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