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