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 "lp_setup_context.h"
34 #include "util/u_math.h"
35 #include "util/u_memory.h"
37 #define NUM_CHANNELS 4
40 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
42 static void constant_coef( struct lp_rast_triangle
*tri
,
47 tri
->inputs
.a0
[slot
][i
] = value
;
48 tri
->inputs
.dadx
[slot
][i
] = 0;
49 tri
->inputs
.dady
[slot
][i
] = 0;
53 * Compute a0, dadx and dady for a linearly interpolated coefficient,
56 static void linear_coef( struct lp_rast_triangle
*tri
,
65 float a1
= v1
[vert_attr
][i
];
66 float a2
= v2
[vert_attr
][i
];
67 float a3
= v3
[vert_attr
][i
];
71 float dadx
= (da12
* tri
->dy31
- tri
->dy12
* da31
) * oneoverarea
;
72 float dady
= (da31
* tri
->dx12
- tri
->dx31
* da12
) * oneoverarea
;
74 tri
->inputs
.dadx
[slot
][i
] = dadx
;
75 tri
->inputs
.dady
[slot
][i
] = dady
;
77 /* calculate a0 as the value which would be sampled for the
78 * fragment at (0,0), taking into account that we want to sample at
79 * pixel centers, in other words (0.5, 0.5).
81 * this is neat but unfortunately not a good way to do things for
82 * triangles with very large values of dadx or dady as it will
83 * result in the subtraction and re-addition from a0 of a very
84 * large number, which means we'll end up loosing a lot of the
85 * fractional bits and precision from a0. the way to fix this is
86 * to define a0 as the sample at a pixel center somewhere near vmin
87 * instead - i'll switch to this later.
89 tri
->inputs
.a0
[slot
][i
] = (v1
[vert_attr
][i
] -
90 (dadx
* (v1
[0][0] - 0.5f
) +
91 dady
* (v1
[0][1] - 0.5f
)));
96 * Compute a0, dadx and dady for a perspective-corrected interpolant,
98 * We basically multiply the vertex value by 1/w before computing
99 * the plane coefficients (a0, dadx, dady).
100 * Later, when we compute the value at a particular fragment position we'll
101 * divide the interpolated value by the interpolated W at that fragment.
103 static void perspective_coef( struct lp_rast_triangle
*tri
,
106 const float (*v1
)[4],
107 const float (*v2
)[4],
108 const float (*v3
)[4],
112 /* premultiply by 1/w (v[0][3] is always 1/w):
114 float a1
= v1
[vert_attr
][i
] * v1
[0][3];
115 float a2
= v2
[vert_attr
][i
] * v2
[0][3];
116 float a3
= v3
[vert_attr
][i
] * v3
[0][3];
117 float da12
= a1
- a2
;
118 float da31
= a3
- a1
;
119 float dadx
= (da12
* tri
->dy31
- tri
->dy12
* da31
) * oneoverarea
;
120 float dady
= (da31
* tri
->dx12
- tri
->dx31
* da12
) * oneoverarea
;
122 tri
->inputs
.dadx
[slot
][i
] = dadx
;
123 tri
->inputs
.dady
[slot
][i
] = dady
;
124 tri
->inputs
.a0
[slot
][i
] = (a1
-
125 (dadx
* (v1
[0][0] - 0.5f
) +
126 dady
* (v1
[0][1] - 0.5f
)));
131 * Special coefficient setup for gl_FragCoord.
132 * X and Y are trivial, though Y has to be inverted for OpenGL.
133 * Z and W are copied from position_coef which should have already been computed.
134 * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
137 setup_fragcoord_coef(struct lp_rast_triangle
*tri
,
140 const float (*v1
)[4],
141 const float (*v2
)[4],
142 const float (*v3
)[4])
145 tri
->inputs
.a0
[slot
][0] = 0.0;
146 tri
->inputs
.dadx
[slot
][0] = 1.0;
147 tri
->inputs
.dady
[slot
][0] = 0.0;
149 tri
->inputs
.a0
[slot
][1] = 0.0;
150 tri
->inputs
.dadx
[slot
][1] = 0.0;
151 tri
->inputs
.dady
[slot
][1] = 1.0;
153 linear_coef(tri
, oneoverarea
, slot
, v1
, v2
, v3
, 0, 2);
155 linear_coef(tri
, oneoverarea
, slot
, v1
, v2
, v3
, 0, 3);
159 static void setup_facing_coef( struct lp_rast_triangle
*tri
,
163 constant_coef( tri
, slot
, 1.0f
- frontface
, 0 );
164 constant_coef( tri
, slot
, 0.0f
, 1 ); /* wasted */
165 constant_coef( tri
, slot
, 0.0f
, 2 ); /* wasted */
166 constant_coef( tri
, slot
, 0.0f
, 3 ); /* wasted */
171 * Compute the tri->coef[] array dadx, dady, a0 values.
173 static void setup_tri_coefficients( struct setup_context
*setup
,
174 struct lp_rast_triangle
*tri
,
176 const float (*v1
)[4],
177 const float (*v2
)[4],
178 const float (*v3
)[4],
181 struct lp_bins
*bins
= lp_setup_get_current_bins(setup
);
184 /* Allocate space for the a0, dadx and dady arrays
188 bytes
= (setup
->fs
.nr_inputs
+ 1) * 4 * sizeof(float);
189 tri
->inputs
.a0
= lp_bin_alloc_aligned( bins
, bytes
, 16 );
190 tri
->inputs
.dadx
= lp_bin_alloc_aligned( bins
, bytes
, 16 );
191 tri
->inputs
.dady
= lp_bin_alloc_aligned( bins
, bytes
, 16 );
194 /* The internal position input is in slot zero:
196 setup_fragcoord_coef(tri
, oneoverarea
, 0, v1
, v2
, v3
);
198 /* setup interpolation for all the remaining attributes:
200 for (slot
= 0; slot
< setup
->fs
.nr_inputs
; slot
++) {
201 unsigned vert_attr
= setup
->fs
.input
[slot
].src_index
;
204 switch (setup
->fs
.input
[slot
].interp
) {
205 case LP_INTERP_CONSTANT
:
206 for (i
= 0; i
< NUM_CHANNELS
; i
++)
207 constant_coef(tri
, slot
+1, v3
[vert_attr
][i
], i
);
210 case LP_INTERP_LINEAR
:
211 for (i
= 0; i
< NUM_CHANNELS
; i
++)
212 linear_coef(tri
, oneoverarea
, slot
+1, v1
, v2
, v3
, vert_attr
, i
);
215 case LP_INTERP_PERSPECTIVE
:
216 for (i
= 0; i
< NUM_CHANNELS
; i
++)
217 perspective_coef(tri
, oneoverarea
, slot
+1, v1
, v2
, v3
, vert_attr
, i
);
220 case LP_INTERP_POSITION
:
221 /* XXX: fix me - duplicates the values in slot zero.
223 setup_fragcoord_coef(tri
, oneoverarea
, slot
+1, v1
, v2
, v3
);
226 case LP_INTERP_FACING
:
227 setup_facing_coef(tri
, slot
+1, frontface
);
238 static inline int subpixel_snap( float a
)
240 return util_iround(FIXED_ONE
* a
);
244 #define MIN3(a,b,c) MIN2(MIN2(a,b),c)
245 #define MAX3(a,b,c) MAX2(MAX2(a,b),c)
248 * Do basic setup for triangle rasterization and determine which
249 * framebuffer tiles are touched. Put the triangle in the bins for the
250 * tiles which we overlap.
253 do_triangle_ccw(struct setup_context
*setup
,
254 const float (*v1
)[4],
255 const float (*v2
)[4],
256 const float (*v3
)[4],
257 boolean frontfacing
)
259 /* x/y positions in fixed point */
260 const int x1
= subpixel_snap(v1
[0][0]);
261 const int x2
= subpixel_snap(v2
[0][0]);
262 const int x3
= subpixel_snap(v3
[0][0]);
263 const int y1
= subpixel_snap(v1
[0][1]);
264 const int y2
= subpixel_snap(v2
[0][1]);
265 const int y3
= subpixel_snap(v3
[0][1]);
267 struct lp_bins
*bins
= lp_setup_get_current_bins(setup
);
268 struct lp_rast_triangle
*tri
= lp_bin_alloc( bins
, sizeof *tri
);
269 float area
, oneoverarea
;
270 int minx
, maxx
, miny
, maxy
;
280 area
= (tri
->dx12
* tri
->dy31
-
281 tri
->dx31
* tri
->dy12
);
283 /* Cull non-ccw and zero-sized triangles.
285 * XXX: subject to overflow??
288 lp_bin_putback_data( bins
, sizeof *tri
);
292 /* Bounding rectangle (in pixels) */
293 tri
->minx
= (MIN3(x1
, x2
, x3
) + 0xf) >> FIXED_ORDER
;
294 tri
->maxx
= (MAX3(x1
, x2
, x3
) + 0xf) >> FIXED_ORDER
;
295 tri
->miny
= (MIN3(y1
, y2
, y3
) + 0xf) >> FIXED_ORDER
;
296 tri
->maxy
= (MAX3(y1
, y2
, y3
) + 0xf) >> FIXED_ORDER
;
298 if (tri
->miny
== tri
->maxy
||
299 tri
->minx
== tri
->maxx
) {
300 lp_bin_putback_data( bins
, sizeof *tri
);
306 oneoverarea
= ((float)FIXED_ONE
) / (float)area
;
308 /* Setup parameter interpolants:
310 setup_tri_coefficients( setup
, tri
, oneoverarea
, v1
, v2
, v3
, frontfacing
);
312 /* half-edge constants, will be interated over the whole
315 tri
->c1
= tri
->dy12
* x1
- tri
->dx12
* y1
;
316 tri
->c2
= tri
->dy23
* x2
- tri
->dx23
* y2
;
317 tri
->c3
= tri
->dy31
* x3
- tri
->dx31
* y3
;
319 /* correct for top-left fill convention:
321 if (tri
->dy12
< 0 || (tri
->dy12
== 0 && tri
->dx12
> 0)) tri
->c1
++;
322 if (tri
->dy23
< 0 || (tri
->dy23
== 0 && tri
->dx23
> 0)) tri
->c2
++;
323 if (tri
->dy31
< 0 || (tri
->dy31
== 0 && tri
->dx31
> 0)) tri
->c3
++;
325 tri
->dy12
*= FIXED_ONE
;
326 tri
->dy23
*= FIXED_ONE
;
327 tri
->dy31
*= FIXED_ONE
;
329 tri
->dx12
*= FIXED_ONE
;
330 tri
->dx23
*= FIXED_ONE
;
331 tri
->dx31
*= FIXED_ONE
;
333 /* find trivial reject offsets for each edge for a single-pixel
334 * sized block. These will be scaled up at each recursive level to
335 * match the active blocksize. Scaling in this way works best if
336 * the blocks are square.
339 if (tri
->dy12
< 0) tri
->eo1
-= tri
->dy12
;
340 if (tri
->dx12
> 0) tri
->eo1
+= tri
->dx12
;
343 if (tri
->dy23
< 0) tri
->eo2
-= tri
->dy23
;
344 if (tri
->dx23
> 0) tri
->eo2
+= tri
->dx23
;
347 if (tri
->dy31
< 0) tri
->eo3
-= tri
->dy31
;
348 if (tri
->dx31
> 0) tri
->eo3
+= tri
->dx31
;
350 /* Calculate trivial accept offsets from the above.
352 tri
->ei1
= tri
->dx12
- tri
->dy12
- tri
->eo1
;
353 tri
->ei2
= tri
->dx23
- tri
->dy23
- tri
->eo2
;
354 tri
->ei3
= tri
->dx31
- tri
->dy31
- tri
->eo3
;
357 int xstep1
= -tri
->dy12
;
358 int xstep2
= -tri
->dy23
;
359 int xstep3
= -tri
->dy31
;
361 int ystep1
= tri
->dx12
;
362 int ystep2
= tri
->dx23
;
363 int ystep3
= tri
->dx31
;
372 for (iy
= 0; iy
< 4; iy
++) {
377 for (ix
= 0; ix
< 4; ix
++, i
++) {
378 tri
->step
[0][i
] = cx1
;
379 tri
->step
[1][i
] = cx2
;
380 tri
->step
[2][i
] = cx3
;
393 * All fields of 'tri' are now set. The remaining code here is
394 * concerned with binning.
397 /* Convert to tile coordinates:
399 minx
= tri
->minx
/ TILE_SIZE
;
400 miny
= tri
->miny
/ TILE_SIZE
;
401 maxx
= tri
->maxx
/ TILE_SIZE
;
402 maxy
= tri
->maxy
/ TILE_SIZE
;
404 /* Determine which tile(s) intersect the triangle's bounding box
406 if (miny
== maxy
&& minx
== maxx
)
408 /* Triangle is contained in a single tile:
410 lp_bin_command( bins
, minx
, miny
, lp_rast_triangle
,
411 lp_rast_arg_triangle(tri
) );
416 tri
->dx12
* miny
* TILE_SIZE
-
417 tri
->dy12
* minx
* TILE_SIZE
);
419 tri
->dx23
* miny
* TILE_SIZE
-
420 tri
->dy23
* minx
* TILE_SIZE
);
422 tri
->dx31
* miny
* TILE_SIZE
-
423 tri
->dy31
* minx
* TILE_SIZE
);
425 int ei1
= tri
->ei1
<< TILE_ORDER
;
426 int ei2
= tri
->ei2
<< TILE_ORDER
;
427 int ei3
= tri
->ei3
<< TILE_ORDER
;
429 int eo1
= tri
->eo1
<< TILE_ORDER
;
430 int eo2
= tri
->eo2
<< TILE_ORDER
;
431 int eo3
= tri
->eo3
<< TILE_ORDER
;
433 int xstep1
= -(tri
->dy12
<< TILE_ORDER
);
434 int xstep2
= -(tri
->dy23
<< TILE_ORDER
);
435 int xstep3
= -(tri
->dy31
<< TILE_ORDER
);
437 int ystep1
= tri
->dx12
<< TILE_ORDER
;
438 int ystep2
= tri
->dx23
<< TILE_ORDER
;
439 int ystep3
= tri
->dx31
<< TILE_ORDER
;
443 /* Trivially accept or reject blocks, else jump to per-pixel
446 for (y
= miny
; y
<= maxy
; y
++)
453 for (x
= minx
; x
<= maxx
; x
++)
463 else if (cx1
+ ei1
> 0 &&
468 /* triangle covers the whole tile- shade whole tile */
469 lp_bin_command( bins
, x
, y
,
471 lp_rast_arg_inputs(&tri
->inputs
) );
476 /* shade partial tile */
477 lp_bin_command( bins
, x
, y
,
479 lp_rast_arg_triangle(tri
) );
482 /* Iterate cx values across the region:
489 /* Iterate c values down the region:
498 static void triangle_cw( struct setup_context
*setup
,
499 const float (*v0
)[4],
500 const float (*v1
)[4],
501 const float (*v2
)[4] )
503 do_triangle_ccw( setup
, v1
, v0
, v2
, !setup
->ccw_is_frontface
);
506 static void triangle_ccw( struct setup_context
*setup
,
507 const float (*v0
)[4],
508 const float (*v1
)[4],
509 const float (*v2
)[4] )
511 do_triangle_ccw( setup
, v0
, v1
, v2
, setup
->ccw_is_frontface
);
514 static void triangle_both( struct setup_context
*setup
,
515 const float (*v0
)[4],
516 const float (*v1
)[4],
517 const float (*v2
)[4] )
519 /* edge vectors e = v0 - v2, f = v1 - v2 */
520 const float ex
= v0
[0][0] - v2
[0][0];
521 const float ey
= v0
[0][1] - v2
[0][1];
522 const float fx
= v1
[0][0] - v2
[0][0];
523 const float fy
= v1
[0][1] - v2
[0][1];
525 /* det = cross(e,f).z */
526 if (ex
* fy
- ey
* fx
< 0)
527 triangle_ccw( setup
, v0
, v1
, v2
);
529 triangle_cw( setup
, v0
, v1
, v2
);
532 static void triangle_nop( struct setup_context
*setup
,
533 const float (*v0
)[4],
534 const float (*v1
)[4],
535 const float (*v2
)[4] )
541 lp_setup_choose_triangle( struct setup_context
*setup
)
543 switch (setup
->cullmode
) {
544 case PIPE_WINDING_NONE
:
545 setup
->triangle
= triangle_both
;
547 case PIPE_WINDING_CCW
:
548 setup
->triangle
= triangle_cw
;
550 case PIPE_WINDING_CW
:
551 setup
->triangle
= triangle_ccw
;
554 setup
->triangle
= triangle_nop
;