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