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_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
  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    struct llvmpipe_query *query;
 100
 101    pipe_semaphore work_ready;
 102    pipe_semaphore work_done;
 103 };
 104
 105
 106 /**
 107  * This is the state required while rasterizing tiles.
 108  * Note that this contains per-thread information too.
 109  * The tile size is TILE_SIZE x TILE_SIZE pixels.
 110  */
 111 struct lp_rasterizer
 112 {
 113    boolean exit_flag;
 114    boolean no_rast;  /**< For debugging/profiling */
 115
 116    /** The incoming queue of scenes ready to rasterize */
 117    struct lp_scene_queue *full_scenes;
 118
 119    /** The scene currently being rasterized by the threads */
 120    struct lp_scene *curr_scene;
 121
 122    /** A task object for each rasterization thread */
 123    struct lp_rasterizer_task tasks[LP_MAX_THREADS];
 124
 125    unsigned num_threads;
 126    pipe_thread threads[LP_MAX_THREADS];
 127
 128    /** For synchronizing the rasterization threads */
 129    pipe_barrier barrier;
 130 };
 131
 132
 133 void
 134 lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
 135                          const struct lp_rast_shader_inputs *inputs,
 136                          unsigned x, unsigned y,
 137                          unsigned mask);
 138
 139
 140
 141 /**
 142  * Get the pointer to a 4x4 depth/stencil block.
 143  * We'll map the z/stencil buffer on demand here.
 144  * Note that this may be called even when there's no z/stencil buffer - return
 145  * NULL in that case.
 146  * \param x, y location of 4x4 block in window coords
 147  */
 148 static INLINE void *
 149 lp_rast_get_depth_block_pointer(struct lp_rasterizer_task *task,
 150                                 unsigned x, unsigned y)
 151 {
 152    const struct lp_scene *scene = task->scene;
 153    void *depth;
 154
 155    assert(x < scene->tiles_x * TILE_SIZE);
 156    assert(y < scene->tiles_y * TILE_SIZE);
 157    assert((x % TILE_VECTOR_WIDTH) == 0);
 158    assert((y % TILE_VECTOR_HEIGHT) == 0);
 159
 160    if (!scene->zsbuf.map) {
 161       /* Either out of memory or no zsbuf.  Can't tell without access
 162        * to the state.  Just use dummy tile memory, but don't print
 163        * the oom warning as this most likely because there is no
 164        * zsbuf.
 165        */
 166       return lp_dummy_tile;
 167    }
 168
 169    depth = (scene->zsbuf.map +
 170             scene->zsbuf.stride * y +
 171             scene->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT);
 172
 173    assert(lp_check_alignment(depth, 16));
 174    return depth;
 175 }
 176
 177
 178 /**
 179  * Get pointer to the swizzled color tile
 180  */
 181 static INLINE uint8_t *
 182 lp_rast_get_color_tile_pointer(struct lp_rasterizer_task *task,
 183                                unsigned buf, enum lp_texture_usage usage)
 184 {
 185    const struct lp_scene *scene = task->scene;
 186
 187    assert(task->x < scene->tiles_x * TILE_SIZE);
 188    assert(task->y < scene->tiles_y * TILE_SIZE);
 189    assert(task->x % TILE_SIZE == 0);
 190    assert(task->y % TILE_SIZE == 0);
 191    assert(buf < scene->fb.nr_cbufs);
 192    assert(scene->cbufs[buf].unswizzled == 0);
 193
 194    if (!task->color_tiles[buf]) {
 195       struct pipe_surface *cbuf = scene->fb.cbufs[buf];
 196       struct llvmpipe_resource *lpt;
 197       assert(cbuf);
 198       lpt = llvmpipe_resource(cbuf->texture);
 199       task->color_tiles[buf] = lp_swizzled_cbuf[task->thread_index][buf];
 200
 201       if (usage != LP_TEX_USAGE_WRITE_ALL) {
 202          llvmpipe_swizzle_cbuf_tile(lpt,
 203                                     cbuf->u.tex.first_layer,
 204                                     cbuf->u.tex.level,
 205                                     task->x, task->y,
 206                                     task->color_tiles[buf]);
 207       }
 208    }
 209
 210    return task->color_tiles[buf];
 211 }
 212
 213
 214 /**
 215  * Get pointer to the unswizzled color tile
 216  */
 217 static INLINE uint8_t *
 218 lp_rast_get_unswizzled_color_tile_pointer(struct lp_rasterizer_task *task,
 219                                           unsigned buf, enum lp_texture_usage usage)
 220 {
 221    const struct lp_scene *scene = task->scene;
 222    unsigned format_bytes;
 223
 224    assert(task->x < scene->tiles_x * TILE_SIZE);
 225    assert(task->y < scene->tiles_y * TILE_SIZE);
 226    assert(task->x % TILE_SIZE == 0);
 227    assert(task->y % TILE_SIZE == 0);
 228    assert(buf < scene->fb.nr_cbufs);
 229    assert(scene->cbufs[buf].unswizzled);
 230
 231    if (!task->color_tiles[buf]) {
 232       struct pipe_surface *cbuf = scene->fb.cbufs[buf];
 233       assert(cbuf);
 234
 235       format_bytes = util_format_description(cbuf->format)->block.bits / 8;
 236       task->color_tiles[buf] = scene->cbufs[buf].map + scene->cbufs[buf].stride * task->y + format_bytes * task->x;
 237    }
 238
 239    return task->color_tiles[buf];
 240 }
 241
 242
 243 /**
 244  * Get the pointer to a 4x4 color block (within a 64x64 tile).
 245  * We'll map the color buffer on demand here.
 246  * Note that this may be called even when there's no color buffers - return
 247  * NULL in that case.
 248  * \param x, y location of 4x4 block in window coords
 249  */
 250 static INLINE uint8_t *
 251 lp_rast_get_color_block_pointer(struct lp_rasterizer_task *task,
 252                                 unsigned buf, unsigned x, unsigned y)
 253 {
 254    unsigned px, py, pixel_offset;
 255    uint8_t *color;
 256
 257    assert(x < task->scene->tiles_x * TILE_SIZE);
 258    assert(y < task->scene->tiles_y * TILE_SIZE);
 259    assert((x % TILE_VECTOR_WIDTH) == 0);
 260    assert((y % TILE_VECTOR_HEIGHT) == 0);
 261    assert(buf < task->scene->fb.nr_cbufs);
 262    assert(task->scene->cbufs[buf].unswizzled == 0);
 263
 264    color = lp_rast_get_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE);
 265    assert(color);
 266
 267    px = x % TILE_SIZE;
 268    py = y % TILE_SIZE;
 269    pixel_offset = tile_pixel_offset(px, py, 0);
 270
 271    color = color + pixel_offset;
 272
 273    assert(lp_check_alignment(color, 16));
 274    return color;
 275 }
 276
 277
 278 /**
 279  * Get the pointer to an unswizzled 4x4 color block (within an unswizzled 64x64 tile).
 280  * \param x, y location of 4x4 block in window coords
 281  */
 282 static INLINE uint8_t *
 283 lp_rast_get_unswizzled_color_block_pointer(struct lp_rasterizer_task *task,
 284                                            unsigned buf, unsigned x, unsigned y)
 285 {
 286    unsigned px, py, pixel_offset, format_bytes;
 287    uint8_t *color;
 288
 289    assert(x < task->scene->tiles_x * TILE_SIZE);
 290    assert(y < task->scene->tiles_y * TILE_SIZE);
 291    assert((x % TILE_VECTOR_WIDTH) == 0);
 292    assert((y % TILE_VECTOR_HEIGHT) == 0);
 293    assert(buf < task->scene->fb.nr_cbufs);
 294    assert(task->scene->cbufs[buf].unswizzled);
 295
 296    format_bytes = util_format_description(task->scene->fb.cbufs[buf]->format)->block.bits / 8;
 297
 298    color = lp_rast_get_unswizzled_color_tile_pointer(task, buf, LP_TEX_USAGE_READ_WRITE);
 299    assert(color);
 300
 301    px = x % TILE_SIZE;
 302    py = y % TILE_SIZE;
 303    pixel_offset = px * format_bytes + py * task->scene->cbufs[buf].stride;
 304
 305    color = color + pixel_offset;
 306
 307    assert(lp_check_alignment(color, llvmpipe_get_format_alignment(task->scene->fb.cbufs[buf]->format)));
 308    return color;
 309 }
 310
 311
 312
 313 /**
 314  * Shade all pixels in a 4x4 block.  The fragment code omits the
 315  * triangle in/out tests.
 316  * \param x, y location of 4x4 block in window coords
 317  */
 318 static INLINE void
 319 lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
 320                          const struct lp_rast_shader_inputs *inputs,
 321                          unsigned x, unsigned y )
 322 {
 323    const struct lp_scene *scene = task->scene;
 324    const struct lp_rast_state *state = task->state;
 325    struct lp_fragment_shader_variant *variant = state->variant;
 326    uint8_t *color[PIPE_MAX_COLOR_BUFS];
 327    unsigned stride[PIPE_MAX_COLOR_BUFS];
 328    void *depth;
 329    unsigned i;
 330
 331    /* color buffer */
 332    for (i = 0; i < scene->fb.nr_cbufs; i++) {
 333       stride[i] = scene->cbufs[i].stride;
 334
 335       if (scene->cbufs[i].unswizzled) {
 336          color[i] = lp_rast_get_unswizzled_color_block_pointer(task, i, x, y);
 337       } else {
 338          color[i] = lp_rast_get_color_block_pointer(task, i, x, y);
 339       }
 340    }
 341
 342    depth = lp_rast_get_depth_block_pointer(task, x, y);
 343
 344    /* run shader on 4x4 block */
 345    BEGIN_JIT_CALL(state, task);
 346    variant->jit_function[RAST_WHOLE]( &state->jit_context,
 347                                       x, y,
 348                                       inputs->frontfacing,
 349                                       GET_A0(inputs),
 350                                       GET_DADX(inputs),
 351                                       GET_DADY(inputs),
 352                                       color,
 353                                       depth,
 354                                       0xffff,
 355                                       &task->vis_counter,
 356                                       stride );
 357    END_JIT_CALL();
 358 }
 359
 360 void lp_rast_triangle_1( struct lp_rasterizer_task *,
 361                          const union lp_rast_cmd_arg );
 362 void lp_rast_triangle_2( struct lp_rasterizer_task *,
 363                          const union lp_rast_cmd_arg );
 364 void lp_rast_triangle_3( struct lp_rasterizer_task *,
 365                          const union lp_rast_cmd_arg );
 366 void lp_rast_triangle_4( struct lp_rasterizer_task *,
 367                          const union lp_rast_cmd_arg );
 368 void lp_rast_triangle_5( struct lp_rasterizer_task *,
 369                          const union lp_rast_cmd_arg );
 370 void lp_rast_triangle_6( struct lp_rasterizer_task *,
 371                          const union lp_rast_cmd_arg );
 372 void lp_rast_triangle_7( struct lp_rasterizer_task *,
 373                          const union lp_rast_cmd_arg );
 374 void lp_rast_triangle_8( struct lp_rasterizer_task *,
 375                          const union lp_rast_cmd_arg );
 376
 377 void lp_rast_triangle_3_4(struct lp_rasterizer_task *,
 378                           const union lp_rast_cmd_arg );
 379
 380 void lp_rast_triangle_3_16( struct lp_rasterizer_task *,
 381                             const union lp_rast_cmd_arg );
 382
 383 void lp_rast_triangle_4_16( struct lp_rasterizer_task *,
 384                             const union lp_rast_cmd_arg );
 385
 386 void
 387 lp_rast_set_state(struct lp_rasterizer_task *task,
 388                   const union lp_rast_cmd_arg arg);
 389
 390 void
 391 lp_debug_bin( const struct cmd_bin *bin );
 392
 393 #endif