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 #include "translate/translate.h"
  44
  45 struct fetch_shade_emit;
  46
  47
  48 /* Prototype fetch, shade, emit-hw-verts all in one go.
  49  */
  50 struct fetch_shade_emit {
  51    struct draw_pt_middle_end base;
  52    struct draw_context *draw;
  53
  54
  55    /* Temporaries:
  56     */
  57    const float *constants;
  58    unsigned pitch[PIPE_MAX_ATTRIBS];
  59    const ubyte *src[PIPE_MAX_ATTRIBS];
  60    unsigned prim;
  61
  62    struct draw_vs_varient_key key;
  63    struct draw_vs_varient *active;
  64
  65
  66    const struct vertex_info *vinfo;
  67 };
  68
  69
  70
  71
  72 static void fse_prepare( struct draw_pt_middle_end *middle,
  73                          unsigned prim,
  74                          unsigned opt,
  75                          unsigned *max_vertices )
  76 {
  77    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
  78    struct draw_context *draw = fse->draw;
  79    unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
  80    const struct vertex_info *vinfo;
  81    unsigned i;
  82    unsigned nr_vbs = 0;
  83
  84
  85    if (!draw->render->set_primitive( draw->render,
  86                                      prim )) {
  87       assert(0);
  88       return;
  89    }
  90
  91    /* Must do this after set_primitive() above:
  92     */
  93    fse->vinfo = vinfo = draw->render->get_vertex_info(draw->render);
  94
  95
  96
  97    fse->key.output_stride = vinfo->size * 4;
  98    fse->key.nr_outputs = vinfo->num_attribs;
  99    fse->key.nr_inputs = num_vs_inputs;
 100
 101    fse->key.nr_elements = MAX2(fse->key.nr_outputs,     /* outputs - translate to hw format */
 102                                fse->key.nr_inputs);     /* inputs - fetch from api format */
 103
 104    fse->key.viewport = (!draw->rasterizer->bypass_vs_clip_and_viewport &&
 105                         !draw->identity_viewport);
 106    fse->key.clip = !draw->bypass_clipping;
 107    fse->key.const_vbuffers = 0;
 108
 109    memset(fse->key.element, 0,
 110           fse->key.nr_elements * sizeof(fse->key.element[0]));
 111
 112    for (i = 0; i < num_vs_inputs; i++) {
 113       const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
 114       fse->key.element[i].in.format = src->src_format;
 115
 116       /* Consider ignoring these, ie make generated programs
 117        * independent of this state:
 118        */
 119       fse->key.element[i].in.buffer = src->vertex_buffer_index;
 120       fse->key.element[i].in.offset = src->src_offset;
 121       nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1);
 122    }
 123
 124    for (i = 0; i < 5 && i < nr_vbs; i++) {
 125       if (draw->pt.vertex_buffer[i].stride == 0)
 126          fse->key.const_vbuffers |= (1<<i);
 127    }
 128
 129    if (0) debug_printf("%s: lookup const_vbuffers: %x\n", __FUNCTION__, fse->key.const_vbuffers);
 130
 131    {
 132       unsigned dst_offset = 0;
 133
 134       for (i = 0; i < vinfo->num_attribs; i++) {
 135          unsigned emit_sz = 0;
 136
 137          switch (vinfo->attrib[i].emit) {
 138          case EMIT_4F:
 139             emit_sz = 4 * sizeof(float);
 140             break;
 141          case EMIT_3F:
 142             emit_sz = 3 * sizeof(float);
 143             break;
 144          case EMIT_2F:
 145             emit_sz = 2 * sizeof(float);
 146             break;
 147          case EMIT_1F:
 148             emit_sz = 1 * sizeof(float);
 149             break;
 150          case EMIT_1F_PSIZE:
 151             emit_sz = 1 * sizeof(float);
 152             break;
 153          case EMIT_4UB:
 154             emit_sz = 4 * sizeof(ubyte);
 155             break;
 156          default:
 157             assert(0);
 158             break;
 159          }
 160
 161          /* The elements in the key correspond to vertex shader output
 162           * numbers, not to positions in the hw vertex description --
 163           * that's handled by the output_offset field.
 164           */
 165          fse->key.element[i].out.format = vinfo->attrib[i].emit;
 166          fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index;
 167          fse->key.element[i].out.offset = dst_offset;
 168
 169          dst_offset += emit_sz;
 170          assert(fse->key.output_stride >= dst_offset);
 171       }
 172    }
 173
 174
 175    fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader,
 176                                          &fse->key );
 177
 178    if (!fse->active) {
 179       assert(0);
 180       return ;
 181    }
 182
 183    if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__,
 184                        fse->active->key.const_vbuffers);
 185
 186    /* Now set buffer pointers:
 187     */
 188    for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
 189       fse->active->set_buffer( fse->active,
 190                                i,
 191                                ((const ubyte *) draw->pt.user.vbuffer[i] +
 192                                 draw->pt.vertex_buffer[i].buffer_offset),
 193                               draw->pt.vertex_buffer[i].stride );
 194    }
 195
 196    *max_vertices = (draw->render->max_vertex_buffer_bytes /
 197                     (vinfo->size * 4));
 198
 199    /* Return an even number of verts.
 200     * This prevents "parity" errors when splitting long triangle strips which
 201     * can lead to front/back culling mix-ups.
 202     * Every other triangle in a strip has an alternate front/back orientation
 203     * so splitting at an odd position can cause the orientation of subsequent
 204     * triangles to get reversed.
 205     */
 206    *max_vertices = *max_vertices & ~1;
 207
 208    /* Probably need to do this somewhere (or fix exec shader not to
 209     * need it):
 210     */
 211    if (1) {
 212       struct draw_vertex_shader *vs = draw->vs.vertex_shader;
 213       vs->prepare(vs, draw);
 214    }
 215 }
 216
 217
 218
 219 static void fse_run_linear( struct draw_pt_middle_end *middle,
 220                             unsigned start,
 221                             unsigned count )
 222 {
 223    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 224    struct draw_context *draw = fse->draw;
 225    char *hw_verts;
 226
 227    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 228     */
 229    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 230
 231    if (count >= UNDEFINED_VERTEX_ID)
 232       goto fail;
 233
 234    if (!draw->render->allocate_vertices( draw->render,
 235                                          (ushort)fse->key.output_stride,
 236                                          (ushort)count ))
 237       goto fail;
 238
 239    hw_verts = draw->render->map_vertices( draw->render );
 240    if (!hw_verts)
 241       goto fail;
 242
 243    /* Single routine to fetch vertices, run shader and emit HW verts.
 244     * Clipping is done elsewhere -- either by the API or on hardware,
 245     * or for some other reason not required...
 246     */
 247    fse->active->run_linear( fse->active,
 248                             start, count,
 249                             hw_verts );
 250
 251
 252    if (0) {
 253       unsigned i;
 254       for (i = 0; i < count; i++) {
 255          debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n", __FUNCTION__, i,
 256                       fse->key.output_stride,
 257                       fse->key.output_stride * i);
 258
 259          draw_dump_emitted_vertex( fse->vinfo,
 260                                    (const uint8_t *)hw_verts + fse->key.output_stride * i );
 261       }
 262    }
 263
 264    draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
 265
 266    /* Draw arrays path to avoid re-emitting index list again and
 267     * again.
 268     */
 269    draw->render->draw_arrays( draw->render,
 270                               0,
 271                               count );
 272
 273
 274    draw->render->release_vertices( draw->render );
 275
 276    return;
 277
 278 fail:
 279    assert(0);
 280    return;
 281 }
 282
 283
 284 static void
 285 fse_run(struct draw_pt_middle_end *middle,
 286         const unsigned *fetch_elts,
 287         unsigned fetch_count,
 288         const ushort *draw_elts,
 289         unsigned draw_count )
 290 {
 291    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 292    struct draw_context *draw = fse->draw;
 293    void *hw_verts;
 294
 295    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 296     */
 297    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 298
 299    if (fetch_count >= UNDEFINED_VERTEX_ID)
 300       goto fail;
 301
 302    if (!draw->render->allocate_vertices( draw->render,
 303                                          (ushort)fse->key.output_stride,
 304                                          (ushort)fetch_count ))
 305       goto fail;
 306
 307    hw_verts = draw->render->map_vertices( draw->render );
 308    if (!hw_verts)
 309       goto fail;
 310
 311
 312    /* Single routine to fetch vertices, run shader and emit HW verts.
 313     */
 314    fse->active->run_elts( fse->active,
 315                           fetch_elts,
 316                           fetch_count,
 317                           hw_verts );
 318
 319
 320    if (0) {
 321       unsigned i;
 322       for (i = 0; i < fetch_count; i++) {
 323          debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i);
 324          draw_dump_emitted_vertex( fse->vinfo,
 325                                    (const uint8_t *)hw_verts +
 326                                    fse->key.output_stride * i );
 327       }
 328    }
 329
 330    draw->render->unmap_vertices( draw->render, 0, (ushort)(fetch_count - 1) );
 331
 332    draw->render->draw( draw->render,
 333                        draw_elts,
 334                        draw_count );
 335
 336
 337    draw->render->release_vertices( draw->render );
 338    return;
 339
 340 fail:
 341    assert(0);
 342    return;
 343 }
 344
 345
 346
 347 static boolean fse_run_linear_elts( struct draw_pt_middle_end *middle,
 348                                  unsigned start,
 349                                  unsigned count,
 350                                  const ushort *draw_elts,
 351                                  unsigned draw_count )
 352 {
 353    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 354    struct draw_context *draw = fse->draw;
 355    char *hw_verts;
 356
 357    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 358     */
 359    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 360
 361    if (count >= UNDEFINED_VERTEX_ID)
 362       return FALSE;
 363
 364    if (!draw->render->allocate_vertices( draw->render,
 365                                          (ushort)fse->key.output_stride,
 366                                          (ushort)count ))
 367       return FALSE;
 368
 369    hw_verts = draw->render->map_vertices( draw->render );
 370    if (!hw_verts)
 371       return FALSE;
 372
 373    /* Single routine to fetch vertices, run shader and emit HW verts.
 374     * Clipping is done elsewhere -- either by the API or on hardware,
 375     * or for some other reason not required...
 376     */
 377    fse->active->run_linear( fse->active,
 378                             start, count,
 379                             hw_verts );
 380
 381
 382    draw->render->draw( draw->render,
 383                        draw_elts,
 384                        draw_count );
 385
 386
 387    draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
 388
 389    draw->render->release_vertices( draw->render );
 390
 391    return TRUE;
 392 }
 393
 394
 395
 396 static void fse_finish( struct draw_pt_middle_end *middle )
 397 {
 398 }
 399
 400
 401 static void
 402 fse_destroy( struct draw_pt_middle_end *middle )
 403 {
 404    FREE(middle);
 405 }
 406
 407 struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw )
 408 {
 409    struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
 410    if (!fse)
 411       return NULL;
 412
 413    fse->base.prepare = fse_prepare;
 414    fse->base.run = fse_run;
 415    fse->base.run_linear = fse_run_linear;
 416    fse->base.run_linear_elts = fse_run_linear_elts;
 417    fse->base.finish = fse_finish;
 418    fse->base.destroy = fse_destroy;
 419    fse->draw = draw;
 420
 421    return &fse->base;
 422 }