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
  66
  67
  68 static void fse_prepare( struct draw_pt_middle_end *middle,
  69                          unsigned prim,
  70                          unsigned opt )
  71 {
  72    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
  73    struct draw_context *draw = fse->draw;
  74    unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
  75    const struct vertex_info *vinfo;
  76    unsigned i;
  77    boolean need_psize = 0;
  78
  79
  80    if (!draw->render->set_primitive( draw->render,
  81                                      prim )) {
  82       assert(0);
  83       return;
  84    }
  85
  86    /* Must do this after set_primitive() above:
  87     */
  88    vinfo = draw->render->get_vertex_info(draw->render);
  89
  90
  91
  92    fse->key.output_stride = vinfo->size * 4;
  93    fse->key.nr_outputs = vinfo->num_attribs;
  94    fse->key.nr_inputs = num_vs_inputs;
  95
  96    fse->key.nr_elements = MAX2(fse->key.nr_outputs,     /* outputs - translate to hw format */
  97                                fse->key.nr_inputs);     /* inputs - fetch from api format */
  98
  99    fse->key.viewport = !draw->identity_viewport;
 100    fse->key.clip = !draw->bypass_clipping;
 101    fse->key.pad = 0;
 102
 103    memset(fse->key.element, 0,
 104           fse->key.nr_elements * sizeof(fse->key.element[0]));
 105
 106    for (i = 0; i < num_vs_inputs; i++) {
 107       const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
 108       fse->key.element[i].in.format = src->src_format;
 109
 110       /* Consider ignoring these, ie make generated programs
 111        * independent of this state:
 112        */
 113       fse->key.element[i].in.buffer = src->vertex_buffer_index;
 114       fse->key.element[i].in.offset = src->src_offset;
 115    }
 116
 117
 118    {
 119       unsigned dst_offset = 0;
 120
 121       for (i = 0; i < vinfo->num_attribs; i++) {
 122          unsigned emit_sz = 0;
 123          unsigned output_format = PIPE_FORMAT_NONE;
 124          unsigned vs_output = vinfo->src_index[i];
 125
 126          switch (vinfo->emit[i]) {
 127          case EMIT_4F:
 128             output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 129             emit_sz = 4 * sizeof(float);
 130             break;
 131          case EMIT_3F:
 132             output_format = PIPE_FORMAT_R32G32B32_FLOAT;
 133             emit_sz = 3 * sizeof(float);
 134             break;
 135          case EMIT_2F:
 136             output_format = PIPE_FORMAT_R32G32_FLOAT;
 137             emit_sz = 2 * sizeof(float);
 138             break;
 139          case EMIT_1F:
 140             output_format = PIPE_FORMAT_R32_FLOAT;
 141             emit_sz = 1 * sizeof(float);
 142             break;
 143          case EMIT_1F_PSIZE:
 144             need_psize = 1;
 145             output_format = PIPE_FORMAT_R32_FLOAT;
 146             emit_sz = 1 * sizeof(float);
 147             vs_output = vinfo->num_attribs + 1;
 148             break;
 149          case EMIT_4UB:
 150             output_format = PIPE_FORMAT_B8G8R8A8_UNORM;
 151             emit_sz = 4 * sizeof(ubyte);
 152             break;
 153          default:
 154             assert(0);
 155             break;
 156          }
 157
 158          /* The elements in the key correspond to vertex shader output
 159           * numbers, not to positions in the hw vertex description --
 160           * that's handled by the output_offset field.
 161           */
 162          fse->key.element[vs_output].out.format = output_format;
 163          fse->key.element[vs_output].out.vs_output = vs_output;
 164          fse->key.element[vs_output].out.offset = dst_offset;
 165
 166          dst_offset += emit_sz;
 167          assert(fse->key.output_stride >= dst_offset);
 168       }
 169    }
 170
 171    /* To make psize work, really need to tell the vertex shader to
 172     * copy that value from input->output.  For 'translate' this was
 173     * implicit for all elements.
 174     */
 175 #if 0
 176    if (need_psize) {
 177       unsigned input = num_vs_inputs + 1;
 178       const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
 179       fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT;
 180       fse->key.element[i].input_buffer = 0; //nr_buffers + 1;
 181       fse->key.element[i].input_offset = 0;
 182
 183       fse->key.nr_inputs += 1;
 184       fse->key.nr_elements = MAX2(fse->key.nr_inputs,
 185                                   fse->key.nr_outputs);
 186
 187    }
 188 #endif
 189
 190    /* Would normally look up a vertex shader and peruse its list of
 191     * varients somehow.  We omitted that step and put all the
 192     * hardcoded "shaders" into an array.  We're just making the
 193     * assumption that this happens to be a matching shader...  ie
 194     * you're running isosurf, aren't you?
 195     */
 196    fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader,
 197                                          &fse->key );
 198
 199    if (!fse->active) {
 200       assert(0);
 201       return ;
 202    }
 203
 204    /* Now set buffer pointers:
 205     */
 206    for (i = 0; i < num_vs_inputs; i++) {
 207       unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index;
 208
 209       fse->active->set_input( fse->active,
 210                               i,
 211
 212                               ((const ubyte *) draw->pt.user.vbuffer[buf] +
 213                                draw->pt.vertex_buffer[buf].buffer_offset),
 214
 215                               draw->pt.vertex_buffer[buf].pitch );
 216    }
 217
 218    fse->active->set_constants( fse->active,
 219                                (const float (*)[4])draw->pt.user.constants );
 220
 221    fse->active->set_viewport( fse->active,
 222                               &draw->viewport );
 223
 224    //return TRUE;
 225 }
 226
 227
 228
 229
 230
 231
 232
 233 static void fse_run_linear( struct draw_pt_middle_end *middle,
 234                             unsigned start,
 235                             unsigned count )
 236 {
 237    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 238    struct draw_context *draw = fse->draw;
 239    unsigned alloc_count = align(count, 4);
 240    char *hw_verts;
 241
 242    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 243     */
 244    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 245
 246    hw_verts = draw->render->allocate_vertices( draw->render,
 247                                                (ushort)fse->key.output_stride,
 248                                                (ushort)alloc_count );
 249
 250    if (!hw_verts) {
 251       assert(0);
 252       return;
 253    }
 254
 255    /* Single routine to fetch vertices, run shader and emit HW verts.
 256     * Clipping is done elsewhere -- either by the API or on hardware,
 257     * or for some other reason not required...
 258     */
 259    fse->active->run_linear( fse->active,
 260                             start, count,
 261                             hw_verts );
 262
 263    /* Draw arrays path to avoid re-emitting index list again and
 264     * again.
 265     */
 266    draw->render->draw_arrays( draw->render,
 267                               0,
 268                               count );
 269
 270
 271    draw->render->release_vertices( draw->render,
 272                                    hw_verts,
 273                                    fse->key.output_stride,
 274                                    count );
 275 }
 276
 277
 278 static void
 279 fse_run(struct draw_pt_middle_end *middle,
 280         const unsigned *fetch_elts,
 281         unsigned fetch_count,
 282         const ushort *draw_elts,
 283         unsigned draw_count )
 284 {
 285    struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
 286    struct draw_context *draw = fse->draw;
 287    unsigned alloc_count = align(fetch_count, 4);
 288    void *hw_verts;
 289
 290    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 291     */
 292    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 293
 294    hw_verts = draw->render->allocate_vertices( draw->render,
 295                                                (ushort)fse->key.output_stride,
 296                                                (ushort)alloc_count );
 297    if (!hw_verts) {
 298       assert(0);
 299       return;
 300    }
 301
 302
 303    /* Single routine to fetch vertices, run shader and emit HW verts.
 304     */
 305    fse->active->run_elts( fse->active,
 306                           fetch_elts,
 307                           fetch_count,
 308                           hw_verts );
 309
 310    draw->render->draw( draw->render,
 311                        draw_elts,
 312                        draw_count );
 313
 314    draw->render->release_vertices( draw->render,
 315                                    hw_verts,
 316                                    fse->key.output_stride,
 317                                    fetch_count );
 318
 319 }
 320
 321
 322 static void fse_finish( struct draw_pt_middle_end *middle )
 323 {
 324 }
 325
 326
 327 static void
 328 fse_destroy( struct draw_pt_middle_end *middle )
 329 {
 330    FREE(middle);
 331 }
 332
 333 struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw )
 334 {
 335    struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
 336    if (!fse)
 337       return NULL;
 338
 339    fse->base.prepare = fse_prepare;
 340    fse->base.run = fse_run;
 341    fse->base.run_linear = fse_run_linear;
 342    fse->base.finish = fse_finish;
 343    fse->base.destroy = fse_destroy;
 344    fse->draw = draw;
 345
 346    return &fse->base;
 347 }