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_llvm.h"
  36
  37
  38 struct llvm_middle_end {
  39    struct draw_pt_middle_end base;
  40    struct draw_context *draw;
  41
  42    struct pt_emit *emit;
  43    struct pt_so_emit *so_emit;
  44    struct pt_fetch *fetch;
  45    struct pt_post_vs *post_vs;
  46
  47
  48    unsigned vertex_data_offset;
  49    unsigned vertex_size;
  50    unsigned prim;
  51    unsigned opt;
  52
  53    struct draw_llvm *llvm;
  54    struct draw_llvm_variant *variants;
  55    struct draw_llvm_variant *current_variant;
  56    int nr_variants;
  57 };
  58
  59
  60 static void
  61 llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
  62                          unsigned prim,
  63                          unsigned opt,
  64                          unsigned *max_vertices )
  65 {
  66    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
  67    struct draw_context *draw = fpme->draw;
  68    struct draw_vertex_shader *vs = draw->vs.vertex_shader;
  69    struct draw_llvm_variant_key key;
  70    struct draw_llvm_variant *variant = NULL;
  71    unsigned i;
  72    unsigned instance_id_index = ~0;
  73
  74    /* Add one to num_outputs because the pipeline occasionally tags on
  75     * an additional texcoord, eg for AA lines.
  76     */
  77    unsigned nr = MAX2( vs->info.num_inputs,
  78                        vs->info.num_outputs + 1 );
  79
  80    /* Scan for instanceID system value.
  81     */
  82    for (i = 0; i < vs->info.num_inputs; i++) {
  83       if (vs->info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
  84          instance_id_index = i;
  85          break;
  86       }
  87    }
  88
  89    fpme->prim = prim;
  90    fpme->opt = opt;
  91
  92    /* Always leave room for the vertex header whether we need it or
  93     * not.  It's hard to get rid of it in particular because of the
  94     * viewport code in draw_pt_post_vs.c.
  95     */
  96    fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
  97
  98
  99    /* XXX: it's not really gl rasterization rules we care about here,
 100     * but gl vs dx9 clip spaces.
 101     */
 102    draw_pt_post_vs_prepare( fpme->post_vs,
 103                             (boolean)draw->bypass_clipping,
 104                             (boolean)(draw->identity_viewport),
 105                             (boolean)draw->rasterizer->gl_rasterization_rules,
 106                             (draw->vs.edgeflag_output ? true : false) );
 107
 108    draw_pt_so_emit_prepare( fpme->so_emit, prim );
 109    if (!(opt & PT_PIPELINE)) {
 110       draw_pt_emit_prepare( fpme->emit,
 111                             prim,
 112                             max_vertices );
 113
 114       *max_vertices = MAX2( *max_vertices,
 115                             DRAW_PIPE_MAX_VERTICES );
 116    }
 117    else {
 118       *max_vertices = DRAW_PIPE_MAX_VERTICES;
 119    }
 120
 121    /* return even number */
 122    *max_vertices = *max_vertices & ~1;
 123
 124    draw_llvm_make_variant_key(fpme->llvm, &key);
 125
 126    variant = fpme->variants;
 127    while(variant) {
 128       if(memcmp(&variant->key, &key, sizeof key) == 0)
 129          break;
 130
 131       variant = variant->next;
 132    }
 133
 134    if (!variant) {
 135       variant = draw_llvm_prepare(fpme->llvm, nr);
 136       variant->next = fpme->variants;
 137       fpme->variants = variant;
 138       ++fpme->nr_variants;
 139    }
 140    fpme->current_variant = variant;
 141
 142    /*XXX we only support one constant buffer */
 143    fpme->llvm->jit_context.vs_constants =
 144       draw->pt.user.vs_constants[0];
 145    fpme->llvm->jit_context.gs_constants =
 146       draw->pt.user.gs_constants[0];
 147 }
 148
 149
 150
 151 static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 152                                  const unsigned *fetch_elts,
 153                                  unsigned fetch_count,
 154                                  const ushort *draw_elts,
 155                                  unsigned draw_count )
 156 {
 157    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 158    struct draw_context *draw = fpme->draw;
 159    unsigned opt = fpme->opt;
 160    unsigned alloc_count = align( fetch_count, 4 );
 161
 162    struct vertex_header *pipeline_verts =
 163       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 164
 165    if (!pipeline_verts) {
 166       /* Not much we can do here - just skip the rendering.
 167        */
 168       assert(0);
 169       return;
 170    }
 171
 172    fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
 173                                          pipeline_verts,
 174                                          (const char **)draw->pt.user.vbuffer,
 175                                          fetch_elts,
 176                                          fetch_count,
 177                                          fpme->vertex_size,
 178                                          draw->pt.vertex_buffer );
 179
 180    /* stream output needs to be done before clipping */
 181    draw_pt_so_emit( fpme->so_emit,
 182                     (const float (*)[4])pipeline_verts->data,
 183                     fetch_count,
 184                     fpme->vertex_size );
 185
 186    if (draw_pt_post_vs_run( fpme->post_vs,
 187                             pipeline_verts,
 188                             fetch_count,
 189                             fpme->vertex_size ))
 190    {
 191       opt |= PT_PIPELINE;
 192    }
 193
 194    /* Do we need to run the pipeline?
 195     */
 196    if (opt & PT_PIPELINE) {
 197       draw_pipeline_run( fpme->draw,
 198                          fpme->prim,
 199                          pipeline_verts,
 200                          fetch_count,
 201                          fpme->vertex_size,
 202                          draw_elts,
 203                          draw_count );
 204    }
 205    else {
 206       draw_pt_emit( fpme->emit,
 207                     (const float (*)[4])pipeline_verts->data,
 208                     fetch_count,
 209                     fpme->vertex_size,
 210                     draw_elts,
 211                     draw_count );
 212    }
 213
 214
 215    FREE(pipeline_verts);
 216 }
 217
 218
 219 static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle,
 220                                        unsigned start,
 221                                        unsigned count)
 222 {
 223    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 224    struct draw_context *draw = fpme->draw;
 225    unsigned opt = fpme->opt;
 226    unsigned alloc_count = align( count, 4 );
 227
 228    struct vertex_header *pipeline_verts =
 229       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 230
 231    if (!pipeline_verts) {
 232       /* Not much we can do here - just skip the rendering.
 233        */
 234       assert(0);
 235       return;
 236    }
 237
 238 #if 0
 239    debug_printf("#### Pipeline = %p (data = %p)\n",
 240                 pipeline_verts, pipeline_verts->data);
 241 #endif
 242    fpme->current_variant->jit_func( &fpme->llvm->jit_context,
 243                                     pipeline_verts,
 244                                     (const char **)draw->pt.user.vbuffer,
 245                                     start,
 246                                     count,
 247                                     fpme->vertex_size,
 248                                     draw->pt.vertex_buffer );
 249
 250    /* stream output needs to be done before clipping */
 251    draw_pt_so_emit( fpme->so_emit,
 252                     (const float (*)[4])pipeline_verts->data,
 253                     count,
 254                     fpme->vertex_size );
 255
 256    if (draw_pt_post_vs_run( fpme->post_vs,
 257                             pipeline_verts,
 258                             count,
 259                             fpme->vertex_size ))
 260    {
 261       opt |= PT_PIPELINE;
 262    }
 263
 264    /* Do we need to run the pipeline?
 265     */
 266    if (opt & PT_PIPELINE) {
 267       draw_pipeline_run_linear( fpme->draw,
 268                                 fpme->prim,
 269                                 pipeline_verts,
 270                                 count,
 271                                 fpme->vertex_size);
 272    }
 273    else {
 274       draw_pt_emit_linear( fpme->emit,
 275                            (const float (*)[4])pipeline_verts->data,
 276                            fpme->vertex_size,
 277                            count );
 278    }
 279
 280    FREE(pipeline_verts);
 281 }
 282
 283
 284
 285 static boolean
 286 llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle,
 287                                  unsigned start,
 288                                  unsigned count,
 289                                  const ushort *draw_elts,
 290                                  unsigned draw_count )
 291 {
 292    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 293    struct draw_context *draw = fpme->draw;
 294    unsigned opt = fpme->opt;
 295    unsigned alloc_count = align( count, 4 );
 296
 297    struct vertex_header *pipeline_verts =
 298       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 299
 300    if (!pipeline_verts)
 301       return FALSE;
 302
 303    fpme->current_variant->jit_func( &fpme->llvm->jit_context,
 304                                     pipeline_verts,
 305                                     (const char **)draw->pt.user.vbuffer,
 306                                     start,
 307                                     count,
 308                                     fpme->vertex_size,
 309                                     draw->pt.vertex_buffer );
 310
 311    /* stream output needs to be done before clipping */
 312    draw_pt_so_emit( fpme->so_emit,
 313                     (const float (*)[4])pipeline_verts->data,
 314                     count,
 315                     fpme->vertex_size );
 316
 317    if (draw_pt_post_vs_run( fpme->post_vs,
 318                             pipeline_verts,
 319                             count,
 320                             fpme->vertex_size ))
 321    {
 322       opt |= PT_PIPELINE;
 323    }
 324
 325    /* Do we need to run the pipeline?
 326     */
 327    if (opt & PT_PIPELINE) {
 328       draw_pipeline_run( fpme->draw,
 329                          fpme->prim,
 330                          pipeline_verts,
 331                          count,
 332                          fpme->vertex_size,
 333                          draw_elts,
 334                          draw_count );
 335    }
 336    else {
 337       draw_pt_emit( fpme->emit,
 338                     (const float (*)[4])pipeline_verts->data,
 339                     count,
 340                     fpme->vertex_size,
 341                     draw_elts,
 342                     draw_count );
 343    }
 344
 345    FREE(pipeline_verts);
 346    return TRUE;
 347 }
 348
 349
 350
 351 static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
 352 {
 353    /* nothing to do */
 354 }
 355
 356 static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
 357 {
 358    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 359    struct draw_context *draw = fpme->draw;
 360    struct draw_llvm_variant *variant = NULL;
 361
 362    variant = fpme->variants;
 363    while(variant) {
 364       struct draw_llvm_variant *next = variant->next;
 365
 366       if (variant->function_elts) {
 367          if (variant->function_elts)
 368             LLVMFreeMachineCodeForFunction(draw->engine,
 369                                            variant->function_elts);
 370          LLVMDeleteFunction(variant->function_elts);
 371       }
 372
 373       if (variant->function) {
 374          if (variant->function)
 375             LLVMFreeMachineCodeForFunction(draw->engine,
 376                                            variant->function);
 377          LLVMDeleteFunction(variant->function);
 378       }
 379
 380       FREE(variant);
 381
 382       variant = next;
 383    }
 384    if (fpme->fetch)
 385       draw_pt_fetch_destroy( fpme->fetch );
 386
 387    if (fpme->emit)
 388       draw_pt_emit_destroy( fpme->emit );
 389
 390    if (fpme->so_emit)
 391       draw_pt_so_emit_destroy( fpme->so_emit );
 392
 393    if (fpme->post_vs)
 394       draw_pt_post_vs_destroy( fpme->post_vs );
 395
 396    if (fpme->llvm)
 397       draw_llvm_destroy( fpme->llvm );
 398
 399    FREE(middle);
 400 }
 401
 402
 403 struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit_llvm( struct draw_context *draw )
 404 {
 405    struct llvm_middle_end *fpme = 0;
 406
 407    if (!draw->engine)
 408       return NULL;
 409
 410    fpme = CALLOC_STRUCT( llvm_middle_end );
 411    if (!fpme)
 412       goto fail;
 413
 414    fpme->base.prepare         = llvm_middle_end_prepare;
 415    fpme->base.run             = llvm_middle_end_run;
 416    fpme->base.run_linear      = llvm_middle_end_linear_run;
 417    fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
 418    fpme->base.finish          = llvm_middle_end_finish;
 419    fpme->base.destroy         = llvm_middle_end_destroy;
 420
 421    fpme->draw = draw;
 422
 423    fpme->fetch = draw_pt_fetch_create( draw );
 424    if (!fpme->fetch)
 425       goto fail;
 426
 427    fpme->post_vs = draw_pt_post_vs_create( draw );
 428    if (!fpme->post_vs)
 429       goto fail;
 430
 431    fpme->emit = draw_pt_emit_create( draw );
 432    if (!fpme->emit)
 433       goto fail;
 434
 435    fpme->so_emit = draw_pt_so_emit_create( draw );
 436    if (!fpme->so_emit)
 437       goto fail;
 438
 439    fpme->llvm = draw_llvm_create(draw);
 440    if (!fpme->llvm)
 441       goto fail;
 442
 443    fpme->variants = NULL;
 444    fpme->current_variant = NULL;
 445    fpme->nr_variants = 0;
 446
 447    return &fpme->base;
 448
 449  fail:
 450    if (fpme)
 451       llvm_middle_end_destroy( &fpme->base );
 452
 453    return NULL;
 454 }