1 /**************************************************************************
3 * Copyright 2009 VMware, Inc.
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 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.
26 **************************************************************************/
28 #ifndef LP_RAST_PRIV_H
29 #define LP_RAST_PRIV_H
31 #include "os/os_thread.h"
32 #include "util/u_format.h"
33 #include "gallivm/lp_bld_debug.h"
34 #include "lp_memory.h"
38 #include "lp_texture.h"
39 #include "lp_tile_image.h"
40 #include "lp_limits.h"
43 /* If we crash in a jitted function, we can examine jit_line and jit_state
44 * to get some info. This is not thread-safe, however.
48 struct lp_rasterizer_task;
50 extern const struct lp_rast_state *jit_state;
51 extern const struct lp_rasterizer_task *jit_task;
53 #define BEGIN_JIT_CALL(state, task) \
55 jit_line = __LINE__; \
60 #define END_JIT_CALL() \
68 #define BEGIN_JIT_CALL(X, Y)
69 #define END_JIT_CALL()
78 * Per-thread rasterization state
80 struct lp_rasterizer_task
82 const struct cmd_bin *bin;
83 const struct lp_rast_state *state;
85 struct lp_scene *scene;
86 unsigned x, y; /**< Pos of this tile in framebuffer, in pixels */
88 uint8_t *color_tiles[PIPE_MAX_COLOR_BUFS];
92 struct lp_rasterizer *rast;
95 unsigned thread_index;
97 /* occlude counter for visiable pixels */
100 struct llvmpipe_query *query[PIPE_QUERY_TYPES];
102 pipe_semaphore work_ready;
103 pipe_semaphore work_done;
108 * This is the state required while rasterizing tiles.
109 * Note that this contains per-thread information too.
110 * The tile size is TILE_SIZE x TILE_SIZE pixels.
115 boolean no_rast; /**< For debugging/profiling */
117 /** The incoming queue of scenes ready to rasterize */
118 struct lp_scene_queue *full_scenes;
120 /** The scene currently being rasterized by the threads */
121 struct lp_scene *curr_scene;
123 /** A task object for each rasterization thread */
124 struct lp_rasterizer_task tasks[LP_MAX_THREADS];
126 unsigned num_threads;
127 pipe_thread threads[LP_MAX_THREADS];
129 /** For synchronizing the rasterization threads */
130 pipe_barrier barrier;
135 lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
136 const struct lp_rast_shader_inputs *inputs,
137 unsigned x, unsigned y,
143 * Get the pointer to a 4x4 depth/stencil block.
144 * We'll map the z/stencil buffer on demand here.
145 * Note that this may be called even when there's no z/stencil buffer - return
147 * \param x, y location of 4x4 block in window coords
150 lp_rast_get_depth_block_pointer(struct lp_rasterizer_task *task,
151 unsigned x, unsigned y)
153 const struct lp_scene *scene = task->scene;
156 assert(x < scene->tiles_x * TILE_SIZE);
157 assert(y < scene->tiles_y * TILE_SIZE);
158 assert((x % TILE_VECTOR_WIDTH) == 0);
159 assert((y % TILE_VECTOR_HEIGHT) == 0);
161 if (!scene->zsbuf.map) {
162 /* Either out of memory or no zsbuf. Can't tell without access
163 * to the state. Just use dummy tile memory, but don't print
164 * the oom warning as this most likely because there is no
167 return lp_dummy_tile;
170 depth = (scene->zsbuf.map +
171 scene->zsbuf.stride * y +
172 scene->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT);
174 assert(lp_check_alignment(depth, 16));
180 * Get pointer to the unswizzled color tile
182 static INLINE uint8_t *
183 lp_rast_get_unswizzled_color_tile_pointer(struct lp_rasterizer_task *task,
184 unsigned buf, enum lp_texture_usage usage)
186 const struct lp_scene *scene = task->scene;
187 unsigned format_bytes;
189 assert(task->x < scene->tiles_x * TILE_SIZE);
190 assert(task->y < scene->tiles_y * TILE_SIZE);
191 assert(task->x % TILE_SIZE == 0);
192 assert(task->y % TILE_SIZE == 0);
193 assert(buf < scene->fb.nr_cbufs);
195 if (!task->color_tiles[buf]) {
196 struct pipe_surface *cbuf = scene->fb.cbufs[buf];
199 format_bytes = util_format_description(cbuf->format)->block.bits / 8;
200 task->color_tiles[buf] = scene->cbufs[buf].map + scene->cbufs[buf].stride * task->y + format_bytes * task->x;
203 return task->color_tiles[buf];
208 * Get the pointer to an unswizzled 4x4 color block (within an unswizzled 64x64 tile).
209 * \param x, y location of 4x4 block in window coords
211 static INLINE uint8_t *
212 lp_rast_get_unswizzled_color_block_pointer(struct lp_rasterizer_task *task,
213 unsigned buf, unsigned x, unsigned y)
215 unsigned px, py, pixel_offset, format_bytes;
218 assert(x < task->scene->tiles_x * TILE_SIZE);
219 assert(y < task->scene->tiles_y * TILE_SIZE);
220 assert((x % TILE_VECTOR_WIDTH) == 0);
221 assert((y % TILE_VECTOR_HEIGHT) == 0);
222 assert(buf < task->scene->fb.nr_cbufs);
224 format_bytes = util_format_description(task->scene->fb.cbufs[buf]->format)->block.bits / 8;
226 color = lp_rast_get_unswizzled_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE);
231 pixel_offset = px * format_bytes + py * task->scene->cbufs[buf].stride;
233 color = color + pixel_offset;
235 assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format)));
242 * Shade all pixels in a 4x4 block. The fragment code omits the
243 * triangle in/out tests.
244 * \param x, y location of 4x4 block in window coords
247 lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
248 const struct lp_rast_shader_inputs *inputs,
249 unsigned x, unsigned y )
251 const struct lp_scene *scene = task->scene;
252 const struct lp_rast_state *state = task->state;
253 struct lp_fragment_shader_variant *variant = state->variant;
254 uint8_t *color[PIPE_MAX_COLOR_BUFS];
255 unsigned stride[PIPE_MAX_COLOR_BUFS];
260 for (i = 0; i < scene->fb.nr_cbufs; i++) {
261 stride[i] = scene->cbufs[i].stride;
263 color[i] = lp_rast_get_unswizzled_color_block_pointer(task, i, x, y);
266 depth = lp_rast_get_depth_block_pointer(task, x, y);
268 /* run shader on 4x4 block */
269 BEGIN_JIT_CALL(state, task);
270 variant->jit_function[RAST_WHOLE]( &state->jit_context,
284 void lp_rast_triangle_1( struct lp_rasterizer_task *,
285 const union lp_rast_cmd_arg );
286 void lp_rast_triangle_2( struct lp_rasterizer_task *,
287 const union lp_rast_cmd_arg );
288 void lp_rast_triangle_3( struct lp_rasterizer_task *,
289 const union lp_rast_cmd_arg );
290 void lp_rast_triangle_4( struct lp_rasterizer_task *,
291 const union lp_rast_cmd_arg );
292 void lp_rast_triangle_5( struct lp_rasterizer_task *,
293 const union lp_rast_cmd_arg );
294 void lp_rast_triangle_6( struct lp_rasterizer_task *,
295 const union lp_rast_cmd_arg );
296 void lp_rast_triangle_7( struct lp_rasterizer_task *,
297 const union lp_rast_cmd_arg );
298 void lp_rast_triangle_8( struct lp_rasterizer_task *,
299 const union lp_rast_cmd_arg );
301 void lp_rast_triangle_3_4(struct lp_rasterizer_task *,
302 const union lp_rast_cmd_arg );
304 void lp_rast_triangle_3_16( struct lp_rasterizer_task *,
305 const union lp_rast_cmd_arg );
307 void lp_rast_triangle_4_16( struct lp_rasterizer_task *,
308 const union lp_rast_cmd_arg );
311 lp_rast_set_state(struct lp_rasterizer_task *task,
312 const union lp_rast_cmd_arg arg);
315 lp_debug_bin( const struct cmd_bin *bin );