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 "pipe/p_util.h"
  35 #include "draw/draw_context.h"
  36 #include "draw/draw_private.h"
  37 #include "draw/draw_vbuf.h"
  38 #include "draw/draw_vertex.h"
  39 #include "draw/draw_pt.h"
  40 #include "draw/draw_vs.h"
  41
  42 #include "translate/translate.h"
  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 {
  75    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
  76    struct draw_context *draw = fse->draw;
  77    unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
  78    const struct vertex_info *vinfo;
  79    unsigned i;
  80    boolean need_psize = 0;
  81
  82
  83    if (!draw->render->set_primitive( draw->render,
  84                                      prim )) {
  85       assert(0);
  86       return;
  87    }
  88
  89    /* Must do this after set_primitive() above:
  90     */
  91    fse->vinfo = vinfo = draw->render->get_vertex_info(draw->render);
  92
  93
  94
  95    fse->key.output_stride = vinfo->size * 4;
  96    fse->key.nr_outputs = vinfo->num_attribs;
  97    fse->key.nr_inputs = num_vs_inputs;
  98
  99    fse->key.nr_elements = MAX2(fse->key.nr_outputs,     /* outputs - translate to hw format */
 100                                fse->key.nr_inputs);     /* inputs - fetch from api format */
 101
 102    fse->key.viewport = !draw->identity_viewport;
 103    fse->key.clip = !draw->bypass_clipping;
 104    fse->key.pad = 0;
 105
 106    memset(fse->key.element, 0,
 107           fse->key.nr_elements * sizeof(fse->key.element[0]));
 108
 109    for (i = 0; i < num_vs_inputs; i++) {
 110       const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
 111       fse->key.element[i].in.format = src->src_format;
 112
 113       /* Consider ignoring these, ie make generated programs
 114        * independent of this state:
 115        */
 116       fse->key.element[i].in.buffer = src->vertex_buffer_index;
 117       fse->key.element[i].in.offset = src->src_offset;
 118    }
 119
 120
 121    {
 122       unsigned dst_offset = 0;
 123
 124       for (i = 0; i < vinfo->num_attribs; i++) {
 125          unsigned emit_sz = 0;
 126          unsigned output_format = PIPE_FORMAT_NONE;
 127          unsigned vs_output = vinfo->src_index[i];
 128
 129          switch (vinfo->emit[i]) {
 130          case EMIT_4F:
 131             output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 132             emit_sz = 4 * sizeof(float);
 133             break;
 134          case EMIT_3F:
 135             output_format = PIPE_FORMAT_R32G32B32_FLOAT;
 136             emit_sz = 3 * sizeof(float);
 137             break;
 138          case EMIT_2F:
 139             output_format = PIPE_FORMAT_R32G32_FLOAT;
 140             emit_sz = 2 * sizeof(float);
 141             break;
 142          case EMIT_1F:
 143             output_format = PIPE_FORMAT_R32_FLOAT;
 144             emit_sz = 1 * sizeof(float);
 145             break;
 146          case EMIT_1F_PSIZE:
 147             need_psize = 1;
 148             output_format = PIPE_FORMAT_R32_FLOAT;
 149             emit_sz = 1 * sizeof(float);
 150             vs_output = vinfo->num_attribs + 1;
 151             break;
 152          case EMIT_4UB:
 153             output_format = PIPE_FORMAT_B8G8R8A8_UNORM;
 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[vs_output].out.format = output_format;
 166          fse->key.element[vs_output].out.vs_output = vs_output;
 167          fse->key.element[vs_output].out.offset = dst_offset;
 168
 169          dst_offset += emit_sz;
 170          assert(fse->key.output_stride >= dst_offset);
 171       }
 172    }
 173
 174    /* To make psize work, really need to tell the vertex shader to
 175     * copy that value from input->output.  For 'translate' this was
 176     * implicit for all elements.
 177     */
 178 #if 0
 179    if (need_psize) {
 180       unsigned input = num_vs_inputs + 1;
 181       const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
 182       fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT;
 183       fse->key.element[i].input_buffer = 0; //nr_buffers + 1;
 184       fse->key.element[i].input_offset = 0;
 185
 186       fse->key.nr_inputs += 1;
 187       fse->key.nr_elements = MAX2(fse->key.nr_inputs,
 188                                   fse->key.nr_outputs);
 189
 190    }
 191 #endif
 192
 193    /* Would normally look up a vertex shader and peruse its list of
 194     * varients somehow.  We omitted that step and put all the
 195     * hardcoded "shaders" into an array.  We're just making the
 196     * assumption that this happens to be a matching shader...  ie
 197     * you're running isosurf, aren't you?
 198     */
 199    fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader,
 200                                          &fse->key );
 201
 202    if (!fse->active) {
 203       assert(0);
 204       return ;
 205    }
 206
 207    /* Now set buffer pointers:
 208     */
 209    for (i = 0; i < num_vs_inputs; i++) {
 210       unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index;
 211
 212       fse->active->set_input( fse->active,
 213                               i,
 214
 215                               ((const ubyte *) draw->pt.user.vbuffer[buf] +
 216                                draw->pt.vertex_buffer[buf].buffer_offset),
 217
 218                               draw->pt.vertex_buffer[buf].pitch );
 219    }
 220
 221    fse->active->set_constants( fse->active,
 222                                (const float (*)[4])draw->pt.user.constants );
 223
 224    fse->active->set_viewport( fse->active,
 225                               &draw->viewport );
 226
 227    //return TRUE;
 228 }
 229
 230
 231
 232
 233
 234
 235
 236 static void fse_run_linear( struct draw_pt_middle_end *middle,
 237                             unsigned start,
 238                             unsigned count )
 239 {
 240    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 241    struct draw_context *draw = fse->draw;
 242    unsigned alloc_count = align(count, 4);
 243    char *hw_verts;
 244
 245    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 246     */
 247    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 248
 249    hw_verts = draw->render->allocate_vertices( draw->render,
 250                                                (ushort)fse->key.output_stride,
 251                                                (ushort)alloc_count );
 252
 253    if (!hw_verts) {
 254       assert(0);
 255       return;
 256    }
 257
 258    /* Single routine to fetch vertices, run shader and emit HW verts.
 259     * Clipping is done elsewhere -- either by the API or on hardware,
 260     * or for some other reason not required...
 261     */
 262    fse->active->run_linear( fse->active,
 263                             start, count,
 264                             hw_verts );
 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    if (0) {
 274       unsigned i;
 275       for (i = 0; i < count; i++) {
 276          debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n", __FUNCTION__, i,
 277                       fse->key.output_stride,
 278                       fse->key.output_stride * i);
 279
 280          draw_dump_emitted_vertex( fse->vinfo,
 281                                    (const uint8_t *)hw_verts + fse->key.output_stride * i );
 282       }
 283    }
 284
 285
 286    draw->render->release_vertices( draw->render,
 287                                    hw_verts,
 288                                    fse->key.output_stride,
 289                                    count );
 290 }
 291
 292
 293 static void
 294 fse_run(struct draw_pt_middle_end *middle,
 295         const unsigned *fetch_elts,
 296         unsigned fetch_count,
 297         const ushort *draw_elts,
 298         unsigned draw_count )
 299 {
 300    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 301    struct draw_context *draw = fse->draw;
 302    unsigned alloc_count = align(fetch_count, 4);
 303    void *hw_verts;
 304
 305    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 306     */
 307    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 308
 309    hw_verts = draw->render->allocate_vertices( draw->render,
 310                                                (ushort)fse->key.output_stride,
 311                                                (ushort)alloc_count );
 312    if (!hw_verts) {
 313       assert(0);
 314       return;
 315    }
 316
 317
 318    /* Single routine to fetch vertices, run shader and emit HW verts.
 319     */
 320    fse->active->run_elts( fse->active,
 321                           fetch_elts,
 322                           fetch_count,
 323                           hw_verts );
 324
 325    draw->render->draw( draw->render,
 326                        draw_elts,
 327                        draw_count );
 328
 329    if (0) {
 330       unsigned i;
 331       for (i = 0; i < fetch_count; i++) {
 332          debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i);
 333          draw_dump_emitted_vertex( fse->vinfo,
 334                                    (const uint8_t *)hw_verts +
 335                                    fse->key.output_stride * i );
 336       }
 337    }
 338
 339
 340    draw->render->release_vertices( draw->render,
 341                                    hw_verts,
 342                                    fse->key.output_stride,
 343                                    fetch_count );
 344
 345 }
 346
 347
 348 static void fse_finish( struct draw_pt_middle_end *middle )
 349 {
 350 }
 351
 352
 353 static void
 354 fse_destroy( struct draw_pt_middle_end *middle )
 355 {
 356    FREE(middle);
 357 }
 358
 359 struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw )
 360 {
 361    struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
 362    if (!fse)
 363       return NULL;
 364
 365    fse->base.prepare = fse_prepare;
 366    fse->base.run = fse_run;
 367    fse->base.run_linear = fse_run_linear;
 368    fse->base.finish = fse_finish;
 369    fse->base.destroy = fse_destroy;
 370    fse->draw = draw;
 371
 372    return &fse->base;
 373 }