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_limits.h"
42 #define TILE_VECTOR_HEIGHT 4
43 #define TILE_VECTOR_WIDTH 4
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.
50 struct lp_rasterizer_task
;
52 extern const struct lp_rast_state
*jit_state
;
53 extern const struct lp_rasterizer_task
*jit_task
;
55 #define BEGIN_JIT_CALL(state, task) \
57 jit_line = __LINE__; \
62 #define END_JIT_CALL() \
70 #define BEGIN_JIT_CALL(X, Y)
71 #define END_JIT_CALL()
80 * Per-thread rasterization state
82 struct lp_rasterizer_task
84 const struct cmd_bin
*bin
;
85 const struct lp_rast_state
*state
;
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 */
91 uint8_t *color_tiles
[PIPE_MAX_COLOR_BUFS
];
95 struct lp_rasterizer
*rast
;
98 unsigned thread_index
;
100 /** Non-interpolated passthru state and occlude counter for visible pixels */
101 struct lp_jit_thread_data thread_data
;
102 uint64_t ps_invocations
;
103 uint8_t ps_inv_multiplier
;
105 pipe_semaphore work_ready
;
106 pipe_semaphore work_done
;
111 * This is the state required while rasterizing tiles.
112 * Note that this contains per-thread information too.
113 * The tile size is TILE_SIZE x TILE_SIZE pixels.
118 boolean no_rast
; /**< For debugging/profiling */
120 /** The incoming queue of scenes ready to rasterize */
121 struct lp_scene_queue
*full_scenes
;
123 /** The scene currently being rasterized by the threads */
124 struct lp_scene
*curr_scene
;
126 /** A task object for each rasterization thread */
127 struct lp_rasterizer_task tasks
[LP_MAX_THREADS
];
129 unsigned num_threads
;
130 pipe_thread threads
[LP_MAX_THREADS
];
132 /** For synchronizing the rasterization threads */
133 pipe_barrier barrier
;
138 lp_rast_shade_quads_mask(struct lp_rasterizer_task
*task
,
139 const struct lp_rast_shader_inputs
*inputs
,
140 unsigned x
, unsigned y
,
145 * Get the pointer to a 4x4 color block (within a 64x64 tile).
146 * \param x, y location of 4x4 block in window coords
148 static inline uint8_t *
149 lp_rast_get_color_block_pointer(struct lp_rasterizer_task
*task
,
150 unsigned buf
, unsigned x
, unsigned y
,
153 unsigned px
, py
, pixel_offset
;
156 assert(x
< task
->scene
->tiles_x
* TILE_SIZE
);
157 assert(y
< task
->scene
->tiles_y
* TILE_SIZE
);
158 assert((x
% TILE_VECTOR_WIDTH
) == 0);
159 assert((y
% TILE_VECTOR_HEIGHT
) == 0);
160 assert(buf
< task
->scene
->fb
.nr_cbufs
);
162 assert(task
->color_tiles
[buf
]);
165 * We don't actually benefit from having per tile cbuf/zsbuf pointers,
166 * it's just extra work - the mul/add would be exactly the same anyway.
167 * Fortunately the extra work (modulo) here is very cheap at least...
172 pixel_offset
= px
* task
->scene
->cbufs
[buf
].format_bytes
+
173 py
* task
->scene
->cbufs
[buf
].stride
;
174 color
= task
->color_tiles
[buf
] + pixel_offset
;
177 color
+= layer
* task
->scene
->cbufs
[buf
].layer_stride
;
180 assert(lp_check_alignment(color
, llvmpipe_get_format_alignment(task
->scene
->fb
.cbufs
[buf
]->format
)));
186 * Get the pointer to a 4x4 depth block (within a 64x64 tile).
187 * \param x, y location of 4x4 block in window coords
189 static inline uint8_t *
190 lp_rast_get_depth_block_pointer(struct lp_rasterizer_task
*task
,
191 unsigned x
, unsigned y
, unsigned layer
)
193 unsigned px
, py
, pixel_offset
;
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);
201 assert(task
->depth_tile
);
206 pixel_offset
= px
* task
->scene
->zsbuf
.format_bytes
+
207 py
* task
->scene
->zsbuf
.stride
;
208 depth
= task
->depth_tile
+ pixel_offset
;
211 depth
+= layer
* task
->scene
->zsbuf
.layer_stride
;
214 assert(lp_check_alignment(depth
, llvmpipe_get_format_alignment(task
->scene
->fb
.zsbuf
->format
)));
221 * Shade all pixels in a 4x4 block. The fragment code omits the
222 * triangle in/out tests.
223 * \param x, y location of 4x4 block in window coords
226 lp_rast_shade_quads_all( struct lp_rasterizer_task
*task
,
227 const struct lp_rast_shader_inputs
*inputs
,
228 unsigned x
, unsigned y
)
230 const struct lp_scene
*scene
= task
->scene
;
231 const struct lp_rast_state
*state
= task
->state
;
232 struct lp_fragment_shader_variant
*variant
= state
->variant
;
233 uint8_t *color
[PIPE_MAX_COLOR_BUFS
];
234 unsigned stride
[PIPE_MAX_COLOR_BUFS
];
235 uint8_t *depth
= NULL
;
236 unsigned depth_stride
= 0;
240 for (i
= 0; i
< scene
->fb
.nr_cbufs
; i
++) {
241 if (scene
->fb
.cbufs
[i
]) {
242 stride
[i
] = scene
->cbufs
[i
].stride
;
243 color
[i
] = lp_rast_get_color_block_pointer(task
, i
, x
, y
,
252 if (scene
->zsbuf
.map
) {
253 depth
= lp_rast_get_depth_block_pointer(task
, x
, y
, inputs
->layer
);
254 depth_stride
= scene
->zsbuf
.stride
;
258 * The rasterizer may produce fragments outside our
259 * allocated 4x4 blocks hence need to filter them out here.
261 if ((x
% TILE_SIZE
) < task
->width
&& (y
% TILE_SIZE
) < task
->height
) {
262 /* not very accurate would need a popcount on the mask */
263 /* always count this not worth bothering? */
264 task
->ps_invocations
+= 1 * variant
->ps_inv_multiplier
;
266 /* Propagate non-interpolated raster state. */
267 task
->thread_data
.raster_state
.viewport_index
= inputs
->viewport_index
;
269 /* run shader on 4x4 block */
270 BEGIN_JIT_CALL(state
, task
);
271 variant
->jit_function
[RAST_WHOLE
]( &state
->jit_context
,
287 void lp_rast_triangle_1( struct lp_rasterizer_task
*,
288 const union lp_rast_cmd_arg
);
289 void lp_rast_triangle_2( struct lp_rasterizer_task
*,
290 const union lp_rast_cmd_arg
);
291 void lp_rast_triangle_3( struct lp_rasterizer_task
*,
292 const union lp_rast_cmd_arg
);
293 void lp_rast_triangle_4( struct lp_rasterizer_task
*,
294 const union lp_rast_cmd_arg
);
295 void lp_rast_triangle_5( struct lp_rasterizer_task
*,
296 const union lp_rast_cmd_arg
);
297 void lp_rast_triangle_6( struct lp_rasterizer_task
*,
298 const union lp_rast_cmd_arg
);
299 void lp_rast_triangle_7( struct lp_rasterizer_task
*,
300 const union lp_rast_cmd_arg
);
301 void lp_rast_triangle_8( struct lp_rasterizer_task
*,
302 const union lp_rast_cmd_arg
);
304 void lp_rast_triangle_3_4(struct lp_rasterizer_task
*,
305 const union lp_rast_cmd_arg
);
307 void lp_rast_triangle_3_16( struct lp_rasterizer_task
*,
308 const union lp_rast_cmd_arg
);
310 void lp_rast_triangle_4_16( struct lp_rasterizer_task
*,
311 const union lp_rast_cmd_arg
);
314 void lp_rast_triangle_32_1( struct lp_rasterizer_task
*,
315 const union lp_rast_cmd_arg
);
316 void lp_rast_triangle_32_2( struct lp_rasterizer_task
*,
317 const union lp_rast_cmd_arg
);
318 void lp_rast_triangle_32_3( struct lp_rasterizer_task
*,
319 const union lp_rast_cmd_arg
);
320 void lp_rast_triangle_32_4( struct lp_rasterizer_task
*,
321 const union lp_rast_cmd_arg
);
322 void lp_rast_triangle_32_5( struct lp_rasterizer_task
*,
323 const union lp_rast_cmd_arg
);
324 void lp_rast_triangle_32_6( struct lp_rasterizer_task
*,
325 const union lp_rast_cmd_arg
);
326 void lp_rast_triangle_32_7( struct lp_rasterizer_task
*,
327 const union lp_rast_cmd_arg
);
328 void lp_rast_triangle_32_8( struct lp_rasterizer_task
*,
329 const union lp_rast_cmd_arg
);
331 void lp_rast_triangle_32_3_4(struct lp_rasterizer_task
*,
332 const union lp_rast_cmd_arg
);
334 void lp_rast_triangle_32_3_16( struct lp_rasterizer_task
*,
335 const union lp_rast_cmd_arg
);
337 void lp_rast_triangle_32_4_16( struct lp_rasterizer_task
*,
338 const union lp_rast_cmd_arg
);
341 lp_rast_set_state(struct lp_rasterizer_task
*task
,
342 const union lp_rast_cmd_arg arg
);
345 lp_debug_bin( const struct cmd_bin
*bin
, int x
, int y
);