src/gallium/auxiliary/draw/draw_pt_fetch_shade_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
  34 #include "util/u_math.h"
  35 #include "util/u_memory.h"
  36 #include "draw/draw_context.h"
  37 #include "draw/draw_private.h"
  38 #include "draw/draw_vbuf.h"
  39 #include "draw/draw_vertex.h"
  40 #include "draw/draw_pt.h"
  41 #include "draw/draw_vs.h"
  42
  43
  44 struct fetch_shade_emit;
  45
  46
  47 /* Prototype fetch, shade, emit-hw-verts all in one go.
  48  */
  49 struct fetch_shade_emit {
  50    struct draw_pt_middle_end base;
  51    struct draw_context *draw;
  52
  53
  54    /* Temporaries:
  55     */
  56    const float *constants;
  57    unsigned pitch[PIPE_MAX_ATTRIBS];
  58    const ubyte *src[PIPE_MAX_ATTRIBS];
  59    unsigned prim;
  60
  61    struct draw_vs_varient_key key;
  62    struct draw_vs_varient *active;
  63
  64
  65    const struct vertex_info *vinfo;
  66 };
  67
  68
  69
  70
  71 static void fse_prepare( struct draw_pt_middle_end *middle,
  72                          unsigned prim,
  73                          unsigned opt,
  74                          unsigned *max_vertices )
  75 {
  76    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
  77    struct draw_context *draw = fse->draw;
  78    unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
  79    const struct vertex_info *vinfo;
  80    unsigned i;
  81    unsigned nr_vbs = 0;
  82
  83
  84    if (!draw->render->set_primitive( draw->render,
  85                                      prim )) {
  86       assert(0);
  87       return;
  88    }
  89
  90    /* Must do this after set_primitive() above:
  91     */
  92    fse->vinfo = vinfo = draw->render->get_vertex_info(draw->render);
  93
  94
  95
  96    fse->key.output_stride = vinfo->size * 4;
  97    fse->key.nr_outputs = vinfo->num_attribs;
  98    fse->key.nr_inputs = num_vs_inputs;
  99
 100    fse->key.nr_elements = MAX2(fse->key.nr_outputs,     /* outputs - translate to hw format */
 101                                fse->key.nr_inputs);     /* inputs - fetch from api format */
 102
 103    fse->key.viewport = (!draw->rasterizer->bypass_vs_clip_and_viewport &&
 104                         !draw->identity_viewport);
 105    fse->key.clip = !draw->bypass_clipping;
 106    fse->key.const_vbuffers = 0;
 107
 108    memset(fse->key.element, 0,
 109           fse->key.nr_elements * sizeof(fse->key.element[0]));
 110
 111    for (i = 0; i < num_vs_inputs; i++) {
 112       const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
 113       fse->key.element[i].in.format = src->src_format;
 114
 115       /* Consider ignoring these, ie make generated programs
 116        * independent of this state:
 117        */
 118       fse->key.element[i].in.buffer = src->vertex_buffer_index;
 119       fse->key.element[i].in.offset = src->src_offset;
 120       nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1);
 121    }
 122
 123    for (i = 0; i < 5 && i < nr_vbs; i++) {
 124       if (draw->pt.vertex_buffer[i].stride == 0)
 125          fse->key.const_vbuffers |= (1<<i);
 126    }
 127
 128    if (0) debug_printf("%s: lookup const_vbuffers: %x\n", __FUNCTION__, fse->key.const_vbuffers);
 129
 130    {
 131       unsigned dst_offset = 0;
 132
 133       for (i = 0; i < vinfo->num_attribs; i++) {
 134          unsigned emit_sz = 0;
 135
 136          switch (vinfo->attrib[i].emit) {
 137          case EMIT_4F:
 138             emit_sz = 4 * sizeof(float);
 139             break;
 140          case EMIT_3F:
 141             emit_sz = 3 * sizeof(float);
 142             break;
 143          case EMIT_2F:
 144             emit_sz = 2 * sizeof(float);
 145             break;
 146          case EMIT_1F:
 147             emit_sz = 1 * sizeof(float);
 148             break;
 149          case EMIT_1F_PSIZE:
 150             emit_sz = 1 * sizeof(float);
 151             break;
 152          case EMIT_4UB:
 153             emit_sz = 4 * sizeof(ubyte);
 154             break;
 155          default:
 156             assert(0);
 157             break;
 158          }
 159
 160          /* The elements in the key correspond to vertex shader output
 161           * numbers, not to positions in the hw vertex description --
 162           * that's handled by the output_offset field.
 163           */
 164          fse->key.element[i].out.format = vinfo->attrib[i].emit;
 165          fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index;
 166          fse->key.element[i].out.offset = dst_offset;
 167
 168          dst_offset += emit_sz;
 169          assert(fse->key.output_stride >= dst_offset);
 170       }
 171    }
 172
 173
 174    fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader,
 175                                          &fse->key );
 176
 177    if (!fse->active) {
 178       assert(0);
 179       return ;
 180    }
 181
 182    if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__,
 183                        fse->active->key.const_vbuffers);
 184
 185    /* Now set buffer pointers:
 186     */
 187    for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
 188       fse->active->set_buffer( fse->active,
 189                                i,
 190                                ((const ubyte *) draw->pt.user.vbuffer[i] +
 191                                 draw->pt.vertex_buffer[i].buffer_offset),
 192                               draw->pt.vertex_buffer[i].stride );
 193    }
 194
 195    *max_vertices = (draw->render->max_vertex_buffer_bytes /
 196                     (vinfo->size * 4));
 197
 198    /* Return an even number of verts.
 199     * This prevents "parity" errors when splitting long triangle strips which
 200     * can lead to front/back culling mix-ups.
 201     * Every other triangle in a strip has an alternate front/back orientation
 202     * so splitting at an odd position can cause the orientation of subsequent
 203     * triangles to get reversed.
 204     */
 205    *max_vertices = *max_vertices & ~1;
 206
 207    /* Probably need to do this somewhere (or fix exec shader not to
 208     * need it):
 209     */
 210    if (1) {
 211       struct draw_vertex_shader *vs = draw->vs.vertex_shader;
 212       vs->prepare(vs, draw);
 213    }
 214 }
 215
 216
 217
 218 static void fse_run_linear( struct draw_pt_middle_end *middle,
 219                             unsigned start,
 220                             unsigned count )
 221 {
 222    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 223    struct draw_context *draw = fse->draw;
 224    char *hw_verts;
 225
 226    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 227     */
 228    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 229
 230    if (count >= UNDEFINED_VERTEX_ID)
 231       goto fail;
 232
 233    if (!draw->render->allocate_vertices( draw->render,
 234                                          (ushort)fse->key.output_stride,
 235                                          (ushort)count ))
 236       goto fail;
 237
 238    hw_verts = draw->render->map_vertices( draw->render );
 239    if (!hw_verts)
 240       goto fail;
 241
 242    /* Single routine to fetch vertices, run shader and emit HW verts.
 243     * Clipping is done elsewhere -- either by the API or on hardware,
 244     * or for some other reason not required...
 245     */
 246    fse->active->run_linear( fse->active,
 247                             start, count,
 248                             hw_verts );
 249
 250
 251    if (0) {
 252       unsigned i;
 253       for (i = 0; i < count; i++) {
 254          debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n", __FUNCTION__, i,
 255                       fse->key.output_stride,
 256                       fse->key.output_stride * i);
 257
 258          draw_dump_emitted_vertex( fse->vinfo,
 259                                    (const uint8_t *)hw_verts + fse->key.output_stride * i );
 260       }
 261    }
 262
 263    draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
 264
 265    /* Draw arrays path to avoid re-emitting index list again and
 266     * again.
 267     */
 268    draw->render->draw_arrays( draw->render,
 269                               0,
 270                               count );
 271
 272
 273    draw->render->release_vertices( draw->render );
 274
 275    return;
 276
 277 fail:
 278    assert(0);
 279    return;
 280 }
 281
 282
 283 static void
 284 fse_run(struct draw_pt_middle_end *middle,
 285         const unsigned *fetch_elts,
 286         unsigned fetch_count,
 287         const ushort *draw_elts,
 288         unsigned draw_count )
 289 {
 290    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 291    struct draw_context *draw = fse->draw;
 292    void *hw_verts;
 293
 294    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 295     */
 296    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 297
 298    if (fetch_count >= UNDEFINED_VERTEX_ID)
 299       goto fail;
 300
 301    if (!draw->render->allocate_vertices( draw->render,
 302                                          (ushort)fse->key.output_stride,
 303                                          (ushort)fetch_count ))
 304       goto fail;
 305
 306    hw_verts = draw->render->map_vertices( draw->render );
 307    if (!hw_verts)
 308       goto fail;
 309
 310
 311    /* Single routine to fetch vertices, run shader and emit HW verts.
 312     */
 313    fse->active->run_elts( fse->active,
 314                           fetch_elts,
 315                           fetch_count,
 316                           hw_verts );
 317
 318
 319    if (0) {
 320       unsigned i;
 321       for (i = 0; i < fetch_count; i++) {
 322          debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i);
 323          draw_dump_emitted_vertex( fse->vinfo,
 324                                    (const uint8_t *)hw_verts +
 325                                    fse->key.output_stride * i );
 326       }
 327    }
 328
 329    draw->render->unmap_vertices( draw->render, 0, (ushort)(fetch_count - 1) );
 330
 331    draw->render->draw( draw->render,
 332                        draw_elts,
 333                        draw_count );
 334
 335
 336    draw->render->release_vertices( draw->render );
 337    return;
 338
 339 fail:
 340    assert(0);
 341    return;
 342 }
 343
 344
 345
 346 static boolean fse_run_linear_elts( struct draw_pt_middle_end *middle,
 347                                  unsigned start,
 348                                  unsigned count,
 349                                  const ushort *draw_elts,
 350                                  unsigned draw_count )
 351 {
 352    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 353    struct draw_context *draw = fse->draw;
 354    char *hw_verts;
 355
 356    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 357     */
 358    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 359
 360    if (count >= UNDEFINED_VERTEX_ID)
 361       return FALSE;
 362
 363    if (!draw->render->allocate_vertices( draw->render,
 364                                          (ushort)fse->key.output_stride,
 365                                          (ushort)count ))
 366       return FALSE;
 367
 368    hw_verts = draw->render->map_vertices( draw->render );
 369    if (!hw_verts)
 370       return FALSE;
 371
 372    /* Single routine to fetch vertices, run shader and emit HW verts.
 373     * Clipping is done elsewhere -- either by the API or on hardware,
 374     * or for some other reason not required...
 375     */
 376    fse->active->run_linear( fse->active,
 377                             start, count,
 378                             hw_verts );
 379
 380
 381    draw->render->draw( draw->render,
 382                        draw_elts,
 383                        draw_count );
 384
 385
 386    draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
 387
 388    draw->render->release_vertices( draw->render );
 389
 390    return TRUE;
 391 }
 392
 393
 394
 395 static void fse_finish( struct draw_pt_middle_end *middle )
 396 {
 397 }
 398
 399
 400 static void
 401 fse_destroy( struct draw_pt_middle_end *middle )
 402 {
 403    FREE(middle);
 404 }
 405
 406 struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw )
 407 {
 408    struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
 409    if (!fse)
 410       return NULL;
 411
 412    fse->base.prepare = fse_prepare;
 413    fse->base.run = fse_run;
 414    fse->base.run_linear = fse_run_linear;
 415    fse->base.run_linear_elts = fse_run_linear_elts;
 416    fse->base.finish = fse_finish;
 417    fse->base.destroy = fse_destroy;
 418    fse->draw = draw;
 419
 420    return &fse->base;
 421 }