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
134 void
135 lp_rast_shade_quads_mask_sample(struct lp_rasterizer_task *task,
136 const struct lp_rast_shader_inputs *inputs,
137 unsigned x, unsigned y,
138 uint64_t mask);
139 void
140 lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
141 const struct lp_rast_shader_inputs *inputs,
142 unsigned x, unsigned y,
143 unsigned mask);
144
145
146 /**
147 * Get the pointer to a 4x4 color block (within a 64x64 tile).
148 * \param x, y location of 4x4 block in window coords
149 */
150 static inline uint8_t *
lp_rast_get_color_block_pointer(struct lp_rasterizer_task * task,unsigned buf,unsigned x,unsigned y,unsigned layer)151 lp_rast_get_color_block_pointer(struct lp_rasterizer_task *task,
152 unsigned buf, unsigned x, unsigned y,
153 unsigned layer)
154 {
155 unsigned px, py, pixel_offset;
156 uint8_t *color;
157
158 assert(x < task->scene->tiles_x * TILE_SIZE);
159 assert(y < task->scene->tiles_y * TILE_SIZE);
160 assert((x % TILE_VECTOR_WIDTH) == 0);
161 assert((y % TILE_VECTOR_HEIGHT) == 0);
162 assert(buf < task->scene->fb.nr_cbufs);
163
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 px = x % TILE_SIZE;
172 py = y % TILE_SIZE;
173
174 pixel_offset = px * task->scene->cbufs[buf].format_bytes +
175 py * task->scene->cbufs[buf].stride;
176 color = task->color_tiles[buf] + pixel_offset;
177
178 if (layer) {
179 color += layer * task->scene->cbufs[buf].layer_stride;
180 }
181
182 assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format)));
183 return color;
184 }
185
186
187 /**
188 * Get the pointer to a 4x4 depth block (within a 64x64 tile).
189 * \param x, y location of 4x4 block in window coords
190 */
191 static inline uint8_t *
lp_rast_get_depth_block_pointer(struct lp_rasterizer_task * task,unsigned x,unsigned y,unsigned layer)192 lp_rast_get_depth_block_pointer(struct lp_rasterizer_task *task,
193 unsigned x, unsigned y, unsigned layer)
194 {
195 unsigned px, py, pixel_offset;
196 uint8_t *depth;
197
198 assert(x < task->scene->tiles_x * TILE_SIZE);
199 assert(y < task->scene->tiles_y * TILE_SIZE);
200 assert((x % TILE_VECTOR_WIDTH) == 0);
201 assert((y % TILE_VECTOR_HEIGHT) == 0);
202
203 assert(task->depth_tile);
204
205 px = x % TILE_SIZE;
206 py = y % TILE_SIZE;
207
208 pixel_offset = px * task->scene->zsbuf.format_bytes +
209 py * task->scene->zsbuf.stride;
210 depth = task->depth_tile + pixel_offset;
211
212 if (layer) {
213 depth += layer * task->scene->zsbuf.layer_stride;
214 }
215
216 assert(lp_check_alignment(depth, llvmpipe_get_format_alignment(task->scene->fb.zsbuf->format)));
217 return depth;
218 }
219
220
221
222 /**
223 * Shade all pixels in a 4x4 block. The fragment code omits the
224 * triangle in/out tests.
225 * \param x, y location of 4x4 block in window coords
226 */
227 static inline void
lp_rast_shade_quads_all(struct lp_rasterizer_task * task,const struct lp_rast_shader_inputs * inputs,unsigned x,unsigned y)228 lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
229 const struct lp_rast_shader_inputs *inputs,
230 unsigned x, unsigned y )
231 {
232 const struct lp_scene *scene = task->scene;
233 const struct lp_rast_state *state = task->state;
234 struct lp_fragment_shader_variant *variant = state->variant;
235 uint8_t *color[PIPE_MAX_COLOR_BUFS];
236 unsigned stride[PIPE_MAX_COLOR_BUFS];
237 unsigned sample_stride[PIPE_MAX_COLOR_BUFS];
238 uint8_t *depth = NULL;
239 unsigned depth_stride = 0;
240 unsigned depth_sample_stride = 0;
241 unsigned i;
242
243 /* color buffer */
244 for (i = 0; i < scene->fb.nr_cbufs; i++) {
245 if (scene->fb.cbufs[i]) {
246 stride[i] = scene->cbufs[i].stride;
247 sample_stride[i] = scene->cbufs[i].sample_stride;
248 color[i] = lp_rast_get_color_block_pointer(task, i, x, y,
249 inputs->layer);
250 }
251 else {
252 stride[i] = 0;
253 sample_stride[i] = 0;
254 color[i] = NULL;
255 }
256 }
257
258 if (scene->zsbuf.map) {
259 depth = lp_rast_get_depth_block_pointer(task, x, y, inputs->layer);
260 depth_sample_stride = scene->zsbuf.sample_stride;
261 depth_stride = scene->zsbuf.stride;
262 }
263
264 uint64_t mask = 0;
265 for (unsigned i = 0; i < scene->fb_max_samples; i++)
266 mask |= (uint64_t)0xffff << (16 * i);
267
268 /*
269 * The rasterizer may produce fragments outside our
270 * allocated 4x4 blocks hence need to filter them out here.
271 */
272 if ((x % TILE_SIZE) < task->width && (y % TILE_SIZE) < task->height) {
273 /* Propagate non-interpolated raster state. */
274 task->thread_data.raster_state.viewport_index = inputs->viewport_index;
275
276 /* run shader on 4x4 block */
277 BEGIN_JIT_CALL(state, task);
278 variant->jit_function[RAST_WHOLE]( &state->jit_context,
279 x, y,
280 inputs->frontfacing,
281 GET_A0(inputs),
282 GET_DADX(inputs),
283 GET_DADY(inputs),
284 color,
285 depth,
286 mask,
287 &task->thread_data,
288 stride,
289 depth_stride,
290 sample_stride,
291 depth_sample_stride);
292 END_JIT_CALL();
293 }
294 }
295
296 void lp_rast_triangle_1( struct lp_rasterizer_task *,
297 const union lp_rast_cmd_arg );
298 void lp_rast_triangle_2( struct lp_rasterizer_task *,
299 const union lp_rast_cmd_arg );
300 void lp_rast_triangle_3( struct lp_rasterizer_task *,
301 const union lp_rast_cmd_arg );
302 void lp_rast_triangle_4( struct lp_rasterizer_task *,
303 const union lp_rast_cmd_arg );
304 void lp_rast_triangle_5( struct lp_rasterizer_task *,
305 const union lp_rast_cmd_arg );
306 void lp_rast_triangle_6( struct lp_rasterizer_task *,
307 const union lp_rast_cmd_arg );
308 void lp_rast_triangle_7( struct lp_rasterizer_task *,
309 const union lp_rast_cmd_arg );
310 void lp_rast_triangle_8( struct lp_rasterizer_task *,
311 const union lp_rast_cmd_arg );
312
313 void lp_rast_triangle_3_4(struct lp_rasterizer_task *,
314 const union lp_rast_cmd_arg );
315
316 void lp_rast_triangle_3_16( struct lp_rasterizer_task *,
317 const union lp_rast_cmd_arg );
318
319 void lp_rast_triangle_4_16( struct lp_rasterizer_task *,
320 const union lp_rast_cmd_arg );
321
322
323 void lp_rast_triangle_32_1( struct lp_rasterizer_task *,
324 const union lp_rast_cmd_arg );
325 void lp_rast_triangle_32_2( struct lp_rasterizer_task *,
326 const union lp_rast_cmd_arg );
327 void lp_rast_triangle_32_3( struct lp_rasterizer_task *,
328 const union lp_rast_cmd_arg );
329 void lp_rast_triangle_32_4( struct lp_rasterizer_task *,
330 const union lp_rast_cmd_arg );
331 void lp_rast_triangle_32_5( struct lp_rasterizer_task *,
332 const union lp_rast_cmd_arg );
333 void lp_rast_triangle_32_6( struct lp_rasterizer_task *,
334 const union lp_rast_cmd_arg );
335 void lp_rast_triangle_32_7( struct lp_rasterizer_task *,
336 const union lp_rast_cmd_arg );
337 void lp_rast_triangle_32_8( struct lp_rasterizer_task *,
338 const union lp_rast_cmd_arg );
339
340 void lp_rast_triangle_32_3_4(struct lp_rasterizer_task *,
341 const union lp_rast_cmd_arg );
342
343 void lp_rast_triangle_32_3_16( struct lp_rasterizer_task *,
344 const union lp_rast_cmd_arg );
345
346 void lp_rast_triangle_32_4_16( struct lp_rasterizer_task *,
347 const union lp_rast_cmd_arg );
348
349 void lp_rast_triangle_ms_1( struct lp_rasterizer_task *,
350 const union lp_rast_cmd_arg );
351 void lp_rast_triangle_ms_2( struct lp_rasterizer_task *,
352 const union lp_rast_cmd_arg );
353 void lp_rast_triangle_ms_3( struct lp_rasterizer_task *,
354 const union lp_rast_cmd_arg );
355 void lp_rast_triangle_ms_4( struct lp_rasterizer_task *,
356 const union lp_rast_cmd_arg );
357 void lp_rast_triangle_ms_5( struct lp_rasterizer_task *,
358 const union lp_rast_cmd_arg );
359 void lp_rast_triangle_ms_6( struct lp_rasterizer_task *,
360 const union lp_rast_cmd_arg );
361 void lp_rast_triangle_ms_7( struct lp_rasterizer_task *,
362 const union lp_rast_cmd_arg );
363 void lp_rast_triangle_ms_8( struct lp_rasterizer_task *,
364 const union lp_rast_cmd_arg );
365
366 void lp_rast_triangle_ms_3_4(struct lp_rasterizer_task *,
367 const union lp_rast_cmd_arg );
368
369 void lp_rast_triangle_ms_3_16( struct lp_rasterizer_task *,
370 const union lp_rast_cmd_arg );
371
372 void lp_rast_triangle_ms_4_16( struct lp_rasterizer_task *,
373 const union lp_rast_cmd_arg );
374
375 void lp_rast_triangle_ms_32_1( struct lp_rasterizer_task *,
376 const union lp_rast_cmd_arg );
377 void lp_rast_triangle_ms_32_2( struct lp_rasterizer_task *,
378 const union lp_rast_cmd_arg );
379 void lp_rast_triangle_ms_32_3( struct lp_rasterizer_task *,
380 const union lp_rast_cmd_arg );
381 void lp_rast_triangle_ms_32_4( struct lp_rasterizer_task *,
382 const union lp_rast_cmd_arg );
383 void lp_rast_triangle_ms_32_5( struct lp_rasterizer_task *,
384 const union lp_rast_cmd_arg );
385 void lp_rast_triangle_ms_32_6( struct lp_rasterizer_task *,
386 const union lp_rast_cmd_arg );
387 void lp_rast_triangle_ms_32_7( struct lp_rasterizer_task *,
388 const union lp_rast_cmd_arg );
389 void lp_rast_triangle_ms_32_8( struct lp_rasterizer_task *,
390 const union lp_rast_cmd_arg );
391
392 void lp_rast_triangle_ms_32_3_4(struct lp_rasterizer_task *,
393 const union lp_rast_cmd_arg );
394
395 void lp_rast_triangle_ms_32_3_16( struct lp_rasterizer_task *,
396 const union lp_rast_cmd_arg );
397
398 void lp_rast_triangle_ms_32_4_16( struct lp_rasterizer_task *,
399 const union lp_rast_cmd_arg );
400
401 void
402 lp_rast_set_state(struct lp_rasterizer_task *task,
403 const union lp_rast_cmd_arg arg);
404
405 void
406 lp_debug_bin( const struct cmd_bin *bin, int x, int y );
407
408 #endif
409