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"
39 #define NUM_CHANNELS 4
53 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
55 static void constant_coef( struct lp_setup_context
*setup
,
56 struct lp_rast_triangle
*tri
,
61 tri
->inputs
.a0
[slot
][i
] = value
;
62 tri
->inputs
.dadx
[slot
][i
] = 0.0f
;
63 tri
->inputs
.dady
[slot
][i
] = 0.0f
;
68 * Compute a0, dadx and dady for a linearly interpolated coefficient,
71 static void linear_coef( struct lp_setup_context
*setup
,
72 struct lp_rast_triangle
*tri
,
73 struct lp_line_info
*info
,
78 float a1
= info
->v1
[vert_attr
][i
];
79 float a2
= info
->v2
[vert_attr
][i
];
82 float dadx
= da21
* info
->dx
* info
->oneoverarea
;
83 float dady
= da21
* info
->dy
* info
->oneoverarea
;
85 tri
->inputs
.dadx
[slot
][i
] = dadx
;
86 tri
->inputs
.dady
[slot
][i
] = dady
;
88 tri
->inputs
.a0
[slot
][i
] = (a1
-
89 (dadx
* (info
->v1
[0][0] - setup
->pixel_offset
) +
90 dady
* (info
->v1
[0][1] - setup
->pixel_offset
)));
95 * Compute a0, dadx and dady for a perspective-corrected interpolant,
97 * We basically multiply the vertex value by 1/w before computing
98 * the plane coefficients (a0, dadx, dady).
99 * Later, when we compute the value at a particular fragment position we'll
100 * divide the interpolated value by the interpolated W at that fragment.
102 static void perspective_coef( struct lp_setup_context
*setup
,
103 struct lp_rast_triangle
*tri
,
104 struct lp_line_info
*info
,
109 /* premultiply by 1/w (v[0][3] is always 1/w):
111 float a1
= info
->v1
[vert_attr
][i
] * info
->v1
[0][3];
112 float a2
= info
->v2
[vert_attr
][i
] * info
->v2
[0][3];
114 float da21
= a1
- a2
;
115 float dadx
= da21
* info
->dx
* info
->oneoverarea
;
116 float dady
= da21
* info
->dy
* info
->oneoverarea
;
118 tri
->inputs
.dadx
[slot
][i
] = dadx
;
119 tri
->inputs
.dady
[slot
][i
] = dady
;
121 tri
->inputs
.a0
[slot
][i
] = (a1
-
122 (dadx
* (info
->v1
[0][0] - setup
->pixel_offset
) +
123 dady
* (info
->v1
[0][1] - setup
->pixel_offset
)));
127 setup_fragcoord_coef( struct lp_setup_context
*setup
,
128 struct lp_rast_triangle
*tri
,
129 struct lp_line_info
*info
,
134 if (usage_mask
& TGSI_WRITEMASK_X
) {
135 tri
->inputs
.a0
[slot
][0] = 0.0;
136 tri
->inputs
.dadx
[slot
][0] = 1.0;
137 tri
->inputs
.dady
[slot
][0] = 0.0;
141 if (usage_mask
& TGSI_WRITEMASK_Y
) {
142 tri
->inputs
.a0
[slot
][1] = 0.0;
143 tri
->inputs
.dadx
[slot
][1] = 0.0;
144 tri
->inputs
.dady
[slot
][1] = 1.0;
148 if (usage_mask
& TGSI_WRITEMASK_Z
) {
149 linear_coef(setup
, tri
, info
, slot
, 0, 2);
153 if (usage_mask
& TGSI_WRITEMASK_W
) {
154 linear_coef(setup
, tri
, info
, slot
, 0, 3);
159 * Compute the tri->coef[] array dadx, dady, a0 values.
161 static void setup_line_coefficients( struct lp_setup_context
*setup
,
162 struct lp_rast_triangle
*tri
,
163 struct lp_line_info
*info
)
165 unsigned fragcoord_usage_mask
= TGSI_WRITEMASK_XYZ
;
168 /* setup interpolation for all the remaining attributes:
170 for (slot
= 0; slot
< setup
->fs
.nr_inputs
; slot
++) {
171 unsigned vert_attr
= setup
->fs
.input
[slot
].src_index
;
172 unsigned usage_mask
= setup
->fs
.input
[slot
].usage_mask
;
175 switch (setup
->fs
.input
[slot
].interp
) {
176 case LP_INTERP_CONSTANT
:
177 if (setup
->flatshade_first
) {
178 for (i
= 0; i
< NUM_CHANNELS
; i
++)
179 if (usage_mask
& (1 << i
))
180 constant_coef(setup
, tri
, slot
+1, info
->v1
[vert_attr
][i
], i
);
183 for (i
= 0; i
< NUM_CHANNELS
; i
++)
184 if (usage_mask
& (1 << i
))
185 constant_coef(setup
, tri
, slot
+1, info
->v2
[vert_attr
][i
], i
);
189 case LP_INTERP_LINEAR
:
190 for (i
= 0; i
< NUM_CHANNELS
; i
++)
191 if (usage_mask
& (1 << i
))
192 linear_coef(setup
, tri
, info
, slot
+1, vert_attr
, i
);
195 case LP_INTERP_PERSPECTIVE
:
196 for (i
= 0; i
< NUM_CHANNELS
; i
++)
197 if (usage_mask
& (1 << i
))
198 perspective_coef(setup
, tri
, info
, slot
+1, vert_attr
, i
);
199 fragcoord_usage_mask
|= TGSI_WRITEMASK_W
;
202 case LP_INTERP_POSITION
:
204 * The generated pixel interpolators will pick up the coeffs from
205 * slot 0, so all need to ensure that the usage mask is covers all
208 fragcoord_usage_mask
|= usage_mask
;
216 /* The internal position input is in slot zero:
218 setup_fragcoord_coef(setup
, tri
, info
, 0,
219 fragcoord_usage_mask
);
224 static INLINE
int subpixel_snap( float a
)
226 return util_iround(FIXED_ONE
* a
);
231 * Print line vertex attribs (for debug).
234 print_line(struct lp_setup_context
*setup
,
235 const float (*v1
)[4],
236 const float (*v2
)[4])
240 debug_printf("llvmpipe line\n");
241 for (i
= 0; i
< 1 + setup
->fs
.nr_inputs
; i
++) {
242 debug_printf(" v1[%d]: %f %f %f %f\n", i
,
243 v1
[i
][0], v1
[i
][1], v1
[i
][2], v1
[i
][3]);
245 for (i
= 0; i
< 1 + setup
->fs
.nr_inputs
; i
++) {
246 debug_printf(" v2[%d]: %f %f %f %f\n", i
,
247 v2
[i
][0], v2
[i
][1], v2
[i
][2], v2
[i
][3]);
252 static INLINE boolean
sign(float x
){
257 /* Used on positive floats only:
259 static INLINE
float fracf(float f
)
261 return f
- floorf(f
);
267 try_setup_line( struct lp_setup_context
*setup
,
268 const float (*v1
)[4],
269 const float (*v2
)[4])
271 struct lp_scene
*scene
= setup
->scene
;
272 struct lp_rast_triangle
*line
;
273 struct lp_line_info info
;
274 float width
= MAX2(1.0, setup
->line_width
);
282 /* linewidth should be interpreted as integer */
283 int fixed_width
= util_iround(width
) * FIXED_ONE
;
287 float x_offset_end
=0;
288 float y_offset_end
=0;
298 boolean will_draw_start
;
299 boolean will_draw_end
;
302 print_line(setup
, v1
, v2
);
304 if (setup
->scissor_test
) {
312 dx
= v1
[0][0] - v2
[0][0];
313 dy
= v1
[0][1] - v2
[0][1];
316 if (fabsf(dx
) >= fabsf(dy
)) {
317 float dydx
= dy
/ dx
;
319 x1diff
= v1
[0][0] - (float) floor(v1
[0][0]) - 0.5;
320 y1diff
= v1
[0][1] - (float) floor(v1
[0][1]) - 0.5;
321 x2diff
= v2
[0][0] - (float) floor(v2
[0][0]) - 0.5;
322 y2diff
= v2
[0][1] - (float) floor(v2
[0][1]) - 0.5;
324 if (y2diff
==-0.5 && dy
<0){
329 * Diamond exit rule test for starting point
331 if (fabsf(x1diff
) + fabsf(y1diff
) < 0.5) {
334 else if (sign(x1diff
) == sign(-dx
)) {
337 else if (sign(-y1diff
) != sign(dy
)) {
341 /* do intersection test */
342 float yintersect
= fracf(v1
[0][1]) + x1diff
* dydx
;
343 draw_start
= (yintersect
< 1.0 && yintersect
> 0.0);
348 * Diamond exit rule test for ending point
350 if (fabsf(x2diff
) + fabsf(y2diff
) < 0.5) {
353 else if (sign(x2diff
) != sign(-dx
)) {
356 else if (sign(-y2diff
) == sign(dy
)) {
360 /* do intersection test */
361 float yintersect
= fracf(v2
[0][1]) + x2diff
* dydx
;
362 draw_end
= (yintersect
< 1.0 && yintersect
> 0.0);
365 /* Are we already drawing start/end?
367 will_draw_start
= sign(-x1diff
) != sign(dx
);
368 will_draw_end
= (sign(x2diff
) == sign(-dx
)) || x2diff
==0;
371 /* if v2 is to the right of v1, swap pointers */
372 const float (*temp
)[4] = v1
;
377 /* Otherwise shift planes appropriately */
378 if (will_draw_start
!= draw_start
) {
379 x_offset_end
= - x1diff
- 0.5;
380 y_offset_end
= x_offset_end
* dydx
;
383 if (will_draw_end
!= draw_end
) {
384 x_offset
= - x2diff
- 0.5;
385 y_offset
= x_offset
* dydx
;
390 /* Otherwise shift planes appropriately */
391 if (will_draw_start
!= draw_start
) {
392 x_offset
= - x1diff
+ 0.5;
393 y_offset
= x_offset
* dydx
;
395 if (will_draw_end
!= draw_end
) {
396 x_offset_end
= - x2diff
+ 0.5;
397 y_offset_end
= x_offset_end
* dydx
;
401 /* x/y positions in fixed point */
402 x
[0] = subpixel_snap(v1
[0][0] + x_offset
- setup
->pixel_offset
);
403 x
[1] = subpixel_snap(v2
[0][0] + x_offset_end
- setup
->pixel_offset
);
404 x
[2] = subpixel_snap(v2
[0][0] + x_offset_end
- setup
->pixel_offset
);
405 x
[3] = subpixel_snap(v1
[0][0] + x_offset
- setup
->pixel_offset
);
407 y
[0] = subpixel_snap(v1
[0][1] + y_offset
- setup
->pixel_offset
) - fixed_width
/2;
408 y
[1] = subpixel_snap(v2
[0][1] + y_offset_end
- setup
->pixel_offset
) - fixed_width
/2;
409 y
[2] = subpixel_snap(v2
[0][1] + y_offset_end
- setup
->pixel_offset
) + fixed_width
/2;
410 y
[3] = subpixel_snap(v1
[0][1] + y_offset
- setup
->pixel_offset
) + fixed_width
/2;
414 const float dxdy
= dx
/ dy
;
417 x1diff
= v1
[0][0] - (float) floor(v1
[0][0]) - 0.5;
418 y1diff
= v1
[0][1] - (float) floor(v1
[0][1]) - 0.5;
419 x2diff
= v2
[0][0] - (float) floor(v2
[0][0]) - 0.5;
420 y2diff
= v2
[0][1] - (float) floor(v2
[0][1]) - 0.5;
422 if (x2diff
==-0.5 && dx
<0) {
427 * Diamond exit rule test for starting point
429 if (fabsf(x1diff
) + fabsf(y1diff
) < 0.5) {
432 else if (sign(-y1diff
) == sign(dy
)) {
435 else if (sign(x1diff
) != sign(-dx
)) {
439 /* do intersection test */
440 float xintersect
= fracf(v1
[0][0]) + y1diff
* dxdy
;
441 draw_start
= (xintersect
< 1.0 && xintersect
> 0.0);
445 * Diamond exit rule test for ending point
447 if (fabsf(x2diff
) + fabsf(y2diff
) < 0.5) {
450 else if (sign(-y2diff
) != sign(dy
) ) {
453 else if (sign(x2diff
) == sign(-dx
) ) {
457 /* do intersection test */
458 float xintersect
= fracf(v2
[0][0]) + y2diff
* dxdy
;
459 draw_end
= (xintersect
< 1.0 && xintersect
> 0.0);
462 /* Are we already drawing start/end?
464 will_draw_start
= sign(y1diff
) == sign(dy
);
465 will_draw_end
= (sign(-y2diff
) == sign(dy
)) || y2diff
==0;
468 /* if v2 is on top of v1, swap pointers */
469 const float (*temp
)[4] = v1
;
475 /* Otherwise shift planes appropriately */
476 if (will_draw_start
!= draw_start
) {
477 y_offset_end
= - y1diff
+ 0.5;
478 x_offset_end
= y_offset_end
* dxdy
;
480 if (will_draw_end
!= draw_end
) {
481 y_offset
= - y2diff
+ 0.5;
482 x_offset
= y_offset
* dxdy
;
486 /* Otherwise shift planes appropriately */
487 if (will_draw_start
!= draw_start
) {
488 y_offset
= - y1diff
- 0.5;
489 x_offset
= y_offset
* dxdy
;
492 if (will_draw_end
!= draw_end
) {
493 y_offset_end
= - y2diff
- 0.5;
494 x_offset_end
= y_offset_end
* dxdy
;
498 /* x/y positions in fixed point */
499 x
[0] = subpixel_snap(v1
[0][0] + x_offset
- setup
->pixel_offset
) - fixed_width
/2;
500 x
[1] = subpixel_snap(v2
[0][0] + x_offset_end
- setup
->pixel_offset
) - fixed_width
/2;
501 x
[2] = subpixel_snap(v2
[0][0] + x_offset_end
- setup
->pixel_offset
) + fixed_width
/2;
502 x
[3] = subpixel_snap(v1
[0][0] + x_offset
- setup
->pixel_offset
) + fixed_width
/2;
504 y
[0] = subpixel_snap(v1
[0][1] + y_offset
- setup
->pixel_offset
);
505 y
[1] = subpixel_snap(v2
[0][1] + y_offset_end
- setup
->pixel_offset
);
506 y
[2] = subpixel_snap(v2
[0][1] + y_offset_end
- setup
->pixel_offset
);
507 y
[3] = subpixel_snap(v1
[0][1] + y_offset
- setup
->pixel_offset
);
515 /* Bounding rectangle (in pixels) */
517 /* Yes this is necessary to accurately calculate bounding boxes
518 * with the two fill-conventions we support. GL (normally) ends
519 * up needing a bottom-left fill convention, which requires
520 * slightly different rounding.
522 int adj
= (setup
->pixel_offset
!= 0) ? 1 : 0;
524 bbox
.x0
= (MIN4(x
[0], x
[1], x
[2], x
[3]) + (FIXED_ONE
-1)) >> FIXED_ORDER
;
525 bbox
.x1
= (MAX4(x
[0], x
[1], x
[2], x
[3]) + (FIXED_ONE
-1)) >> FIXED_ORDER
;
526 bbox
.y0
= (MIN4(y
[0], y
[1], y
[2], y
[3]) + (FIXED_ONE
-1) + adj
) >> FIXED_ORDER
;
527 bbox
.y1
= (MAX4(y
[0], y
[1], y
[2], y
[3]) + (FIXED_ONE
-1) + adj
) >> FIXED_ORDER
;
529 /* Inclusive coordinates:
535 if (bbox
.x1
< bbox
.x0
||
537 if (0) debug_printf("empty bounding box\n");
538 LP_COUNT(nr_culled_tris
);
542 if (!u_rect_test_intersection(&setup
->draw_region
, &bbox
)) {
543 if (0) debug_printf("offscreen\n");
544 LP_COUNT(nr_culled_tris
);
548 u_rect_find_intersection(&setup
->draw_region
, &bbox
);
550 line
= lp_setup_alloc_triangle(scene
,
558 line
->v
[0][0] = v1
[0][0];
559 line
->v
[1][0] = v2
[0][0];
560 line
->v
[0][1] = v1
[0][1];
561 line
->v
[1][1] = v2
[0][1];
564 /* calculate the deltas */
565 line
->plane
[0].dcdy
= x
[0] - x
[1];
566 line
->plane
[1].dcdy
= x
[1] - x
[2];
567 line
->plane
[2].dcdy
= x
[2] - x
[3];
568 line
->plane
[3].dcdy
= x
[3] - x
[0];
570 line
->plane
[0].dcdx
= y
[0] - y
[1];
571 line
->plane
[1].dcdx
= y
[1] - y
[2];
572 line
->plane
[2].dcdx
= y
[2] - y
[3];
573 line
->plane
[3].dcdx
= y
[3] - y
[0];
576 info
.oneoverarea
= 1.0f
/ (dx
* dx
+ dy
* dy
);
582 /* Setup parameter interpolants:
584 setup_line_coefficients( setup
, line
, &info
);
586 line
->inputs
.facing
= 1.0F
;
587 line
->inputs
.state
= setup
->fs
.stored
;
588 line
->inputs
.disable
= FALSE
;
589 line
->inputs
.opaque
= FALSE
;
591 for (i
= 0; i
< 4; i
++) {
592 struct lp_rast_plane
*plane
= &line
->plane
[i
];
594 /* half-edge constants, will be interated over the whole render
597 plane
->c
= plane
->dcdx
* x
[i
] - plane
->dcdy
* y
[i
];
600 /* correct for top-left vs. bottom-left fill convention.
602 * note that we're overloading gl_rasterization_rules to mean
603 * both (0.5,0.5) pixel centers *and* bottom-left filling
606 * GL actually has a top-left filling convention, but GL's
607 * notion of "top" differs from gallium's...
609 * Also, sometimes (in FBO cases) GL will render upside down
610 * to its usual method, in which case it will probably want
611 * to use the opposite, top-left convention.
613 if (plane
->dcdx
< 0) {
614 /* both fill conventions want this - adjust for left edges */
617 else if (plane
->dcdx
== 0) {
618 if (setup
->pixel_offset
== 0) {
619 /* correct for top-left fill convention:
621 if (plane
->dcdy
> 0) plane
->c
++;
624 /* correct for bottom-left fill convention:
626 if (plane
->dcdy
< 0) plane
->c
++;
630 plane
->dcdx
*= FIXED_ONE
;
631 plane
->dcdy
*= FIXED_ONE
;
633 /* find trivial reject offsets for each edge for a single-pixel
634 * sized block. These will be scaled up at each recursive level to
635 * match the active blocksize. Scaling in this way works best if
636 * the blocks are square.
639 if (plane
->dcdx
< 0) plane
->eo
-= plane
->dcdx
;
640 if (plane
->dcdy
> 0) plane
->eo
+= plane
->dcdy
;
642 /* Calculate trivial accept offsets from the above.
644 plane
->ei
= plane
->dcdy
- plane
->dcdx
- plane
->eo
;
649 * When rasterizing scissored tris, use the intersection of the
650 * triangle bounding box and the scissor rect to generate the
653 * This permits us to cut off the triangle "tails" that are present
654 * in the intermediate recursive levels caused when two of the
655 * triangles edges don't diverge quickly enough to trivially reject
656 * exterior blocks from the triangle.
658 * It's not really clear if it's worth worrying about these tails,
659 * but since we generate the planes for each scissored tri, it's
660 * free to trim them in this case.
662 * Note that otherwise, the scissor planes only vary in 'C' value,
663 * and even then only on state-changes. Could alternatively store
664 * these planes elsewhere.
666 if (nr_planes
== 8) {
667 line
->plane
[4].dcdx
= -1;
668 line
->plane
[4].dcdy
= 0;
669 line
->plane
[4].c
= 1-bbox
.x0
;
670 line
->plane
[4].ei
= 0;
671 line
->plane
[4].eo
= 1;
673 line
->plane
[5].dcdx
= 1;
674 line
->plane
[5].dcdy
= 0;
675 line
->plane
[5].c
= bbox
.x1
+1;
676 line
->plane
[5].ei
= -1;
677 line
->plane
[5].eo
= 0;
679 line
->plane
[6].dcdx
= 0;
680 line
->plane
[6].dcdy
= 1;
681 line
->plane
[6].c
= 1-bbox
.y0
;
682 line
->plane
[6].ei
= 0;
683 line
->plane
[6].eo
= 1;
685 line
->plane
[7].dcdx
= 0;
686 line
->plane
[7].dcdy
= -1;
687 line
->plane
[7].c
= bbox
.y1
+1;
688 line
->plane
[7].ei
= -1;
689 line
->plane
[7].eo
= 0;
692 return lp_setup_bin_triangle(setup
, line
, &bbox
, nr_planes
);
696 static void lp_setup_line( struct lp_setup_context
*setup
,
697 const float (*v0
)[4],
698 const float (*v1
)[4] )
700 if (!try_setup_line( setup
, v0
, v1
))
702 lp_setup_flush_and_restart(setup
);
704 if (!try_setup_line( setup
, v0
, v1
))
710 void lp_setup_choose_line( struct lp_setup_context
*setup
)
712 setup
->line
= lp_setup_line
;