src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 VMware, Inc.
   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 VMWARE 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 /** cast wrapper */
  57 static inline struct fetch_pipeline_middle_end *
  58 fetch_pipeline_middle_end(struct draw_pt_middle_end *middle)
  59 {
  60    return (struct fetch_pipeline_middle_end *) middle;
  61 }
  62
  63
  64 /**
  65  * Prepare/validate middle part of the vertex pipeline.
  66  * NOTE: if you change this function, also look at the LLVM
  67  * function llvm_middle_end_prepare() for similar changes.
  68  */
  69 static void
  70 fetch_pipeline_prepare(struct draw_pt_middle_end *middle,
  71                        unsigned prim,
  72                        unsigned opt,
  73                        unsigned *max_vertices)
  74 {
  75    struct fetch_pipeline_middle_end *fpme = fetch_pipeline_middle_end(middle);
  76    struct draw_context *draw = fpme->draw;
  77    struct draw_vertex_shader *vs = draw->vs.vertex_shader;
  78    struct draw_geometry_shader *gs = draw->gs.geometry_shader;
  79    unsigned i;
  80    unsigned instance_id_index = ~0;
  81    const unsigned gs_out_prim = (gs ? gs->output_primitive :
  82                                  u_assembled_prim(prim));
  83    unsigned nr_vs_outputs = draw_total_vs_outputs(draw);
  84    unsigned nr = MAX2(vs->info.num_inputs, nr_vs_outputs);
  85    unsigned point_clip = draw->rasterizer->fill_front == PIPE_POLYGON_MODE_POINT ||
  86                          gs_out_prim == PIPE_PRIM_POINTS;
  87
  88    if (gs) {
  89       nr = MAX2(nr, gs->info.num_outputs + 1);
  90    }
  91
  92    /* Scan for instanceID system value.
  93     */
  94    for (i = 0; i < vs->info.num_inputs; i++) {
  95       if (vs->info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
  96          instance_id_index = i;
  97          break;
  98       }
  99    }
 100
 101    fpme->input_prim = prim;
 102    fpme->opt = opt;
 103
 104    /* Always leave room for the vertex header whether we need it or
 105     * not.  It's hard to get rid of it in particular because of the
 106     * viewport code in draw_pt_post_vs.c.
 107     */
 108    fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
 109
 110    draw_pt_fetch_prepare( fpme->fetch,
 111                           vs->info.num_inputs,
 112                           fpme->vertex_size,
 113                           instance_id_index );
 114    draw_pt_post_vs_prepare( fpme->post_vs,
 115                             draw->clip_xy,
 116                             draw->clip_z,
 117                             draw->clip_user,
 118                             point_clip ? draw->guard_band_points_xy :
 119                                          draw->guard_band_xy,
 120                             draw->bypass_viewport,
 121                             draw->rasterizer->clip_halfz,
 122                             (draw->vs.edgeflag_output ? TRUE : FALSE) );
 123
 124    draw_pt_so_emit_prepare( fpme->so_emit, FALSE );
 125
 126    if (!(opt & PT_PIPELINE)) {
 127       draw_pt_emit_prepare( fpme->emit,
 128                             gs_out_prim,
 129                             max_vertices );
 130
 131       *max_vertices = MAX2( *max_vertices, 4096 );
 132    }
 133    else {
 134       /* limit max fetches by limiting max_vertices */
 135       *max_vertices = 4096;
 136    }
 137
 138    /* No need to prepare the shader.
 139     */
 140    vs->prepare(vs, draw);
 141
 142    /* Make sure that the vertex size didn't change at any point above */
 143    assert(nr_vs_outputs == draw_total_vs_outputs(draw));
 144 }
 145
 146
 147 static void
 148 fetch_pipeline_bind_parameters(struct draw_pt_middle_end *middle)
 149 {
 150    /* No-op since the vertex shader executor and drawing pipeline
 151     * just grab the constants, viewport, etc. from the draw context state.
 152     */
 153 }
 154
 155
 156 static void fetch( struct pt_fetch *fetch,
 157                    const struct draw_fetch_info *fetch_info,
 158                    char *output)
 159 {
 160    if (fetch_info->linear) {
 161       draw_pt_fetch_run_linear( fetch,
 162                                 fetch_info->start,
 163                                 fetch_info->count,
 164                                 output );
 165    }
 166    else {
 167       draw_pt_fetch_run( fetch,
 168                          fetch_info->elts,
 169                          fetch_info->count,
 170                          output );
 171    }
 172 }
 173
 174
 175 static void pipeline(struct fetch_pipeline_middle_end *fpme,
 176                      const struct draw_vertex_info *vert_info,
 177                      const struct draw_prim_info *prim_info)
 178 {
 179    if (prim_info->linear)
 180       draw_pipeline_run_linear( fpme->draw,
 181                                 vert_info,
 182                                 prim_info);
 183    else
 184       draw_pipeline_run( fpme->draw,
 185                          vert_info,
 186                          prim_info );
 187 }
 188
 189
 190 static void
 191 emit(struct pt_emit *emit,
 192      const struct draw_vertex_info *vert_info,
 193      const struct draw_prim_info *prim_info)
 194 {
 195    if (prim_info->linear) {
 196       draw_pt_emit_linear(emit, vert_info, prim_info);
 197    }
 198    else {
 199       draw_pt_emit(emit, vert_info, prim_info);
 200    }
 201 }
 202
 203
 204 static void
 205 draw_vertex_shader_run(struct draw_vertex_shader *vshader,
 206                        const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
 207                        unsigned const_size[PIPE_MAX_CONSTANT_BUFFERS],
 208                        const struct draw_fetch_info *fetch_info,
 209                        const struct draw_vertex_info *input_verts,
 210                        struct draw_vertex_info *output_verts)
 211 {
 212    output_verts->vertex_size = input_verts->vertex_size;
 213    output_verts->stride = input_verts->vertex_size;
 214    output_verts->count = input_verts->count;
 215    output_verts->verts =
 216       (struct vertex_header *)MALLOC(output_verts->vertex_size *
 217                                      align(output_verts->count, 4));
 218
 219    vshader->run_linear(vshader,
 220                        (const float (*)[4])input_verts->verts->data,
 221                        (      float (*)[4])output_verts->verts->data,
 222                        constants,
 223                        const_size,
 224                        input_verts->count,
 225                        input_verts->vertex_size,
 226                        input_verts->vertex_size,
 227                        fetch_info->elts);
 228 }
 229
 230
 231 static void
 232 fetch_pipeline_generic(struct draw_pt_middle_end *middle,
 233                        const struct draw_fetch_info *fetch_info,
 234                        const struct draw_prim_info *in_prim_info)
 235 {
 236    struct fetch_pipeline_middle_end *fpme = fetch_pipeline_middle_end(middle);
 237    struct draw_context *draw = fpme->draw;
 238    struct draw_vertex_shader *vshader = draw->vs.vertex_shader;
 239    struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
 240    struct draw_prim_info gs_prim_info[TGSI_MAX_VERTEX_STREAMS];
 241    struct draw_vertex_info fetched_vert_info;
 242    struct draw_vertex_info vs_vert_info;
 243    struct draw_vertex_info gs_vert_info[TGSI_MAX_VERTEX_STREAMS];
 244    struct draw_vertex_info *vert_info;
 245    struct draw_prim_info ia_prim_info;
 246    struct draw_vertex_info ia_vert_info;
 247    const struct draw_prim_info *prim_info = in_prim_info;
 248    boolean free_prim_info = FALSE;
 249    unsigned opt = fpme->opt;
 250    int num_vertex_streams = 1;
 251
 252    fetched_vert_info.count = fetch_info->count;
 253    fetched_vert_info.vertex_size = fpme->vertex_size;
 254    fetched_vert_info.stride = fpme->vertex_size;
 255    fetched_vert_info.verts =
 256       (struct vertex_header *)MALLOC(fpme->vertex_size *
 257                                      align(fetch_info->count,  4));
 258    if (!fetched_vert_info.verts) {
 259       assert(0);
 260       return;
 261    }
 262    if (draw->collect_statistics) {
 263       draw->statistics.ia_vertices += prim_info->count;
 264       draw->statistics.ia_primitives +=
 265          u_decomposed_prims_for_vertices(prim_info->prim, fetch_info->count);
 266       draw->statistics.vs_invocations += fetch_info->count;
 267    }
 268
 269    /* Fetch into our vertex buffer.
 270     */
 271    fetch( fpme->fetch, fetch_info, (char *)fetched_vert_info.verts );
 272
 273    vert_info = &fetched_vert_info;
 274
 275    /* Run the shader, note that this overwrites the data[] parts of
 276     * the pipeline verts.
 277     * Need fetch info to get vertex id correct.
 278     */
 279    if (fpme->opt & PT_SHADE) {
 280       draw_vertex_shader_run(vshader,
 281                              draw->pt.user.vs_constants,
 282                              draw->pt.user.vs_constants_size,
 283                              fetch_info,
 284                              vert_info,
 285                              &vs_vert_info);
 286
 287       FREE(vert_info->verts);
 288       vert_info = &vs_vert_info;
 289    }
 290
 291    /* Finished with fetch:
 292     */
 293    fetch_info = NULL;
 294
 295    if ((fpme->opt & PT_SHADE) && gshader) {
 296       draw_geometry_shader_run(gshader,
 297                                draw->pt.user.gs_constants,
 298                                draw->pt.user.gs_constants_size,
 299                                vert_info,
 300                                prim_info,
 301                                &vshader->info,
 302                                gs_vert_info,
 303                                gs_prim_info);
 304
 305       FREE(vert_info->verts);
 306       vert_info = &gs_vert_info[0];
 307       prim_info = &gs_prim_info[0];
 308       num_vertex_streams = gshader->num_vertex_streams;
 309
 310       /*
 311        * pt emit can only handle ushort number of vertices (see
 312        * render->allocate_vertices).
 313        * vsplit guarantees there's never more than 4096, however GS can
 314        * easily blow this up (by a factor of 256 (or even 1024) max).
 315        */
 316       if (vert_info->count > 65535) {
 317          opt |= PT_PIPELINE;
 318       }
 319    } else {
 320       if (draw_prim_assembler_is_required(draw, prim_info, vert_info)) {
 321          draw_prim_assembler_run(draw, prim_info, vert_info,
 322                                  &ia_prim_info, &ia_vert_info);
 323
 324          if (ia_vert_info.count) {
 325             FREE(vert_info->verts);
 326             vert_info = &ia_vert_info;
 327             prim_info = &ia_prim_info;
 328             free_prim_info = TRUE;
 329          }
 330       }
 331    }
 332    if (prim_info->count == 0) {
 333       debug_printf("GS/IA didn't emit any vertices!\n");
 334
 335       FREE(vert_info->verts);
 336       if (free_prim_info) {
 337          FREE(prim_info->primitive_lengths);
 338       }
 339       return;
 340    }
 341
 342
 343    /* Stream output needs to be done before clipping.
 344     *
 345     * XXX: Stream output surely needs to respect the prim_info->elt
 346     *      lists.
 347     */
 348    draw_pt_so_emit( fpme->so_emit, num_vertex_streams, vert_info, prim_info );
 349
 350    draw_stats_clipper_primitives(draw, prim_info);
 351
 352    /*
 353     * if there's no position, need to stop now, or the latter stages
 354     * will try to access non-existent position output.
 355     */
 356    if (draw_current_shader_position_output(draw) != -1) {
 357
 358       if (draw_pt_post_vs_run( fpme->post_vs, vert_info, prim_info ))
 359       {
 360          opt |= PT_PIPELINE;
 361       }
 362
 363       /* Do we need to run the pipeline?
 364        */
 365       if (opt & PT_PIPELINE) {
 366          pipeline( fpme, vert_info, prim_info );
 367       }
 368       else {
 369          emit( fpme->emit, vert_info, prim_info );
 370       }
 371    }
 372    FREE(vert_info->verts);
 373    if (free_prim_info) {
 374       FREE(prim_info->primitive_lengths);
 375    }
 376 }
 377
 378
 379 static inline unsigned
 380 prim_type(unsigned prim, unsigned flags)
 381 {
 382    if (flags & DRAW_LINE_LOOP_AS_STRIP)
 383       return PIPE_PRIM_LINE_STRIP;
 384    else
 385       return prim;
 386 }
 387
 388
 389 static void
 390 fetch_pipeline_run(struct draw_pt_middle_end *middle,
 391                    const unsigned *fetch_elts,
 392                    unsigned fetch_count,
 393                    const ushort *draw_elts,
 394                    unsigned draw_count,
 395                    unsigned prim_flags)
 396 {
 397    struct fetch_pipeline_middle_end *fpme = fetch_pipeline_middle_end(middle);
 398    struct draw_fetch_info fetch_info;
 399    struct draw_prim_info prim_info;
 400
 401    fetch_info.linear = FALSE;
 402    fetch_info.start = 0;
 403    fetch_info.elts = fetch_elts;
 404    fetch_info.count = fetch_count;
 405
 406    prim_info.linear = FALSE;
 407    prim_info.start = 0;
 408    prim_info.count = draw_count;
 409    prim_info.elts = draw_elts;
 410    prim_info.prim = prim_type(fpme->input_prim, prim_flags);
 411    prim_info.flags = prim_flags;
 412    prim_info.primitive_count = 1;
 413    prim_info.primitive_lengths = &draw_count;
 414
 415    fetch_pipeline_generic( middle, &fetch_info, &prim_info );
 416 }
 417
 418
 419 static void
 420 fetch_pipeline_linear_run(struct draw_pt_middle_end *middle,
 421                           unsigned start,
 422                           unsigned count,
 423                           unsigned prim_flags)
 424 {
 425    struct fetch_pipeline_middle_end *fpme = fetch_pipeline_middle_end(middle);
 426    struct draw_fetch_info fetch_info;
 427    struct draw_prim_info prim_info;
 428
 429    fetch_info.linear = TRUE;
 430    fetch_info.start = start;
 431    fetch_info.count = count;
 432    fetch_info.elts = NULL;
 433
 434    prim_info.linear = TRUE;
 435    prim_info.start = 0;
 436    prim_info.count = count;
 437    prim_info.elts = NULL;
 438    prim_info.prim = prim_type(fpme->input_prim, prim_flags);
 439    prim_info.flags = prim_flags;
 440    prim_info.primitive_count = 1;
 441    prim_info.primitive_lengths = &count;
 442
 443    fetch_pipeline_generic( middle, &fetch_info, &prim_info );
 444 }
 445
 446
 447
 448 static boolean
 449 fetch_pipeline_linear_run_elts(struct draw_pt_middle_end *middle,
 450                                unsigned start,
 451                                unsigned count,
 452                                const ushort *draw_elts,
 453                                unsigned draw_count,
 454                                unsigned prim_flags )
 455 {
 456    struct fetch_pipeline_middle_end *fpme = fetch_pipeline_middle_end(middle);
 457    struct draw_fetch_info fetch_info;
 458    struct draw_prim_info prim_info;
 459
 460    fetch_info.linear = TRUE;
 461    fetch_info.start = start;
 462    fetch_info.count = count;
 463    fetch_info.elts = NULL;
 464
 465    prim_info.linear = FALSE;
 466    prim_info.start = 0;
 467    prim_info.count = draw_count;
 468    prim_info.elts = draw_elts;
 469    prim_info.prim = prim_type(fpme->input_prim, prim_flags);
 470    prim_info.flags = prim_flags;
 471    prim_info.primitive_count = 1;
 472    prim_info.primitive_lengths = &draw_count;
 473
 474    fetch_pipeline_generic( middle, &fetch_info, &prim_info );
 475
 476    return TRUE;
 477 }
 478
 479
 480 static void
 481 fetch_pipeline_finish( struct draw_pt_middle_end *middle )
 482 {
 483    /* nothing to do */
 484 }
 485
 486
 487 static void
 488 fetch_pipeline_destroy( struct draw_pt_middle_end *middle )
 489 {
 490    struct fetch_pipeline_middle_end *fpme = fetch_pipeline_middle_end(middle);
 491
 492    if (fpme->fetch)
 493       draw_pt_fetch_destroy( fpme->fetch );
 494
 495    if (fpme->emit)
 496       draw_pt_emit_destroy( fpme->emit );
 497
 498    if (fpme->so_emit)
 499       draw_pt_so_emit_destroy( fpme->so_emit );
 500
 501    if (fpme->post_vs)
 502       draw_pt_post_vs_destroy( fpme->post_vs );
 503
 504    FREE(middle);
 505 }
 506
 507
 508 struct draw_pt_middle_end *
 509 draw_pt_fetch_pipeline_or_emit(struct draw_context *draw)
 510 {
 511    struct fetch_pipeline_middle_end *fpme =
 512       CALLOC_STRUCT( fetch_pipeline_middle_end );
 513    if (!fpme)
 514       goto fail;
 515
 516    fpme->base.prepare        = fetch_pipeline_prepare;
 517    fpme->base.bind_parameters  = fetch_pipeline_bind_parameters;
 518    fpme->base.run            = fetch_pipeline_run;
 519    fpme->base.run_linear     = fetch_pipeline_linear_run;
 520    fpme->base.run_linear_elts = fetch_pipeline_linear_run_elts;
 521    fpme->base.finish         = fetch_pipeline_finish;
 522    fpme->base.destroy        = fetch_pipeline_destroy;
 523
 524    fpme->draw = draw;
 525
 526    fpme->fetch = draw_pt_fetch_create( draw );
 527    if (!fpme->fetch)
 528       goto fail;
 529
 530    fpme->post_vs = draw_pt_post_vs_create( draw );
 531    if (!fpme->post_vs)
 532       goto fail;
 533
 534    fpme->emit = draw_pt_emit_create( draw );
 535    if (!fpme->emit)
 536       goto fail;
 537
 538    fpme->so_emit = draw_pt_so_emit_create( draw );
 539    if (!fpme->so_emit)
 540       goto fail;
 541
 542    return &fpme->base;
 543
 544  fail:
 545    if (fpme)
 546       fetch_pipeline_destroy( &fpme->base );
 547
 548    return NULL;
 549 }