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"
55 float dx
; /**< X(v1) - X(v0), used only during setup */
56 float dy
; /**< Y(v1) - Y(v0), used only during setup */
57 float dxdy
; /**< dx/dy */
58 float sx
, sy
; /**< first sample point coord */
59 int lines
; /**< number of lines on this edge */
64 * Max number of quads (2x2 pixel blocks) to process per batch.
65 * This can't be arbitrarily increased since we depend on some 32-bit
66 * bitmasks (two bits per quad).
72 * Triangle setup info.
73 * Also used for line drawing (taking some liberties).
75 struct setup_context
{
76 struct softpipe_context
*softpipe
;
78 /* Vertices are just an array of floats making up each attribute in
79 * turn. Currently fixed at 4 floats, but should change in time.
80 * Codegen will help cope with this.
82 const float (*vmax
)[4];
83 const float (*vmid
)[4];
84 const float (*vmin
)[4];
85 const float (*vprovoke
)[4];
96 struct quad_header quad
[MAX_QUADS
];
97 struct quad_header
*quad_ptrs
[MAX_QUADS
];
100 struct tgsi_interp_coef coef
[PIPE_MAX_SHADER_INPUTS
];
101 struct tgsi_interp_coef posCoef
; /* For Z, W */
104 int left
[2]; /**< [0] = row0, [1] = row1 */
110 uint numFragsEmitted
; /**< per primitive */
111 uint numFragsWritten
; /**< per primitive */
114 unsigned winding
; /* which winding to cull */
115 unsigned nr_vertex_attrs
;
122 * Do triangle cull test using tri determinant (sign indicates orientation)
123 * \return true if triangle is to be culled.
125 static INLINE boolean
126 cull_tri(const struct setup_context
*setup
, float det
)
129 /* if (det < 0 then Z points toward camera and triangle is
130 * counter-clockwise winding.
132 unsigned winding
= (det
< 0) ? PIPE_WINDING_CCW
: PIPE_WINDING_CW
;
134 if ((winding
& setup
->winding
) == 0)
146 * Clip setup->quad against the scissor/surface bounds.
149 quad_clip(struct setup_context
*setup
, struct quad_header
*quad
)
151 const struct pipe_scissor_state
*cliprect
= &setup
->softpipe
->cliprect
;
152 const int minx
= (int) cliprect
->minx
;
153 const int maxx
= (int) cliprect
->maxx
;
154 const int miny
= (int) cliprect
->miny
;
155 const int maxy
= (int) cliprect
->maxy
;
157 if (quad
->input
.x0
>= maxx
||
158 quad
->input
.y0
>= maxy
||
159 quad
->input
.x0
+ 1 < minx
||
160 quad
->input
.y0
+ 1 < miny
) {
161 /* totally clipped */
162 quad
->inout
.mask
= 0x0;
165 if (quad
->input
.x0
< minx
)
166 quad
->inout
.mask
&= (MASK_BOTTOM_RIGHT
| MASK_TOP_RIGHT
);
167 if (quad
->input
.y0
< miny
)
168 quad
->inout
.mask
&= (MASK_BOTTOM_LEFT
| MASK_BOTTOM_RIGHT
);
169 if (quad
->input
.x0
== maxx
- 1)
170 quad
->inout
.mask
&= (MASK_BOTTOM_LEFT
| MASK_TOP_LEFT
);
171 if (quad
->input
.y0
== maxy
- 1)
172 quad
->inout
.mask
&= (MASK_TOP_LEFT
| MASK_TOP_RIGHT
);
177 * Emit a quad (pass to next stage) with clipping.
180 clip_emit_quad(struct setup_context
*setup
, struct quad_header
*quad
)
182 quad_clip( setup
, quad
);
184 if (quad
->inout
.mask
) {
185 struct softpipe_context
*sp
= setup
->softpipe
;
187 sp
->quad
.first
->run( sp
->quad
.first
, &quad
, 1 );
194 * Given an X or Y coordinate, return the block/quad coordinate that it
212 * Render a horizontal span of quads
215 flush_spans(struct setup_context
*setup
)
217 const int step
= MAX_QUADS
;
218 const int xleft0
= setup
->span
.left
[0];
219 const int xleft1
= setup
->span
.left
[1];
220 const int xright0
= setup
->span
.right
[0];
221 const int xright1
= setup
->span
.right
[1];
222 struct quad_stage
*pipe
= setup
->softpipe
->quad
.first
;
224 const int minleft
= block_x(MIN2(xleft0
, xleft1
));
225 const int maxright
= MAX2(xright0
, xright1
);
228 /* process quads in horizontal chunks of 16 */
229 for (x
= minleft
; x
< maxright
; x
+= step
) {
230 unsigned skip_left0
= CLAMP(xleft0
- x
, 0, step
);
231 unsigned skip_left1
= CLAMP(xleft1
- x
, 0, step
);
232 unsigned skip_right0
= CLAMP(x
+ step
- xright0
, 0, step
);
233 unsigned skip_right1
= CLAMP(x
+ step
- xright1
, 0, step
);
237 unsigned skipmask_left0
= (1U << skip_left0
) - 1U;
238 unsigned skipmask_left1
= (1U << skip_left1
) - 1U;
240 /* These calculations fail when step == 32 and skip_right == 0.
242 unsigned skipmask_right0
= ~0U << (unsigned)(step
- skip_right0
);
243 unsigned skipmask_right1
= ~0U << (unsigned)(step
- skip_right1
);
245 unsigned mask0
= ~skipmask_left0
& ~skipmask_right0
;
246 unsigned mask1
= ~skipmask_left1
& ~skipmask_right1
;
250 unsigned quadmask
= (mask0
& 3) | ((mask1
& 3) << 2);
252 setup
->quad
[q
].input
.x0
= lx
;
253 setup
->quad
[q
].input
.y0
= setup
->span
.y
;
254 setup
->quad
[q
].input
.facing
= setup
->facing
;
255 setup
->quad
[q
].inout
.mask
= quadmask
;
256 setup
->quad_ptrs
[q
] = &setup
->quad
[q
];
262 } while (mask0
| mask1
);
264 pipe
->run( pipe
, setup
->quad_ptrs
, q
);
270 setup
->span
.right
[0] = 0;
271 setup
->span
.right
[1] = 0;
272 setup
->span
.left
[0] = 1000000; /* greater than right[0] */
273 setup
->span
.left
[1] = 1000000; /* greater than right[1] */
279 print_vertex(const struct setup_context
*setup
,
283 debug_printf(" Vertex: (%p)\n", (void *) v
);
284 for (i
= 0; i
< setup
->nr_vertex_attrs
; i
++) {
285 debug_printf(" %d: %f %f %f %f\n", i
,
286 v
[i
][0], v
[i
][1], v
[i
][2], v
[i
][3]);
287 if (util_is_inf_or_nan(v
[i
][0])) {
288 debug_printf(" NaN!\n");
296 * Sort the vertices from top to bottom order, setting up the triangle
297 * edge fields (ebot, emaj, etop).
298 * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise
301 setup_sort_vertices(struct setup_context
*setup
,
303 const float (*v0
)[4],
304 const float (*v1
)[4],
305 const float (*v2
)[4])
307 if (setup
->softpipe
->rasterizer
->flatshade_first
)
308 setup
->vprovoke
= v0
;
310 setup
->vprovoke
= v2
;
312 /* determine bottom to top order of vertices */
359 setup
->ebot
.dx
= setup
->vmid
[0][0] - setup
->vmin
[0][0];
360 setup
->ebot
.dy
= setup
->vmid
[0][1] - setup
->vmin
[0][1];
361 setup
->emaj
.dx
= setup
->vmax
[0][0] - setup
->vmin
[0][0];
362 setup
->emaj
.dy
= setup
->vmax
[0][1] - setup
->vmin
[0][1];
363 setup
->etop
.dx
= setup
->vmax
[0][0] - setup
->vmid
[0][0];
364 setup
->etop
.dy
= setup
->vmax
[0][1] - setup
->vmid
[0][1];
367 * Compute triangle's area. Use 1/area to compute partial
368 * derivatives of attributes later.
370 * The area will be the same as prim->det, but the sign may be
371 * different depending on how the vertices get sorted above.
373 * To determine whether the primitive is front or back facing we
374 * use the prim->det value because its sign is correct.
377 const float area
= (setup
->emaj
.dx
* setup
->ebot
.dy
-
378 setup
->ebot
.dx
* setup
->emaj
.dy
);
380 setup
->oneoverarea
= 1.0f
/ area
;
383 debug_printf("%s one-over-area %f area %f det %f\n",
384 __FUNCTION__, setup->oneoverarea, area, det );
386 if (util_is_inf_or_nan(setup
->oneoverarea
))
390 /* We need to know if this is a front or back-facing triangle for:
391 * - the GLSL gl_FrontFacing fragment attribute (bool)
392 * - two-sided stencil test
393 * 0 = front-facing, 1 = back-facing
397 (setup
->softpipe
->rasterizer
->front_winding
== PIPE_WINDING_CW
));
399 /* Prepare pixel offset for rasterisation:
400 * - pixel center (0.5, 0.5) for GL, or
401 * - assume (0.0, 0.0) for other APIs.
403 if (setup
->softpipe
->rasterizer
->gl_rasterization_rules
) {
404 setup
->pixel_offset
= 0.5f
;
406 setup
->pixel_offset
= 0.0f
;
413 /* Apply cylindrical wrapping to v0, v1, v2 coordinates, if enabled.
414 * Input coordinates must be in [0, 1] range, otherwise results are undefined.
415 * Some combinations of coordinates produce invalid results,
416 * but this behaviour is acceptable.
419 tri_apply_cylindrical_wrap(float v0
,
422 uint cylindrical_wrap
,
425 if (cylindrical_wrap
) {
432 else if (delta
< -0.5f
) {
440 else if (delta
< -0.5f
) {
448 else if (delta
< -0.5f
) {
460 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
461 * The value value comes from vertex[slot][i].
462 * The result will be put into setup->coef[slot].a0[i].
463 * \param slot which attribute slot
464 * \param i which component of the slot (0..3)
467 const_coeff(struct setup_context
*setup
,
468 struct tgsi_interp_coef
*coef
,
469 uint vertSlot
, uint i
)
476 /* need provoking vertex info!
478 coef
->a0
[i
] = setup
->vprovoke
[vertSlot
][i
];
483 * Compute a0, dadx and dady for a linearly interpolated coefficient,
485 * v[0], v[1] and v[2] are vmin, vmid and vmax, respectively.
488 tri_linear_coeff(struct setup_context
*setup
,
489 struct tgsi_interp_coef
*coef
,
493 float botda
= v
[1] - v
[0];
494 float majda
= v
[2] - v
[0];
495 float a
= setup
->ebot
.dy
* majda
- botda
* setup
->emaj
.dy
;
496 float b
= setup
->emaj
.dx
* botda
- majda
* setup
->ebot
.dx
;
497 float dadx
= a
* setup
->oneoverarea
;
498 float dady
= b
* setup
->oneoverarea
;
502 coef
->dadx
[i
] = dadx
;
503 coef
->dady
[i
] = dady
;
505 /* calculate a0 as the value which would be sampled for the
506 * fragment at (0,0), taking into account that we want to sample at
507 * pixel centers, in other words (pixel_offset, pixel_offset).
509 * this is neat but unfortunately not a good way to do things for
510 * triangles with very large values of dadx or dady as it will
511 * result in the subtraction and re-addition from a0 of a very
512 * large number, which means we'll end up loosing a lot of the
513 * fractional bits and precision from a0. the way to fix this is
514 * to define a0 as the sample at a pixel center somewhere near vmin
515 * instead - i'll switch to this later.
517 coef
->a0
[i
] = (v
[0] -
518 (dadx
* (setup
->vmin
[0][0] - setup
->pixel_offset
) +
519 dady
* (setup
->vmin
[0][1] - setup
->pixel_offset
)));
522 debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
524 setup->coef[slot].a0[i],
525 setup->coef[slot].dadx[i],
526 setup->coef[slot].dady[i]);
532 * Compute a0, dadx and dady for a perspective-corrected interpolant,
534 * We basically multiply the vertex value by 1/w before computing
535 * the plane coefficients (a0, dadx, dady).
536 * Later, when we compute the value at a particular fragment position we'll
537 * divide the interpolated value by the interpolated W at that fragment.
538 * v[0], v[1] and v[2] are vmin, vmid and vmax, respectively.
541 tri_persp_coeff(struct setup_context
*setup
,
542 struct tgsi_interp_coef
*coef
,
546 /* premultiply by 1/w (v[0][3] is always W):
548 float mina
= v
[0] * setup
->vmin
[0][3];
549 float mida
= v
[1] * setup
->vmid
[0][3];
550 float maxa
= v
[2] * setup
->vmax
[0][3];
551 float botda
= mida
- mina
;
552 float majda
= maxa
- mina
;
553 float a
= setup
->ebot
.dy
* majda
- botda
* setup
->emaj
.dy
;
554 float b
= setup
->emaj
.dx
* botda
- majda
* setup
->ebot
.dx
;
555 float dadx
= a
* setup
->oneoverarea
;
556 float dady
= b
* setup
->oneoverarea
;
559 debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i,
560 setup->vmin[vertSlot][i],
561 setup->vmid[vertSlot][i],
562 setup->vmax[vertSlot][i]
567 coef
->dadx
[i
] = dadx
;
568 coef
->dady
[i
] = dady
;
569 coef
->a0
[i
] = (mina
-
570 (dadx
* (setup
->vmin
[0][0] - setup
->pixel_offset
) +
571 dady
* (setup
->vmin
[0][1] - setup
->pixel_offset
)));
576 * Special coefficient setup for gl_FragCoord.
577 * X and Y are trivial, though Y may have to be inverted for OpenGL.
578 * Z and W are copied from posCoef which should have already been computed.
579 * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
582 setup_fragcoord_coeff(struct setup_context
*setup
, uint slot
)
584 struct sp_fragment_shader
* spfs
= setup
->softpipe
->fs
;
586 setup
->coef
[slot
].a0
[0] = spfs
->pixel_center_integer
? 0.0 : 0.5;
587 setup
->coef
[slot
].dadx
[0] = 1.0;
588 setup
->coef
[slot
].dady
[0] = 0.0;
590 setup
->coef
[slot
].a0
[1] =
591 (spfs
->origin_lower_left
? setup
->softpipe
->framebuffer
.height
: 0)
592 + (spfs
->pixel_center_integer
? 0.0 : 0.5);
593 setup
->coef
[slot
].dadx
[1] = 0.0;
594 setup
->coef
[slot
].dady
[1] = spfs
->origin_lower_left
? -1.0 : 1.0;
596 setup
->coef
[slot
].a0
[2] = setup
->posCoef
.a0
[2];
597 setup
->coef
[slot
].dadx
[2] = setup
->posCoef
.dadx
[2];
598 setup
->coef
[slot
].dady
[2] = setup
->posCoef
.dady
[2];
600 setup
->coef
[slot
].a0
[3] = setup
->posCoef
.a0
[3];
601 setup
->coef
[slot
].dadx
[3] = setup
->posCoef
.dadx
[3];
602 setup
->coef
[slot
].dady
[3] = setup
->posCoef
.dady
[3];
608 * Compute the setup->coef[] array dadx, dady, a0 values.
609 * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized.
612 setup_tri_coefficients(struct setup_context
*setup
)
614 struct softpipe_context
*softpipe
= setup
->softpipe
;
615 const struct sp_fragment_shader
*spfs
= softpipe
->fs
;
616 const struct vertex_info
*vinfo
= softpipe_get_vertex_info(softpipe
);
620 /* z and w are done by linear interpolation:
622 v
[0] = setup
->vmin
[0][2];
623 v
[1] = setup
->vmid
[0][2];
624 v
[2] = setup
->vmax
[0][2];
625 tri_linear_coeff(setup
, &setup
->posCoef
, 2, v
);
627 v
[0] = setup
->vmin
[0][3];
628 v
[1] = setup
->vmid
[0][3];
629 v
[2] = setup
->vmax
[0][3];
630 tri_linear_coeff(setup
, &setup
->posCoef
, 3, v
);
632 /* setup interpolation for all the remaining attributes:
634 for (fragSlot
= 0; fragSlot
< spfs
->info
.num_inputs
; fragSlot
++) {
635 const uint vertSlot
= vinfo
->attrib
[fragSlot
].src_index
;
638 switch (vinfo
->attrib
[fragSlot
].interp_mode
) {
639 case INTERP_CONSTANT
:
640 for (j
= 0; j
< NUM_CHANNELS
; j
++)
641 const_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
644 for (j
= 0; j
< NUM_CHANNELS
; j
++) {
645 tri_apply_cylindrical_wrap(setup
->vmin
[vertSlot
][j
],
646 setup
->vmid
[vertSlot
][j
],
647 setup
->vmax
[vertSlot
][j
],
648 spfs
->info
.input_cylindrical_wrap
[fragSlot
] & (1 << j
),
650 tri_linear_coeff(setup
, &setup
->coef
[fragSlot
], j
, v
);
653 case INTERP_PERSPECTIVE
:
654 for (j
= 0; j
< NUM_CHANNELS
; j
++) {
655 tri_apply_cylindrical_wrap(setup
->vmin
[vertSlot
][j
],
656 setup
->vmid
[vertSlot
][j
],
657 setup
->vmax
[vertSlot
][j
],
658 spfs
->info
.input_cylindrical_wrap
[fragSlot
] & (1 << j
),
660 tri_persp_coeff(setup
, &setup
->coef
[fragSlot
], j
, v
);
664 setup_fragcoord_coeff(setup
, fragSlot
);
670 if (spfs
->info
.input_semantic_name
[fragSlot
] == TGSI_SEMANTIC_FACE
) {
671 /* convert 0 to 1.0 and 1 to -1.0 */
672 setup
->coef
[fragSlot
].a0
[0] = setup
->facing
* -2.0f
+ 1.0f
;
673 setup
->coef
[fragSlot
].dadx
[0] = 0.0;
674 setup
->coef
[fragSlot
].dady
[0] = 0.0;
681 setup_tri_edges(struct setup_context
*setup
)
683 float vmin_x
= setup
->vmin
[0][0] + setup
->pixel_offset
;
684 float vmid_x
= setup
->vmid
[0][0] + setup
->pixel_offset
;
686 float vmin_y
= setup
->vmin
[0][1] - setup
->pixel_offset
;
687 float vmid_y
= setup
->vmid
[0][1] - setup
->pixel_offset
;
688 float vmax_y
= setup
->vmax
[0][1] - setup
->pixel_offset
;
690 setup
->emaj
.sy
= ceilf(vmin_y
);
691 setup
->emaj
.lines
= (int) ceilf(vmax_y
- setup
->emaj
.sy
);
692 setup
->emaj
.dxdy
= setup
->emaj
.dy
? setup
->emaj
.dx
/ setup
->emaj
.dy
: .0f
;
693 setup
->emaj
.sx
= vmin_x
+ (setup
->emaj
.sy
- vmin_y
) * setup
->emaj
.dxdy
;
695 setup
->etop
.sy
= ceilf(vmid_y
);
696 setup
->etop
.lines
= (int) ceilf(vmax_y
- setup
->etop
.sy
);
697 setup
->etop
.dxdy
= setup
->etop
.dy
? setup
->etop
.dx
/ setup
->etop
.dy
: .0f
;
698 setup
->etop
.sx
= vmid_x
+ (setup
->etop
.sy
- vmid_y
) * setup
->etop
.dxdy
;
700 setup
->ebot
.sy
= ceilf(vmin_y
);
701 setup
->ebot
.lines
= (int) ceilf(vmid_y
- setup
->ebot
.sy
);
702 setup
->ebot
.dxdy
= setup
->ebot
.dy
? setup
->ebot
.dx
/ setup
->ebot
.dy
: .0f
;
703 setup
->ebot
.sx
= vmin_x
+ (setup
->ebot
.sy
- vmin_y
) * setup
->ebot
.dxdy
;
708 * Render the upper or lower half of a triangle.
709 * Scissoring/cliprect is applied here too.
712 subtriangle(struct setup_context
*setup
,
717 const struct pipe_scissor_state
*cliprect
= &setup
->softpipe
->cliprect
;
718 const int minx
= (int) cliprect
->minx
;
719 const int maxx
= (int) cliprect
->maxx
;
720 const int miny
= (int) cliprect
->miny
;
721 const int maxy
= (int) cliprect
->maxy
;
722 int y
, start_y
, finish_y
;
723 int sy
= (int)eleft
->sy
;
725 assert((int)eleft
->sy
== (int) eright
->sy
);
728 /* clip top/bottom */
733 finish_y
= sy
+ lines
;
741 debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);
744 for (y
= start_y
; y
< finish_y
; y
++) {
746 /* avoid accumulating adds as floats don't have the precision to
747 * accurately iterate large triangle edges that way. luckily we
748 * can just multiply these days.
750 * this is all drowned out by the attribute interpolation anyway.
752 int left
= (int)(eleft
->sx
+ y
* eleft
->dxdy
);
753 int right
= (int)(eright
->sx
+ y
* eright
->dxdy
);
755 /* clip left/right */
763 if (block(_y
) != setup
->span
.y
) {
765 setup
->span
.y
= block(_y
);
768 setup
->span
.left
[_y
&1] = left
;
769 setup
->span
.right
[_y
&1] = right
;
774 /* save the values so that emaj can be restarted:
776 eleft
->sx
+= lines
* eleft
->dxdy
;
777 eright
->sx
+= lines
* eright
->dxdy
;
784 * Recalculate prim's determinant. This is needed as we don't have
785 * get this information through the vbuf_render interface & we must
789 calc_det(const float (*v0
)[4],
790 const float (*v1
)[4],
791 const float (*v2
)[4])
793 /* edge vectors e = v0 - v2, f = v1 - v2 */
794 const float ex
= v0
[0][0] - v2
[0][0];
795 const float ey
= v0
[0][1] - v2
[0][1];
796 const float fx
= v1
[0][0] - v2
[0][0];
797 const float fy
= v1
[0][1] - v2
[0][1];
799 /* det = cross(e,f).z */
800 return ex
* fy
- ey
* fx
;
805 * Do setup for triangle rasterization, then render the triangle.
808 sp_setup_tri(struct setup_context
*setup
,
809 const float (*v0
)[4],
810 const float (*v1
)[4],
811 const float (*v2
)[4])
816 debug_printf("Setup triangle:\n");
817 print_vertex(setup
, v0
);
818 print_vertex(setup
, v1
);
819 print_vertex(setup
, v2
);
822 if (setup
->softpipe
->no_rast
)
825 det
= calc_det(v0
, v1
, v2
);
827 debug_printf("%s\n", __FUNCTION__ );
831 setup
->numFragsEmitted
= 0;
832 setup
->numFragsWritten
= 0;
835 if (cull_tri( setup
, det
))
838 if (!setup_sort_vertices( setup
, det
, v0
, v1
, v2
))
840 setup_tri_coefficients( setup
);
841 setup_tri_edges( setup
);
843 assert(setup
->softpipe
->reduced_prim
== PIPE_PRIM_TRIANGLES
);
846 setup
->span
.right
[0] = 0;
847 setup
->span
.right
[1] = 0;
848 /* setup->span.z_mode = tri_z_mode( setup->ctx ); */
850 /* init_constant_attribs( setup ); */
852 if (setup
->oneoverarea
< 0.0) {
855 subtriangle( setup
, &setup
->emaj
, &setup
->ebot
, setup
->ebot
.lines
);
856 subtriangle( setup
, &setup
->emaj
, &setup
->etop
, setup
->etop
.lines
);
861 subtriangle( setup
, &setup
->ebot
, &setup
->emaj
, setup
->ebot
.lines
);
862 subtriangle( setup
, &setup
->etop
, &setup
->emaj
, setup
->etop
.lines
);
865 flush_spans( setup
);
868 printf("Tri: %u frags emitted, %u written\n",
869 setup
->numFragsEmitted
,
870 setup
->numFragsWritten
);
875 /* Apply cylindrical wrapping to v0, v1 coordinates, if enabled.
876 * Input coordinates must be in [0, 1] range, otherwise results are undefined.
879 line_apply_cylindrical_wrap(float v0
,
881 uint cylindrical_wrap
,
884 if (cylindrical_wrap
) {
891 else if (delta
< -0.5f
) {
902 * Compute a0, dadx and dady for a linearly interpolated coefficient,
904 * v[0] and v[1] are vmin and vmax, respectively.
907 line_linear_coeff(const struct setup_context
*setup
,
908 struct tgsi_interp_coef
*coef
,
912 const float da
= v
[1] - v
[0];
913 const float dadx
= da
* setup
->emaj
.dx
* setup
->oneoverarea
;
914 const float dady
= da
* setup
->emaj
.dy
* setup
->oneoverarea
;
915 coef
->dadx
[i
] = dadx
;
916 coef
->dady
[i
] = dady
;
917 coef
->a0
[i
] = (v
[0] -
918 (dadx
* (setup
->vmin
[0][0] - setup
->pixel_offset
) +
919 dady
* (setup
->vmin
[0][1] - setup
->pixel_offset
)));
924 * Compute a0, dadx and dady for a perspective-corrected interpolant,
926 * v[0] and v[1] are vmin and vmax, respectively.
929 line_persp_coeff(const struct setup_context
*setup
,
930 struct tgsi_interp_coef
*coef
,
934 const float a0
= v
[0] * setup
->vmin
[0][3];
935 const float a1
= v
[1] * setup
->vmax
[0][3];
936 const float da
= a1
- a0
;
937 const float dadx
= da
* setup
->emaj
.dx
* setup
->oneoverarea
;
938 const float dady
= da
* setup
->emaj
.dy
* setup
->oneoverarea
;
939 coef
->dadx
[i
] = dadx
;
940 coef
->dady
[i
] = dady
;
942 (dadx
* (setup
->vmin
[0][0] - setup
->pixel_offset
) +
943 dady
* (setup
->vmin
[0][1] - setup
->pixel_offset
)));
948 * Compute the setup->coef[] array dadx, dady, a0 values.
949 * Must be called after setup->vmin,vmax are initialized.
952 setup_line_coefficients(struct setup_context
*setup
,
953 const float (*v0
)[4],
954 const float (*v1
)[4])
956 struct softpipe_context
*softpipe
= setup
->softpipe
;
957 const struct sp_fragment_shader
*spfs
= softpipe
->fs
;
958 const struct vertex_info
*vinfo
= softpipe_get_vertex_info(softpipe
);
963 /* use setup->vmin, vmax to point to vertices */
964 if (softpipe
->rasterizer
->flatshade_first
)
965 setup
->vprovoke
= v0
;
967 setup
->vprovoke
= v1
;
971 setup
->emaj
.dx
= setup
->vmax
[0][0] - setup
->vmin
[0][0];
972 setup
->emaj
.dy
= setup
->vmax
[0][1] - setup
->vmin
[0][1];
974 /* NOTE: this is not really area but something proportional to it */
975 area
= setup
->emaj
.dx
* setup
->emaj
.dx
+ setup
->emaj
.dy
* setup
->emaj
.dy
;
976 if (area
== 0.0f
|| util_is_inf_or_nan(area
))
978 setup
->oneoverarea
= 1.0f
/ area
;
980 /* z and w are done by linear interpolation:
982 v
[0] = setup
->vmin
[0][2];
983 v
[1] = setup
->vmax
[0][2];
984 line_linear_coeff(setup
, &setup
->posCoef
, 2, v
);
986 v
[0] = setup
->vmin
[0][3];
987 v
[1] = setup
->vmax
[0][3];
988 line_linear_coeff(setup
, &setup
->posCoef
, 3, v
);
990 /* setup interpolation for all the remaining attributes:
992 for (fragSlot
= 0; fragSlot
< spfs
->info
.num_inputs
; fragSlot
++) {
993 const uint vertSlot
= vinfo
->attrib
[fragSlot
].src_index
;
996 switch (vinfo
->attrib
[fragSlot
].interp_mode
) {
997 case INTERP_CONSTANT
:
998 for (j
= 0; j
< NUM_CHANNELS
; j
++)
999 const_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
1002 for (j
= 0; j
< NUM_CHANNELS
; j
++) {
1003 line_apply_cylindrical_wrap(setup
->vmin
[vertSlot
][j
],
1004 setup
->vmax
[vertSlot
][j
],
1005 spfs
->info
.input_cylindrical_wrap
[fragSlot
] & (1 << j
),
1007 line_linear_coeff(setup
, &setup
->coef
[fragSlot
], j
, v
);
1010 case INTERP_PERSPECTIVE
:
1011 for (j
= 0; j
< NUM_CHANNELS
; j
++) {
1012 line_apply_cylindrical_wrap(setup
->vmin
[vertSlot
][j
],
1013 setup
->vmax
[vertSlot
][j
],
1014 spfs
->info
.input_cylindrical_wrap
[fragSlot
] & (1 << j
),
1016 line_persp_coeff(setup
, &setup
->coef
[fragSlot
], j
, v
);
1020 setup_fragcoord_coeff(setup
, fragSlot
);
1026 if (spfs
->info
.input_semantic_name
[fragSlot
] == TGSI_SEMANTIC_FACE
) {
1027 /* convert 0 to 1.0 and 1 to -1.0 */
1028 setup
->coef
[fragSlot
].a0
[0] = setup
->facing
* -2.0f
+ 1.0f
;
1029 setup
->coef
[fragSlot
].dadx
[0] = 0.0;
1030 setup
->coef
[fragSlot
].dady
[0] = 0.0;
1038 * Plot a pixel in a line segment.
1041 plot(struct setup_context
*setup
, int x
, int y
)
1043 const int iy
= y
& 1;
1044 const int ix
= x
& 1;
1045 const int quadX
= x
- ix
;
1046 const int quadY
= y
- iy
;
1047 const int mask
= (1 << ix
) << (2 * iy
);
1049 if (quadX
!= setup
->quad
[0].input
.x0
||
1050 quadY
!= setup
->quad
[0].input
.y0
)
1052 /* flush prev quad, start new quad */
1054 if (setup
->quad
[0].input
.x0
!= -1)
1055 clip_emit_quad( setup
, &setup
->quad
[0] );
1057 setup
->quad
[0].input
.x0
= quadX
;
1058 setup
->quad
[0].input
.y0
= quadY
;
1059 setup
->quad
[0].inout
.mask
= 0x0;
1062 setup
->quad
[0].inout
.mask
|= mask
;
1067 * Do setup for line rasterization, then render the line.
1068 * Single-pixel width, no stipple, etc. We rely on the 'draw' module
1069 * to handle stippling and wide lines.
1072 sp_setup_line(struct setup_context
*setup
,
1073 const float (*v0
)[4],
1074 const float (*v1
)[4])
1076 int x0
= (int) v0
[0][0];
1077 int x1
= (int) v1
[0][0];
1078 int y0
= (int) v0
[0][1];
1079 int y1
= (int) v1
[0][1];
1085 debug_printf("Setup line:\n");
1086 print_vertex(setup
, v0
);
1087 print_vertex(setup
, v1
);
1090 if (setup
->softpipe
->no_rast
)
1093 if (dx
== 0 && dy
== 0)
1096 if (!setup_line_coefficients(setup
, v0
, v1
))
1099 assert(v0
[0][0] < 1.0e9
);
1100 assert(v0
[0][1] < 1.0e9
);
1101 assert(v1
[0][0] < 1.0e9
);
1102 assert(v1
[0][1] < 1.0e9
);
1105 dx
= -dx
; /* make positive */
1113 dy
= -dy
; /* make positive */
1122 assert(setup
->softpipe
->reduced_prim
== PIPE_PRIM_LINES
);
1124 setup
->quad
[0].input
.x0
= setup
->quad
[0].input
.y0
= -1;
1125 setup
->quad
[0].inout
.mask
= 0x0;
1127 /* XXX temporary: set coverage to 1.0 so the line appears
1128 * if AA mode happens to be enabled.
1130 setup
->quad
[0].input
.coverage
[0] =
1131 setup
->quad
[0].input
.coverage
[1] =
1132 setup
->quad
[0].input
.coverage
[2] =
1133 setup
->quad
[0].input
.coverage
[3] = 1.0;
1136 /*** X-major line ***/
1138 const int errorInc
= dy
+ dy
;
1139 int error
= errorInc
- dx
;
1140 const int errorDec
= error
- dx
;
1142 for (i
= 0; i
< dx
; i
++) {
1143 plot(setup
, x0
, y0
);
1156 /*** Y-major line ***/
1158 const int errorInc
= dx
+ dx
;
1159 int error
= errorInc
- dy
;
1160 const int errorDec
= error
- dy
;
1162 for (i
= 0; i
< dy
; i
++) {
1163 plot(setup
, x0
, y0
);
1176 /* draw final quad */
1177 if (setup
->quad
[0].inout
.mask
) {
1178 clip_emit_quad( setup
, &setup
->quad
[0] );
1184 point_persp_coeff(const struct setup_context
*setup
,
1185 const float (*vert
)[4],
1186 struct tgsi_interp_coef
*coef
,
1187 uint vertSlot
, uint i
)
1190 coef
->dadx
[i
] = 0.0F
;
1191 coef
->dady
[i
] = 0.0F
;
1192 coef
->a0
[i
] = vert
[vertSlot
][i
] * vert
[0][3];
1197 * Do setup for point rasterization, then render the point.
1198 * Round or square points...
1199 * XXX could optimize a lot for 1-pixel points.
1202 sp_setup_point(struct setup_context
*setup
,
1203 const float (*v0
)[4])
1205 struct softpipe_context
*softpipe
= setup
->softpipe
;
1206 const struct sp_fragment_shader
*spfs
= softpipe
->fs
;
1207 const int sizeAttr
= setup
->softpipe
->psize_slot
;
1209 = sizeAttr
> 0 ? v0
[sizeAttr
][0]
1210 : setup
->softpipe
->rasterizer
->point_size
;
1211 const float halfSize
= 0.5F
* size
;
1212 const boolean round
= (boolean
) setup
->softpipe
->rasterizer
->point_smooth
;
1213 const float x
= v0
[0][0]; /* Note: data[0] is always position */
1214 const float y
= v0
[0][1];
1215 const struct vertex_info
*vinfo
= softpipe_get_vertex_info(softpipe
);
1219 debug_printf("Setup point:\n");
1220 print_vertex(setup
, v0
);
1223 if (softpipe
->no_rast
)
1226 assert(setup
->softpipe
->reduced_prim
== PIPE_PRIM_POINTS
);
1228 /* For points, all interpolants are constant-valued.
1229 * However, for point sprites, we'll need to setup texcoords appropriately.
1230 * XXX: which coefficients are the texcoords???
1231 * We may do point sprites as textured quads...
1233 * KW: We don't know which coefficients are texcoords - ultimately
1234 * the choice of what interpolation mode to use for each attribute
1235 * should be determined by the fragment program, using
1236 * per-attribute declaration statements that include interpolation
1237 * mode as a parameter. So either the fragment program will have
1238 * to be adjusted for pointsprite vs normal point behaviour, or
1239 * otherwise a special interpolation mode will have to be defined
1240 * which matches the required behaviour for point sprites. But -
1241 * the latter is not a feature of normal hardware, and as such
1242 * probably should be ruled out on that basis.
1244 setup
->vprovoke
= v0
;
1247 const_coeff(setup
, &setup
->posCoef
, 0, 2);
1248 const_coeff(setup
, &setup
->posCoef
, 0, 3);
1250 for (fragSlot
= 0; fragSlot
< spfs
->info
.num_inputs
; fragSlot
++) {
1251 const uint vertSlot
= vinfo
->attrib
[fragSlot
].src_index
;
1254 switch (vinfo
->attrib
[fragSlot
].interp_mode
) {
1255 case INTERP_CONSTANT
:
1258 for (j
= 0; j
< NUM_CHANNELS
; j
++)
1259 const_coeff(setup
, &setup
->coef
[fragSlot
], vertSlot
, j
);
1261 case INTERP_PERSPECTIVE
:
1262 for (j
= 0; j
< NUM_CHANNELS
; j
++)
1263 point_persp_coeff(setup
, setup
->vprovoke
,
1264 &setup
->coef
[fragSlot
], vertSlot
, j
);
1267 setup_fragcoord_coeff(setup
, fragSlot
);
1273 if (spfs
->info
.input_semantic_name
[fragSlot
] == TGSI_SEMANTIC_FACE
) {
1274 /* convert 0 to 1.0 and 1 to -1.0 */
1275 setup
->coef
[fragSlot
].a0
[0] = setup
->facing
* -2.0f
+ 1.0f
;
1276 setup
->coef
[fragSlot
].dadx
[0] = 0.0;
1277 setup
->coef
[fragSlot
].dady
[0] = 0.0;
1282 if (halfSize
<= 0.5 && !round
) {
1283 /* special case for 1-pixel points */
1284 const int ix
= ((int) x
) & 1;
1285 const int iy
= ((int) y
) & 1;
1286 setup
->quad
[0].input
.x0
= (int) x
- ix
;
1287 setup
->quad
[0].input
.y0
= (int) y
- iy
;
1288 setup
->quad
[0].inout
.mask
= (1 << ix
) << (2 * iy
);
1289 clip_emit_quad( setup
, &setup
->quad
[0] );
1293 /* rounded points */
1294 const int ixmin
= block((int) (x
- halfSize
));
1295 const int ixmax
= block((int) (x
+ halfSize
));
1296 const int iymin
= block((int) (y
- halfSize
));
1297 const int iymax
= block((int) (y
+ halfSize
));
1298 const float rmin
= halfSize
- 0.7071F
; /* 0.7071 = sqrt(2)/2 */
1299 const float rmax
= halfSize
+ 0.7071F
;
1300 const float rmin2
= MAX2(0.0F
, rmin
* rmin
);
1301 const float rmax2
= rmax
* rmax
;
1302 const float cscale
= 1.0F
/ (rmax2
- rmin2
);
1305 for (iy
= iymin
; iy
<= iymax
; iy
+= 2) {
1306 for (ix
= ixmin
; ix
<= ixmax
; ix
+= 2) {
1307 float dx
, dy
, dist2
, cover
;
1309 setup
->quad
[0].inout
.mask
= 0x0;
1311 dx
= (ix
+ 0.5f
) - x
;
1312 dy
= (iy
+ 0.5f
) - y
;
1313 dist2
= dx
* dx
+ dy
* dy
;
1314 if (dist2
<= rmax2
) {
1315 cover
= 1.0F
- (dist2
- rmin2
) * cscale
;
1316 setup
->quad
[0].input
.coverage
[QUAD_TOP_LEFT
] = MIN2(cover
, 1.0f
);
1317 setup
->quad
[0].inout
.mask
|= MASK_TOP_LEFT
;
1320 dx
= (ix
+ 1.5f
) - x
;
1321 dy
= (iy
+ 0.5f
) - y
;
1322 dist2
= dx
* dx
+ dy
* dy
;
1323 if (dist2
<= rmax2
) {
1324 cover
= 1.0F
- (dist2
- rmin2
) * cscale
;
1325 setup
->quad
[0].input
.coverage
[QUAD_TOP_RIGHT
] = MIN2(cover
, 1.0f
);
1326 setup
->quad
[0].inout
.mask
|= MASK_TOP_RIGHT
;
1329 dx
= (ix
+ 0.5f
) - x
;
1330 dy
= (iy
+ 1.5f
) - y
;
1331 dist2
= dx
* dx
+ dy
* dy
;
1332 if (dist2
<= rmax2
) {
1333 cover
= 1.0F
- (dist2
- rmin2
) * cscale
;
1334 setup
->quad
[0].input
.coverage
[QUAD_BOTTOM_LEFT
] = MIN2(cover
, 1.0f
);
1335 setup
->quad
[0].inout
.mask
|= MASK_BOTTOM_LEFT
;
1338 dx
= (ix
+ 1.5f
) - x
;
1339 dy
= (iy
+ 1.5f
) - y
;
1340 dist2
= dx
* dx
+ dy
* dy
;
1341 if (dist2
<= rmax2
) {
1342 cover
= 1.0F
- (dist2
- rmin2
) * cscale
;
1343 setup
->quad
[0].input
.coverage
[QUAD_BOTTOM_RIGHT
] = MIN2(cover
, 1.0f
);
1344 setup
->quad
[0].inout
.mask
|= MASK_BOTTOM_RIGHT
;
1347 if (setup
->quad
[0].inout
.mask
) {
1348 setup
->quad
[0].input
.x0
= ix
;
1349 setup
->quad
[0].input
.y0
= iy
;
1350 clip_emit_quad( setup
, &setup
->quad
[0] );
1357 const int xmin
= (int) (x
+ 0.75 - halfSize
);
1358 const int ymin
= (int) (y
+ 0.25 - halfSize
);
1359 const int xmax
= xmin
+ (int) size
;
1360 const int ymax
= ymin
+ (int) size
;
1361 /* XXX could apply scissor to xmin,ymin,xmax,ymax now */
1362 const int ixmin
= block(xmin
);
1363 const int ixmax
= block(xmax
- 1);
1364 const int iymin
= block(ymin
);
1365 const int iymax
= block(ymax
- 1);
1369 debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax);
1371 for (iy
= iymin
; iy
<= iymax
; iy
+= 2) {
1374 /* above the top edge */
1375 rowMask
&= (MASK_BOTTOM_LEFT
| MASK_BOTTOM_RIGHT
);
1377 if (iy
+ 1 >= ymax
) {
1378 /* below the bottom edge */
1379 rowMask
&= (MASK_TOP_LEFT
| MASK_TOP_RIGHT
);
1382 for (ix
= ixmin
; ix
<= ixmax
; ix
+= 2) {
1383 uint mask
= rowMask
;
1386 /* fragment is past left edge of point, turn off left bits */
1387 mask
&= (MASK_BOTTOM_RIGHT
| MASK_TOP_RIGHT
);
1389 if (ix
+ 1 >= xmax
) {
1390 /* past the right edge */
1391 mask
&= (MASK_BOTTOM_LEFT
| MASK_TOP_LEFT
);
1394 setup
->quad
[0].inout
.mask
= mask
;
1395 setup
->quad
[0].input
.x0
= ix
;
1396 setup
->quad
[0].input
.y0
= iy
;
1397 clip_emit_quad( setup
, &setup
->quad
[0] );
1406 * Called by vbuf code just before we start buffering primitives.
1409 sp_setup_prepare(struct setup_context
*setup
)
1411 struct softpipe_context
*sp
= setup
->softpipe
;
1414 softpipe_update_derived(sp
);
1417 /* Note: nr_attrs is only used for debugging (vertex printing) */
1418 setup
->nr_vertex_attrs
= draw_num_shader_outputs(sp
->draw
);
1420 sp
->quad
.first
->begin( sp
->quad
.first
);
1422 if (sp
->reduced_api_prim
== PIPE_PRIM_TRIANGLES
&&
1423 sp
->rasterizer
->fill_cw
== PIPE_POLYGON_MODE_FILL
&&
1424 sp
->rasterizer
->fill_ccw
== PIPE_POLYGON_MODE_FILL
) {
1425 /* we'll do culling */
1426 setup
->winding
= sp
->rasterizer
->cull_mode
;
1429 /* 'draw' will do culling */
1430 setup
->winding
= PIPE_WINDING_NONE
;
1436 sp_setup_destroy_context(struct setup_context
*setup
)
1443 * Create a new primitive setup/render stage.
1445 struct setup_context
*
1446 sp_setup_create_context(struct softpipe_context
*softpipe
)
1448 struct setup_context
*setup
= CALLOC_STRUCT(setup_context
);
1451 setup
->softpipe
= softpipe
;
1453 for (i
= 0; i
< MAX_QUADS
; i
++) {
1454 setup
->quad
[i
].coef
= setup
->coef
;
1455 setup
->quad
[i
].posCoef
= &setup
->posCoef
;
1458 setup
->span
.left
[0] = 1000000; /* greater than right[0] */
1459 setup
->span
.left
[1] = 1000000; /* greater than right[1] */