1 /**************************************************************************
2 *
3 * Copyright 2009 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 #ifndef LP_RAST_PRIV_H
29 #define LP_RAST_PRIV_H
30
31 #include "util/format/u_format.h"
32 #include "util/u_thread.h"
33 #include "gallivm/lp_bld_debug.h"
34 #include "lp_memory.h"
35 #include "lp_rast.h"
36 #include "lp_scene.h"
37 #include "lp_state.h"
38 #include "lp_texture.h"
39 #include "lp_limits.h"
40
41
42 #define TILE_VECTOR_HEIGHT 4
43 #define TILE_VECTOR_WIDTH 4
44
45 /* If we crash in a jitted function, we can examine jit_line and jit_state
46 * to get some info. This is not thread-safe, however.
47 */
48 #ifdef DEBUG
49
50 struct lp_rasterizer_task;
51 extern int jit_line;
52 extern const struct lp_rast_state *jit_state;
53 extern const struct lp_rasterizer_task *jit_task;
54
55 #define BEGIN_JIT_CALL(state, task) \
56 do { \
57 jit_line = __LINE__; \
58 jit_state = state; \
59 jit_task = task; \
60 } while (0)
61
62 #define END_JIT_CALL() \
63 do { \
64 jit_line = 0; \
65 jit_state = NULL; \
66 } while (0)
67
68 #else
69
70 #define BEGIN_JIT_CALL(X, Y)
71 #define END_JIT_CALL()
72
73 #endif
74
75
76 struct lp_rasterizer;
77 struct cmd_bin;
78
79 /**
80 * Per-thread rasterization state
81 */
82 struct lp_rasterizer_task
83 {
84 const struct cmd_bin *bin;
85 const struct lp_rast_state *state;
86
87 struct lp_scene *scene;
88 unsigned x, y; /**< Pos of this tile in framebuffer, in pixels */
89 unsigned width, height; /**< width, height of current tile, in pixels */
90
91 uint8_t *color_tiles[PIPE_MAX_COLOR_BUFS];
92 uint8_t *depth_tile;
93
94 /** "back" pointer */
95 struct lp_rasterizer *rast;
96
97 /** "my" index */
98 unsigned thread_index;
99
100 /** Non-interpolated passthru state and occlude counter for visible pixels */
101 struct lp_jit_thread_data thread_data;
102
103 pipe_semaphore work_ready;
104 pipe_semaphore work_done;
105 };
106
107
108 /**
109 * This is the state required while rasterizing tiles.
110 * Note that this contains per-thread information too.
111 * The tile size is TILE_SIZE x TILE_SIZE pixels.
112 */
113 struct lp_rasterizer
114 {
115 boolean exit_flag;
116 boolean no_rast; /**< For debugging/profiling */
117
118 /** The incoming queue of scenes ready to rasterize */
119 struct lp_scene_queue *full_scenes;
120
121 /** The scene currently being rasterized by the threads */
122 struct lp_scene *curr_scene;
123
124 /** A task object for each rasterization thread */
125 struct lp_rasterizer_task tasks[LP_MAX_THREADS];
126
127 unsigned num_threads;
128 thrd_t threads[LP_MAX_THREADS];
129
130 /** For synchronizing the rasterization threads */
131 util_barrier barrier;
132
133 struct lp_fence *last_fence;
134 };
135
136
137 void
138 lp_rast_shade_quads_mask_sample(struct lp_rasterizer_task *task,
139 const struct lp_rast_shader_inputs *inputs,
140 unsigned x, unsigned y,
141 uint64_t mask);
142
143 void
144 lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
145 const struct lp_rast_shader_inputs *inputs,
146 unsigned x, unsigned y,
147 unsigned mask);
148
149
150 /**
151 * Get the pointer to a 4x4 color block (within a 64x64 tile).
152 * \param x, y location of 4x4 block in window coords
153 */
154 static inline uint8_t *
lp_rast_get_color_block_pointer(struct lp_rasterizer_task * task,unsigned buf,unsigned x,unsigned y,unsigned layer)155 lp_rast_get_color_block_pointer(struct lp_rasterizer_task *task,
156 unsigned buf, unsigned x, unsigned y,
157 unsigned layer)
158 {
159 assert(x < task->scene->tiles_x * TILE_SIZE);
160 assert(y < task->scene->tiles_y * TILE_SIZE);
161 assert((x % TILE_VECTOR_WIDTH) == 0);
162 assert((y % TILE_VECTOR_HEIGHT) == 0);
163 assert(buf < task->scene->fb.nr_cbufs);
164 assert(task->color_tiles[buf]);
165
166 /*
167 * We don't actually benefit from having per tile cbuf/zsbuf pointers,
168 * it's just extra work - the mul/add would be exactly the same anyway.
169 * Fortunately the extra work (modulo) here is very cheap at least...
170 */
171 unsigned px = x % TILE_SIZE;
172 unsigned py = y % TILE_SIZE;
173
174 unsigned pixel_offset = px * task->scene->cbufs[buf].format_bytes +
175 py * task->scene->cbufs[buf].stride;
176 uint8_t *color = task->color_tiles[buf] + pixel_offset;
177
178 if (layer) {
179 assert(layer <= task->scene->fb_max_layer);
180 color += layer * task->scene->cbufs[buf].layer_stride;
181 }
182
183 assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format)));
184 return color;
185 }
186
187
188 /**
189 * Get the pointer to a 4x4 depth block (within a 64x64 tile).
190 * \param x, y location of 4x4 block in window coords
191 */
192 static inline uint8_t *
lp_rast_get_depth_block_pointer(struct lp_rasterizer_task * task,unsigned x,unsigned y,unsigned layer)193 lp_rast_get_depth_block_pointer(struct lp_rasterizer_task *task,
194 unsigned x, unsigned y, unsigned layer)
195 {
196 assert(x < task->scene->tiles_x * TILE_SIZE);
197 assert(y < task->scene->tiles_y * TILE_SIZE);
198 assert((x % TILE_VECTOR_WIDTH) == 0);
199 assert((y % TILE_VECTOR_HEIGHT) == 0);
200 assert(task->depth_tile);
201
202 unsigned px = x % TILE_SIZE;
203 unsigned py = y % TILE_SIZE;
204
205 unsigned pixel_offset = px * task->scene->zsbuf.format_bytes +
206 py * task->scene->zsbuf.stride;
207 uint8_t *depth = task->depth_tile + pixel_offset;
208
209 if (layer) {
210 depth += layer * task->scene->zsbuf.layer_stride;
211 }
212
213 assert(lp_check_alignment(depth, llvmpipe_get_format_alignment(task->scene->fb.zsbuf->format)));
214 return depth;
215 }
216
217
218
219 /**
220 * Shade all pixels in a 4x4 block. The fragment code omits the
221 * triangle in/out tests.
222 * \param x, y location of 4x4 block in window coords
223 */
224 static inline void
lp_rast_shade_quads_all(struct lp_rasterizer_task * task,const struct lp_rast_shader_inputs * inputs,unsigned x,unsigned y)225 lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
226 const struct lp_rast_shader_inputs *inputs,
227 unsigned x, unsigned y )
228 {
229 const struct lp_scene *scene = task->scene;
230 const struct lp_rast_state *state = task->state;
231 struct lp_fragment_shader_variant *variant = state->variant;
232 uint8_t *color[PIPE_MAX_COLOR_BUFS];
233 unsigned stride[PIPE_MAX_COLOR_BUFS];
234 unsigned sample_stride[PIPE_MAX_COLOR_BUFS];
235 uint8_t *depth = NULL;
236 unsigned depth_stride = 0;
237 unsigned depth_sample_stride = 0;
238
239 /* color buffer */
240 for (unsigned i = 0; i < scene->fb.nr_cbufs; i++) {
241 if (scene->fb.cbufs[i]) {
242 stride[i] = scene->cbufs[i].stride;
243 sample_stride[i] = scene->cbufs[i].sample_stride;
244 color[i] = lp_rast_get_color_block_pointer(task, i, x, y,
245 inputs->layer + inputs->view_index);
246 }
247 else {
248 stride[i] = 0;
249 sample_stride[i] = 0;
250 color[i] = NULL;
251 }
252 }
253
254 if (scene->zsbuf.map) {
255 depth = lp_rast_get_depth_block_pointer(task, x, y, inputs->layer + inputs->view_index);
256 depth_sample_stride = scene->zsbuf.sample_stride;
257 depth_stride = scene->zsbuf.stride;
258 }
259
260 uint64_t mask = 0;
261 for (unsigned i = 0; i < scene->fb_max_samples; i++)
262 mask |= (uint64_t)0xffff << (16 * i);
263
264 /*
265 * The rasterizer may produce fragments outside our
266 * allocated 4x4 blocks hence need to filter them out here.
267 */
268 if ((x % TILE_SIZE) < task->width && (y % TILE_SIZE) < task->height) {
269 /* Propagate non-interpolated raster state. */
270 task->thread_data.raster_state.viewport_index = inputs->viewport_index;
271 task->thread_data.raster_state.view_index = inputs->view_index;
272
273 /* run shader on 4x4 block */
274 BEGIN_JIT_CALL(state, task);
275 variant->jit_function[RAST_WHOLE]( &state->jit_context,
276 x, y,
277 inputs->frontfacing,
278 GET_A0(inputs),
279 GET_DADX(inputs),
280 GET_DADY(inputs),
281 color,
282 depth,
283 mask,
284 &task->thread_data,
285 stride,
286 depth_stride,
287 sample_stride,
288 depth_sample_stride);
289 END_JIT_CALL();
290 }
291 }
292
293 void
294 lp_rast_triangle_1(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
295
296 void
297 lp_rast_triangle_2(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
298
299 void
300 lp_rast_triangle_3(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
301
302 void
303 lp_rast_triangle_4(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
304
305 void
306 lp_rast_triangle_5(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
307
308 void
309 lp_rast_triangle_6(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
310
311 void
312 lp_rast_triangle_7(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
313
314 void
315 lp_rast_triangle_8(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
316
317 void
318 lp_rast_triangle_3_4(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
319
320 void
321 lp_rast_triangle_3_16(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
322
323 void
324 lp_rast_triangle_4_16(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
325
326 void
327 lp_rast_triangle_32_1(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
328
329 void
330 lp_rast_triangle_32_2(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
331
332 void
333 lp_rast_triangle_32_3(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
334
335 void
336 lp_rast_triangle_32_4(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
337
338 void
339 lp_rast_triangle_32_5(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
340
341 void
342 lp_rast_triangle_32_6(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
343
344 void
345 lp_rast_triangle_32_7(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
346
347 void
348 lp_rast_triangle_32_8(struct lp_rasterizer_task *, const union lp_rast_cmd_arg);
349
350 void
351 lp_rast_triangle_32_3_4(struct lp_rasterizer_task *,
352 const union lp_rast_cmd_arg);
353
354 void
355 lp_rast_triangle_32_3_16(struct lp_rasterizer_task *,
356 const union lp_rast_cmd_arg);
357
358 void
359 lp_rast_triangle_32_4_16(struct lp_rasterizer_task *,
360 const union lp_rast_cmd_arg);
361
362 void
363 lp_rast_rectangle(struct lp_rasterizer_task *,
364 const union lp_rast_cmd_arg);
365
366 void
367 lp_rast_triangle_ms_1(struct lp_rasterizer_task *,
368 const union lp_rast_cmd_arg);
369
370 void
371 lp_rast_triangle_ms_2(struct lp_rasterizer_task *,
372 const union lp_rast_cmd_arg);
373
374 void
375 lp_rast_triangle_ms_3(struct lp_rasterizer_task *,
376 const union lp_rast_cmd_arg);
377
378 void
379 lp_rast_triangle_ms_4(struct lp_rasterizer_task *,
380 const union lp_rast_cmd_arg);
381
382 void
383 lp_rast_triangle_ms_5(struct lp_rasterizer_task *,
384 const union lp_rast_cmd_arg);
385
386 void
387 lp_rast_triangle_ms_6(struct lp_rasterizer_task *,
388 const union lp_rast_cmd_arg);
389
390 void
391 lp_rast_triangle_ms_7(struct lp_rasterizer_task *,
392 const union lp_rast_cmd_arg);
393
394 void
395 lp_rast_triangle_ms_8(struct lp_rasterizer_task *,
396 const union lp_rast_cmd_arg);
397
398
399 void
400 lp_rast_triangle_ms_3_4(struct lp_rasterizer_task *,
401 const union lp_rast_cmd_arg);
402
403
404 void
405 lp_rast_triangle_ms_3_16(struct lp_rasterizer_task *,
406 const union lp_rast_cmd_arg);
407
408
409 void
410 lp_rast_triangle_ms_4_16(struct lp_rasterizer_task *,
411 const union lp_rast_cmd_arg);
412
413
414 void
415 lp_rast_triangle_ms_32_1(struct lp_rasterizer_task *,
416 const union lp_rast_cmd_arg);
417
418 void
419 lp_rast_triangle_ms_32_2(struct lp_rasterizer_task *,
420 const union lp_rast_cmd_arg);
421
422 void
423 lp_rast_triangle_ms_32_3(struct lp_rasterizer_task *,
424 const union lp_rast_cmd_arg);
425
426 void
427 lp_rast_triangle_ms_32_4(struct lp_rasterizer_task *,
428 const union lp_rast_cmd_arg);
429
430 void
431 lp_rast_triangle_ms_32_5(struct lp_rasterizer_task *,
432 const union lp_rast_cmd_arg);
433
434 void
435 lp_rast_triangle_ms_32_6(struct lp_rasterizer_task *,
436 const union lp_rast_cmd_arg);
437
438 void
439 lp_rast_triangle_ms_32_7(struct lp_rasterizer_task *,
440 const union lp_rast_cmd_arg);
441
442 void
443 lp_rast_triangle_ms_32_8(struct lp_rasterizer_task *,
444 const union lp_rast_cmd_arg);
445
446 void
447 lp_rast_triangle_ms_32_3_4(struct lp_rasterizer_task *,
448 const union lp_rast_cmd_arg);
449
450 void
451 lp_rast_triangle_ms_32_3_16(struct lp_rasterizer_task *,
452 const union lp_rast_cmd_arg);
453
454 void
455 lp_rast_triangle_ms_32_4_16(struct lp_rasterizer_task *,
456 const union lp_rast_cmd_arg);
457
458 void
459 lp_rast_set_state(struct lp_rasterizer_task *task,
460 const union lp_rast_cmd_arg arg);
461
462 void
463 lp_debug_bin(const struct cmd_bin *bin, int x, int y);
464
465 void
466 lp_linear_rasterize_bin(struct lp_rasterizer_task *task,
467 const struct cmd_bin *bin);
468
469 void
470 lp_rast_linear_rect_fallback(struct lp_rasterizer_task *task,
471 const struct lp_rast_shader_inputs *inputs,
472 const struct u_rect *box);
473
474 #endif
475