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
57 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
59 static void constant_coef( struct lp_setup_context
*setup
,
60 struct lp_line_info
*info
,
65 info
->a0
[slot
][i
] = value
;
66 info
->dadx
[slot
][i
] = 0.0f
;
67 info
->dady
[slot
][i
] = 0.0f
;
72 * Compute a0, dadx and dady for a linearly interpolated coefficient,
75 static void linear_coef( struct lp_setup_context
*setup
,
76 struct lp_line_info
*info
,
81 float a1
= info
->v1
[vert_attr
][i
];
82 float a2
= info
->v2
[vert_attr
][i
];
85 float dadx
= da21
* info
->dx
* info
->oneoverarea
;
86 float dady
= da21
* info
->dy
* info
->oneoverarea
;
88 info
->dadx
[slot
][i
] = dadx
;
89 info
->dady
[slot
][i
] = dady
;
91 info
->a0
[slot
][i
] = (a1
-
92 (dadx
* (info
->v1
[0][0] - setup
->pixel_offset
) +
93 dady
* (info
->v1
[0][1] - setup
->pixel_offset
)));
98 * Compute a0, dadx and dady for a perspective-corrected interpolant,
100 * We basically multiply the vertex value by 1/w before computing
101 * the plane coefficients (a0, dadx, dady).
102 * Later, when we compute the value at a particular fragment position we'll
103 * divide the interpolated value by the interpolated W at that fragment.
105 static void perspective_coef( struct lp_setup_context
*setup
,
106 struct lp_line_info
*info
,
111 /* premultiply by 1/w (v[0][3] is always 1/w):
113 float a1
= info
->v1
[vert_attr
][i
] * info
->v1
[0][3];
114 float a2
= info
->v2
[vert_attr
][i
] * info
->v2
[0][3];
116 float da21
= a1
- a2
;
117 float dadx
= da21
* info
->dx
* info
->oneoverarea
;
118 float dady
= da21
* info
->dy
* info
->oneoverarea
;
120 info
->dadx
[slot
][i
] = dadx
;
121 info
->dady
[slot
][i
] = dady
;
123 info
->a0
[slot
][i
] = (a1
-
124 (dadx
* (info
->v1
[0][0] - setup
->pixel_offset
) +
125 dady
* (info
->v1
[0][1] - setup
->pixel_offset
)));
129 setup_fragcoord_coef( struct lp_setup_context
*setup
,
130 struct lp_line_info
*info
,
135 if (usage_mask
& TGSI_WRITEMASK_X
) {
136 info
->a0
[slot
][0] = 0.0;
137 info
->dadx
[slot
][0] = 1.0;
138 info
->dady
[slot
][0] = 0.0;
142 if (usage_mask
& TGSI_WRITEMASK_Y
) {
143 info
->a0
[slot
][1] = 0.0;
144 info
->dadx
[slot
][1] = 0.0;
145 info
->dady
[slot
][1] = 1.0;
149 if (usage_mask
& TGSI_WRITEMASK_Z
) {
150 linear_coef(setup
, info
, slot
, 0, 2);
154 if (usage_mask
& TGSI_WRITEMASK_W
) {
155 linear_coef(setup
, info
, slot
, 0, 3);
160 * Compute the tri->coef[] array dadx, dady, a0 values.
162 static void setup_line_coefficients( struct lp_setup_context
*setup
,
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
, info
, 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
, info
, 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
, 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
, 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
;
211 case LP_INTERP_FACING
:
212 for (i
= 0; i
< NUM_CHANNELS
; i
++)
213 if (usage_mask
& (1 << i
))
214 constant_coef(setup
, info
, slot
+1, 1.0, i
);
222 /* The internal position input is in slot zero:
224 setup_fragcoord_coef(setup
, info
, 0,
225 fragcoord_usage_mask
);
230 static INLINE
int subpixel_snap( float a
)
232 return util_iround(FIXED_ONE
* a
);
237 * Print line vertex attribs (for debug).
240 print_line(struct lp_setup_context
*setup
,
241 const float (*v1
)[4],
242 const float (*v2
)[4])
246 debug_printf("llvmpipe line\n");
247 for (i
= 0; i
< 1 + setup
->fs
.nr_inputs
; i
++) {
248 debug_printf(" v1[%d]: %f %f %f %f\n", i
,
249 v1
[i
][0], v1
[i
][1], v1
[i
][2], v1
[i
][3]);
251 for (i
= 0; i
< 1 + setup
->fs
.nr_inputs
; i
++) {
252 debug_printf(" v2[%d]: %f %f %f %f\n", i
,
253 v2
[i
][0], v2
[i
][1], v2
[i
][2], v2
[i
][3]);
258 static INLINE boolean
sign(float x
){
263 /* Used on positive floats only:
265 static INLINE
float fracf(float f
)
267 return f
- floorf(f
);
273 try_setup_line( struct lp_setup_context
*setup
,
274 const float (*v1
)[4],
275 const float (*v2
)[4])
277 struct lp_scene
*scene
= setup
->scene
;
278 struct lp_rast_triangle
*line
;
279 struct lp_rast_plane
*plane
;
280 struct lp_line_info info
;
281 float width
= MAX2(1.0, setup
->line_width
);
289 /* linewidth should be interpreted as integer */
290 int fixed_width
= util_iround(width
) * FIXED_ONE
;
294 float x_offset_end
=0;
295 float y_offset_end
=0;
306 boolean will_draw_start
;
307 boolean will_draw_end
;
310 print_line(setup
, v1
, v2
);
312 if (setup
->scissor_test
) {
320 dx
= v1
[0][0] - v2
[0][0];
321 dy
= v1
[0][1] - v2
[0][1];
322 area
= (dx
* dx
+ dy
* dy
);
324 LP_COUNT(nr_culled_tris
);
328 info
.oneoverarea
= 1.0f
/ area
;
336 if (fabsf(dx
) >= fabsf(dy
)) {
337 float dydx
= dy
/ dx
;
339 x1diff
= v1
[0][0] - (float) floor(v1
[0][0]) - 0.5;
340 y1diff
= v1
[0][1] - (float) floor(v1
[0][1]) - 0.5;
341 x2diff
= v2
[0][0] - (float) floor(v2
[0][0]) - 0.5;
342 y2diff
= v2
[0][1] - (float) floor(v2
[0][1]) - 0.5;
344 if (y2diff
==-0.5 && dy
<0){
349 * Diamond exit rule test for starting point
351 if (fabsf(x1diff
) + fabsf(y1diff
) < 0.5) {
354 else if (sign(x1diff
) == sign(-dx
)) {
357 else if (sign(-y1diff
) != sign(dy
)) {
361 /* do intersection test */
362 float yintersect
= fracf(v1
[0][1]) + x1diff
* dydx
;
363 draw_start
= (yintersect
< 1.0 && yintersect
> 0.0);
368 * Diamond exit rule test for ending point
370 if (fabsf(x2diff
) + fabsf(y2diff
) < 0.5) {
373 else if (sign(x2diff
) != sign(-dx
)) {
376 else if (sign(-y2diff
) == sign(dy
)) {
380 /* do intersection test */
381 float yintersect
= fracf(v2
[0][1]) + x2diff
* dydx
;
382 draw_end
= (yintersect
< 1.0 && yintersect
> 0.0);
385 /* Are we already drawing start/end?
387 will_draw_start
= sign(-x1diff
) != sign(dx
);
388 will_draw_end
= (sign(x2diff
) == sign(-dx
)) || x2diff
==0;
391 /* if v2 is to the right of v1, swap pointers */
392 const float (*temp
)[4] = v1
;
397 /* Otherwise shift planes appropriately */
398 if (will_draw_start
!= draw_start
) {
399 x_offset_end
= - x1diff
- 0.5;
400 y_offset_end
= x_offset_end
* dydx
;
403 if (will_draw_end
!= draw_end
) {
404 x_offset
= - x2diff
- 0.5;
405 y_offset
= x_offset
* dydx
;
410 /* Otherwise shift planes appropriately */
411 if (will_draw_start
!= draw_start
) {
412 x_offset
= - x1diff
+ 0.5;
413 y_offset
= x_offset
* dydx
;
415 if (will_draw_end
!= draw_end
) {
416 x_offset_end
= - x2diff
+ 0.5;
417 y_offset_end
= x_offset_end
* dydx
;
421 /* x/y positions in fixed point */
422 x
[0] = subpixel_snap(v1
[0][0] + x_offset
- setup
->pixel_offset
);
423 x
[1] = subpixel_snap(v2
[0][0] + x_offset_end
- setup
->pixel_offset
);
424 x
[2] = subpixel_snap(v2
[0][0] + x_offset_end
- setup
->pixel_offset
);
425 x
[3] = subpixel_snap(v1
[0][0] + x_offset
- setup
->pixel_offset
);
427 y
[0] = subpixel_snap(v1
[0][1] + y_offset
- setup
->pixel_offset
) - fixed_width
/2;
428 y
[1] = subpixel_snap(v2
[0][1] + y_offset_end
- setup
->pixel_offset
) - fixed_width
/2;
429 y
[2] = subpixel_snap(v2
[0][1] + y_offset_end
- setup
->pixel_offset
) + fixed_width
/2;
430 y
[3] = subpixel_snap(v1
[0][1] + y_offset
- setup
->pixel_offset
) + fixed_width
/2;
434 const float dxdy
= dx
/ dy
;
437 x1diff
= v1
[0][0] - (float) floor(v1
[0][0]) - 0.5;
438 y1diff
= v1
[0][1] - (float) floor(v1
[0][1]) - 0.5;
439 x2diff
= v2
[0][0] - (float) floor(v2
[0][0]) - 0.5;
440 y2diff
= v2
[0][1] - (float) floor(v2
[0][1]) - 0.5;
442 if (x2diff
==-0.5 && dx
<0) {
447 * Diamond exit rule test for starting point
449 if (fabsf(x1diff
) + fabsf(y1diff
) < 0.5) {
452 else if (sign(-y1diff
) == sign(dy
)) {
455 else if (sign(x1diff
) != sign(-dx
)) {
459 /* do intersection test */
460 float xintersect
= fracf(v1
[0][0]) + y1diff
* dxdy
;
461 draw_start
= (xintersect
< 1.0 && xintersect
> 0.0);
465 * Diamond exit rule test for ending point
467 if (fabsf(x2diff
) + fabsf(y2diff
) < 0.5) {
470 else if (sign(-y2diff
) != sign(dy
) ) {
473 else if (sign(x2diff
) == sign(-dx
) ) {
477 /* do intersection test */
478 float xintersect
= fracf(v2
[0][0]) + y2diff
* dxdy
;
479 draw_end
= (xintersect
< 1.0 && xintersect
>= 0.0);
482 /* Are we already drawing start/end?
484 will_draw_start
= sign(y1diff
) == sign(dy
);
485 will_draw_end
= (sign(-y2diff
) == sign(dy
)) || y2diff
==0;
488 /* if v2 is on top of v1, swap pointers */
489 const float (*temp
)[4] = v1
;
495 /* Otherwise shift planes appropriately */
496 if (will_draw_start
!= draw_start
) {
497 y_offset_end
= - y1diff
+ 0.5;
498 x_offset_end
= y_offset_end
* dxdy
;
500 if (will_draw_end
!= draw_end
) {
501 y_offset
= - y2diff
+ 0.5;
502 x_offset
= y_offset
* dxdy
;
506 /* Otherwise shift planes appropriately */
507 if (will_draw_start
!= draw_start
) {
508 y_offset
= - y1diff
- 0.5;
509 x_offset
= y_offset
* dxdy
;
512 if (will_draw_end
!= draw_end
) {
513 y_offset_end
= - y2diff
- 0.5;
514 x_offset_end
= y_offset_end
* dxdy
;
518 /* x/y positions in fixed point */
519 x
[0] = subpixel_snap(v1
[0][0] + x_offset
- setup
->pixel_offset
) - fixed_width
/2;
520 x
[1] = subpixel_snap(v2
[0][0] + x_offset_end
- setup
->pixel_offset
) - fixed_width
/2;
521 x
[2] = subpixel_snap(v2
[0][0] + x_offset_end
- setup
->pixel_offset
) + fixed_width
/2;
522 x
[3] = subpixel_snap(v1
[0][0] + x_offset
- setup
->pixel_offset
) + fixed_width
/2;
524 y
[0] = subpixel_snap(v1
[0][1] + y_offset
- setup
->pixel_offset
);
525 y
[1] = subpixel_snap(v2
[0][1] + y_offset_end
- setup
->pixel_offset
);
526 y
[2] = subpixel_snap(v2
[0][1] + y_offset_end
- setup
->pixel_offset
);
527 y
[3] = subpixel_snap(v1
[0][1] + y_offset
- setup
->pixel_offset
);
535 /* Bounding rectangle (in pixels) */
537 /* Yes this is necessary to accurately calculate bounding boxes
538 * with the two fill-conventions we support. GL (normally) ends
539 * up needing a bottom-left fill convention, which requires
540 * slightly different rounding.
542 int adj
= (setup
->pixel_offset
!= 0) ? 1 : 0;
544 bbox
.x0
= (MIN4(x
[0], x
[1], x
[2], x
[3]) + (FIXED_ONE
-1)) >> FIXED_ORDER
;
545 bbox
.x1
= (MAX4(x
[0], x
[1], x
[2], x
[3]) + (FIXED_ONE
-1)) >> FIXED_ORDER
;
546 bbox
.y0
= (MIN4(y
[0], y
[1], y
[2], y
[3]) + (FIXED_ONE
-1) + adj
) >> FIXED_ORDER
;
547 bbox
.y1
= (MAX4(y
[0], y
[1], y
[2], y
[3]) + (FIXED_ONE
-1) + adj
) >> FIXED_ORDER
;
549 /* Inclusive coordinates:
555 if (bbox
.x1
< bbox
.x0
||
557 if (0) debug_printf("empty bounding box\n");
558 LP_COUNT(nr_culled_tris
);
562 if (!u_rect_test_intersection(&setup
->draw_region
, &bbox
)) {
563 if (0) debug_printf("offscreen\n");
564 LP_COUNT(nr_culled_tris
);
568 u_rect_find_intersection(&setup
->draw_region
, &bbox
);
570 line
= lp_setup_alloc_triangle(scene
,
578 line
->v
[0][0] = v1
[0][0];
579 line
->v
[1][0] = v2
[0][0];
580 line
->v
[0][1] = v1
[0][1];
581 line
->v
[1][1] = v2
[0][1];
584 /* calculate the deltas */
585 plane
= GET_PLANES(line
);
586 plane
[0].dcdy
= x
[0] - x
[1];
587 plane
[1].dcdy
= x
[1] - x
[2];
588 plane
[2].dcdy
= x
[2] - x
[3];
589 plane
[3].dcdy
= x
[3] - x
[0];
591 plane
[0].dcdx
= y
[0] - y
[1];
592 plane
[1].dcdx
= y
[1] - y
[2];
593 plane
[2].dcdx
= y
[2] - y
[3];
594 plane
[3].dcdx
= y
[3] - y
[0];
597 /* Setup parameter interpolants:
599 info
.a0
= GET_A0(&line
->inputs
);
600 info
.dadx
= GET_DADX(&line
->inputs
);
601 info
.dady
= GET_DADY(&line
->inputs
);
602 setup_line_coefficients(setup
, &info
);
604 line
->inputs
.frontfacing
= TRUE
;
605 line
->inputs
.disable
= FALSE
;
606 line
->inputs
.opaque
= FALSE
;
608 for (i
= 0; i
< 4; i
++) {
610 /* half-edge constants, will be interated over the whole render
613 plane
[i
].c
= plane
[i
].dcdx
* x
[i
] - plane
[i
].dcdy
* y
[i
];
616 /* correct for top-left vs. bottom-left fill convention.
618 * note that we're overloading gl_rasterization_rules to mean
619 * both (0.5,0.5) pixel centers *and* bottom-left filling
622 * GL actually has a top-left filling convention, but GL's
623 * notion of "top" differs from gallium's...
625 * Also, sometimes (in FBO cases) GL will render upside down
626 * to its usual method, in which case it will probably want
627 * to use the opposite, top-left convention.
629 if (plane
[i
].dcdx
< 0) {
630 /* both fill conventions want this - adjust for left edges */
633 else if (plane
[i
].dcdx
== 0) {
634 if (setup
->pixel_offset
== 0) {
635 /* correct for top-left fill convention:
637 if (plane
[i
].dcdy
> 0) plane
[i
].c
++;
640 /* correct for bottom-left fill convention:
642 if (plane
[i
].dcdy
< 0) plane
[i
].c
++;
646 plane
[i
].dcdx
*= FIXED_ONE
;
647 plane
[i
].dcdy
*= FIXED_ONE
;
649 /* find trivial reject offsets for each edge for a single-pixel
650 * sized block. These will be scaled up at each recursive level to
651 * match the active blocksize. Scaling in this way works best if
652 * the blocks are square.
655 if (plane
[i
].dcdx
< 0) plane
[i
].eo
-= plane
[i
].dcdx
;
656 if (plane
[i
].dcdy
> 0) plane
[i
].eo
+= plane
[i
].dcdy
;
658 /* Calculate trivial accept offsets from the above.
660 plane
[i
].ei
= plane
[i
].dcdy
- plane
[i
].dcdx
- plane
[i
].eo
;
665 * When rasterizing scissored tris, use the intersection of the
666 * triangle bounding box and the scissor rect to generate the
669 * This permits us to cut off the triangle "tails" that are present
670 * in the intermediate recursive levels caused when two of the
671 * triangles edges don't diverge quickly enough to trivially reject
672 * exterior blocks from the triangle.
674 * It's not really clear if it's worth worrying about these tails,
675 * but since we generate the planes for each scissored tri, it's
676 * free to trim them in this case.
678 * Note that otherwise, the scissor planes only vary in 'C' value,
679 * and even then only on state-changes. Could alternatively store
680 * these planes elsewhere.
682 if (nr_planes
== 8) {
685 plane
[4].c
= 1-bbox
.x0
;
691 plane
[5].c
= bbox
.x1
+1;
697 plane
[6].c
= 1-bbox
.y0
;
703 plane
[7].c
= bbox
.y1
+1;
708 return lp_setup_bin_triangle(setup
, line
, &bbox
, nr_planes
);
712 static void lp_setup_line( struct lp_setup_context
*setup
,
713 const float (*v0
)[4],
714 const float (*v1
)[4] )
716 if (!try_setup_line( setup
, v0
, v1
))
718 if (!lp_setup_flush_and_restart(setup
))
721 if (!try_setup_line( setup
, v0
, v1
))
727 void lp_setup_choose_line( struct lp_setup_context
*setup
)
729 setup
->line
= lp_setup_line
;