1 /**************************************************************************
2 *
3 * Copyright 2010, VMware Inc.
4 * All Rights Reserved.
5 *
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:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
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 VMWARE 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.
25 *
26 **************************************************************************/
27
28 /*
29 * Binning code for points
30 */
31
32 #include "lp_setup_context.h"
33 #include "util/u_math.h"
34 #include "util/u_memory.h"
35 #include "lp_perf.h"
36 #include "lp_rast.h"
37 #include "lp_state_fs.h"
38 #include "lp_state_setup.h"
39 #include "tgsi/tgsi_scan.h"
40
41 #define NUM_CHANNELS 4
42
43 struct point_info {
44 /* x,y deltas */
45 int dy01, dy12;
46 int dx01, dx12;
47
48 const float (*v0)[4];
49
50 float (*a0)[4];
51 float (*dadx)[4];
52 float (*dady)[4];
53 };
54
55
56 /**
57 * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
58 */
59 static void
constant_coef(struct lp_setup_context * setup,struct point_info * info,unsigned slot,const float value,unsigned i)60 constant_coef(struct lp_setup_context *setup,
61 struct point_info *info,
62 unsigned slot,
63 const float value,
64 unsigned i)
65 {
66 info->a0[slot][i] = value;
67 info->dadx[slot][i] = 0.0f;
68 info->dady[slot][i] = 0.0f;
69 }
70
71
72 static void
point_persp_coeff(struct lp_setup_context * setup,const struct point_info * info,unsigned slot,unsigned i)73 point_persp_coeff(struct lp_setup_context *setup,
74 const struct point_info *info,
75 unsigned slot,
76 unsigned i)
77 {
78 /*
79 * Fragment shader expects pre-multiplied w for LP_INTERP_PERSPECTIVE. A
80 * better stratergy would be to take the primitive in consideration when
81 * generating the fragment shader key, and therefore avoid the per-fragment
82 * perspective divide.
83 */
84
85 float w0 = info->v0[0][3];
86
87 assert(i < 4);
88
89 info->a0[slot][i] = info->v0[slot][i]*w0;
90 info->dadx[slot][i] = 0.0f;
91 info->dady[slot][i] = 0.0f;
92 }
93
94
95 /**
96 * Setup automatic texcoord coefficients (for sprite rendering).
97 * \param slot the vertex attribute slot to setup
98 * \param i the attribute channel in [0,3]
99 * \param sprite_coord_origin one of PIPE_SPRITE_COORD_x
100 * \param perspective does the shader expects pre-multiplied w, i.e.,
101 * LP_INTERP_PERSPECTIVE is specified in the shader key
102 */
103 static void
texcoord_coef(struct lp_setup_context * setup,const struct point_info * info,unsigned slot,unsigned i,unsigned sprite_coord_origin,boolean perspective)104 texcoord_coef(struct lp_setup_context *setup,
105 const struct point_info *info,
106 unsigned slot,
107 unsigned i,
108 unsigned sprite_coord_origin,
109 boolean perspective)
110 {
111 float w0 = info->v0[0][3];
112
113 assert(i < 4);
114
115 if (i == 0) {
116 float dadx = FIXED_ONE / (float)info->dx12;
117 float dady = 0.0f;
118 float x0 = info->v0[0][0] - setup->pixel_offset;
119 float y0 = info->v0[0][1] - setup->pixel_offset;
120
121 info->dadx[slot][0] = dadx;
122 info->dady[slot][0] = dady;
123 info->a0[slot][0] = 0.5 - (dadx * x0 + dady * y0);
124
125 if (perspective) {
126 info->dadx[slot][0] *= w0;
127 info->dady[slot][0] *= w0;
128 info->a0[slot][0] *= w0;
129 }
130 }
131 else if (i == 1) {
132 float dadx = 0.0f;
133 float dady = FIXED_ONE / (float)info->dx12;
134 float x0 = info->v0[0][0] - setup->pixel_offset;
135 float y0 = info->v0[0][1] - setup->pixel_offset;
136
137 if (sprite_coord_origin == PIPE_SPRITE_COORD_LOWER_LEFT) {
138 dady = -dady;
139 }
140
141 info->dadx[slot][1] = dadx;
142 info->dady[slot][1] = dady;
143 info->a0[slot][1] = 0.5 - (dadx * x0 + dady * y0);
144
145 if (perspective) {
146 info->dadx[slot][1] *= w0;
147 info->dady[slot][1] *= w0;
148 info->a0[slot][1] *= w0;
149 }
150 }
151 else if (i == 2) {
152 info->a0[slot][2] = 0.0f;
153 info->dadx[slot][2] = 0.0f;
154 info->dady[slot][2] = 0.0f;
155 }
156 else {
157 info->a0[slot][3] = perspective ? w0 : 1.0f;
158 info->dadx[slot][3] = 0.0f;
159 info->dady[slot][3] = 0.0f;
160 }
161 }
162
163
164 /**
165 * Special coefficient setup for gl_FragCoord.
166 * X and Y are trivial
167 * Z and W are copied from position_coef which should have already been computed.
168 * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
169 */
170 static void
setup_point_fragcoord_coef(struct lp_setup_context * setup,struct point_info * info,unsigned slot,unsigned usage_mask)171 setup_point_fragcoord_coef(struct lp_setup_context *setup,
172 struct point_info *info,
173 unsigned slot,
174 unsigned usage_mask)
175 {
176 /*X*/
177 if (usage_mask & TGSI_WRITEMASK_X) {
178 info->a0[slot][0] = 0.0;
179 info->dadx[slot][0] = 1.0;
180 info->dady[slot][0] = 0.0;
181 }
182
183 /*Y*/
184 if (usage_mask & TGSI_WRITEMASK_Y) {
185 info->a0[slot][1] = 0.0;
186 info->dadx[slot][1] = 0.0;
187 info->dady[slot][1] = 1.0;
188 }
189
190 /*Z*/
191 if (usage_mask & TGSI_WRITEMASK_Z) {
192 constant_coef(setup, info, slot, info->v0[0][2], 2);
193 }
194
195 /*W*/
196 if (usage_mask & TGSI_WRITEMASK_W) {
197 constant_coef(setup, info, slot, info->v0[0][3], 3);
198 }
199 }
200
201
202 /**
203 * Compute the point->coef[] array dadx, dady, a0 values.
204 */
205 static void
setup_point_coefficients(struct lp_setup_context * setup,struct point_info * info)206 setup_point_coefficients( struct lp_setup_context *setup,
207 struct point_info *info)
208 {
209 const struct lp_setup_variant_key *key = &setup->setup.variant->key;
210 const struct lp_fragment_shader *shader = setup->fs.current.variant->shader;
211 unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ;
212 unsigned slot;
213
214 /* setup interpolation for all the remaining attributes:
215 */
216 for (slot = 0; slot < key->num_inputs; slot++) {
217 unsigned vert_attr = key->inputs[slot].src_index;
218 unsigned usage_mask = key->inputs[slot].usage_mask;
219 enum lp_interp interp = key->inputs[slot].interp;
220 boolean perspective = !!(interp == LP_INTERP_PERSPECTIVE);
221 unsigned i;
222
223 if (perspective & usage_mask) {
224 fragcoord_usage_mask |= TGSI_WRITEMASK_W;
225 }
226
227 switch (interp) {
228 case LP_INTERP_POSITION:
229 /*
230 * The generated pixel interpolators will pick up the coeffs from
231 * slot 0, so all need to ensure that the usage mask is covers all
232 * usages.
233 */
234 fragcoord_usage_mask |= usage_mask;
235 break;
236
237 case LP_INTERP_LINEAR:
238 /* Sprite tex coords may use linear interpolation someday */
239 /* fall-through */
240 case LP_INTERP_PERSPECTIVE:
241 /* check if the sprite coord flag is set for this attribute.
242 * If so, set it up so it up so x and y vary from 0 to 1.
243 */
244 if (shader->info.base.input_semantic_name[slot] == TGSI_SEMANTIC_GENERIC) {
245 unsigned semantic_index = shader->info.base.input_semantic_index[slot];
246 /* Note that sprite_coord enable is a bitfield of
247 * PIPE_MAX_SHADER_OUTPUTS bits.
248 */
249 if (semantic_index < PIPE_MAX_SHADER_OUTPUTS &&
250 (setup->sprite_coord_enable & (1 << semantic_index))) {
251 for (i = 0; i < NUM_CHANNELS; i++) {
252 if (usage_mask & (1 << i)) {
253 texcoord_coef(setup, info, slot + 1, i,
254 setup->sprite_coord_origin,
255 perspective);
256 }
257 }
258 break;
259 }
260 }
261 /* fall-through */
262 case LP_INTERP_CONSTANT:
263 for (i = 0; i < NUM_CHANNELS; i++) {
264 if (usage_mask & (1 << i)) {
265 if (perspective) {
266 point_persp_coeff(setup, info, slot+1, i);
267 }
268 else {
269 constant_coef(setup, info, slot+1, info->v0[vert_attr][i], i);
270 }
271 }
272 }
273 break;
274
275 case LP_INTERP_FACING:
276 for (i = 0; i < NUM_CHANNELS; i++)
277 if (usage_mask & (1 << i))
278 constant_coef(setup, info, slot+1, 1.0, i);
279 break;
280
281 default:
282 assert(0);
283 break;
284 }
285 }
286
287 /* The internal position input is in slot zero:
288 */
289 setup_point_fragcoord_coef(setup, info, 0,
290 fragcoord_usage_mask);
291 }
292
293
294 static INLINE int
subpixel_snap(float a)295 subpixel_snap(float a)
296 {
297 return util_iround(FIXED_ONE * a);
298 }
299
300
301 static boolean
try_setup_point(struct lp_setup_context * setup,const float (* v0)[4])302 try_setup_point( struct lp_setup_context *setup,
303 const float (*v0)[4] )
304 {
305 /* x/y positions in fixed point */
306 const struct lp_setup_variant_key *key = &setup->setup.variant->key;
307 const int sizeAttr = setup->psize;
308 const float size
309 = (setup->point_size_per_vertex && sizeAttr > 0) ? v0[sizeAttr][0]
310 : setup->point_size;
311
312 /* Point size as fixed point integer, remove rounding errors
313 * and gives minimum width for very small points
314 */
315 int fixed_width = MAX2(FIXED_ONE,
316 (subpixel_snap(size) + FIXED_ONE/2 - 1) & ~(FIXED_ONE-1));
317
318 const int x0 = subpixel_snap(v0[0][0] - setup->pixel_offset) - fixed_width/2;
319 const int y0 = subpixel_snap(v0[0][1] - setup->pixel_offset) - fixed_width/2;
320
321 struct lp_scene *scene = setup->scene;
322 struct lp_rast_triangle *point;
323 unsigned bytes;
324 struct u_rect bbox;
325 unsigned nr_planes = 4;
326 struct point_info info;
327
328
329 /* Bounding rectangle (in pixels) */
330 {
331 /* Yes this is necessary to accurately calculate bounding boxes
332 * with the two fill-conventions we support. GL (normally) ends
333 * up needing a bottom-left fill convention, which requires
334 * slightly different rounding.
335 */
336 int adj = (setup->pixel_offset != 0) ? 1 : 0;
337
338 bbox.x0 = (x0 + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
339 bbox.x1 = (x0 + fixed_width + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
340 bbox.y0 = (y0 + (FIXED_ONE-1)) >> FIXED_ORDER;
341 bbox.y1 = (y0 + fixed_width + (FIXED_ONE-1)) >> FIXED_ORDER;
342
343 /* Inclusive coordinates:
344 */
345 bbox.x1--;
346 bbox.y1--;
347 }
348
349 if (!u_rect_test_intersection(&setup->draw_region, &bbox)) {
350 if (0) debug_printf("offscreen\n");
351 LP_COUNT(nr_culled_tris);
352 return TRUE;
353 }
354
355 u_rect_find_intersection(&setup->draw_region, &bbox);
356
357 point = lp_setup_alloc_triangle(scene,
358 key->num_inputs,
359 nr_planes,
360 &bytes);
361 if (!point)
362 return FALSE;
363
364 #ifdef DEBUG
365 point->v[0][0] = v0[0][0];
366 point->v[0][1] = v0[0][1];
367 #endif
368
369 info.v0 = v0;
370 info.dx01 = 0;
371 info.dx12 = fixed_width;
372 info.dy01 = fixed_width;
373 info.dy12 = 0;
374 info.a0 = GET_A0(&point->inputs);
375 info.dadx = GET_DADX(&point->inputs);
376 info.dady = GET_DADY(&point->inputs);
377
378 /* Setup parameter interpolants:
379 */
380 setup_point_coefficients(setup, &info);
381
382 point->inputs.frontfacing = TRUE;
383 point->inputs.disable = FALSE;
384 point->inputs.opaque = FALSE;
385
386 {
387 struct lp_rast_plane *plane = GET_PLANES(point);
388
389 plane[0].dcdx = -1;
390 plane[0].dcdy = 0;
391 plane[0].c = 1-bbox.x0;
392 plane[0].eo = 1;
393
394 plane[1].dcdx = 1;
395 plane[1].dcdy = 0;
396 plane[1].c = bbox.x1+1;
397 plane[1].eo = 0;
398
399 plane[2].dcdx = 0;
400 plane[2].dcdy = 1;
401 plane[2].c = 1-bbox.y0;
402 plane[2].eo = 1;
403
404 plane[3].dcdx = 0;
405 plane[3].dcdy = -1;
406 plane[3].c = bbox.y1+1;
407 plane[3].eo = 0;
408 }
409
410 return lp_setup_bin_triangle(setup, point, &bbox, nr_planes);
411 }
412
413
414 static void
lp_setup_point(struct lp_setup_context * setup,const float (* v0)[4])415 lp_setup_point(struct lp_setup_context *setup,
416 const float (*v0)[4])
417 {
418 if (!try_setup_point( setup, v0 ))
419 {
420 if (!lp_setup_flush_and_restart(setup))
421 return;
422
423 if (!try_setup_point( setup, v0 ))
424 return;
425 }
426 }
427
428
429 void
lp_setup_choose_point(struct lp_setup_context * setup)430 lp_setup_choose_point( struct lp_setup_context *setup )
431 {
432 setup->point = lp_setup_point;
433 }
434
435
436