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