src/gallium/drivers/llvmpipe/lp_rast_priv.h

   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_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
  90    uint8_t *color_tiles[PIPE_MAX_COLOR_BUFS];
  91    uint8_t *depth_tile;
  92
  93    /** "back" pointer */
  94    struct lp_rasterizer *rast;
  95
  96    /** "my" index */
  97    unsigned thread_index;
  98
  99    /* occlude counter for visible pixels */
 100    struct lp_jit_thread_data thread_data;
 101    uint64_t query_start;
 102    struct llvmpipe_query *query[PIPE_QUERY_TYPES];
 103
 104    pipe_semaphore work_ready;
 105    pipe_semaphore work_done;
 106 };
 107
 108
 109 /**
 110  * This is the state required while rasterizing tiles.
 111  * Note that this contains per-thread information too.
 112  * The tile size is TILE_SIZE x TILE_SIZE pixels.
 113  */
 114 struct lp_rasterizer
 115 {
 116    boolean exit_flag;
 117    boolean no_rast;  /**< For debugging/profiling */
 118
 119    /** The incoming queue of scenes ready to rasterize */
 120    struct lp_scene_queue *full_scenes;
 121
 122    /** The scene currently being rasterized by the threads */
 123    struct lp_scene *curr_scene;
 124
 125    /** A task object for each rasterization thread */
 126    struct lp_rasterizer_task tasks[LP_MAX_THREADS];
 127
 128    unsigned num_threads;
 129    pipe_thread threads[LP_MAX_THREADS];
 130
 131    /** For synchronizing the rasterization threads */
 132    pipe_barrier barrier;
 133 };
 134
 135
 136 void
 137 lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
 138                          const struct lp_rast_shader_inputs *inputs,
 139                          unsigned x, unsigned y,
 140                          unsigned mask);
 141
 142
 143
 144 /**
 145  * Get pointer to the unswizzled color tile
 146  */
 147 static INLINE uint8_t *
 148 lp_rast_get_unswizzled_color_tile_pointer(struct lp_rasterizer_task *task,
 149                                           unsigned buf, enum lp_texture_usage usage)
 150 {
 151    const struct lp_scene *scene = task->scene;
 152    unsigned format_bytes;
 153
 154    assert(task->x < scene->tiles_x * TILE_SIZE);
 155    assert(task->y < scene->tiles_y * TILE_SIZE);
 156    assert(task->x % TILE_SIZE == 0);
 157    assert(task->y % TILE_SIZE == 0);
 158    assert(buf < scene->fb.nr_cbufs);
 159
 160    if (!task->color_tiles[buf]) {
 161       struct pipe_surface *cbuf = scene->fb.cbufs[buf];
 162       assert(cbuf);
 163
 164       format_bytes = util_format_get_blocksize(cbuf->format);
 165       task->color_tiles[buf] = scene->cbufs[buf].map + scene->cbufs[buf].stride * task->y + format_bytes * task->x;
 166    }
 167
 168    return task->color_tiles[buf];
 169 }
 170
 171
 172 /**
 173  * Get pointer to the unswizzled depth tile
 174  */
 175 static INLINE uint8_t *
 176 lp_rast_get_unswizzled_depth_tile_pointer(struct lp_rasterizer_task *task,
 177                                           enum lp_texture_usage usage)
 178 {
 179    const struct lp_scene *scene = task->scene;
 180    unsigned format_bytes;
 181
 182    assert(task->x < scene->tiles_x * TILE_SIZE);
 183    assert(task->y < scene->tiles_y * TILE_SIZE);
 184    assert(task->x % TILE_SIZE == 0);
 185    assert(task->y % TILE_SIZE == 0);
 186
 187    if (!task->depth_tile) {
 188       struct pipe_surface *dbuf = scene->fb.zsbuf;
 189       assert(dbuf);
 190
 191       format_bytes = util_format_get_blocksize(dbuf->format);
 192       task->depth_tile = scene->zsbuf.map + scene->zsbuf.stride * task->y + format_bytes * task->x;
 193    }
 194
 195    return task->depth_tile;
 196 }
 197
 198
 199 /**
 200  * Get the pointer to an unswizzled 4x4 color block (within an unswizzled 64x64 tile).
 201  * \param x, y location of 4x4 block in window coords
 202  */
 203 static INLINE uint8_t *
 204 lp_rast_get_unswizzled_color_block_pointer(struct lp_rasterizer_task *task,
 205                                            unsigned buf, unsigned x, unsigned y)
 206 {
 207    unsigned px, py, pixel_offset, format_bytes;
 208    uint8_t *color;
 209
 210    assert(x < task->scene->tiles_x * TILE_SIZE);
 211    assert(y < task->scene->tiles_y * TILE_SIZE);
 212    assert((x % TILE_VECTOR_WIDTH) == 0);
 213    assert((y % TILE_VECTOR_HEIGHT) == 0);
 214    assert(buf < task->scene->fb.nr_cbufs);
 215
 216    format_bytes = util_format_get_blocksize(task->scene->fb.cbufs[buf]->format);
 217
 218    color = lp_rast_get_unswizzled_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE);
 219    assert(color);
 220
 221    px = x % TILE_SIZE;
 222    py = y % TILE_SIZE;
 223    pixel_offset = px * format_bytes + py * task->scene->cbufs[buf].stride;
 224
 225    color = color + pixel_offset;
 226
 227    assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format)));
 228    return color;
 229 }
 230
 231
 232 /**
 233  * Get the pointer to an unswizzled 4x4 depth block (within an unswizzled 64x64 tile).
 234  * \param x, y location of 4x4 block in window coords
 235  */
 236 static INLINE uint8_t *
 237 lp_rast_get_unswizzled_depth_block_pointer(struct lp_rasterizer_task *task,
 238                                            unsigned x, unsigned y)
 239 {
 240    unsigned px, py, pixel_offset, format_bytes;
 241    uint8_t *depth;
 242
 243    assert(x < task->scene->tiles_x * TILE_SIZE);
 244    assert(y < task->scene->tiles_y * TILE_SIZE);
 245    assert((x % TILE_VECTOR_WIDTH) == 0);
 246    assert((y % TILE_VECTOR_HEIGHT) == 0);
 247
 248    format_bytes = util_format_get_blocksize(task->scene->fb.zsbuf->format);
 249
 250    depth = lp_rast_get_unswizzled_depth_tile_pointer(task, LP_TEX_USAGE_READ_WRITE);
 251    assert(depth);
 252
 253    px = x % TILE_SIZE;
 254    py = y % TILE_SIZE;
 255    pixel_offset = px * format_bytes + py * task->scene->zsbuf.stride;
 256
 257    depth = depth + pixel_offset;
 258
 259    assert(lp_check_alignment(depth, llvmpipe_get_format_alignment(task->scene->fb.zsbuf->format)));
 260    return depth;
 261 }
 262
 263
 264
 265 /**
 266  * Shade all pixels in a 4x4 block.  The fragment code omits the
 267  * triangle in/out tests.
 268  * \param x, y location of 4x4 block in window coords
 269  */
 270 static INLINE void
 271 lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
 272                          const struct lp_rast_shader_inputs *inputs,
 273                          unsigned x, unsigned y )
 274 {
 275    const struct lp_scene *scene = task->scene;
 276    const struct lp_rast_state *state = task->state;
 277    struct lp_fragment_shader_variant *variant = state->variant;
 278    uint8_t *color[PIPE_MAX_COLOR_BUFS];
 279    unsigned stride[PIPE_MAX_COLOR_BUFS];
 280    void *depth = NULL;
 281    unsigned depth_stride = 0;
 282    unsigned i;
 283
 284    /* color buffer */
 285    for (i = 0; i < scene->fb.nr_cbufs; i++) {
 286       stride[i] = scene->cbufs[i].stride;
 287
 288       color[i] = lp_rast_get_unswizzled_color_block_pointer(task, i, x, y);
 289    }
 290
 291    if (scene->zsbuf.map) {
 292       depth = lp_rast_get_unswizzled_depth_block_pointer(task, x, y);
 293       depth_stride = scene->zsbuf.stride;
 294    }
 295
 296    /* run shader on 4x4 block */
 297    BEGIN_JIT_CALL(state, task);
 298    variant->jit_function[RAST_WHOLE]( &state->jit_context,
 299                                       x, y,
 300                                       inputs->frontfacing,
 301                                       GET_A0(inputs),
 302                                       GET_DADX(inputs),
 303                                       GET_DADY(inputs),
 304                                       color,
 305                                       depth,
 306                                       0xffff,
 307                                       &task->thread_data,
 308                                       stride,
 309                                       depth_stride);
 310    END_JIT_CALL();
 311 }
 312
 313 void lp_rast_triangle_1( struct lp_rasterizer_task *,
 314                          const union lp_rast_cmd_arg );
 315 void lp_rast_triangle_2( struct lp_rasterizer_task *,
 316                          const union lp_rast_cmd_arg );
 317 void lp_rast_triangle_3( struct lp_rasterizer_task *,
 318                          const union lp_rast_cmd_arg );
 319 void lp_rast_triangle_4( struct lp_rasterizer_task *,
 320                          const union lp_rast_cmd_arg );
 321 void lp_rast_triangle_5( struct lp_rasterizer_task *,
 322                          const union lp_rast_cmd_arg );
 323 void lp_rast_triangle_6( struct lp_rasterizer_task *,
 324                          const union lp_rast_cmd_arg );
 325 void lp_rast_triangle_7( struct lp_rasterizer_task *,
 326                          const union lp_rast_cmd_arg );
 327 void lp_rast_triangle_8( struct lp_rasterizer_task *,
 328                          const union lp_rast_cmd_arg );
 329
 330 void lp_rast_triangle_3_4(struct lp_rasterizer_task *,
 331                           const union lp_rast_cmd_arg );
 332
 333 void lp_rast_triangle_3_16( struct lp_rasterizer_task *,
 334                             const union lp_rast_cmd_arg );
 335
 336 void lp_rast_triangle_4_16( struct lp_rasterizer_task *,
 337                             const union lp_rast_cmd_arg );
 338
 339 void
 340 lp_rast_set_state(struct lp_rasterizer_task *task,
 341                   const union lp_rast_cmd_arg arg);
 342
 343 void
 344 lp_debug_bin( const struct cmd_bin *bin );
 345
 346 #endif