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