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 * \brief Primitive rasterization/rendering (points, lines, triangles)
31 * \author Keith Whitwell <keith@tungstengraphics.com>
35 #include "sp_context.h"
37 #include "sp_quad_pipe.h"
40 #include "draw/draw_context.h"
41 #include "draw/draw_vertex.h"
42 #include "pipe/p_shader_tokens.h"
43 #include "util/u_math.h"
44 #include "util/u_memory.h"
54 float dx
; /**< X(v1) - X(v0), used only during setup */
55 float dy
; /**< Y(v1) - Y(v0), used only during setup */
56 float dxdy
; /**< dx/dy */
57 float sx
, sy
; /**< first sample point coord */
58 int lines
; /**< number of lines on this edge */
66 * Triangle setup info (derived from draw_stage).
67 * Also used for line drawing (taking some liberties).
69 struct setup_context
{
70 struct softpipe_context
*softpipe
;
72 /* Vertices are just an array of floats making up each attribute in
73 * turn. Currently fixed at 4 floats, but should change in time.
74 * Codegen will help cope with this.
76 const float (*vmax
)[4];
77 const float (*vmid
)[4];
78 const float (*vmin
)[4];
79 const float (*vprovoke
)[4];
90 struct quad_header quad
[MAX_QUADS
];
91 struct quad_header
*quad_ptrs
[MAX_QUADS
];
94 struct tgsi_interp_coef coef
[PIPE_MAX_SHADER_INPUTS
];
95 struct tgsi_interp_coef posCoef
; /* For Z, W */
98 int left
[2]; /**< [0] = row0, [1] = row1 */
104 uint numFragsEmitted
; /**< per primitive */
105 uint numFragsWritten
; /**< per primitive */
108 unsigned winding
; /* which winding to cull */
109 unsigned nr_vertex_attrs
;
116 * Do triangle cull test using tri determinant (sign indicates orientation)
117 * \return true if triangle is to be culled.
119 static INLINE boolean
120 cull_tri(const struct setup_context
*setup
, float det
)
123 /* if (det < 0 then Z points toward camera and triangle is
124 * counter-clockwise winding.
126 unsigned winding
= (det
< 0) ? PIPE_WINDING_CCW
: PIPE_WINDING_CW
;
128 if ((winding
& setup
->winding
) == 0)
140 * Clip setup->quad against the scissor/surface bounds.
143 quad_clip( struct setup_context
*setup
, struct quad_header
*quad
)
145 const struct pipe_scissor_state
*cliprect
= &setup
->softpipe
->cliprect
;
146 const int minx
= (int) cliprect
->minx
;
147 const int maxx
= (int) cliprect
->maxx
;
148 const int miny
= (int) cliprect
->miny
;
149 const int maxy
= (int) cliprect
->maxy
;
151 if (quad
->input
.x0
>= maxx
||
152 quad
->input
.y0
>= maxy
||
153 quad
->input
.x0
+ 1 < minx
||
154 quad
->input
.y0
+ 1 < miny
) {
155 /* totally clipped */
156 quad
->inout
.mask
= 0x0;
159 if (quad
->input
.x0
< minx
)
160 quad
->inout
.mask
&= (MASK_BOTTOM_RIGHT
| MASK_TOP_RIGHT
);
161 if (quad
->input
.y0
< miny
)
162 quad
->inout
.mask
&= (MASK_BOTTOM_LEFT
| MASK_BOTTOM_RIGHT
);
163 if (quad
->input
.x0
== maxx
- 1)
164 quad
->inout
.mask
&= (MASK_BOTTOM_LEFT
| MASK_TOP_LEFT
);
165 if (quad
->input
.y0
== maxy
- 1)
166 quad
->inout
.mask
&= (MASK_TOP_LEFT
| MASK_TOP_RIGHT
);
171 * Emit a quad (pass to next stage) with clipping.
174 clip_emit_quad( struct setup_context
*setup
, struct quad_header
*quad
)
176 quad_clip( setup
, quad
);
178 if (quad
->inout
.mask
) {
179 struct softpipe_context
*sp
= setup
->softpipe
;
181 sp
->quad
.first
->run( sp
->quad
.first
, &quad
, 1 );
188 * Given an X or Y coordinate, return the block/quad coordinate that it
191 static INLINE
int block( int x
)
196 static INLINE
int block_x( int x
)
203 * Render a horizontal span of quads
205 static void flush_spans( struct setup_context
*setup
)
208 const int xleft0
= setup
->span
.left
[0];
209 const int xleft1
= setup
->span
.left
[1];
210 const int xright0
= setup
->span
.right
[0];
211 const int xright1
= setup
->span
.right
[1];
212 struct quad_stage
*pipe
= setup
->softpipe
->quad
.first
;
215 int minleft
= block_x(MIN2(xleft0
, xleft1
));
216 int maxright
= MAX2(xright0
, xright1
);
219 for (x
= minleft
; x
< maxright
; x
+= step
) {
220 unsigned skip_left0
= CLAMP(xleft0
- x
, 0, step
);
221 unsigned skip_left1
= CLAMP(xleft1
- x
, 0, step
);
222 unsigned skip_right0
= CLAMP(x
+ step
- xright0
, 0, step
);
223 unsigned skip_right1
= CLAMP(x
+ step
- xright1
, 0, step
);
227 unsigned skipmask_left0
= (1U << skip_left0
) - 1U;
228 unsigned skipmask_left1
= (1U << skip_left1
) - 1U;
230 /* These calculations fail when step == 32 and skip_right == 0.
232 unsigned skipmask_right0
= ~0U << (unsigned)(step
- skip_right0
);
233 unsigned skipmask_right1
= ~0U << (unsigned)(step
- skip_right1
);
235 unsigned mask0
= ~skipmask_left0
& ~skipmask_right0
;
236 unsigned mask1
= ~skipmask_left1
& ~skipmask_right1
;
240 unsigned quadmask
= (mask0
& 3) | ((mask1
& 3) << 2);
242 setup
->quad
[q
].input
.x0
= lx
;
243 setup
->quad
[q
].input
.y0
= setup
->span
.y
;
244 setup
->quad
[q
].input
.facing
= setup
->facing
;
245 setup
->quad
[q
].inout
.mask
= quadmask
;
246 setup
->quad_ptrs
[q
] = &setup
->quad
[q
];
252 } while (mask0
| mask1
);
254 pipe
->run( pipe
, setup
->quad_ptrs
, q
);
260 setup
->span
.right
[0] = 0;
261 setup
->span
.right
[1] = 0;
262 setup
->span
.left
[0] = 1000000; /* greater than right[0] */
263 setup
->span
.left
[1] = 1000000; /* greater than right[1] */
268 static void print_vertex(const struct setup_context
*setup
,
272 debug_printf(" Vertex: (%p)\n", (void *) v
);
273 for (i
= 0; i
< setup
->nr_vertex_attrs
; i
++) {
274 debug_printf(" %d: %f %f %f %f\n", i
,
275 v
[i
][0], v
[i
][1], v
[i
][2], v
[i
][3]);
276 if (util_is_inf_or_nan(v
[i
][0])) {
277 debug_printf(" NaN!\n");
284 * Sort the vertices from top to bottom order, setting up the triangle
285 * edge fields (ebot, emaj, etop).
286 * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise
288 static boolean
setup_sort_vertices( struct setup_context
*setup
,
290 const float (*v0
)[4],
291 const float (*v1
)[4],
292 const float (*v2
)[4] )
294 setup
->vprovoke
= v2
;
296 /* determine bottom to top order of vertices */
343 setup
->ebot
.dx
= setup
->vmid
[0][0] - setup
->vmin
[0][0];
344 setup
->ebot
.dy
= setup
->vmid
[0][1] - setup
->vmin
[0][1];
345 setup
->emaj
.dx
= setup
->vmax
[0][0] - setup
->vmin
[0][0];
346 setup
->emaj
.dy
= setup
->vmax
[0][1] - setup
->vmin
[0][1];
347 setup
->etop
.dx
= setup
->vmax
[0][0] - setup
->vmid
[0][0];
348 setup
->etop
.dy
= setup
->vmax
[0][1] - setup
->vmid
[0][1];
351 * Compute triangle's area. Use 1/area to compute partial
352 * derivatives of attributes later.
354 * The area will be the same as prim->det, but the sign may be
355 * different depending on how the vertices get sorted above.
357 * To determine whether the primitive is front or back facing we
358 * use the prim->det value because its sign is correct.
361 const float area
= (setup
->emaj
.dx
* setup
->ebot
.dy
-
362 setup
->ebot
.dx
* setup
->emaj
.dy
);
364 setup
->oneoverarea
= 1.0f
/ area
;
367 debug_printf("%s one-over-area %f area %f det %f\n",
368 __FUNCTION__, setup->oneoverarea, area, det );
370 if (util_is_inf_or_nan(setup
->oneoverarea
))
374 /* We need to know if this is a front or back-facing triangle for:
375 * - the GLSL gl_FrontFacing fragment attribute (bool)
376 * - two-sided stencil test
380 (setup
->softpipe
->rasterizer
->front_winding
== PIPE_WINDING_CW
));
382 /* Prepare pixel offset for rasterisation:
383 * - pixel center (0.5, 0.5) for GL, or
384 * - assume (0.0, 0.0) for other APIs.
386 if (setup
->softpipe
->rasterizer
->gl_rasterization_rules
) {
387 setup
->pixel_offset
= 0.5f
;
389 setup
->pixel_offset
= 0.0f
;
397 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
398 * The value value comes from vertex[slot][i].
399 * The result will be put into setup->coef[slot].a0[i].
400 * \param slot which attribute slot
401 * \param i which component of the slot (0..3)
403 static void const_coeff( struct setup_context
*setup
,
404 struct tgsi_interp_coef
*coef
,
405 uint vertSlot
, uint i
)
412 /* need provoking vertex info!
414 coef
->a0
[i
] = setup
->vprovoke
[vertSlot
][i
];
419 * Compute a0, dadx and dady for a linearly interpolated coefficient,
422 static void tri_linear_coeff( struct setup_context
*setup
,
423 struct tgsi_interp_coef
*coef
,
424 uint vertSlot
, uint i
)
426 float botda
= setup
->vmid
[vertSlot
][i
] - setup
->vmin
[vertSlot
][i
];
427 float majda
= setup
->vmax
[vertSlot
][i
] - setup
->vmin
[vertSlot
][i
];
428 float a
= setup
->ebot
.dy
* majda
- botda
* setup
->emaj
.dy
;
429 float b
= setup
->emaj
.dx
* botda
- majda
* setup
->ebot
.dx
;
430 float dadx
= a
* setup
->oneoverarea
;
431 float dady
= b
* setup
->oneoverarea
;
435 coef
->dadx
[i
] = dadx
;
436 coef
->dady
[i
] = dady
;
438 /* calculate a0 as the value which would be sampled for the
439 * fragment at (0,0), taking into account that we want to sample at
440 * pixel centers, in other words (pixel_offset, pixel_offset).
442 * this is neat but unfortunately not a good way to do things for
443 * triangles with very large values of dadx or dady as it will
444 * result in the subtraction and re-addition from a0 of a very
445 * large number, which means we'll end up loosing a lot of the
446 * fractional bits and precision from a0. the way to fix this is
447 * to define a0 as the sample at a pixel center somewhere near vmin
448 * instead - i'll switch to this later.
450 coef
->a0
[i
] = (setup
->vmin
[vertSlot
][i
] -
451 (dadx
* (setup
->vmin
[0][0] - setup
->pixel_offset
) +
452 dady
* (setup
->vmin
[0][1] - setup
->pixel_offset
)));
455 debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
457 setup->coef[slot].a0[i],
458 setup->coef[slot].dadx[i],
459 setup->coef[slot].dady[i]);
465 * Compute a0, dadx and dady for a perspective-corrected interpolant,
467 * We basically multiply the vertex value by 1/w before computing
468 * the plane coefficients (a0, dadx, dady).
469 * Later, when we compute the value at a particular fragment position we'll
470 * divide the interpolated value by the interpolated W at that fragment.
472 static void tri_persp_coeff( struct setup_context
*setup
,
473 struct tgsi_interp_coef
*coef
,
474 uint vertSlot
, uint i
)
476 /* premultiply by 1/w (v[0][3] is always W):
478 float mina
= setup
->vmin
[vertSlot
][i
] * setup
->vmin
[0][3];
479 float mida
= setup
->vmid
[vertSlot
][i
] * setup
->vmid
[0][3];
480 float maxa
= setup
->vmax
[vertSlot
][i
] * setup
->vmax
[0][3];
481 float botda
= mida
- mina
;
482 float majda
= maxa
- mina
;
483 float a
= setup
->ebot
.dy
* majda
- botda
* setup
->emaj
.dy
;
484 float b
= setup
->emaj
.dx
* botda
- majda
* setup
->ebot
.dx
;
485 float dadx
= a
* setup
->oneoverarea
;
486 float dady
= b
* setup
->oneoverarea
;
489 debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i,
490 setup->vmin[vertSlot][i],
491 setup->vmid[vertSlot][i],
492 setup->vmax[vertSlot][i]
497 coef
->dadx
[i
] = dadx
;
498 coef
->dady
[i
] = dady
;
499 coef
->a0
[i
] = (mina
-
500 (dadx
* (setup
->vmin
[0][0] - setup
->pixel_offset
) +
501 dady
* (setup
->vmin
[0][1] - setup
->pixel_offset
)));
506 * Special coefficient setup for gl_FragCoord.
507 * X and Y are trivial, though Y has to be inverted for OpenGL.
508 * Z and W are copied from posCoef which should have already been computed.
509 * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
512 setup_fragcoord_coeff(struct setup_context
*setup
, uint slot
)
515 setup
->coef
[slot
].a0
[0] = 0;
516 setup
->coef
[slot
].dadx
[0] = 1.0;
517 setup
->coef
[slot
].dady
[0] = 0.0;
519 setup
->coef
[slot
].a0
[1] = 0.0;
520 setup
->coef
[slot
].dadx
[1] = 0.0;
521 setup
->coef
[slot
].dady
[1] = 1.0;
523 setup
->coef
[slot
].a0
[2] = setup
->posCoef
.a0
[2];
524 setup
->coef
[slot
].dadx
[2] = setup
->posCoef
.dadx
[2];
525 setup
->coef
[slot
].dady
[2] = setup
->posCoef
.dady
[2];
527 setup
->coef
[slot
].a0
[3] = setup
->posCoef
.a0
[3];
528 setup
->coef
[slot
].dadx
[3] = setup
->posCoef
.dadx
[3];
529 setup
->coef
[slot
].dady
[3] = setup
->posCoef
.dady
[3];
535 * Compute the setup->coef[] array dadx, dady, a0 values.
536 * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized.
538 static void setup_tri_coefficients( struct setup_context
*setup
)
540 struct softpipe_context
*softpipe
= setup
->softpipe
;
541 const struct sp_fragment_shader
*spfs
= softpipe
->fs
;
542 const struct vertex_info
*vinfo
= softpipe_get_vertex_info(softpipe
);
545 /* z and w are done by linear interpolation:
547 tri_linear_coeff(setup
, &setup
->posCoef
, 0, 2);
548 tri_linear_coeff(setup
, &setup
->posCoef
, 0, 3);
550 /* setup interpolation for all the remaining attributes:
552 for (fragSlot
= 0; fragSlot
< spfs
->info
.num_inputs
; fragSlot
++) {
553 const uint vertSlot
= vinfo
->attrib
[fragSlot
].src_index
;
556 switch (vinfo
->attrib
[fragSlot
].interp_mode
) {
557 case INTERP_CONSTANT
:
558 for (j
= 0; j
< NUM_CHANNELS
; j
++)
559 const_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
562 for (j
= 0; j
< NUM_CHANNELS
; j
++)
563 tri_linear_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
565 case INTERP_PERSPECTIVE
:
566 for (j
= 0; j
< NUM_CHANNELS
; j
++)
567 tri_persp_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
570 setup_fragcoord_coeff(setup
, fragSlot
);
576 if (spfs
->info
.input_semantic_name
[fragSlot
] == TGSI_SEMANTIC_FACE
) {
577 setup
->coef
[fragSlot
].a0
[0] = 1.0f
- setup
->facing
;
578 setup
->coef
[fragSlot
].dadx
[0] = 0.0;
579 setup
->coef
[fragSlot
].dady
[0] = 0.0;
586 static void setup_tri_edges( struct setup_context
*setup
)
588 float vmin_x
= setup
->vmin
[0][0] + setup
->pixel_offset
;
589 float vmid_x
= setup
->vmid
[0][0] + setup
->pixel_offset
;
591 float vmin_y
= setup
->vmin
[0][1] - setup
->pixel_offset
;
592 float vmid_y
= setup
->vmid
[0][1] - setup
->pixel_offset
;
593 float vmax_y
= setup
->vmax
[0][1] - setup
->pixel_offset
;
595 setup
->emaj
.sy
= ceilf(vmin_y
);
596 setup
->emaj
.lines
= (int) ceilf(vmax_y
- setup
->emaj
.sy
);
597 setup
->emaj
.dxdy
= setup
->emaj
.dx
/ setup
->emaj
.dy
;
598 setup
->emaj
.sx
= vmin_x
+ (setup
->emaj
.sy
- vmin_y
) * setup
->emaj
.dxdy
;
600 setup
->etop
.sy
= ceilf(vmid_y
);
601 setup
->etop
.lines
= (int) ceilf(vmax_y
- setup
->etop
.sy
);
602 setup
->etop
.dxdy
= setup
->etop
.dx
/ setup
->etop
.dy
;
603 setup
->etop
.sx
= vmid_x
+ (setup
->etop
.sy
- vmid_y
) * setup
->etop
.dxdy
;
605 setup
->ebot
.sy
= ceilf(vmin_y
);
606 setup
->ebot
.lines
= (int) ceilf(vmid_y
- setup
->ebot
.sy
);
607 setup
->ebot
.dxdy
= setup
->ebot
.dx
/ setup
->ebot
.dy
;
608 setup
->ebot
.sx
= vmin_x
+ (setup
->ebot
.sy
- vmin_y
) * setup
->ebot
.dxdy
;
613 * Render the upper or lower half of a triangle.
614 * Scissoring/cliprect is applied here too.
616 static void subtriangle( struct setup_context
*setup
,
621 const struct pipe_scissor_state
*cliprect
= &setup
->softpipe
->cliprect
;
622 const int minx
= (int) cliprect
->minx
;
623 const int maxx
= (int) cliprect
->maxx
;
624 const int miny
= (int) cliprect
->miny
;
625 const int maxy
= (int) cliprect
->maxy
;
626 int y
, start_y
, finish_y
;
627 int sy
= (int)eleft
->sy
;
629 assert((int)eleft
->sy
== (int) eright
->sy
);
631 /* clip top/bottom */
636 finish_y
= sy
+ lines
;
644 debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);
647 for (y
= start_y
; y
< finish_y
; y
++) {
649 /* avoid accumulating adds as floats don't have the precision to
650 * accurately iterate large triangle edges that way. luckily we
651 * can just multiply these days.
653 * this is all drowned out by the attribute interpolation anyway.
655 int left
= (int)(eleft
->sx
+ y
* eleft
->dxdy
);
656 int right
= (int)(eright
->sx
+ y
* eright
->dxdy
);
658 /* clip left/right */
666 if (block(_y
) != setup
->span
.y
) {
668 setup
->span
.y
= block(_y
);
671 setup
->span
.left
[_y
&1] = left
;
672 setup
->span
.right
[_y
&1] = right
;
677 /* save the values so that emaj can be restarted:
679 eleft
->sx
+= lines
* eleft
->dxdy
;
680 eright
->sx
+= lines
* eright
->dxdy
;
687 * Recalculate prim's determinant. This is needed as we don't have
688 * get this information through the vbuf_render interface & we must
692 calc_det( const float (*v0
)[4],
693 const float (*v1
)[4],
694 const float (*v2
)[4] )
696 /* edge vectors e = v0 - v2, f = v1 - v2 */
697 const float ex
= v0
[0][0] - v2
[0][0];
698 const float ey
= v0
[0][1] - v2
[0][1];
699 const float fx
= v1
[0][0] - v2
[0][0];
700 const float fy
= v1
[0][1] - v2
[0][1];
702 /* det = cross(e,f).z */
703 return ex
* fy
- ey
* fx
;
708 * Do setup for triangle rasterization, then render the triangle.
710 void sp_setup_tri( struct setup_context
*setup
,
711 const float (*v0
)[4],
712 const float (*v1
)[4],
713 const float (*v2
)[4] )
718 debug_printf("Setup triangle:\n");
719 print_vertex(setup
, v0
);
720 print_vertex(setup
, v1
);
721 print_vertex(setup
, v2
);
724 if (setup
->softpipe
->no_rast
)
727 det
= calc_det(v0
, v1
, v2
);
729 debug_printf("%s\n", __FUNCTION__ );
733 setup
->numFragsEmitted
= 0;
734 setup
->numFragsWritten
= 0;
737 if (cull_tri( setup
, det
))
740 if (!setup_sort_vertices( setup
, det
, v0
, v1
, v2
))
742 setup_tri_coefficients( setup
);
743 setup_tri_edges( setup
);
745 assert(setup
->softpipe
->reduced_prim
== PIPE_PRIM_TRIANGLES
);
748 setup
->span
.right
[0] = 0;
749 setup
->span
.right
[1] = 0;
750 /* setup->span.z_mode = tri_z_mode( setup->ctx ); */
752 /* init_constant_attribs( setup ); */
754 if (setup
->oneoverarea
< 0.0) {
757 subtriangle( setup
, &setup
->emaj
, &setup
->ebot
, setup
->ebot
.lines
);
758 subtriangle( setup
, &setup
->emaj
, &setup
->etop
, setup
->etop
.lines
);
763 subtriangle( setup
, &setup
->ebot
, &setup
->emaj
, setup
->ebot
.lines
);
764 subtriangle( setup
, &setup
->etop
, &setup
->emaj
, setup
->etop
.lines
);
767 flush_spans( setup
);
770 printf("Tri: %u frags emitted, %u written\n",
771 setup
->numFragsEmitted
,
772 setup
->numFragsWritten
);
779 * Compute a0, dadx and dady for a linearly interpolated coefficient,
783 line_linear_coeff(const struct setup_context
*setup
,
784 struct tgsi_interp_coef
*coef
,
785 uint vertSlot
, uint i
)
787 const float da
= setup
->vmax
[vertSlot
][i
] - setup
->vmin
[vertSlot
][i
];
788 const float dadx
= da
* setup
->emaj
.dx
* setup
->oneoverarea
;
789 const float dady
= da
* setup
->emaj
.dy
* setup
->oneoverarea
;
790 coef
->dadx
[i
] = dadx
;
791 coef
->dady
[i
] = dady
;
792 coef
->a0
[i
] = (setup
->vmin
[vertSlot
][i
] -
793 (dadx
* (setup
->vmin
[0][0] - setup
->pixel_offset
) +
794 dady
* (setup
->vmin
[0][1] - setup
->pixel_offset
)));
799 * Compute a0, dadx and dady for a perspective-corrected interpolant,
803 line_persp_coeff(const struct setup_context
*setup
,
804 struct tgsi_interp_coef
*coef
,
805 uint vertSlot
, uint i
)
807 /* XXX double-check/verify this arithmetic */
808 const float a0
= setup
->vmin
[vertSlot
][i
] * setup
->vmin
[0][3];
809 const float a1
= setup
->vmax
[vertSlot
][i
] * setup
->vmax
[0][3];
810 const float da
= a1
- a0
;
811 const float dadx
= da
* setup
->emaj
.dx
* setup
->oneoverarea
;
812 const float dady
= da
* setup
->emaj
.dy
* setup
->oneoverarea
;
813 coef
->dadx
[i
] = dadx
;
814 coef
->dady
[i
] = dady
;
815 coef
->a0
[i
] = (setup
->vmin
[vertSlot
][i
] -
816 (dadx
* (setup
->vmin
[0][0] - setup
->pixel_offset
) +
817 dady
* (setup
->vmin
[0][1] - setup
->pixel_offset
)));
822 * Compute the setup->coef[] array dadx, dady, a0 values.
823 * Must be called after setup->vmin,vmax are initialized.
825 static INLINE boolean
826 setup_line_coefficients(struct setup_context
*setup
,
827 const float (*v0
)[4],
828 const float (*v1
)[4])
830 struct softpipe_context
*softpipe
= setup
->softpipe
;
831 const struct sp_fragment_shader
*spfs
= softpipe
->fs
;
832 const struct vertex_info
*vinfo
= softpipe_get_vertex_info(softpipe
);
836 /* use setup->vmin, vmax to point to vertices */
837 if (softpipe
->rasterizer
->flatshade_first
)
838 setup
->vprovoke
= v0
;
840 setup
->vprovoke
= v1
;
844 setup
->emaj
.dx
= setup
->vmax
[0][0] - setup
->vmin
[0][0];
845 setup
->emaj
.dy
= setup
->vmax
[0][1] - setup
->vmin
[0][1];
847 /* NOTE: this is not really area but something proportional to it */
848 area
= setup
->emaj
.dx
* setup
->emaj
.dx
+ setup
->emaj
.dy
* setup
->emaj
.dy
;
849 if (area
== 0.0f
|| util_is_inf_or_nan(area
))
851 setup
->oneoverarea
= 1.0f
/ area
;
853 /* z and w are done by linear interpolation:
855 line_linear_coeff(setup
, &setup
->posCoef
, 0, 2);
856 line_linear_coeff(setup
, &setup
->posCoef
, 0, 3);
858 /* setup interpolation for all the remaining attributes:
860 for (fragSlot
= 0; fragSlot
< spfs
->info
.num_inputs
; fragSlot
++) {
861 const uint vertSlot
= vinfo
->attrib
[fragSlot
].src_index
;
864 switch (vinfo
->attrib
[fragSlot
].interp_mode
) {
865 case INTERP_CONSTANT
:
866 for (j
= 0; j
< NUM_CHANNELS
; j
++)
867 const_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
870 for (j
= 0; j
< NUM_CHANNELS
; j
++)
871 line_linear_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
873 case INTERP_PERSPECTIVE
:
874 for (j
= 0; j
< NUM_CHANNELS
; j
++)
875 line_persp_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
878 setup_fragcoord_coeff(setup
, fragSlot
);
884 if (spfs
->info
.input_semantic_name
[fragSlot
] == TGSI_SEMANTIC_FACE
) {
885 setup
->coef
[fragSlot
].a0
[0] = 1.0f
- setup
->facing
;
886 setup
->coef
[fragSlot
].dadx
[0] = 0.0;
887 setup
->coef
[fragSlot
].dady
[0] = 0.0;
895 * Plot a pixel in a line segment.
898 plot(struct setup_context
*setup
, int x
, int y
)
900 const int iy
= y
& 1;
901 const int ix
= x
& 1;
902 const int quadX
= x
- ix
;
903 const int quadY
= y
- iy
;
904 const int mask
= (1 << ix
) << (2 * iy
);
906 if (quadX
!= setup
->quad
[0].input
.x0
||
907 quadY
!= setup
->quad
[0].input
.y0
)
909 /* flush prev quad, start new quad */
911 if (setup
->quad
[0].input
.x0
!= -1)
912 clip_emit_quad( setup
, &setup
->quad
[0] );
914 setup
->quad
[0].input
.x0
= quadX
;
915 setup
->quad
[0].input
.y0
= quadY
;
916 setup
->quad
[0].inout
.mask
= 0x0;
919 setup
->quad
[0].inout
.mask
|= mask
;
924 * Do setup for line rasterization, then render the line.
925 * Single-pixel width, no stipple, etc. We rely on the 'draw' module
926 * to handle stippling and wide lines.
929 sp_setup_line(struct setup_context
*setup
,
930 const float (*v0
)[4],
931 const float (*v1
)[4])
933 int x0
= (int) v0
[0][0];
934 int x1
= (int) v1
[0][0];
935 int y0
= (int) v0
[0][1];
936 int y1
= (int) v1
[0][1];
942 debug_printf("Setup line:\n");
943 print_vertex(setup
, v0
);
944 print_vertex(setup
, v1
);
947 if (setup
->softpipe
->no_rast
)
950 if (dx
== 0 && dy
== 0)
953 if (!setup_line_coefficients(setup
, v0
, v1
))
956 assert(v0
[0][0] < 1.0e9
);
957 assert(v0
[0][1] < 1.0e9
);
958 assert(v1
[0][0] < 1.0e9
);
959 assert(v1
[0][1] < 1.0e9
);
962 dx
= -dx
; /* make positive */
970 dy
= -dy
; /* make positive */
979 assert(setup
->softpipe
->reduced_prim
== PIPE_PRIM_LINES
);
981 setup
->quad
[0].input
.x0
= setup
->quad
[0].input
.y0
= -1;
982 setup
->quad
[0].inout
.mask
= 0x0;
984 /* XXX temporary: set coverage to 1.0 so the line appears
985 * if AA mode happens to be enabled.
987 setup
->quad
[0].input
.coverage
[0] =
988 setup
->quad
[0].input
.coverage
[1] =
989 setup
->quad
[0].input
.coverage
[2] =
990 setup
->quad
[0].input
.coverage
[3] = 1.0;
993 /*** X-major line ***/
995 const int errorInc
= dy
+ dy
;
996 int error
= errorInc
- dx
;
997 const int errorDec
= error
- dx
;
999 for (i
= 0; i
< dx
; i
++) {
1000 plot(setup
, x0
, y0
);
1013 /*** Y-major line ***/
1015 const int errorInc
= dx
+ dx
;
1016 int error
= errorInc
- dy
;
1017 const int errorDec
= error
- dy
;
1019 for (i
= 0; i
< dy
; i
++) {
1020 plot(setup
, x0
, y0
);
1033 /* draw final quad */
1034 if (setup
->quad
[0].inout
.mask
) {
1035 clip_emit_quad( setup
, &setup
->quad
[0] );
1041 point_persp_coeff(const struct setup_context
*setup
,
1042 const float (*vert
)[4],
1043 struct tgsi_interp_coef
*coef
,
1044 uint vertSlot
, uint i
)
1047 coef
->dadx
[i
] = 0.0F
;
1048 coef
->dady
[i
] = 0.0F
;
1049 coef
->a0
[i
] = vert
[vertSlot
][i
] * vert
[0][3];
1054 * Do setup for point rasterization, then render the point.
1055 * Round or square points...
1056 * XXX could optimize a lot for 1-pixel points.
1059 sp_setup_point( struct setup_context
*setup
,
1060 const float (*v0
)[4] )
1062 struct softpipe_context
*softpipe
= setup
->softpipe
;
1063 const struct sp_fragment_shader
*spfs
= softpipe
->fs
;
1064 const int sizeAttr
= setup
->softpipe
->psize_slot
;
1066 = sizeAttr
> 0 ? v0
[sizeAttr
][0]
1067 : setup
->softpipe
->rasterizer
->point_size
;
1068 const float halfSize
= 0.5F
* size
;
1069 const boolean round
= (boolean
) setup
->softpipe
->rasterizer
->point_smooth
;
1070 const float x
= v0
[0][0]; /* Note: data[0] is always position */
1071 const float y
= v0
[0][1];
1072 const struct vertex_info
*vinfo
= softpipe_get_vertex_info(softpipe
);
1076 debug_printf("Setup point:\n");
1077 print_vertex(setup
, v0
);
1080 if (softpipe
->no_rast
)
1083 assert(setup
->softpipe
->reduced_prim
== PIPE_PRIM_POINTS
);
1085 /* For points, all interpolants are constant-valued.
1086 * However, for point sprites, we'll need to setup texcoords appropriately.
1087 * XXX: which coefficients are the texcoords???
1088 * We may do point sprites as textured quads...
1090 * KW: We don't know which coefficients are texcoords - ultimately
1091 * the choice of what interpolation mode to use for each attribute
1092 * should be determined by the fragment program, using
1093 * per-attribute declaration statements that include interpolation
1094 * mode as a parameter. So either the fragment program will have
1095 * to be adjusted for pointsprite vs normal point behaviour, or
1096 * otherwise a special interpolation mode will have to be defined
1097 * which matches the required behaviour for point sprites. But -
1098 * the latter is not a feature of normal hardware, and as such
1099 * probably should be ruled out on that basis.
1101 setup
->vprovoke
= v0
;
1104 const_coeff(setup
, &setup
->posCoef
, 0, 2);
1105 const_coeff(setup
, &setup
->posCoef
, 0, 3);
1107 for (fragSlot
= 0; fragSlot
< spfs
->info
.num_inputs
; fragSlot
++) {
1108 const uint vertSlot
= vinfo
->attrib
[fragSlot
].src_index
;
1111 switch (vinfo
->attrib
[fragSlot
].interp_mode
) {
1112 case INTERP_CONSTANT
:
1115 for (j
= 0; j
< NUM_CHANNELS
; j
++)
1116 const_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
1118 case INTERP_PERSPECTIVE
:
1119 for (j
= 0; j
< NUM_CHANNELS
; j
++)
1120 point_persp_coeff(setup
, setup
->vprovoke
,
1121 &setup
->coef
[fragSlot
], vertSlot
, j
);
1124 setup_fragcoord_coeff(setup
, fragSlot
);
1130 if (spfs
->info
.input_semantic_name
[fragSlot
] == TGSI_SEMANTIC_FACE
) {
1131 setup
->coef
[fragSlot
].a0
[0] = 1.0f
- setup
->facing
;
1132 setup
->coef
[fragSlot
].dadx
[0] = 0.0;
1133 setup
->coef
[fragSlot
].dady
[0] = 0.0;
1138 if (halfSize
<= 0.5 && !round
) {
1139 /* special case for 1-pixel points */
1140 const int ix
= ((int) x
) & 1;
1141 const int iy
= ((int) y
) & 1;
1142 setup
->quad
[0].input
.x0
= (int) x
- ix
;
1143 setup
->quad
[0].input
.y0
= (int) y
- iy
;
1144 setup
->quad
[0].inout
.mask
= (1 << ix
) << (2 * iy
);
1145 clip_emit_quad( setup
, &setup
->quad
[0] );
1149 /* rounded points */
1150 const int ixmin
= block((int) (x
- halfSize
));
1151 const int ixmax
= block((int) (x
+ halfSize
));
1152 const int iymin
= block((int) (y
- halfSize
));
1153 const int iymax
= block((int) (y
+ halfSize
));
1154 const float rmin
= halfSize
- 0.7071F
; /* 0.7071 = sqrt(2)/2 */
1155 const float rmax
= halfSize
+ 0.7071F
;
1156 const float rmin2
= MAX2(0.0F
, rmin
* rmin
);
1157 const float rmax2
= rmax
* rmax
;
1158 const float cscale
= 1.0F
/ (rmax2
- rmin2
);
1161 for (iy
= iymin
; iy
<= iymax
; iy
+= 2) {
1162 for (ix
= ixmin
; ix
<= ixmax
; ix
+= 2) {
1163 float dx
, dy
, dist2
, cover
;
1165 setup
->quad
[0].inout
.mask
= 0x0;
1167 dx
= (ix
+ 0.5f
) - x
;
1168 dy
= (iy
+ 0.5f
) - y
;
1169 dist2
= dx
* dx
+ dy
* dy
;
1170 if (dist2
<= rmax2
) {
1171 cover
= 1.0F
- (dist2
- rmin2
) * cscale
;
1172 setup
->quad
[0].input
.coverage
[QUAD_TOP_LEFT
] = MIN2(cover
, 1.0f
);
1173 setup
->quad
[0].inout
.mask
|= MASK_TOP_LEFT
;
1176 dx
= (ix
+ 1.5f
) - x
;
1177 dy
= (iy
+ 0.5f
) - y
;
1178 dist2
= dx
* dx
+ dy
* dy
;
1179 if (dist2
<= rmax2
) {
1180 cover
= 1.0F
- (dist2
- rmin2
) * cscale
;
1181 setup
->quad
[0].input
.coverage
[QUAD_TOP_RIGHT
] = MIN2(cover
, 1.0f
);
1182 setup
->quad
[0].inout
.mask
|= MASK_TOP_RIGHT
;
1185 dx
= (ix
+ 0.5f
) - x
;
1186 dy
= (iy
+ 1.5f
) - y
;
1187 dist2
= dx
* dx
+ dy
* dy
;
1188 if (dist2
<= rmax2
) {
1189 cover
= 1.0F
- (dist2
- rmin2
) * cscale
;
1190 setup
->quad
[0].input
.coverage
[QUAD_BOTTOM_LEFT
] = MIN2(cover
, 1.0f
);
1191 setup
->quad
[0].inout
.mask
|= MASK_BOTTOM_LEFT
;
1194 dx
= (ix
+ 1.5f
) - x
;
1195 dy
= (iy
+ 1.5f
) - y
;
1196 dist2
= dx
* dx
+ dy
* dy
;
1197 if (dist2
<= rmax2
) {
1198 cover
= 1.0F
- (dist2
- rmin2
) * cscale
;
1199 setup
->quad
[0].input
.coverage
[QUAD_BOTTOM_RIGHT
] = MIN2(cover
, 1.0f
);
1200 setup
->quad
[0].inout
.mask
|= MASK_BOTTOM_RIGHT
;
1203 if (setup
->quad
[0].inout
.mask
) {
1204 setup
->quad
[0].input
.x0
= ix
;
1205 setup
->quad
[0].input
.y0
= iy
;
1206 clip_emit_quad( setup
, &setup
->quad
[0] );
1213 const int xmin
= (int) (x
+ 0.75 - halfSize
);
1214 const int ymin
= (int) (y
+ 0.25 - halfSize
);
1215 const int xmax
= xmin
+ (int) size
;
1216 const int ymax
= ymin
+ (int) size
;
1217 /* XXX could apply scissor to xmin,ymin,xmax,ymax now */
1218 const int ixmin
= block(xmin
);
1219 const int ixmax
= block(xmax
- 1);
1220 const int iymin
= block(ymin
);
1221 const int iymax
= block(ymax
- 1);
1225 debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax);
1227 for (iy
= iymin
; iy
<= iymax
; iy
+= 2) {
1230 /* above the top edge */
1231 rowMask
&= (MASK_BOTTOM_LEFT
| MASK_BOTTOM_RIGHT
);
1233 if (iy
+ 1 >= ymax
) {
1234 /* below the bottom edge */
1235 rowMask
&= (MASK_TOP_LEFT
| MASK_TOP_RIGHT
);
1238 for (ix
= ixmin
; ix
<= ixmax
; ix
+= 2) {
1239 uint mask
= rowMask
;
1242 /* fragment is past left edge of point, turn off left bits */
1243 mask
&= (MASK_BOTTOM_RIGHT
| MASK_TOP_RIGHT
);
1245 if (ix
+ 1 >= xmax
) {
1246 /* past the right edge */
1247 mask
&= (MASK_BOTTOM_LEFT
| MASK_TOP_LEFT
);
1250 setup
->quad
[0].inout
.mask
= mask
;
1251 setup
->quad
[0].input
.x0
= ix
;
1252 setup
->quad
[0].input
.y0
= iy
;
1253 clip_emit_quad( setup
, &setup
->quad
[0] );
1260 void sp_setup_prepare( struct setup_context
*setup
)
1262 struct softpipe_context
*sp
= setup
->softpipe
;
1265 softpipe_update_derived(sp
);
1268 /* Note: nr_attrs is only used for debugging (vertex printing) */
1269 setup
->nr_vertex_attrs
= draw_num_shader_outputs(sp
->draw
);
1271 sp
->quad
.first
->begin( sp
->quad
.first
);
1273 if (sp
->reduced_api_prim
== PIPE_PRIM_TRIANGLES
&&
1274 sp
->rasterizer
->fill_cw
== PIPE_POLYGON_MODE_FILL
&&
1275 sp
->rasterizer
->fill_ccw
== PIPE_POLYGON_MODE_FILL
) {
1276 /* we'll do culling */
1277 setup
->winding
= sp
->rasterizer
->cull_mode
;
1280 /* 'draw' will do culling */
1281 setup
->winding
= PIPE_WINDING_NONE
;
1287 void sp_setup_destroy_context( struct setup_context
*setup
)
1294 * Create a new primitive setup/render stage.
1296 struct setup_context
*sp_setup_create_context( struct softpipe_context
*softpipe
)
1298 struct setup_context
*setup
= CALLOC_STRUCT(setup_context
);
1301 setup
->softpipe
= softpipe
;
1303 for (i
= 0; i
< MAX_QUADS
; i
++) {
1304 setup
->quad
[i
].coef
= setup
->coef
;
1305 setup
->quad
[i
].posCoef
= &setup
->posCoef
;
1308 setup
->span
.left
[0] = 1000000; /* greater than right[0] */
1309 setup
->span
.left
[1] = 1000000; /* greater than right[1] */