src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2010 VMWare, Inc.
   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 #include "util/u_math.h"
  29 #include "util/u_memory.h"
  30 #include "draw/draw_context.h"
  31 #include "draw/draw_vbuf.h"
  32 #include "draw/draw_vertex.h"
  33 #include "draw/draw_pt.h"
  34 #include "draw/draw_vs.h"
  35 #include "draw/draw_gs.h"
  36 #include "draw/draw_llvm.h"
  37
  38 #include "translate/translate.h"
  39
  40
  41 struct llvm_middle_end {
  42    struct draw_pt_middle_end base;
  43    struct draw_context *draw;
  44
  45    struct pt_emit *emit;
  46    struct pt_fetch *fetch;
  47    struct pt_post_vs *post_vs;
  48
  49
  50    unsigned vertex_data_offset;
  51    unsigned vertex_size;
  52    unsigned prim;
  53    unsigned opt;
  54
  55    struct draw_llvm *llvm;
  56 };
  57
  58
  59 static void
  60 llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
  61                          unsigned prim,
  62                          unsigned opt,
  63                          unsigned *max_vertices )
  64 {
  65    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
  66    struct draw_context *draw = fpme->draw;
  67    struct draw_vertex_shader *vs = draw->vs.vertex_shader;
  68    struct draw_geometry_shader *gs = draw->gs.geometry_shader;
  69    unsigned i;
  70    unsigned instance_id_index = ~0;
  71
  72    /* Add one to num_outputs because the pipeline occasionally tags on
  73     * an additional texcoord, eg for AA lines.
  74     */
  75    unsigned nr = MAX2( vs->info.num_inputs,
  76                        vs->info.num_outputs + 1 );
  77
  78    /* Scan for instanceID system value.
  79     */
  80    for (i = 0; i < vs->info.num_inputs; i++) {
  81       if (vs->info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
  82          instance_id_index = i;
  83          break;
  84       }
  85    }
  86
  87    fpme->prim = prim;
  88    fpme->opt = opt;
  89
  90    /* Always leave room for the vertex header whether we need it or
  91     * not.  It's hard to get rid of it in particular because of the
  92     * viewport code in draw_pt_post_vs.c.
  93     */
  94    fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
  95
  96
  97
  98    draw_pt_fetch_prepare( fpme->fetch,
  99                           vs->info.num_inputs,
 100                           fpme->vertex_size,
 101                           instance_id_index );
 102    if (opt & PT_SHADE) {
 103       vs->prepare(vs, draw);
 104       draw_geometry_shader_prepare(gs, draw);
 105    }
 106
 107
 108    /* XXX: it's not really gl rasterization rules we care about here,
 109     * but gl vs dx9 clip spaces.
 110     */
 111    draw_pt_post_vs_prepare( fpme->post_vs,
 112                             (boolean)draw->bypass_clipping,
 113                             (boolean)(draw->identity_viewport ||
 114                             draw->rasterizer->bypass_vs_clip_and_viewport),
 115                             (boolean)draw->rasterizer->gl_rasterization_rules,
 116                             (draw->vs.edgeflag_output ? true : false) );
 117
 118    if (!(opt & PT_PIPELINE)) {
 119       draw_pt_emit_prepare( fpme->emit,
 120                             prim,
 121                             max_vertices );
 122
 123       *max_vertices = MAX2( *max_vertices,
 124                             DRAW_PIPE_MAX_VERTICES );
 125    }
 126    else {
 127       *max_vertices = DRAW_PIPE_MAX_VERTICES;
 128    }
 129
 130    /* return even number */
 131    *max_vertices = *max_vertices & ~1;
 132
 133    draw_llvm_prepare(fpme->llvm);
 134 }
 135
 136
 137
 138 static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 139                                  const unsigned *fetch_elts,
 140                                  unsigned fetch_count,
 141                                  const ushort *draw_elts,
 142                                  unsigned draw_count )
 143 {
 144    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 145    struct draw_context *draw = fpme->draw;
 146    struct draw_vertex_shader *vshader = draw->vs.vertex_shader;
 147    struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
 148    unsigned opt = fpme->opt;
 149    unsigned alloc_count = align( fetch_count, 4 );
 150
 151    struct vertex_header *pipeline_verts =
 152       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 153
 154    if (!pipeline_verts) {
 155       /* Not much we can do here - just skip the rendering.
 156        */
 157       assert(0);
 158       return;
 159    }
 160
 161    /* Fetch into our vertex buffer
 162     */
 163    draw_pt_fetch_run( fpme->fetch,
 164                       fetch_elts,
 165                       fetch_count,
 166                       (char *)pipeline_verts );
 167
 168    /* Run the shader, note that this overwrites the data[] parts of
 169     * the pipeline verts.  If there is no shader, eg if
 170     * bypass_vs_clip_and_viewport, then the inputs == outputs, and are
 171     * already in the correct place.*/
 172    if (opt & PT_SHADE)
 173    {
 174       vshader->run_linear(vshader,
 175                           (const float (*)[4])pipeline_verts->data,
 176                           (      float (*)[4])pipeline_verts->data,
 177                           draw->pt.user.vs_constants,
 178                           fetch_count,
 179                           fpme->vertex_size,
 180                           fpme->vertex_size);
 181       if (gshader)
 182          draw_geometry_shader_run(gshader,
 183                                   (const float (*)[4])pipeline_verts->data,
 184                                   (      float (*)[4])pipeline_verts->data,
 185                                   draw->pt.user.gs_constants,
 186                                   fetch_count,
 187                                   fpme->vertex_size,
 188                                   fpme->vertex_size);
 189    }
 190
 191    if (draw_pt_post_vs_run( fpme->post_vs,
 192                             pipeline_verts,
 193                             fetch_count,
 194                             fpme->vertex_size ))
 195    {
 196       opt |= PT_PIPELINE;
 197    }
 198
 199    /* Do we need to run the pipeline?
 200     */
 201    if (opt & PT_PIPELINE) {
 202       draw_pipeline_run( fpme->draw,
 203                          fpme->prim,
 204                          pipeline_verts,
 205                          fetch_count,
 206                          fpme->vertex_size,
 207                          draw_elts,
 208                          draw_count );
 209    }
 210    else {
 211       draw_pt_emit( fpme->emit,
 212                     (const float (*)[4])pipeline_verts->data,
 213                     fetch_count,
 214                     fpme->vertex_size,
 215                     draw_elts,
 216                     draw_count );
 217    }
 218
 219
 220    FREE(pipeline_verts);
 221 }
 222
 223
 224 static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle,
 225                                        unsigned start,
 226                                        unsigned count)
 227 {
 228    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 229    struct draw_context *draw = fpme->draw;
 230    unsigned opt = fpme->opt;
 231    unsigned alloc_count = align( count, 4 );
 232
 233    struct vertex_header *pipeline_verts =
 234       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 235
 236    if (!pipeline_verts) {
 237       /* Not much we can do here - just skip the rendering.
 238        */
 239       assert(0);
 240       return;
 241    }
 242
 243    fpme->llvm->jit_func( &fpme->llvm->jit_context,
 244                          pipeline_verts,
 245                          start,
 246                          count,
 247                          fpme->vertex_size );
 248
 249    if (draw_pt_post_vs_run( fpme->post_vs,
 250                             pipeline_verts,
 251                             count,
 252                             fpme->vertex_size ))
 253    {
 254       opt |= PT_PIPELINE;
 255    }
 256
 257    /* Do we need to run the pipeline?
 258     */
 259    if (opt & PT_PIPELINE) {
 260       draw_pipeline_run_linear( fpme->draw,
 261                                 fpme->prim,
 262                                 pipeline_verts,
 263                                 count,
 264                                 fpme->vertex_size);
 265    }
 266    else {
 267       draw_pt_emit_linear( fpme->emit,
 268                            (const float (*)[4])pipeline_verts->data,
 269                            fpme->vertex_size,
 270                            count );
 271    }
 272
 273    FREE(pipeline_verts);
 274 }
 275
 276
 277
 278 static boolean
 279 llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle,
 280                                  unsigned start,
 281                                  unsigned count,
 282                                  const ushort *draw_elts,
 283                                  unsigned draw_count )
 284 {
 285    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 286    struct draw_context *draw = fpme->draw;
 287    struct draw_vertex_shader *shader = draw->vs.vertex_shader;
 288    struct draw_geometry_shader *geometry_shader = draw->gs.geometry_shader;
 289    unsigned opt = fpme->opt;
 290    unsigned alloc_count = align( count, 4 );
 291
 292    struct vertex_header *pipeline_verts =
 293       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 294
 295    if (!pipeline_verts)
 296       return FALSE;
 297
 298    /* Fetch into our vertex buffer
 299     */
 300    draw_pt_fetch_run_linear( fpme->fetch,
 301                              start,
 302                              count,
 303                              (char *)pipeline_verts );
 304
 305    /* Run the shader, note that this overwrites the data[] parts of
 306     * the pipeline verts.  If there is no shader, ie if
 307     * bypass_vs_clip_and_viewport, then the inputs == outputs, and are
 308     * already in the correct place.
 309     */
 310    if (opt & PT_SHADE)
 311    {
 312       shader->run_linear(shader,
 313                          (const float (*)[4])pipeline_verts->data,
 314                          (      float (*)[4])pipeline_verts->data,
 315                          draw->pt.user.vs_constants,
 316                          count,
 317                          fpme->vertex_size,
 318                          fpme->vertex_size);
 319
 320       if (geometry_shader)
 321          draw_geometry_shader_run(geometry_shader,
 322                                   (const float (*)[4])pipeline_verts->data,
 323                                   (      float (*)[4])pipeline_verts->data,
 324                                   draw->pt.user.gs_constants,
 325                                   count,
 326                                   fpme->vertex_size,
 327                                   fpme->vertex_size);
 328    }
 329
 330    if (draw_pt_post_vs_run( fpme->post_vs,
 331                             pipeline_verts,
 332                             count,
 333                             fpme->vertex_size ))
 334    {
 335       opt |= PT_PIPELINE;
 336    }
 337
 338    /* Do we need to run the pipeline?
 339     */
 340    if (opt & PT_PIPELINE) {
 341       draw_pipeline_run( fpme->draw,
 342                          fpme->prim,
 343                          pipeline_verts,
 344                          count,
 345                          fpme->vertex_size,
 346                          draw_elts,
 347                          draw_count );
 348    }
 349    else {
 350       draw_pt_emit( fpme->emit,
 351                     (const float (*)[4])pipeline_verts->data,
 352                     count,
 353                     fpme->vertex_size,
 354                     draw_elts,
 355                     draw_count );
 356    }
 357
 358    FREE(pipeline_verts);
 359    return TRUE;
 360 }
 361
 362
 363
 364 static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
 365 {
 366    /* nothing to do */
 367 }
 368
 369 static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
 370 {
 371    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 372
 373    if (fpme->fetch)
 374       draw_pt_fetch_destroy( fpme->fetch );
 375
 376    if (fpme->emit)
 377       draw_pt_emit_destroy( fpme->emit );
 378
 379    if (fpme->post_vs)
 380       draw_pt_post_vs_destroy( fpme->post_vs );
 381
 382    if (fpme->llvm)
 383       draw_llvm_destroy( fpme->llvm );
 384
 385    FREE(middle);
 386 }
 387
 388
 389 struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit_llvm( struct draw_context *draw )
 390 {
 391    struct llvm_middle_end *fpme = 0;
 392
 393    if (!draw->engine)
 394       return NULL;
 395
 396    fpme = CALLOC_STRUCT( llvm_middle_end );
 397    if (!fpme)
 398       goto fail;
 399
 400    fpme->base.prepare         = llvm_middle_end_prepare;
 401    fpme->base.run             = llvm_middle_end_run;
 402    fpme->base.run_linear      = llvm_middle_end_linear_run;
 403    fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
 404    fpme->base.finish          = llvm_middle_end_finish;
 405    fpme->base.destroy         = llvm_middle_end_destroy;
 406
 407    fpme->draw = draw;
 408
 409    fpme->fetch = draw_pt_fetch_create( draw );
 410    if (!fpme->fetch)
 411       goto fail;
 412
 413    fpme->post_vs = draw_pt_post_vs_create( draw );
 414    if (!fpme->post_vs)
 415       goto fail;
 416
 417    fpme->emit = draw_pt_emit_create( draw );
 418    if (!fpme->emit)
 419       goto fail;
 420
 421    fpme->llvm = draw_llvm_create(draw);
 422    if (!fpme->llvm)
 423       goto fail;
 424
 425    return &fpme->base;
 426
 427  fail:
 428    if (fpme)
 429       llvm_middle_end_destroy( &fpme->base );
 430
 431    return NULL;
 432 }