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_tile_image.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
  48 struct lp_rasterizer_task;
  49 extern int jit_line;
  50 extern const struct lp_rast_state *jit_state;
  51 extern const struct lp_rasterizer_task *jit_task;
  52
  53 #define BEGIN_JIT_CALL(state, task)                  \
  54    do { \
  55       jit_line = __LINE__; \
  56       jit_state = state; \
  57       jit_task = task; \
  58    } while (0)
  59
  60 #define END_JIT_CALL() \
  61    do { \
  62       jit_line = 0; \
  63       jit_state = NULL; \
  64    } while (0)
  65
  66 #else
  67
  68 #define BEGIN_JIT_CALL(X, Y)
  69 #define END_JIT_CALL()
  70
  71 #endif
  72
  73
  74 struct lp_rasterizer;
  75 struct cmd_bin;
  76
  77 /**
  78  * Per-thread rasterization state
  79  */
  80 struct lp_rasterizer_task
  81 {
  82    const struct cmd_bin *bin;
  83    const struct lp_rast_state *state;
  84
  85    struct lp_scene *scene;
  86    unsigned x, y;          /**< Pos of this tile in framebuffer, in pixels */
  87
  88    uint8_t *color_tiles[PIPE_MAX_COLOR_BUFS];
  89    uint8_t *depth_tile;
  90
  91    /** "back" pointer */
  92    struct lp_rasterizer *rast;
  93
  94    /** "my" index */
  95    unsigned thread_index;
  96
  97    /* occlude counter for visible pixels */
  98    struct lp_jit_thread_data thread_data;
  99    uint64_t query_start;
 100    struct llvmpipe_query *query[PIPE_QUERY_TYPES];
 101
 102    pipe_semaphore work_ready;
 103    pipe_semaphore work_done;
 104 };
 105
 106
 107 /**
 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.
 111  */
 112 struct lp_rasterizer
 113 {
 114    boolean exit_flag;
 115    boolean no_rast;  /**< For debugging/profiling */
 116
 117    /** The incoming queue of scenes ready to rasterize */
 118    struct lp_scene_queue *full_scenes;
 119
 120    /** The scene currently being rasterized by the threads */
 121    struct lp_scene *curr_scene;
 122
 123    /** A task object for each rasterization thread */
 124    struct lp_rasterizer_task tasks[LP_MAX_THREADS];
 125
 126    unsigned num_threads;
 127    pipe_thread threads[LP_MAX_THREADS];
 128
 129    /** For synchronizing the rasterization threads */
 130    pipe_barrier barrier;
 131 };
 132
 133
 134 void
 135 lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
 136                          const struct lp_rast_shader_inputs *inputs,
 137                          unsigned x, unsigned y,
 138                          unsigned mask);
 139
 140
 141
 142 /**
 143  * Get pointer to the unswizzled color tile
 144  */
 145 static INLINE uint8_t *
 146 lp_rast_get_unswizzled_color_tile_pointer(struct lp_rasterizer_task *task,
 147                                           unsigned buf, enum lp_texture_usage usage)
 148 {
 149    const struct lp_scene *scene = task->scene;
 150    unsigned format_bytes;
 151
 152    assert(task->x < scene->tiles_x * TILE_SIZE);
 153    assert(task->y < scene->tiles_y * TILE_SIZE);
 154    assert(task->x % TILE_SIZE == 0);
 155    assert(task->y % TILE_SIZE == 0);
 156    assert(buf < scene->fb.nr_cbufs);
 157
 158    if (!task->color_tiles[buf]) {
 159       struct pipe_surface *cbuf = scene->fb.cbufs[buf];
 160       assert(cbuf);
 161
 162       format_bytes = util_format_get_blocksize(cbuf->format);
 163       task->color_tiles[buf] = scene->cbufs[buf].map + scene->cbufs[buf].stride * task->y + format_bytes * task->x;
 164    }
 165
 166    return task->color_tiles[buf];
 167 }
 168
 169
 170 /**
 171  * Get pointer to the unswizzled depth tile
 172  */
 173 static INLINE uint8_t *
 174 lp_rast_get_unswizzled_depth_tile_pointer(struct lp_rasterizer_task *task,
 175                                           enum lp_texture_usage usage)
 176 {
 177    const struct lp_scene *scene = task->scene;
 178    unsigned format_bytes;
 179
 180    assert(task->x < scene->tiles_x * TILE_SIZE);
 181    assert(task->y < scene->tiles_y * TILE_SIZE);
 182    assert(task->x % TILE_SIZE == 0);
 183    assert(task->y % TILE_SIZE == 0);
 184
 185    if (!task->depth_tile) {
 186       struct pipe_surface *dbuf = scene->fb.zsbuf;
 187       assert(dbuf);
 188
 189       format_bytes = util_format_get_blocksize(dbuf->format);
 190       task->depth_tile = scene->zsbuf.map + scene->zsbuf.stride * task->y + format_bytes * task->x;
 191    }
 192
 193    return task->depth_tile;
 194 }
 195
 196
 197 /**
 198  * Get the pointer to an unswizzled 4x4 color block (within an unswizzled 64x64 tile).
 199  * \param x, y location of 4x4 block in window coords
 200  */
 201 static INLINE uint8_t *
 202 lp_rast_get_unswizzled_color_block_pointer(struct lp_rasterizer_task *task,
 203                                            unsigned buf, unsigned x, unsigned y)
 204 {
 205    unsigned px, py, pixel_offset, format_bytes;
 206    uint8_t *color;
 207
 208    assert(x < task->scene->tiles_x * TILE_SIZE);
 209    assert(y < task->scene->tiles_y * TILE_SIZE);
 210    assert((x % TILE_VECTOR_WIDTH) == 0);
 211    assert((y % TILE_VECTOR_HEIGHT) == 0);
 212    assert(buf < task->scene->fb.nr_cbufs);
 213
 214    format_bytes = util_format_get_blocksize(task->scene->fb.cbufs[buf]->format);
 215
 216    color = lp_rast_get_unswizzled_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE);
 217    assert(color);
 218
 219    px = x % TILE_SIZE;
 220    py = y % TILE_SIZE;
 221    pixel_offset = px * format_bytes + py * task->scene->cbufs[buf].stride;
 222
 223    color = color + pixel_offset;
 224
 225    assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format)));
 226    return color;
 227 }
 228
 229
 230 /**
 231  * Get the pointer to an unswizzled 4x4 depth block (within an unswizzled 64x64 tile).
 232  * \param x, y location of 4x4 block in window coords
 233  */
 234 static INLINE uint8_t *
 235 lp_rast_get_unswizzled_depth_block_pointer(struct lp_rasterizer_task *task,
 236                                            unsigned x, unsigned y)
 237 {
 238    unsigned px, py, pixel_offset, format_bytes;
 239    uint8_t *depth;
 240
 241    assert(x < task->scene->tiles_x * TILE_SIZE);
 242    assert(y < task->scene->tiles_y * TILE_SIZE);
 243    assert((x % TILE_VECTOR_WIDTH) == 0);
 244    assert((y % TILE_VECTOR_HEIGHT) == 0);
 245
 246    format_bytes = util_format_get_blocksize(task->scene->fb.zsbuf->format);
 247
 248    depth = lp_rast_get_unswizzled_depth_tile_pointer(task, LP_TEX_USAGE_READ_WRITE);
 249    assert(depth);
 250
 251    px = x % TILE_SIZE;
 252    py = y % TILE_SIZE;
 253    pixel_offset = px * format_bytes + py * task->scene->zsbuf.stride;
 254
 255    depth = depth + pixel_offset;
 256
 257    assert(lp_check_alignment(depth, llvmpipe_get_format_alignment(task->scene->fb.zsbuf->format)));
 258    return depth;
 259 }
 260
 261
 262
 263 /**
 264  * Shade all pixels in a 4x4 block.  The fragment code omits the
 265  * triangle in/out tests.
 266  * \param x, y location of 4x4 block in window coords
 267  */
 268 static INLINE void
 269 lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
 270                          const struct lp_rast_shader_inputs *inputs,
 271                          unsigned x, unsigned y )
 272 {
 273    const struct lp_scene *scene = task->scene;
 274    const struct lp_rast_state *state = task->state;
 275    struct lp_fragment_shader_variant *variant = state->variant;
 276    uint8_t *color[PIPE_MAX_COLOR_BUFS];
 277    unsigned stride[PIPE_MAX_COLOR_BUFS];
 278    void *depth = NULL;
 279    unsigned depth_stride = 0;
 280    unsigned i;
 281
 282    /* color buffer */
 283    for (i = 0; i < scene->fb.nr_cbufs; i++) {
 284       stride[i] = scene->cbufs[i].stride;
 285
 286       color[i] = lp_rast_get_unswizzled_color_block_pointer(task, i, x, y);
 287    }
 288
 289    if (scene->zsbuf.map) {
 290       depth = lp_rast_get_unswizzled_depth_block_pointer(task, x, y);
 291       depth_stride = scene->zsbuf.stride;
 292    }
 293
 294    /* run shader on 4x4 block */
 295    BEGIN_JIT_CALL(state, task);
 296    variant->jit_function[RAST_WHOLE]( &state->jit_context,
 297                                       x, y,
 298                                       inputs->frontfacing,
 299                                       GET_A0(inputs),
 300                                       GET_DADX(inputs),
 301                                       GET_DADY(inputs),
 302                                       color,
 303                                       depth,
 304                                       0xffff,
 305                                       &task->thread_data,
 306                                       stride,
 307                                       depth_stride);
 308    END_JIT_CALL();
 309 }
 310
 311 void lp_rast_triangle_1( struct lp_rasterizer_task *,
 312                          const union lp_rast_cmd_arg );
 313 void lp_rast_triangle_2( struct lp_rasterizer_task *,
 314                          const union lp_rast_cmd_arg );
 315 void lp_rast_triangle_3( struct lp_rasterizer_task *,
 316                          const union lp_rast_cmd_arg );
 317 void lp_rast_triangle_4( struct lp_rasterizer_task *,
 318                          const union lp_rast_cmd_arg );
 319 void lp_rast_triangle_5( struct lp_rasterizer_task *,
 320                          const union lp_rast_cmd_arg );
 321 void lp_rast_triangle_6( struct lp_rasterizer_task *,
 322                          const union lp_rast_cmd_arg );
 323 void lp_rast_triangle_7( struct lp_rasterizer_task *,
 324                          const union lp_rast_cmd_arg );
 325 void lp_rast_triangle_8( struct lp_rasterizer_task *,
 326                          const union lp_rast_cmd_arg );
 327
 328 void lp_rast_triangle_3_4(struct lp_rasterizer_task *,
 329                           const union lp_rast_cmd_arg );
 330
 331 void lp_rast_triangle_3_16( struct lp_rasterizer_task *,
 332                             const union lp_rast_cmd_arg );
 333
 334 void lp_rast_triangle_4_16( struct lp_rasterizer_task *,
 335                             const union lp_rast_cmd_arg );
 336
 337 void
 338 lp_rast_set_state(struct lp_rasterizer_task *task,
 339                   const union lp_rast_cmd_arg arg);
 340
 341 void
 342 lp_debug_bin( const struct cmd_bin *bin );
 343
 344 #endif