src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.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 #include "util/u_math.h"
  29 #include "util/u_memory.h"
  30 #include "util/u_prim.h"
  31 #include "draw/draw_context.h"
  32 #include "draw/draw_vbuf.h"
  33 #include "draw/draw_vertex.h"
  34 #include "draw/draw_prim_assembler.h"
  35 #include "draw/draw_pt.h"
  36 #include "draw/draw_vs.h"
  37 #include "draw/draw_gs.h"
  38
  39
  40 struct fetch_pipeline_middle_end {
  41    struct draw_pt_middle_end base;
  42    struct draw_context *draw;
  43
  44    struct pt_emit *emit;
  45    struct pt_so_emit *so_emit;
  46    struct pt_fetch *fetch;
  47    struct pt_post_vs *post_vs;
  48
  49    unsigned vertex_data_offset;
  50    unsigned vertex_size;
  51    unsigned input_prim;
  52    unsigned opt;
  53 };
  54
  55
  56 /**
  57  * Prepare/validate middle part of the vertex pipeline.
  58  * NOTE: if you change this function, also look at the LLVM
  59  * function llvm_middle_end_prepare() for similar changes.
  60  */
  61 static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
  62                                     unsigned prim,
  63                                     unsigned opt,
  64                                     unsigned *max_vertices )
  65 {
  66    struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
  67    struct draw_context *draw = fpme->draw;
  68    struct draw_vertex_shader *vs = draw->vs.vertex_shader;
  69    struct draw_geometry_shader *gs = draw->gs.geometry_shader;
  70    unsigned i;
  71    unsigned instance_id_index = ~0;
  72
  73    const unsigned gs_out_prim = (gs ? gs->output_primitive :
  74                                  u_assembled_prim(prim));
  75
  76    /* Add one to num_outputs because the pipeline occasionally tags on
  77     * an additional texcoord, eg for AA lines.
  78     */
  79    unsigned nr = MAX2( vs->info.num_inputs,
  80                        vs->info.num_outputs + 1 );
  81
  82    if (gs) {
  83       nr = MAX2(nr, gs->info.num_outputs + 1);
  84    }
  85
  86    /* Scan for instanceID system value.
  87     */
  88    for (i = 0; i < vs->info.num_inputs; i++) {
  89       if (vs->info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
  90          instance_id_index = i;
  91          break;
  92       }
  93    }
  94
  95    fpme->input_prim = prim;
  96    fpme->opt = opt;
  97
  98    /* Always leave room for the vertex header whether we need it or
  99     * not.  It's hard to get rid of it in particular because of the
 100     * viewport code in draw_pt_post_vs.c.
 101     */
 102    fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
 103
 104
 105
 106    draw_pt_fetch_prepare( fpme->fetch,
 107                           vs->info.num_inputs,
 108                           fpme->vertex_size,
 109                           instance_id_index );
 110    draw_pt_post_vs_prepare( fpme->post_vs,
 111                             draw->clip_xy,
 112                             draw->clip_z,
 113                             draw->clip_user,
 114                             draw->guard_band_xy,
 115                             draw->identity_viewport,
 116                             draw->rasterizer->clip_halfz,
 117                             (draw->vs.edgeflag_output ? TRUE : FALSE) );
 118
 119    draw_pt_so_emit_prepare( fpme->so_emit, FALSE );
 120
 121    if (!(opt & PT_PIPELINE)) {
 122       draw_pt_emit_prepare( fpme->emit,
 123                             gs_out_prim,
 124                             max_vertices );
 125
 126       *max_vertices = MAX2( *max_vertices, 4096 );
 127    }
 128    else {
 129       /* limit max fetches by limiting max_vertices */
 130       *max_vertices = 4096;
 131    }
 132
 133    /* No need to prepare the shader.
 134     */
 135    vs->prepare(vs, draw);
 136 }
 137
 138
 139 static void
 140 fetch_pipeline_bind_parameters(struct draw_pt_middle_end *middle)
 141 {
 142    /* No-op since the vertex shader executor and drawing pipeline
 143     * just grab the constants, viewport, etc. from the draw context state.
 144     */
 145 }
 146
 147
 148 static void fetch( struct pt_fetch *fetch,
 149                    const struct draw_fetch_info *fetch_info,
 150                    char *output)
 151 {
 152    if (fetch_info->linear) {
 153       draw_pt_fetch_run_linear( fetch,
 154                                 fetch_info->start,
 155                                 fetch_info->count,
 156                                 output );
 157    }
 158    else {
 159       draw_pt_fetch_run( fetch,
 160                          fetch_info->elts,
 161                          fetch_info->count,
 162                          output );
 163    }
 164 }
 165
 166
 167 static void pipeline(struct fetch_pipeline_middle_end *fpme,
 168                      const struct draw_vertex_info *vert_info,
 169                      const struct draw_prim_info *prim_info)
 170 {
 171    if (prim_info->linear)
 172       draw_pipeline_run_linear( fpme->draw,
 173                                 vert_info,
 174                                 prim_info);
 175    else
 176       draw_pipeline_run( fpme->draw,
 177                          vert_info,
 178                          prim_info );
 179 }
 180
 181 static void emit(struct pt_emit *emit,
 182                  const struct draw_vertex_info *vert_info,
 183                  const struct draw_prim_info *prim_info)
 184 {
 185    if (prim_info->linear) {
 186       draw_pt_emit_linear(emit, vert_info, prim_info);
 187    }
 188    else {
 189       draw_pt_emit(emit, vert_info, prim_info);
 190    }
 191 }
 192
 193
 194 static void draw_vertex_shader_run(struct draw_vertex_shader *vshader,
 195                                    const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
 196                                    unsigned const_size[PIPE_MAX_CONSTANT_BUFFERS],
 197                                    const struct draw_vertex_info *input_verts,
 198                                    struct draw_vertex_info *output_verts )
 199 {
 200    output_verts->vertex_size = input_verts->vertex_size;
 201    output_verts->stride = input_verts->vertex_size;
 202    output_verts->count = input_verts->count;
 203    output_verts->verts =
 204       (struct vertex_header *)MALLOC(output_verts->vertex_size *
 205                                      align(output_verts->count, 4));
 206
 207    vshader->run_linear(vshader,
 208                        (const float (*)[4])input_verts->verts->data,
 209                        (      float (*)[4])output_verts->verts->data,
 210                        constants,
 211                        const_size,
 212                        input_verts->count,
 213                        input_verts->vertex_size,
 214                        input_verts->vertex_size);
 215 }
 216
 217 static void fetch_pipeline_generic( struct draw_pt_middle_end *middle,
 218                                     const struct draw_fetch_info *fetch_info,
 219                                     const struct draw_prim_info *in_prim_info )
 220 {
 221    struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
 222    struct draw_context *draw = fpme->draw;
 223    struct draw_vertex_shader *vshader = draw->vs.vertex_shader;
 224    struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
 225    struct draw_prim_info gs_prim_info;
 226    struct draw_vertex_info fetched_vert_info;
 227    struct draw_vertex_info vs_vert_info;
 228    struct draw_vertex_info gs_vert_info;
 229    struct draw_vertex_info *vert_info;
 230    struct draw_prim_info ia_prim_info;
 231    struct draw_vertex_info ia_vert_info;
 232    const struct draw_prim_info *prim_info = in_prim_info;
 233    boolean free_prim_info = FALSE;
 234    unsigned opt = fpme->opt;
 235
 236    fetched_vert_info.count = fetch_info->count;
 237    fetched_vert_info.vertex_size = fpme->vertex_size;
 238    fetched_vert_info.stride = fpme->vertex_size;
 239    fetched_vert_info.verts =
 240       (struct vertex_header *)MALLOC(fpme->vertex_size *
 241                                      align(fetch_info->count,  4));
 242    if (!fetched_vert_info.verts) {
 243       assert(0);
 244       return;
 245    }
 246    if (draw->collect_statistics) {
 247       draw->statistics.ia_vertices += prim_info->count;
 248       draw->statistics.ia_primitives +=
 249          u_decomposed_prims_for_vertices(prim_info->prim, fetch_info->count);
 250       draw->statistics.vs_invocations += fetch_info->count;
 251    }
 252
 253    /* Fetch into our vertex buffer.
 254     */
 255    fetch( fpme->fetch, fetch_info, (char *)fetched_vert_info.verts );
 256
 257    /* Finished with fetch:
 258     */
 259    fetch_info = NULL;
 260    vert_info = &fetched_vert_info;
 261
 262    /* Run the shader, note that this overwrites the data[] parts of
 263     * the pipeline verts.
 264     */
 265    if (fpme->opt & PT_SHADE) {
 266       draw_vertex_shader_run(vshader,
 267                              draw->pt.user.vs_constants,
 268                              draw->pt.user.vs_constants_size,
 269                              vert_info,
 270                              &vs_vert_info);
 271
 272       FREE(vert_info->verts);
 273       vert_info = &vs_vert_info;
 274    }
 275
 276    if ((fpme->opt & PT_SHADE) && gshader) {
 277       draw_geometry_shader_run(gshader,
 278                                draw->pt.user.gs_constants,
 279                                draw->pt.user.gs_constants_size,
 280                                vert_info,
 281                                prim_info,
 282                                &vshader->info,
 283                                &gs_vert_info,
 284                                &gs_prim_info);
 285
 286       FREE(vert_info->verts);
 287       vert_info = &gs_vert_info;
 288       prim_info = &gs_prim_info;
 289    } else {
 290       if (draw_prim_assembler_is_required(draw, prim_info, vert_info)) {
 291          draw_prim_assembler_run(draw, prim_info, vert_info,
 292                                  &ia_prim_info, &ia_vert_info);
 293
 294          if (ia_vert_info.count) {
 295             FREE(vert_info->verts);
 296             vert_info = &ia_vert_info;
 297             prim_info = &ia_prim_info;
 298             free_prim_info = TRUE;
 299          }
 300       }
 301    }
 302    if (prim_info->count == 0) {
 303       debug_printf("GS/IA didn't emit any vertices!\n");
 304
 305       FREE(vert_info->verts);
 306       if (free_prim_info) {
 307          FREE(prim_info->primitive_lengths);
 308       }
 309       return;
 310    }
 311
 312
 313    /* Stream output needs to be done before clipping.
 314     *
 315     * XXX: Stream output surely needs to respect the prim_info->elt
 316     *      lists.
 317     */
 318    draw_pt_so_emit( fpme->so_emit, vert_info, prim_info );
 319
 320    draw_stats_clipper_primitives(draw, prim_info);
 321
 322    /*
 323     * if there's no position, need to stop now, or the latter stages
 324     * will try to access non-existent position output.
 325     */
 326    if (draw_current_shader_position_output(draw) != -1) {
 327
 328       if (draw_pt_post_vs_run( fpme->post_vs, vert_info, prim_info ))
 329       {
 330          opt |= PT_PIPELINE;
 331       }
 332
 333       /* Do we need to run the pipeline?
 334        */
 335       if (opt & PT_PIPELINE) {
 336          pipeline( fpme, vert_info, prim_info );
 337       }
 338       else {
 339          emit( fpme->emit, vert_info, prim_info );
 340       }
 341    }
 342    FREE(vert_info->verts);
 343    if (free_prim_info) {
 344       FREE(prim_info->primitive_lengths);
 345    }
 346 }
 347
 348 static void fetch_pipeline_run( struct draw_pt_middle_end *middle,
 349                                 const unsigned *fetch_elts,
 350                                 unsigned fetch_count,
 351                                 const ushort *draw_elts,
 352                                 unsigned draw_count,
 353                                 unsigned prim_flags )
 354 {
 355    struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
 356    struct draw_fetch_info fetch_info;
 357    struct draw_prim_info prim_info;
 358
 359    fetch_info.linear = FALSE;
 360    fetch_info.start = 0;
 361    fetch_info.elts = fetch_elts;
 362    fetch_info.count = fetch_count;
 363
 364    prim_info.linear = FALSE;
 365    prim_info.start = 0;
 366    prim_info.count = draw_count;
 367    prim_info.elts = draw_elts;
 368    prim_info.prim = fpme->input_prim;
 369    prim_info.flags = prim_flags;
 370    prim_info.primitive_count = 1;
 371    prim_info.primitive_lengths = &draw_count;
 372
 373    fetch_pipeline_generic( middle, &fetch_info, &prim_info );
 374 }
 375
 376
 377 static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle,
 378                                        unsigned start,
 379                                        unsigned count,
 380                                        unsigned prim_flags)
 381 {
 382    struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
 383    struct draw_fetch_info fetch_info;
 384    struct draw_prim_info prim_info;
 385
 386    fetch_info.linear = TRUE;
 387    fetch_info.start = start;
 388    fetch_info.count = count;
 389    fetch_info.elts = NULL;
 390
 391    prim_info.linear = TRUE;
 392    prim_info.start = 0;
 393    prim_info.count = count;
 394    prim_info.elts = NULL;
 395    prim_info.prim = fpme->input_prim;
 396    prim_info.flags = prim_flags;
 397    prim_info.primitive_count = 1;
 398    prim_info.primitive_lengths = &count;
 399
 400    fetch_pipeline_generic( middle, &fetch_info, &prim_info );
 401 }
 402
 403
 404
 405 static boolean fetch_pipeline_linear_run_elts( struct draw_pt_middle_end *middle,
 406                                                unsigned start,
 407                                                unsigned count,
 408                                                const ushort *draw_elts,
 409                                                unsigned draw_count,
 410                                                unsigned prim_flags )
 411 {
 412    struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
 413    struct draw_fetch_info fetch_info;
 414    struct draw_prim_info prim_info;
 415
 416    fetch_info.linear = TRUE;
 417    fetch_info.start = start;
 418    fetch_info.count = count;
 419    fetch_info.elts = NULL;
 420
 421    prim_info.linear = FALSE;
 422    prim_info.start = 0;
 423    prim_info.count = draw_count;
 424    prim_info.elts = draw_elts;
 425    prim_info.prim = fpme->input_prim;
 426    prim_info.flags = prim_flags;
 427    prim_info.primitive_count = 1;
 428    prim_info.primitive_lengths = &draw_count;
 429
 430    fetch_pipeline_generic( middle, &fetch_info, &prim_info );
 431
 432    return TRUE;
 433 }
 434
 435
 436
 437 static void fetch_pipeline_finish( struct draw_pt_middle_end *middle )
 438 {
 439    /* nothing to do */
 440 }
 441
 442 static void fetch_pipeline_destroy( struct draw_pt_middle_end *middle )
 443 {
 444    struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
 445
 446    if (fpme->fetch)
 447       draw_pt_fetch_destroy( fpme->fetch );
 448
 449    if (fpme->emit)
 450       draw_pt_emit_destroy( fpme->emit );
 451
 452    if (fpme->so_emit)
 453       draw_pt_so_emit_destroy( fpme->so_emit );
 454
 455    if (fpme->post_vs)
 456       draw_pt_post_vs_destroy( fpme->post_vs );
 457
 458    FREE(middle);
 459 }
 460
 461
 462 struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit( struct draw_context *draw )
 463 {
 464    struct fetch_pipeline_middle_end *fpme = CALLOC_STRUCT( fetch_pipeline_middle_end );
 465    if (!fpme)
 466       goto fail;
 467
 468    fpme->base.prepare        = fetch_pipeline_prepare;
 469    fpme->base.bind_parameters  = fetch_pipeline_bind_parameters;
 470    fpme->base.run            = fetch_pipeline_run;
 471    fpme->base.run_linear     = fetch_pipeline_linear_run;
 472    fpme->base.run_linear_elts = fetch_pipeline_linear_run_elts;
 473    fpme->base.finish         = fetch_pipeline_finish;
 474    fpme->base.destroy        = fetch_pipeline_destroy;
 475
 476    fpme->draw = draw;
 477
 478    fpme->fetch = draw_pt_fetch_create( draw );
 479    if (!fpme->fetch)
 480       goto fail;
 481
 482    fpme->post_vs = draw_pt_post_vs_create( draw );
 483    if (!fpme->post_vs)
 484       goto fail;
 485
 486    fpme->emit = draw_pt_emit_create( draw );
 487    if (!fpme->emit)
 488       goto fail;
 489
 490    fpme->so_emit = draw_pt_so_emit_create( draw );
 491    if (!fpme->so_emit)
 492       goto fail;
 493
 494    return &fpme->base;
 495
 496  fail:
 497    if (fpme)
 498       fetch_pipeline_destroy( &fpme->base );
 499
 500    return NULL;
 501 }