src/gallium/auxiliary/draw/draw_pt_fetch_emit.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   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 TUNGSTEN GRAPHICS 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  /*
  29   * Authors:
  30   *   Keith Whitwell <keith@tungstengraphics.com>
  31   */
  32
  33 #include "util/u_memory.h"
  34 #include "draw/draw_context.h"
  35 #include "draw/draw_private.h"
  36 #include "draw/draw_vbuf.h"
  37 #include "draw/draw_vertex.h"
  38 #include "draw/draw_pt.h"
  39 #include "translate/translate.h"
  40 #include "translate/translate_cache.h"
  41
  42 /* The simplest 'middle end' in the new vertex code.
  43  *
  44  * The responsibilities of a middle end are to:
  45  *  - perform vertex fetch using
  46  *       - draw vertex element/buffer state
  47  *       - a list of fetch indices we received as an input
  48  *  - run the vertex shader
  49  *  - cliptest,
  50  *  - clip coord calculation
  51  *  - viewport transformation
  52  *  - if necessary, run the primitive pipeline, passing it:
  53  *       - a linear array of vertex_header vertices constructed here
  54  *       - a set of draw indices we received as an input
  55  *  - otherwise, drive the hw backend,
  56  *       - allocate space for hardware format vertices
  57  *       - translate the vertex-shader output vertices to hw format
  58  *       - calling the backend draw functions.
  59  *
  60  * For convenience, we provide a helper function to drive the hardware
  61  * backend given similar inputs to those required to run the pipeline.
  62  *
  63  * In the case of passthrough mode, many of these actions are disabled
  64  * or noops, so we end up doing:
  65  *
  66  *  - perform vertex fetch
  67  *  - drive the hw backend
  68  *
  69  * IE, basically just vertex fetch to post-vs-format vertices,
  70  * followed by a call to the backend helper function.
  71  */
  72
  73
  74 struct fetch_emit_middle_end {
  75    struct draw_pt_middle_end base;
  76    struct draw_context *draw;
  77
  78    struct translate *translate;
  79    const struct vertex_info *vinfo;
  80
  81    /* Cache point size somewhere it's address won't change:
  82     */
  83    float point_size;
  84
  85    struct translate_cache *cache;
  86 };
  87
  88
  89
  90
  91 static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
  92                                 unsigned prim,
  93                                 unsigned opt,
  94                                 unsigned *max_vertices )
  95 {
  96    struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
  97    struct draw_context *draw = feme->draw;
  98    const struct vertex_info *vinfo;
  99    unsigned i, dst_offset;
 100    boolean ok;
 101    struct translate_key key;
 102
 103
 104    ok = draw->render->set_primitive( draw->render,
 105                                      prim );
 106    if (!ok) {
 107       assert(0);
 108       return;
 109    }
 110
 111    /* Must do this after set_primitive() above:
 112     */
 113    vinfo = feme->vinfo = draw->render->get_vertex_info(draw->render);
 114
 115
 116
 117    /* Transform from API vertices to HW vertices, skipping the
 118     * pipeline_vertex intermediate step.
 119     */
 120    dst_offset = 0;
 121    memset(&key, 0, sizeof(key));
 122
 123    for (i = 0; i < vinfo->num_attribs; i++) {
 124       const struct pipe_vertex_element *src = &draw->pt.vertex_element[vinfo->attrib[i].src_index];
 125
 126       unsigned emit_sz = 0;
 127       unsigned input_format = src->src_format;
 128       unsigned input_buffer = src->vertex_buffer_index;
 129       unsigned input_offset = src->src_offset;
 130       unsigned output_format;
 131
 132       switch (vinfo->attrib[i].emit) {
 133       case EMIT_4F:
 134          output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 135          emit_sz = 4 * sizeof(float);
 136          break;
 137       case EMIT_3F:
 138          output_format = PIPE_FORMAT_R32G32B32_FLOAT;
 139          emit_sz = 3 * sizeof(float);
 140          break;
 141       case EMIT_2F:
 142          output_format = PIPE_FORMAT_R32G32_FLOAT;
 143          emit_sz = 2 * sizeof(float);
 144          break;
 145       case EMIT_1F:
 146          output_format = PIPE_FORMAT_R32_FLOAT;
 147          emit_sz = 1 * sizeof(float);
 148          break;
 149       case EMIT_1F_PSIZE:
 150          input_format = PIPE_FORMAT_R32_FLOAT;
 151          input_buffer = draw->pt.nr_vertex_buffers;
 152          input_offset = 0;
 153          output_format = PIPE_FORMAT_R32_FLOAT;
 154          emit_sz = 1 * sizeof(float);
 155          break;
 156       default:
 157          assert(0);
 158          output_format = PIPE_FORMAT_NONE;
 159          emit_sz = 0;
 160          continue;
 161       }
 162
 163       key.element[i].input_format = input_format;
 164       key.element[i].input_buffer = input_buffer;
 165       key.element[i].input_offset = input_offset;
 166       key.element[i].output_format = output_format;
 167       key.element[i].output_offset = dst_offset;
 168
 169       dst_offset += emit_sz;
 170    }
 171
 172    key.nr_elements = vinfo->num_attribs;
 173    key.output_stride = vinfo->size * 4;
 174
 175    /* Don't bother with caching at this stage:
 176     */
 177    if (!feme->translate ||
 178        translate_key_compare(&feme->translate->key, &key) != 0)
 179    {
 180       translate_key_sanitize(&key);
 181       feme->translate = translate_cache_find(feme->cache,
 182                                              &key);
 183
 184
 185       feme->translate->set_buffer(feme->translate,
 186                                   draw->pt.nr_vertex_buffers,
 187                                   &feme->point_size,
 188                                   0);
 189    }
 190
 191    feme->point_size = draw->rasterizer->point_size;
 192
 193    for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
 194       feme->translate->set_buffer(feme->translate,
 195                                   i,
 196                                   ((char *)draw->pt.user.vbuffer[i] +
 197                                    draw->pt.vertex_buffer[i].buffer_offset),
 198                                   draw->pt.vertex_buffer[i].pitch );
 199    }
 200
 201    *max_vertices = (draw->render->max_vertex_buffer_bytes /
 202                     (vinfo->size * 4));
 203
 204    /* Return an even number of verts.
 205     * This prevents "parity" errors when splitting long triangle strips which
 206     * can lead to front/back culling mix-ups.
 207     * Every other triangle in a strip has an alternate front/back orientation
 208     * so splitting at an odd position can cause the orientation of subsequent
 209     * triangles to get reversed.
 210     */
 211    *max_vertices = *max_vertices & ~1;
 212 }
 213
 214
 215
 216
 217
 218 static void fetch_emit_run( struct draw_pt_middle_end *middle,
 219                             const unsigned *fetch_elts,
 220                             unsigned fetch_count,
 221                             const ushort *draw_elts,
 222                             unsigned draw_count )
 223 {
 224    struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
 225    struct draw_context *draw = feme->draw;
 226    void *hw_verts;
 227
 228    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 229     */
 230    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 231
 232    hw_verts = draw->render->allocate_vertices( draw->render,
 233                                                (ushort)feme->translate->key.output_stride,
 234                                                (ushort)fetch_count );
 235    if (!hw_verts) {
 236       assert(0);
 237       return;
 238    }
 239
 240
 241    /* Single routine to fetch vertices and emit HW verts.
 242     */
 243    feme->translate->run_elts( feme->translate,
 244                               fetch_elts,
 245                               fetch_count,
 246                               hw_verts );
 247
 248    if (0) {
 249       unsigned i;
 250       for (i = 0; i < fetch_count; i++) {
 251          debug_printf("\n\nvertex %d:\n", i);
 252          draw_dump_emitted_vertex( feme->vinfo,
 253                                    (const uint8_t *)hw_verts + feme->vinfo->size * 4 * i );
 254       }
 255    }
 256
 257    /* XXX: Draw arrays path to avoid re-emitting index list again and
 258     * again.
 259     */
 260    draw->render->draw( draw->render,
 261                        draw_elts,
 262                        draw_count );
 263
 264    /* Done -- that was easy, wasn't it:
 265     */
 266    draw->render->release_vertices( draw->render,
 267                                    hw_verts,
 268                                    feme->translate->key.output_stride,
 269                                    fetch_count );
 270
 271 }
 272
 273
 274 static void fetch_emit_run_linear( struct draw_pt_middle_end *middle,
 275                                    unsigned start,
 276                                    unsigned count )
 277 {
 278    struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
 279    struct draw_context *draw = feme->draw;
 280    void *hw_verts;
 281
 282    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 283     */
 284    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 285
 286    hw_verts = draw->render->allocate_vertices( draw->render,
 287                                                (ushort)feme->translate->key.output_stride,
 288                                                (ushort)count );
 289    if (!hw_verts) {
 290       assert(0);
 291       return;
 292    }
 293
 294    /* Single routine to fetch vertices and emit HW verts.
 295     */
 296    feme->translate->run( feme->translate,
 297                          start,
 298                          count,
 299                          hw_verts );
 300
 301    if (0) {
 302       unsigned i;
 303       for (i = 0; i < count; i++) {
 304          debug_printf("\n\nvertex %d:\n", i);
 305          draw_dump_emitted_vertex( feme->vinfo,
 306                                    (const uint8_t *)hw_verts + feme->vinfo->size * 4 * i );
 307       }
 308    }
 309
 310    /* XXX: Draw arrays path to avoid re-emitting index list again and
 311     * again.
 312     */
 313    draw->render->draw_arrays( draw->render,
 314                               0, /*start*/
 315                               count );
 316
 317    /* Done -- that was easy, wasn't it:
 318     */
 319    draw->render->release_vertices( draw->render,
 320                                    hw_verts,
 321                                    feme->translate->key.output_stride,
 322                                    count );
 323
 324 }
 325
 326
 327 static boolean fetch_emit_run_linear_elts( struct draw_pt_middle_end *middle,
 328                                         unsigned start,
 329                                         unsigned count,
 330                                         const ushort *draw_elts,
 331                                         unsigned draw_count )
 332 {
 333    struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
 334    struct draw_context *draw = feme->draw;
 335    void *hw_verts;
 336
 337    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 338     */
 339    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 340
 341    hw_verts = draw->render->allocate_vertices( draw->render,
 342                                                (ushort)feme->translate->key.output_stride,
 343                                                (ushort)count );
 344    if (!hw_verts)
 345       return FALSE;
 346
 347    /* Single routine to fetch vertices and emit HW verts.
 348     */
 349    feme->translate->run( feme->translate,
 350                          start,
 351                          count,
 352                          hw_verts );
 353
 354    /* XXX: Draw arrays path to avoid re-emitting index list again and
 355     * again.
 356     */
 357    draw->render->draw( draw->render,
 358                        draw_elts,
 359                        draw_count );
 360
 361    /* Done -- that was easy, wasn't it:
 362     */
 363    draw->render->release_vertices( draw->render,
 364                                    hw_verts,
 365                                    feme->translate->key.output_stride,
 366                                    count );
 367
 368    return TRUE;
 369 }
 370
 371
 372
 373
 374 static void fetch_emit_finish( struct draw_pt_middle_end *middle )
 375 {
 376    /* nothing to do */
 377 }
 378
 379 static void fetch_emit_destroy( struct draw_pt_middle_end *middle )
 380 {
 381    struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
 382
 383    if (feme->cache)
 384       translate_cache_destroy(feme->cache);
 385
 386    FREE(middle);
 387 }
 388
 389
 390 struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw )
 391 {
 392    struct fetch_emit_middle_end *fetch_emit = CALLOC_STRUCT( fetch_emit_middle_end );
 393    if (fetch_emit == NULL)
 394       return NULL;
 395
 396    fetch_emit->cache = translate_cache_create();
 397    if (!fetch_emit->cache) {
 398       FREE(fetch_emit);
 399       return NULL;
 400    }
 401
 402    fetch_emit->base.prepare    = fetch_emit_prepare;
 403    fetch_emit->base.run        = fetch_emit_run;
 404    fetch_emit->base.run_linear = fetch_emit_run_linear;
 405    fetch_emit->base.run_linear_elts = fetch_emit_run_linear_elts;
 406    fetch_emit->base.finish     = fetch_emit_finish;
 407    fetch_emit->base.destroy    = fetch_emit_destroy;
 408
 409    fetch_emit->draw = draw;
 410
 411    return &fetch_emit->base;
 412 }
 413