lp_rast_priv.h revision 9f6e8e1d6b8696a3ee96cba01b2466ba7a1a8ef6 
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 "os/os_thread.h"
32#include "util/u_format.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_tile_soa.h"
40#include "lp_limits.h"
41
42
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.
45 */
46#ifdef DEBUG
47
48extern int jit_line;
49extern const struct lp_rast_state *jit_state;
50
51#define BEGIN_JIT_CALL(state) \
52   do { \
53      jit_line = __LINE__; \
54      jit_state = state; \
55   } while (0)
56
57#define END_JIT_CALL() \
58   do { \
59      jit_line = 0; \
60      jit_state = NULL; \
61   } while (0)
62
63#else
64
65#define BEGIN_JIT_CALL(X)
66#define END_JIT_CALL()
67
68#endif
69
70
71struct lp_rasterizer;
72struct cmd_bin;
73
74/**
75 * Per-thread rasterization state
76 */
77struct lp_rasterizer_task
78{
79   const struct cmd_bin *bin;
80
81   struct lp_scene *scene;
82   unsigned x, y;          /**< Pos of this tile in framebuffer, in pixels */
83
84   uint8_t *color_tiles[PIPE_MAX_COLOR_BUFS];
85   uint8_t *depth_tile;
86
87   /** "back" pointer */
88   struct lp_rasterizer *rast;
89
90   /** "my" index */
91   unsigned thread_index;
92
93   /* occlude counter for visiable pixels */
94   uint32_t vis_counter;
95   struct llvmpipe_query *query;
96
97   pipe_semaphore work_ready;
98   pipe_semaphore work_done;
99};
100
101
102/**
103 * This is the state required while rasterizing tiles.
104 * Note that this contains per-thread information too.
105 * The tile size is TILE_SIZE x TILE_SIZE pixels.
106 */
107struct lp_rasterizer
108{
109   boolean exit_flag;
110
111   /** The incoming queue of scenes ready to rasterize */
112   struct lp_scene_queue *full_scenes;
113
114   /** The scene currently being rasterized by the threads */
115   struct lp_scene *curr_scene;
116
117   /** A task object for each rasterization thread */
118   struct lp_rasterizer_task tasks[LP_MAX_THREADS];
119
120   unsigned num_threads;
121   pipe_thread threads[LP_MAX_THREADS];
122
123   /** For synchronizing the rasterization threads */
124   pipe_barrier barrier;
125};
126
127
128void
129lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
130                         const struct lp_rast_shader_inputs *inputs,
131                         unsigned x, unsigned y,
132                         unsigned mask);
133
134
135
136/**
137 * Get the pointer to a 4x4 depth/stencil block.
138 * We'll map the z/stencil buffer on demand here.
139 * Note that this may be called even when there's no z/stencil buffer - return
140 * NULL in that case.
141 * \param x, y location of 4x4 block in window coords
142 */
143static INLINE void *
144lp_rast_get_depth_block_pointer(struct lp_rasterizer_task *task,
145                                unsigned x, unsigned y)
146{
147   const struct lp_scene *scene = task->scene;
148   void *depth;
149
150   assert((x % TILE_VECTOR_WIDTH) == 0);
151   assert((y % TILE_VECTOR_HEIGHT) == 0);
152
153   if (!scene->zsbuf.map) {
154      /* Either out of memory or no zsbuf.  Can't tell without access
155       * to the state.  Just use dummy tile memory, but don't print
156       * the oom warning as this most likely because there is no
157       * zsbuf.
158       */
159      return lp_dummy_tile;
160   }
161
162   depth = (scene->zsbuf.map +
163            scene->zsbuf.stride * y +
164            scene->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT);
165
166   assert(lp_check_alignment(depth, 16));
167   return depth;
168}
169
170
171/**
172 * Get pointer to the swizzled color tile
173 */
174static INLINE uint8_t *
175lp_rast_get_color_tile_pointer(struct lp_rasterizer_task *task,
176                               unsigned buf, enum lp_texture_usage usage)
177{
178   const struct lp_scene *scene = task->scene;
179
180   assert(task->x % TILE_SIZE == 0);
181   assert(task->y % TILE_SIZE == 0);
182   assert(buf < scene->fb.nr_cbufs);
183
184   if (!task->color_tiles[buf]) {
185      struct pipe_surface *cbuf = scene->fb.cbufs[buf];
186      struct llvmpipe_resource *lpt;
187      assert(cbuf);
188      lpt = llvmpipe_resource(cbuf->texture);
189      task->color_tiles[buf] = lp_swizzled_cbuf[task->thread_index][buf];
190
191      if (usage != LP_TEX_USAGE_WRITE_ALL) {
192         llvmpipe_swizzle_cbuf_tile(lpt,
193                                    cbuf->face + cbuf->zslice,
194                                    cbuf->level,
195                                    task->x, task->y,
196                                    task->color_tiles[buf]);
197      }
198   }
199
200   return task->color_tiles[buf];
201}
202
203
204/**
205 * Get the pointer to a 4x4 color block (within a 64x64 tile).
206 * We'll map the color buffer on demand here.
207 * Note that this may be called even when there's no color buffers - return
208 * NULL in that case.
209 * \param x, y location of 4x4 block in window coords
210 */
211static INLINE uint8_t *
212lp_rast_get_color_block_pointer(struct lp_rasterizer_task *task,
213                                unsigned buf, unsigned x, unsigned y)
214{
215   unsigned px, py, pixel_offset;
216   uint8_t *color;
217
218   assert((x % TILE_VECTOR_WIDTH) == 0);
219   assert((y % TILE_VECTOR_HEIGHT) == 0);
220
221   color = lp_rast_get_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE);
222   assert(color);
223
224   px = x % TILE_SIZE;
225   py = y % TILE_SIZE;
226   pixel_offset = tile_pixel_offset(px, py, 0);
227
228   color = color + pixel_offset;
229
230   assert(lp_check_alignment(color, 16));
231   return color;
232}
233
234
235
236/**
237 * Shade all pixels in a 4x4 block.  The fragment code omits the
238 * triangle in/out tests.
239 * \param x, y location of 4x4 block in window coords
240 */
241static INLINE void
242lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
243                         const struct lp_rast_shader_inputs *inputs,
244                         unsigned x, unsigned y )
245{
246   const struct lp_scene *scene = task->scene;
247   const struct lp_rast_state *state = inputs->state;
248   struct lp_fragment_shader_variant *variant = state->variant;
249   uint8_t *color[PIPE_MAX_COLOR_BUFS];
250   void *depth;
251   unsigned i;
252
253   /* color buffer */
254   for (i = 0; i < scene->fb.nr_cbufs; i++)
255      color[i] = lp_rast_get_color_block_pointer(task, i, x, y);
256
257   depth = lp_rast_get_depth_block_pointer(task, x, y);
258
259   /* run shader on 4x4 block */
260   BEGIN_JIT_CALL(state);
261   variant->jit_function[RAST_WHOLE]( &state->jit_context,
262                                      x, y,
263                                      inputs->facing,
264                                      inputs->a0,
265                                      inputs->dadx,
266                                      inputs->dady,
267                                      color,
268                                      depth,
269                                      0xffff,
270                                      &task->vis_counter );
271   END_JIT_CALL();
272}
273
274void lp_rast_triangle_1( struct lp_rasterizer_task *,
275                         const union lp_rast_cmd_arg );
276void lp_rast_triangle_2( struct lp_rasterizer_task *,
277                         const union lp_rast_cmd_arg );
278void lp_rast_triangle_3( struct lp_rasterizer_task *,
279                         const union lp_rast_cmd_arg );
280void lp_rast_triangle_4( struct lp_rasterizer_task *,
281                         const union lp_rast_cmd_arg );
282void lp_rast_triangle_5( struct lp_rasterizer_task *,
283                         const union lp_rast_cmd_arg );
284void lp_rast_triangle_6( struct lp_rasterizer_task *,
285                         const union lp_rast_cmd_arg );
286void lp_rast_triangle_7( struct lp_rasterizer_task *,
287                         const union lp_rast_cmd_arg );
288void lp_rast_triangle_8( struct lp_rasterizer_task *,
289                         const union lp_rast_cmd_arg );
290
291void lp_rast_triangle_3_16( struct lp_rasterizer_task *,
292                            const union lp_rast_cmd_arg );
293
294
295#endif
296