src/gallium/auxiliary/draw/draw_pt.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  /*
  29   * Authors:
  30   *   Keith Whitwell <keithw@vmware.com>
  31   */
  32
  33 #include "draw/draw_context.h"
  34 #include "draw/draw_gs.h"
  35 #include "draw/draw_tess.h"
  36 #include "draw/draw_private.h"
  37 #include "draw/draw_pt.h"
  38 #include "draw/draw_vbuf.h"
  39 #include "draw/draw_vs.h"
  40 #include "tgsi/tgsi_dump.h"
  41 #include "util/u_math.h"
  42 #include "util/u_prim.h"
  43 #include "util/format/u_format.h"
  44 #include "util/u_draw.h"
  45
  46
  47 DEBUG_GET_ONCE_BOOL_OPTION(draw_fse, "DRAW_FSE", FALSE)
  48 DEBUG_GET_ONCE_BOOL_OPTION(draw_no_fse, "DRAW_NO_FSE", FALSE)
  49
  50 /* Overall we split things into:
  51  *     - frontend -- prepare fetch_elts, draw_elts - eg vsplit
  52  *     - middle   -- fetch, shade, cliptest, viewport
  53  *     - pipeline -- the prim pipeline: clipping, wide lines, etc
  54  *     - backend  -- the vbuf_render provided by the driver.
  55  */
  56 static boolean
  57 draw_pt_arrays(struct draw_context *draw,
  58                unsigned prim,
  59                unsigned start,
  60                unsigned count)
  61 {
  62    struct draw_pt_front_end *frontend = NULL;
  63    struct draw_pt_middle_end *middle = NULL;
  64    unsigned opt = 0;
  65
  66    /* Sanitize primitive length:
  67     */
  68    {
  69       unsigned first, incr;
  70
  71       if (prim == PIPE_PRIM_PATCHES) {
  72          first = draw->pt.vertices_per_patch;
  73          incr = draw->pt.vertices_per_patch;
  74       } else
  75          draw_pt_split_prim(prim, &first, &incr);
  76       count = draw_pt_trim_count(count, first, incr);
  77       if (count < first)
  78          return TRUE;
  79    }
  80
  81    if (!draw->force_passthrough) {
  82       unsigned out_prim = prim;
  83
  84       if (draw->gs.geometry_shader)
  85          out_prim = draw->gs.geometry_shader->output_primitive;
  86       else if (draw->tes.tess_eval_shader)
  87          out_prim = get_tes_output_prim(draw->tes.tess_eval_shader);
  88
  89       if (!draw->render) {
  90          opt |= PT_PIPELINE;
  91       }
  92
  93       if (draw_need_pipeline(draw,
  94                              draw->rasterizer,
  95                              out_prim)) {
  96          opt |= PT_PIPELINE;
  97       }
  98
  99       if ((draw->clip_xy ||
 100            draw->clip_z ||
 101            draw->clip_user) && !draw->pt.test_fse) {
 102          opt |= PT_CLIPTEST;
 103       }
 104
 105       opt |= PT_SHADE;
 106    }
 107
 108    if (draw->pt.middle.llvm) {
 109       middle = draw->pt.middle.llvm;
 110    } else {
 111       if (opt == 0)
 112          middle = draw->pt.middle.fetch_emit;
 113       else if (opt == PT_SHADE && !draw->pt.no_fse)
 114          middle = draw->pt.middle.fetch_shade_emit;
 115       else
 116          middle = draw->pt.middle.general;
 117    }
 118
 119    frontend = draw->pt.frontend;
 120
 121    if (frontend) {
 122       if (draw->pt.prim != prim || draw->pt.opt != opt) {
 123          /* In certain conditions switching primitives requires us to flush
 124           * and validate the different stages. One example is when smooth
 125           * lines are active but first drawn with triangles and then with
 126           * lines.
 127           */
 128          draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
 129          frontend = NULL;
 130       } else if (draw->pt.eltSize != draw->pt.user.eltSize) {
 131          /* Flush draw state if eltSize changed.
 132           * This could be improved so only the frontend is flushed since it
 133           * converts all indices to ushorts and the fetch part of the middle
 134           * always prepares both linear and indexed.
 135           */
 136          frontend->flush( frontend, DRAW_FLUSH_STATE_CHANGE );
 137          frontend = NULL;
 138       }
 139    }
 140
 141    if (!frontend) {
 142       frontend = draw->pt.front.vsplit;
 143
 144       frontend->prepare( frontend, prim, middle, opt );
 145
 146       draw->pt.frontend = frontend;
 147       draw->pt.eltSize = draw->pt.user.eltSize;
 148       draw->pt.prim = prim;
 149       draw->pt.opt = opt;
 150    }
 151
 152    if (draw->pt.rebind_parameters) {
 153       /* update constants, viewport dims, clip planes, etc */
 154       middle->bind_parameters(middle);
 155       draw->pt.rebind_parameters = FALSE;
 156    }
 157
 158    frontend->run( frontend, start, count );
 159
 160    return TRUE;
 161 }
 162
 163 void draw_pt_flush( struct draw_context *draw, unsigned flags )
 164 {
 165    assert(flags);
 166
 167    if (draw->pt.frontend) {
 168       draw->pt.frontend->flush( draw->pt.frontend, flags );
 169
 170       /* don't prepare if we only are flushing the backend */
 171       if (flags & DRAW_FLUSH_STATE_CHANGE)
 172          draw->pt.frontend = NULL;
 173    }
 174
 175    if (flags & DRAW_FLUSH_PARAMETER_CHANGE) {
 176       draw->pt.rebind_parameters = TRUE;
 177    }
 178 }
 179
 180
 181
 182 boolean draw_pt_init( struct draw_context *draw )
 183 {
 184    draw->pt.test_fse = debug_get_option_draw_fse();
 185    draw->pt.no_fse = debug_get_option_draw_no_fse();
 186
 187    draw->pt.front.vsplit = draw_pt_vsplit(draw);
 188    if (!draw->pt.front.vsplit)
 189       return FALSE;
 190
 191    draw->pt.middle.fetch_emit = draw_pt_fetch_emit( draw );
 192    if (!draw->pt.middle.fetch_emit)
 193       return FALSE;
 194
 195    draw->pt.middle.fetch_shade_emit = draw_pt_middle_fse( draw );
 196    if (!draw->pt.middle.fetch_shade_emit)
 197       return FALSE;
 198
 199    draw->pt.middle.general = draw_pt_fetch_pipeline_or_emit( draw );
 200    if (!draw->pt.middle.general)
 201       return FALSE;
 202
 203 #ifdef LLVM_AVAILABLE
 204    if (draw->llvm)
 205       draw->pt.middle.llvm = draw_pt_fetch_pipeline_or_emit_llvm( draw );
 206 #endif
 207
 208    return TRUE;
 209 }
 210
 211
 212 void draw_pt_destroy( struct draw_context *draw )
 213 {
 214    if (draw->pt.middle.llvm) {
 215       draw->pt.middle.llvm->destroy( draw->pt.middle.llvm );
 216       draw->pt.middle.llvm = NULL;
 217    }
 218
 219    if (draw->pt.middle.general) {
 220       draw->pt.middle.general->destroy( draw->pt.middle.general );
 221       draw->pt.middle.general = NULL;
 222    }
 223
 224    if (draw->pt.middle.fetch_emit) {
 225       draw->pt.middle.fetch_emit->destroy( draw->pt.middle.fetch_emit );
 226       draw->pt.middle.fetch_emit = NULL;
 227    }
 228
 229    if (draw->pt.middle.fetch_shade_emit) {
 230       draw->pt.middle.fetch_shade_emit->destroy( draw->pt.middle.fetch_shade_emit );
 231       draw->pt.middle.fetch_shade_emit = NULL;
 232    }
 233
 234    if (draw->pt.front.vsplit) {
 235       draw->pt.front.vsplit->destroy( draw->pt.front.vsplit );
 236       draw->pt.front.vsplit = NULL;
 237    }
 238 }
 239
 240
 241 /**
 242  * Debug- print the first 'count' vertices.
 243  */
 244 static void
 245 draw_print_arrays(struct draw_context *draw, uint prim, int start, uint count)
 246 {
 247    uint i;
 248
 249    debug_printf("Draw arrays(prim = %u, start = %u, count = %u)\n",
 250                 prim, start, count);
 251
 252    for (i = 0; i < count; i++) {
 253       uint ii = 0;
 254       uint j;
 255
 256       if (draw->pt.user.eltSize) {
 257          /* indexed arrays */
 258
 259          switch (draw->pt.user.eltSize) {
 260          case 1:
 261             {
 262                const ubyte *elem = (const ubyte *) draw->pt.user.elts;
 263                ii = elem[start + i];
 264             }
 265             break;
 266          case 2:
 267             {
 268                const ushort *elem = (const ushort *) draw->pt.user.elts;
 269                ii = elem[start + i];
 270             }
 271             break;
 272          case 4:
 273             {
 274                const uint *elem = (const uint *) draw->pt.user.elts;
 275                ii = elem[start + i];
 276             }
 277             break;
 278          default:
 279             assert(0);
 280             return;
 281          }
 282          ii += draw->pt.user.eltBias;
 283          debug_printf("Element[%u + %u] + %i -> Vertex %u:\n", start, i,
 284                       draw->pt.user.eltBias, ii);
 285       }
 286       else {
 287          /* non-indexed arrays */
 288          ii = start + i;
 289          debug_printf("Vertex %u:\n", ii);
 290       }
 291
 292       for (j = 0; j < draw->pt.nr_vertex_elements; j++) {
 293          uint buf = draw->pt.vertex_element[j].vertex_buffer_index;
 294          ubyte *ptr = (ubyte *) draw->pt.user.vbuffer[buf].map;
 295
 296          if (draw->pt.vertex_element[j].instance_divisor) {
 297             ii = draw->instance_id / draw->pt.vertex_element[j].instance_divisor;
 298          }
 299
 300          ptr += draw->pt.vertex_buffer[buf].buffer_offset;
 301          ptr += draw->pt.vertex_buffer[buf].stride * ii;
 302          ptr += draw->pt.vertex_element[j].src_offset;
 303
 304          debug_printf("  Attr %u: ", j);
 305          switch (draw->pt.vertex_element[j].src_format) {
 306          case PIPE_FORMAT_R32_FLOAT:
 307             {
 308                float *v = (float *) ptr;
 309                debug_printf("R %f  @ %p\n", v[0], (void *) v);
 310             }
 311             break;
 312          case PIPE_FORMAT_R32G32_FLOAT:
 313             {
 314                float *v = (float *) ptr;
 315                debug_printf("RG %f %f  @ %p\n", v[0], v[1], (void *) v);
 316             }
 317             break;
 318          case PIPE_FORMAT_R32G32B32_FLOAT:
 319             {
 320                float *v = (float *) ptr;
 321                debug_printf("RGB %f %f %f  @ %p\n", v[0], v[1], v[2], (void *) v);
 322             }
 323             break;
 324          case PIPE_FORMAT_R32G32B32A32_FLOAT:
 325             {
 326                float *v = (float *) ptr;
 327                debug_printf("RGBA %f %f %f %f  @ %p\n", v[0], v[1], v[2], v[3],
 328                             (void *) v);
 329             }
 330             break;
 331          case PIPE_FORMAT_B8G8R8A8_UNORM:
 332             {
 333                ubyte *u = (ubyte *) ptr;
 334                debug_printf("BGRA %d %d %d %d  @ %p\n", u[0], u[1], u[2], u[3],
 335                             (void *) u);
 336             }
 337             break;
 338          case PIPE_FORMAT_A8R8G8B8_UNORM:
 339             {
 340                ubyte *u = (ubyte *) ptr;
 341                debug_printf("ARGB %d %d %d %d  @ %p\n", u[0], u[1], u[2], u[3],
 342                             (void *) u);
 343             }
 344             break;
 345          default:
 346             debug_printf("other format %s (fix me)\n",
 347                      util_format_name(draw->pt.vertex_element[j].src_format));
 348          }
 349       }
 350    }
 351 }
 352
 353
 354 /** Helper code for below */
 355 #define PRIM_RESTART_LOOP(elements) \
 356    do { \
 357       for (j = 0; j < count; j++) {               \
 358          i = draw_overflow_uadd(start, j, MAX_LOOP_IDX);  \
 359          if (i < elt_max && elements[i] == info->restart_index) { \
 360             if (cur_count > 0) { \
 361                /* draw elts up to prev pos */ \
 362                draw_pt_arrays(draw, prim, cur_start, cur_count); \
 363             } \
 364             /* begin new prim at next elt */ \
 365             cur_start = i + 1; \
 366             cur_count = 0; \
 367          } \
 368          else { \
 369             cur_count++; \
 370          } \
 371       } \
 372       if (cur_count > 0) { \
 373          draw_pt_arrays(draw, prim, cur_start, cur_count); \
 374       } \
 375    } while (0)
 376
 377
 378 /**
 379  * For drawing prims with primitive restart enabled.
 380  * Scan for restart indexes and draw the runs of elements/vertices between
 381  * the restarts.
 382  */
 383 static void
 384 draw_pt_arrays_restart(struct draw_context *draw,
 385                        const struct pipe_draw_info *info)
 386 {
 387    const unsigned prim = info->mode;
 388    const unsigned start = info->start;
 389    const unsigned count = info->count;
 390    const unsigned elt_max = draw->pt.user.eltMax;
 391    unsigned i, j, cur_start, cur_count;
 392    /* The largest index within a loop using the i variable as the index.
 393     * Used for overflow detection */
 394    const unsigned MAX_LOOP_IDX = 0xffffffff;
 395
 396    assert(info->primitive_restart);
 397
 398    if (draw->pt.user.eltSize) {
 399       /* indexed prims (draw_elements) */
 400       cur_start = start;
 401       cur_count = 0;
 402
 403       switch (draw->pt.user.eltSize) {
 404       case 1:
 405          {
 406             const ubyte *elt_ub = (const ubyte *) draw->pt.user.elts;
 407             PRIM_RESTART_LOOP(elt_ub);
 408          }
 409          break;
 410       case 2:
 411          {
 412             const ushort *elt_us = (const ushort *) draw->pt.user.elts;
 413             PRIM_RESTART_LOOP(elt_us);
 414          }
 415          break;
 416       case 4:
 417          {
 418             const uint *elt_ui = (const uint *) draw->pt.user.elts;
 419             PRIM_RESTART_LOOP(elt_ui);
 420          }
 421          break;
 422       default:
 423          assert(0 && "bad eltSize in draw_arrays()");
 424       }
 425    }
 426    else {
 427       /* Non-indexed prims (draw_arrays).
 428        * Primitive restart should have been handled in the state tracker.
 429        */
 430       draw_pt_arrays(draw, prim, start, count);
 431    }
 432 }
 433
 434
 435 /**
 436  * Resolve true values within pipe_draw_info.
 437  * If we're rendering from transform feedback/stream output
 438  * buffers both the count and max_index need to be computed
 439  * from the attached stream output target.
 440  */
 441 static void
 442 resolve_draw_info(const struct pipe_draw_info *raw_info,
 443                   struct pipe_draw_info *info,
 444                   struct pipe_vertex_buffer *vertex_buffer)
 445 {
 446    memcpy(info, raw_info, sizeof(struct pipe_draw_info));
 447
 448    if (raw_info->count_from_stream_output) {
 449       struct draw_so_target *target =
 450          (struct draw_so_target *)info->count_from_stream_output;
 451       assert(vertex_buffer != NULL);
 452       info->count = vertex_buffer->stride == 0 ? 0 :
 453                        target->internal_offset / vertex_buffer->stride;
 454
 455       /* Stream output draw can not be indexed */
 456       debug_assert(!info->index_size);
 457       info->max_index = info->count - 1;
 458    }
 459 }
 460
 461 /**
 462  * Draw vertex arrays.
 463  * This is the main entrypoint into the drawing module.  If drawing an indexed
 464  * primitive, the draw_set_indexes() function should have already been called
 465  * to specify the element/index buffer information.
 466  */
 467 void
 468 draw_vbo(struct draw_context *draw,
 469          const struct pipe_draw_info *info)
 470 {
 471    unsigned instance;
 472    unsigned index_limit;
 473    unsigned count;
 474    unsigned fpstate = util_fpstate_get();
 475    struct pipe_draw_info resolved_info;
 476
 477    if (info->instance_count == 0)
 478       return;
 479
 480    /* Make sure that denorms are treated like zeros. This is
 481     * the behavior required by D3D10. OpenGL doesn't care.
 482     */
 483    util_fpstate_set_denorms_to_zero(fpstate);
 484
 485    resolve_draw_info(info, &resolved_info, &(draw->pt.vertex_buffer[0]));
 486    info = &resolved_info;
 487
 488    if (info->index_size)
 489       assert(draw->pt.user.elts);
 490
 491    count = info->count;
 492
 493    draw->pt.user.eltBias = info->index_bias;
 494    draw->pt.user.min_index = info->min_index;
 495    draw->pt.user.max_index = info->max_index;
 496    draw->pt.user.eltSize = info->index_size ? draw->pt.user.eltSizeIB : 0;
 497    draw->pt.user.drawid = info->drawid;
 498
 499    draw->pt.vertices_per_patch = info->vertices_per_patch;
 500
 501    if (0)
 502       debug_printf("draw_vbo(mode=%u start=%u count=%u):\n",
 503                    info->mode, info->start, count);
 504
 505    if (0)
 506       tgsi_dump(draw->vs.vertex_shader->state.tokens, 0);
 507
 508    if (0) {
 509       unsigned int i;
 510       debug_printf("Elements:\n");
 511       for (i = 0; i < draw->pt.nr_vertex_elements; i++) {
 512          debug_printf("  %u: src_offset=%u  inst_div=%u   vbuf=%u  format=%s\n",
 513                       i,
 514                       draw->pt.vertex_element[i].src_offset,
 515                       draw->pt.vertex_element[i].instance_divisor,
 516                       draw->pt.vertex_element[i].vertex_buffer_index,
 517                       util_format_name(draw->pt.vertex_element[i].src_format));
 518       }
 519       debug_printf("Buffers:\n");
 520       for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
 521          debug_printf("  %u: stride=%u offset=%u size=%d ptr=%p\n",
 522                       i,
 523                       draw->pt.vertex_buffer[i].stride,
 524                       draw->pt.vertex_buffer[i].buffer_offset,
 525                       (int) draw->pt.user.vbuffer[i].size,
 526                       draw->pt.user.vbuffer[i].map);
 527       }
 528    }
 529
 530    if (0)
 531       draw_print_arrays(draw, info->mode, info->start, MIN2(count, 20));
 532
 533    index_limit = util_draw_max_index(draw->pt.vertex_buffer,
 534                                      draw->pt.vertex_element,
 535                                      draw->pt.nr_vertex_elements,
 536                                      info);
 537 #ifdef LLVM_AVAILABLE
 538    if (!draw->llvm)
 539 #endif
 540    {
 541       if (index_limit == 0) {
 542          /* one of the buffers is too small to do any valid drawing */
 543          debug_warning("draw: VBO too small to draw anything\n");
 544          util_fpstate_set(fpstate);
 545          return;
 546       }
 547    }
 548
 549    /* If we're collecting stats then make sure we start from scratch */
 550    if (draw->collect_statistics) {
 551       memset(&draw->statistics, 0, sizeof(draw->statistics));
 552    }
 553
 554    draw->pt.max_index = index_limit - 1;
 555    draw->start_index = info->start;
 556
 557    /*
 558     * TODO: We could use draw->pt.max_index to further narrow
 559     * the min_index/max_index hints given by the state tracker.
 560     */
 561
 562    for (instance = 0; instance < info->instance_count; instance++) {
 563       unsigned instance_idx = instance + info->start_instance;
 564       draw->start_instance = info->start_instance;
 565       draw->instance_id = instance;
 566       /* check for overflow */
 567       if (instance_idx < instance ||
 568           instance_idx < draw->start_instance) {
 569          /* if we overflown just set the instance id to the max */
 570          draw->instance_id = 0xffffffff;
 571       }
 572
 573       draw_new_instance(draw);
 574
 575       if (info->primitive_restart) {
 576          draw_pt_arrays_restart(draw, info);
 577       }
 578       else {
 579          draw_pt_arrays(draw, info->mode, info->start, count);
 580       }
 581    }
 582
 583    /* If requested emit the pipeline statistics for this run */
 584    if (draw->collect_statistics) {
 585       draw->render->pipeline_statistics(draw->render, &draw->statistics);
 586    }
 587    util_fpstate_set(fpstate);
 588 }