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 visiable pixels */
  98    uint32_t vis_counter;
  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 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
 146  * NULL in that case.
 147  * \param x, y location of 4x4 block in window coords
 148  */
 149 static INLINE void *
 150 lp_rast_get_depth_block_pointer(struct lp_rasterizer_task *task,
 151                                 unsigned x, unsigned y)
 152 {
 153    const struct lp_scene *scene = task->scene;
 154    void *depth;
 155
 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);
 160
 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
 165        * zsbuf.
 166        */
 167       return lp_dummy_tile;
 168    }
 169
 170    depth = (scene->zsbuf.map +
 171             scene->zsbuf.stride * y +
 172             scene->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT);
 173
 174    assert(lp_check_alignment(depth, 16));
 175    return depth;
 176 }
 177
 178
 179 /**
 180  * Get pointer to the unswizzled color tile
 181  */
 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)
 185 {
 186    const struct lp_scene *scene = task->scene;
 187    unsigned format_bytes;
 188
 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);
 194
 195    if (!task->color_tiles[buf]) {
 196       struct pipe_surface *cbuf = scene->fb.cbufs[buf];
 197       assert(cbuf);
 198
 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;
 201    }
 202
 203    return task->color_tiles[buf];
 204 }
 205
 206
 207 /**
 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
 210  */
 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)
 214 {
 215    unsigned px, py, pixel_offset, format_bytes;
 216    uint8_t *color;
 217
 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);
 223
 224    format_bytes = util_format_description(task->scene->fb.cbufs[buf]->format)->block.bits / 8;
 225
 226    color = lp_rast_get_unswizzled_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE);
 227    assert(color);
 228
 229    px = x % TILE_SIZE;
 230    py = y % TILE_SIZE;
 231    pixel_offset = px * format_bytes + py * task->scene->cbufs[buf].stride;
 232
 233    color = color + pixel_offset;
 234
 235    assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format)));
 236    return color;
 237 }
 238
 239
 240
 241 /**
 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
 245  */
 246 static INLINE void
 247 lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
 248                          const struct lp_rast_shader_inputs *inputs,
 249                          unsigned x, unsigned y )
 250 {
 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];
 256    void *depth;
 257    unsigned i;
 258
 259    /* color buffer */
 260    for (i = 0; i < scene->fb.nr_cbufs; i++) {
 261       stride[i] = scene->cbufs[i].stride;
 262
 263       color[i] = lp_rast_get_unswizzled_color_block_pointer(task, i, x, y);
 264    }
 265
 266    depth = lp_rast_get_depth_block_pointer(task, x, y);
 267
 268    /* run shader on 4x4 block */
 269    BEGIN_JIT_CALL(state, task);
 270    variant->jit_function[RAST_WHOLE]( &state->jit_context,
 271                                       x, y,
 272                                       inputs->frontfacing,
 273                                       GET_A0(inputs),
 274                                       GET_DADX(inputs),
 275                                       GET_DADY(inputs),
 276                                       color,
 277                                       depth,
 278                                       0xffff,
 279                                       &task->vis_counter,
 280                                       stride );
 281    END_JIT_CALL();
 282 }
 283
 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 );
 300
 301 void lp_rast_triangle_3_4(struct lp_rasterizer_task *,
 302                           const union lp_rast_cmd_arg );
 303
 304 void lp_rast_triangle_3_16( struct lp_rasterizer_task *,
 305                             const union lp_rast_cmd_arg );
 306
 307 void lp_rast_triangle_4_16( struct lp_rasterizer_task *,
 308                             const union lp_rast_cmd_arg );
 309
 310 void
 311 lp_rast_set_state(struct lp_rasterizer_task *task,
 312                   const union lp_rast_cmd_arg arg);
 313
 314 void
 315 lp_debug_bin( const struct cmd_bin *bin );
 316
 317 #endif