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                             (boolean)draw->rasterizer->gl_rasterization_rules,
 115                             (draw->vs.edgeflag_output ? true : false) );
 116
 117    if (!(opt & PT_PIPELINE)) {
 118       draw_pt_emit_prepare( fpme->emit,
 119                             prim,
 120                             max_vertices );
 121
 122       *max_vertices = MAX2( *max_vertices,
 123                             DRAW_PIPE_MAX_VERTICES );
 124    }
 125    else {
 126       *max_vertices = DRAW_PIPE_MAX_VERTICES;
 127    }
 128
 129    /* return even number */
 130    *max_vertices = *max_vertices & ~1;
 131
 132    draw_llvm_prepare(fpme->llvm);
 133
 134    /*XXX we only support one constant buffer */
 135    fpme->llvm->jit_context.vs_constants =
 136       draw->pt.user.vs_constants[0];
 137    fpme->llvm->jit_context.gs_constants =
 138       draw->pt.user.gs_constants[0];
 139 }
 140
 141
 142
 143 static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 144                                  const unsigned *fetch_elts,
 145                                  unsigned fetch_count,
 146                                  const ushort *draw_elts,
 147                                  unsigned draw_count )
 148 {
 149    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 150    struct draw_context *draw = fpme->draw;
 151    struct draw_vertex_shader *vshader = draw->vs.vertex_shader;
 152    struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
 153    unsigned opt = fpme->opt;
 154    unsigned alloc_count = align( fetch_count, 4 );
 155
 156    struct vertex_header *pipeline_verts =
 157       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 158
 159    if (!pipeline_verts) {
 160       /* Not much we can do here - just skip the rendering.
 161        */
 162       assert(0);
 163       return;
 164    }
 165
 166    /* Fetch into our vertex buffer
 167     */
 168    draw_pt_fetch_run( fpme->fetch,
 169                       fetch_elts,
 170                       fetch_count,
 171                       (char *)pipeline_verts );
 172
 173    /* Run the shader, note that this overwrites the data[] parts of
 174     * the pipeline verts.  If there is no shader, eg if
 175     * bypass_vs_clip_and_viewport, then the inputs == outputs, and are
 176     * already in the correct place.*/
 177    if (opt & PT_SHADE)
 178    {
 179       vshader->run_linear(vshader,
 180                           (const float (*)[4])pipeline_verts->data,
 181                           (      float (*)[4])pipeline_verts->data,
 182                           draw->pt.user.vs_constants,
 183                           fetch_count,
 184                           fpme->vertex_size,
 185                           fpme->vertex_size);
 186       if (gshader)
 187          draw_geometry_shader_run(gshader,
 188                                   (const float (*)[4])pipeline_verts->data,
 189                                   (      float (*)[4])pipeline_verts->data,
 190                                   draw->pt.user.gs_constants,
 191                                   fetch_count,
 192                                   fpme->vertex_size,
 193                                   fpme->vertex_size);
 194    }
 195
 196    if (draw_pt_post_vs_run( fpme->post_vs,
 197                             pipeline_verts,
 198                             fetch_count,
 199                             fpme->vertex_size ))
 200    {
 201       opt |= PT_PIPELINE;
 202    }
 203
 204    /* Do we need to run the pipeline?
 205     */
 206    if (opt & PT_PIPELINE) {
 207       draw_pipeline_run( fpme->draw,
 208                          fpme->prim,
 209                          pipeline_verts,
 210                          fetch_count,
 211                          fpme->vertex_size,
 212                          draw_elts,
 213                          draw_count );
 214    }
 215    else {
 216       draw_pt_emit( fpme->emit,
 217                     (const float (*)[4])pipeline_verts->data,
 218                     fetch_count,
 219                     fpme->vertex_size,
 220                     draw_elts,
 221                     draw_count );
 222    }
 223
 224
 225    FREE(pipeline_verts);
 226 }
 227
 228
 229 static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle,
 230                                        unsigned start,
 231                                        unsigned count)
 232 {
 233    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 234    struct draw_context *draw = fpme->draw;
 235    unsigned opt = fpme->opt;
 236    unsigned alloc_count = align( count, 4 );
 237
 238    struct vertex_header *pipeline_verts =
 239       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 240
 241    if (!pipeline_verts) {
 242       /* Not much we can do here - just skip the rendering.
 243        */
 244       assert(0);
 245       return;
 246    }
 247
 248    fpme->llvm->jit_func( &fpme->llvm->jit_context,
 249                          pipeline_verts,
 250                          (const char **)draw->pt.user.vbuffer,
 251                          start,
 252                          count,
 253                          fpme->vertex_size );
 254
 255    if (draw_pt_post_vs_run( fpme->post_vs,
 256                             pipeline_verts,
 257                             count,
 258                             fpme->vertex_size ))
 259    {
 260       opt |= PT_PIPELINE;
 261    }
 262
 263    /* Do we need to run the pipeline?
 264     */
 265    if (opt & PT_PIPELINE) {
 266       draw_pipeline_run_linear( fpme->draw,
 267                                 fpme->prim,
 268                                 pipeline_verts,
 269                                 count,
 270                                 fpme->vertex_size);
 271    }
 272    else {
 273       draw_pt_emit_linear( fpme->emit,
 274                            (const float (*)[4])pipeline_verts->data,
 275                            fpme->vertex_size,
 276                            count );
 277    }
 278
 279    FREE(pipeline_verts);
 280 }
 281
 282
 283
 284 static boolean
 285 llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle,
 286                                  unsigned start,
 287                                  unsigned count,
 288                                  const ushort *draw_elts,
 289                                  unsigned draw_count )
 290 {
 291    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 292    struct draw_context *draw = fpme->draw;
 293    struct draw_vertex_shader *shader = draw->vs.vertex_shader;
 294    struct draw_geometry_shader *geometry_shader = draw->gs.geometry_shader;
 295    unsigned opt = fpme->opt;
 296    unsigned alloc_count = align( count, 4 );
 297
 298    struct vertex_header *pipeline_verts =
 299       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 300
 301    if (!pipeline_verts)
 302       return FALSE;
 303
 304    /* Fetch into our vertex buffer
 305     */
 306    draw_pt_fetch_run_linear( fpme->fetch,
 307                              start,
 308                              count,
 309                              (char *)pipeline_verts );
 310
 311    /* Run the shader, note that this overwrites the data[] parts of
 312     * the pipeline verts.  If there is no shader, ie if
 313     * bypass_vs_clip_and_viewport, then the inputs == outputs, and are
 314     * already in the correct place.
 315     */
 316    if (opt & PT_SHADE)
 317    {
 318       shader->run_linear(shader,
 319                          (const float (*)[4])pipeline_verts->data,
 320                          (      float (*)[4])pipeline_verts->data,
 321                          draw->pt.user.vs_constants,
 322                          count,
 323                          fpme->vertex_size,
 324                          fpme->vertex_size);
 325
 326       if (geometry_shader)
 327          draw_geometry_shader_run(geometry_shader,
 328                                   (const float (*)[4])pipeline_verts->data,
 329                                   (      float (*)[4])pipeline_verts->data,
 330                                   draw->pt.user.gs_constants,
 331                                   count,
 332                                   fpme->vertex_size,
 333                                   fpme->vertex_size);
 334    }
 335
 336    if (draw_pt_post_vs_run( fpme->post_vs,
 337                             pipeline_verts,
 338                             count,
 339                             fpme->vertex_size ))
 340    {
 341       opt |= PT_PIPELINE;
 342    }
 343
 344    /* Do we need to run the pipeline?
 345     */
 346    if (opt & PT_PIPELINE) {
 347       draw_pipeline_run( fpme->draw,
 348                          fpme->prim,
 349                          pipeline_verts,
 350                          count,
 351                          fpme->vertex_size,
 352                          draw_elts,
 353                          draw_count );
 354    }
 355    else {
 356       draw_pt_emit( fpme->emit,
 357                     (const float (*)[4])pipeline_verts->data,
 358                     count,
 359                     fpme->vertex_size,
 360                     draw_elts,
 361                     draw_count );
 362    }
 363
 364    FREE(pipeline_verts);
 365    return TRUE;
 366 }
 367
 368
 369
 370 static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
 371 {
 372    /* nothing to do */
 373 }
 374
 375 static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
 376 {
 377    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 378
 379    if (fpme->fetch)
 380       draw_pt_fetch_destroy( fpme->fetch );
 381
 382    if (fpme->emit)
 383       draw_pt_emit_destroy( fpme->emit );
 384
 385    if (fpme->post_vs)
 386       draw_pt_post_vs_destroy( fpme->post_vs );
 387
 388    if (fpme->llvm)
 389       draw_llvm_destroy( fpme->llvm );
 390
 391    FREE(middle);
 392 }
 393
 394
 395 struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit_llvm( struct draw_context *draw )
 396 {
 397    struct llvm_middle_end *fpme = 0;
 398
 399    if (!draw->engine)
 400       return NULL;
 401
 402    fpme = CALLOC_STRUCT( llvm_middle_end );
 403    if (!fpme)
 404       goto fail;
 405
 406    fpme->base.prepare         = llvm_middle_end_prepare;
 407    fpme->base.run             = llvm_middle_end_run;
 408    fpme->base.run_linear      = llvm_middle_end_linear_run;
 409    fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
 410    fpme->base.finish          = llvm_middle_end_finish;
 411    fpme->base.destroy         = llvm_middle_end_destroy;
 412
 413    fpme->draw = draw;
 414
 415    fpme->fetch = draw_pt_fetch_create( draw );
 416    if (!fpme->fetch)
 417       goto fail;
 418
 419    fpme->post_vs = draw_pt_post_vs_create( draw );
 420    if (!fpme->post_vs)
 421       goto fail;
 422
 423    fpme->emit = draw_pt_emit_create( draw );
 424    if (!fpme->emit)
 425       goto fail;
 426
 427    fpme->llvm = draw_llvm_create(draw);
 428    if (!fpme->llvm)
 429       goto fail;
 430
 431    return &fpme->base;
 432
 433  fail:
 434    if (fpme)
 435       llvm_middle_end_destroy( &fpme->base );
 436
 437    return NULL;
 438 }