v0
)[4],
223 const float (*v1
)[4],
224 const float (*v2
)[4],
225 boolean frontfacing
)
227 struct lp_scene
*scene
= setup
->scene
;
228 struct lp_rast_triangle
*tri
;
238 lp_setup_print_triangle(setup
, v0
, v1
, v2
);
240 if (setup
->scissor_test
) {
247 /* x/y positions in fixed point */
248 x
[0] = subpixel_snap(v0
[0][0] - setup
->pixel_offset
);
249 x
[1] = subpixel_snap(v1
[0][0] - setup
->pixel_offset
);
250 x
[2] = subpixel_snap(v2
[0][0] - setup
->pixel_offset
);
251 y
[0] = subpixel_snap(v0
[0][1] - setup
->pixel_offset
);
252 y
[1] = subpixel_snap(v1
[0][1] - setup
->pixel_offset
);
253 y
[2] = subpixel_snap(v2
[0][1] - setup
->pixel_offset
);
256 /* Bounding rectangle (in pixels) */
258 /* Yes this is necessary to accurately calculate bounding boxes
259 * with the two fill-conventions we support. GL (normally) ends
260 * up needing a bottom-left fill convention, which requires
261 * slightly different rounding.
263 int adj
= (setup
->pixel_offset
!= 0) ? 1 : 0;
265 bbox
.x0
= (MIN3(x
[0], x
[1], x
[2]) + (FIXED_ONE
-1)) >> FIXED_ORDER
;
266 bbox
.x1
= (MAX3(x
[0], x
[1], x
[2]) + (FIXED_ONE
-1)) >> FIXED_ORDER
;
267 bbox
.y0
= (MIN3(y
[0], y
[1], y
[2]) + (FIXED_ONE
-1) + adj
) >> FIXED_ORDER
;
268 bbox
.y1
= (MAX3(y
[0], y
[1], y
[2]) + (FIXED_ONE
-1) + adj
) >> FIXED_ORDER
;
270 /* Inclusive coordinates:
276 if (bbox
.x1
< bbox
.x0
||
278 if (0) debug_printf("empty bounding box\n");
279 LP_COUNT(nr_culled_tris
);
283 if (!u_rect_test_intersection(&setup
->draw_region
, &bbox
)) {
284 if (0) debug_printf("offscreen\n");
285 LP_COUNT(nr_culled_tris
);
289 u_rect_find_intersection(&setup
->draw_region
, &bbox
);
291 tri
= lp_setup_alloc_triangle(scene
,
299 tri
->v
[0][0] = v0
[0][0];
300 tri
->v
[1][0] = v1
[0][0];
301 tri
->v
[2][0] = v2
[0][0];
302 tri
->v
[0][1] = v0
[0][1];
303 tri
->v
[1][1] = v1
[0][1];
304 tri
->v
[2][1] = v2
[0][1];
307 tri
->plane
[0].dcdy
= x
[0] - x
[1];
308 tri
->plane
[1].dcdy
= x
[1] - x
[2];
309 tri
->plane
[2].dcdy
= x
[2] - x
[0];
311 tri
->plane
[0].dcdx
= y
[0] - y
[1];
312 tri
->plane
[1].dcdx
= y
[1] - y
[2];
313 tri
->plane
[2].dcdx
= y
[2] - y
[0];
315 area
= (tri
->plane
[0].dcdy
* tri
->plane
[2].dcdx
-
316 tri
->plane
[2].dcdy
* tri
->plane
[0].dcdx
);
320 /* Cull non-ccw and zero-sized triangles.
322 * XXX: subject to overflow??
325 lp_scene_putback_data( scene
, tri_bytes
);
326 LP_COUNT(nr_culled_tris
);
330 /* Setup parameter interpolants:
332 lp_setup_tri_coef( setup
, &tri
->inputs
, v0
, v1
, v2
, frontfacing
);
334 tri
->inputs
.facing
= frontfacing
? 1.0F
: -1.0F
;
335 tri
->inputs
.disable
= FALSE
;
336 tri
->inputs
.opaque
= setup
->fs
.current
.variant
->opaque
;
337 tri
->inputs
.state
= setup
->fs
.stored
;
340 for (i
= 0; i
< 3; i
++) {
341 struct lp_rast_plane
*plane
= &tri
->plane
[i
];
343 /* half-edge constants, will be interated over the whole render
346 plane
->c
= plane
->dcdx
* x
[i
] - plane
->dcdy
* y
[i
];
348 /* correct for top-left vs. bottom-left fill convention.
350 * note that we're overloading gl_rasterization_rules to mean
351 * both (0.5,0.5) pixel centers *and* bottom-left filling
354 * GL actually has a top-left filling convention, but GL's
355 * notion of "top" differs from gallium's...
357 * Also, sometimes (in FBO cases) GL will render upside down
358 * to its usual method, in which case it will probably want
359 * to use the opposite, top-left convention.
361 if (plane
->dcdx
< 0) {
362 /* both fill conventions want this - adjust for left edges */
365 else if (plane
->dcdx
== 0) {
366 if (setup
->pixel_offset
== 0) {
367 /* correct for top-left fill convention:
369 if (plane
->dcdy
> 0) plane
->c
++;
372 /* correct for bottom-left fill convention:
374 if (plane
->dcdy
< 0) plane
->c
++;
378 plane
->dcdx
*= FIXED_ONE
;
379 plane
->dcdy
*= FIXED_ONE
;
381 /* find trivial reject offsets for each edge for a single-pixel
382 * sized block. These will be scaled up at each recursive level to
383 * match the active blocksize. Scaling in this way works best if
384 * the blocks are square.
387 if (plane
->dcdx
< 0) plane
->eo
-= plane
->dcdx
;
388 if (plane
->dcdy
> 0) plane
->eo
+= plane
->dcdy
;
390 /* Calculate trivial accept offsets from the above.
392 plane
->ei
= plane
->dcdy
- plane
->dcdx
- plane
->eo
;
397 * When rasterizing scissored tris, use the intersection of the
398 * triangle bounding box and the scissor rect to generate the
401 * This permits us to cut off the triangle "tails" that are present
402 * in the intermediate recursive levels caused when two of the
403 * triangles edges don't diverge quickly enough to trivially reject
404 * exterior blocks from the triangle.
406 * It's not really clear if it's worth worrying about these tails,
407 * but since we generate the planes for each scissored tri, it's
408 * free to trim them in this case.
410 * Note that otherwise, the scissor planes only vary in 'C' value,
411 * and even then only on state-changes. Could alternatively store
412 * these planes elsewhere.
414 if (nr_planes
== 7) {
415 tri
->plane
[3].dcdx
= -1;
416 tri
->plane
[3].dcdy
= 0;
417 tri
->plane
[3].c
= 1-bbox
.x0
;
418 tri
->plane
[3].ei
= 0;
419 tri
->plane
[3].eo
= 1;
421 tri
->plane
[4].dcdx
= 1;
422 tri
->plane
[4].dcdy
= 0;
423 tri
->plane
[4].c
= bbox
.x1
+1;
424 tri
->plane
[4].ei
= -1;
425 tri
->plane
[4].eo
= 0;
427 tri
->plane
[5].dcdx
= 0;
428 tri
->plane
[5].dcdy
= 1;
429 tri
->plane
[5].c
= 1-bbox
.y0
;
430 tri
->plane
[5].ei
= 0;
431 tri
->plane
[5].eo
= 1;
433 tri
->plane
[6].dcdx
= 0;
434 tri
->plane
[6].dcdy
= -1;
435 tri
->plane
[6].c
= bbox
.y1
+1;
436 tri
->plane
[6].ei
= -1;
437 tri
->plane
[6].eo
= 0;
440 return lp_setup_bin_triangle( setup
, tri
, &bbox
, nr_planes
);
444 * Round to nearest less or equal power of two of the input.
446 * Undefined if no bit set exists, so code should check against 0 first.
448 static INLINE
uint32_t
449 floor_pot(uint32_t n
)
451 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
471 lp_setup_bin_triangle( struct lp_setup_context
*setup
,
472 struct lp_rast_triangle
*tri
,
473 const struct u_rect
*bbox
,
476 struct lp_scene
*scene
= setup
->scene
;
479 /* What is the largest power-of-two boundary this triangle crosses:
481 int dx
= floor_pot((bbox
->x0
^ bbox
->x1
) |
482 (bbox
->y0
^ bbox
->y1
));
484 /* The largest dimension of the rasterized area of the triangle
485 * (aligned to a 4x4 grid), rounded down to the nearest power of two:
487 int sz
= floor_pot((bbox
->x1
- (bbox
->x0
& ~3)) |
488 (bbox
->y1
- (bbox
->y0
& ~3)));
490 if (nr_planes
== 3) {
491 if (sz
< 4 && dx
< 64)
493 /* Triangle is contained in a single 4x4 stamp:
495 int mask
= (bbox
->x0
& 63 & ~3) | ((bbox
->y0
& 63 & ~3) << 8);
497 return lp_scene_bin_command( scene
,
498 bbox
->x0
/64, bbox
->y0
/64,
499 LP_RAST_OP_TRIANGLE_3_4
,
500 lp_rast_arg_triangle(tri
, mask
) );
503 if (sz
< 16 && dx
< 64)
505 int mask
= (bbox
->x0
& 63 & ~3) | ((bbox
->y0
& 63 & ~3) << 8);
507 /* Triangle is contained in a single 16x16 block:
509 return lp_scene_bin_command( scene
,
510 bbox
->x0
/64, bbox
->y0
/64,
511 LP_RAST_OP_TRIANGLE_3_16
,
512 lp_rast_arg_triangle(tri
, mask
) );
517 /* Determine which tile(s) intersect the triangle's bounding box
521 int ix0
= bbox
->x0
/ TILE_SIZE
;
522 int iy0
= bbox
->y0
/ TILE_SIZE
;
524 assert(iy0
== bbox
->y1
/ TILE_SIZE
&&
525 ix0
== bbox
->x1
/ TILE_SIZE
);
527 /* Triangle is contained in a single tile:
529 return lp_scene_bin_command( scene
, ix0
, iy0
,
530 lp_rast_tri_tab
[nr_planes
],
531 lp_rast_arg_triangle(tri
, (1<<nr_planes
)-1) );
538 int xstep
[MAX_PLANES
];
539 int ystep
[MAX_PLANES
];
542 int ix0
= bbox
->x0
/ TILE_SIZE
;
543 int iy0
= bbox
->y0
/ TILE_SIZE
;
544 int ix1
= bbox
->x1
/ TILE_SIZE
;
545 int iy1
= bbox
->y1
/ TILE_SIZE
;
547 for (i
= 0; i
< nr_planes
; i
++) {
548 c
[i
] = (tri
->plane
[i
].c
+
549 tri
->plane
[i
].dcdy
* iy0
* TILE_SIZE
-
550 tri
->plane
[i
].dcdx
* ix0
* TILE_SIZE
);
552 ei
[i
] = tri
->plane
[i
].ei
<< TILE_ORDER
;
553 eo
[i
] = tri
->plane
[i
].eo
<< TILE_ORDER
;
554 xstep
[i
] = -(tri
->plane
[i
].dcdx
<< TILE_ORDER
);
555 ystep
[i
] = tri
->plane
[i
].dcdy
<< TILE_ORDER
;
560 /* Test tile-sized blocks against the triangle.
561 * Discard blocks fully outside the tri. If the block is fully
562 * contained inside the tri, bin an lp_rast_shade_tile command.
563 * Else, bin a lp_rast_triangle command.
565 for (y
= iy0
; y
<= iy1
; y
++)
567 boolean in
= FALSE
; /* are we inside the triangle? */
570 for (i
= 0; i
< nr_planes
; i
++)
573 for (x
= ix0
; x
<= ix1
; x
++)
578 for (i
= 0; i
< nr_planes
; i
++) {
579 int planeout
= cx
[i
] + eo
[i
];
580 int planepartial
= cx
[i
] + ei
[i
] - 1;
581 out
|= (planeout
>> 31);
582 partial
|= (planepartial
>> 31) & (1<<i
);
588 break; /* exiting triangle, all done with this row */
589 LP_COUNT(nr_empty_64
);
592 /* Not trivially accepted by at least one plane -
593 * rasterize/shade partial tile
595 int count
= util_bitcount(partial
);
597 if (!lp_scene_bin_command( scene
, x
, y
,
598 lp_rast_tri_tab
[count
],
599 lp_rast_arg_triangle(tri
, partial
) ))
602 LP_COUNT(nr_partially_covered_64
);
605 /* triangle covers the whole tile- shade whole tile */
606 LP_COUNT(nr_fully_covered_64
);
608 if (!lp_setup_whole_tile(setup
, &tri
->inputs
, x
, y
))
612 /* Iterate cx values across the region:
614 for (i
= 0; i
< nr_planes
; i
++)
618 /* Iterate c values down the region:
620 for (i
= 0; i
< nr_planes
; i
++)
628 /* Need to disable any partially binned triangle. This is easier
629 * than trying to locate all the triangle, shade-tile, etc,
630 * commands which may have been binned.
632 tri
->inputs
.disable
= TRUE
;
638 * Draw triangle if it's CW, cull otherwise.
640 static void triangle_cw( struct lp_setup_context
*setup
,
641 const float (*v0
)[4],
642 const float (*v1
)[4],
643 const float (*v2
)[4] )
645 if (!do_triangle_ccw( setup
, v1
, v0
, v2
, !setup
->ccw_is_frontface
))
647 lp_setup_flush_and_restart(setup
);
649 if (!do_triangle_ccw( setup
, v1
, v0
, v2
, !setup
->ccw_is_frontface
))
656 * Draw triangle if it's CCW, cull otherwise.
658 static void triangle_ccw( struct lp_setup_context
*setup
,
659 const float (*v0
)[4],
660 const float (*v1
)[4],
661 const float (*v2
)[4] )
663 if (!do_triangle_ccw( setup
, v0
, v1
, v2
, setup
->ccw_is_frontface
))
665 lp_setup_flush_and_restart(setup
);
666 if (!do_triangle_ccw( setup
, v0
, v1
, v2
, setup
->ccw_is_frontface
))
674 * Draw triangle whether it's CW or CCW.
676 static void triangle_both( struct lp_setup_context
*setup
,
677 const float (*v0
)[4],
678 const float (*v1
)[4],
679 const float (*v2
)[4] )
681 /* edge vectors e = v0 - v2, f = v1 - v2 */
682 const float ex
= v0
[0][0] - v2
[0][0];
683 const float ey
= v0
[0][1] - v2
[0][1];
684 const float fx
= v1
[0][0] - v2
[0][0];
685 const float fy
= v1
[0][1] - v2
[0][1];
687 /* det = cross(e,f).z */
688 const float det
= ex
* fy
- ey
* fx
;
690 triangle_ccw( setup
, v0
, v1
, v2
);
692 triangle_cw( setup
, v0
, v1
, v2
);
696 static void triangle_nop( struct lp_setup_context
*setup
,
697 const float (*v0
)[4],
698 const float (*v1
)[4],
699 const float (*v2
)[4] )
705 lp_setup_choose_triangle( struct lp_setup_context
*setup
)
707 switch (setup
->cullmode
) {
709 setup
->triangle
= triangle_both
;
712 setup
->triangle
= setup
->ccw_is_frontface
? triangle_ccw
: triangle_cw
;
714 case PIPE_FACE_FRONT
:
715 setup
->triangle
= setup
->ccw_is_frontface
? triangle_cw
: triangle_ccw
;
718 setup
->triangle
= triangle_nop
;