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
  39 struct llvm_middle_end {
  40    struct draw_pt_middle_end base;
  41    struct draw_context *draw;
  42
  43    struct pt_emit *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_geometry_shader *gs = draw->gs.geometry_shader;
  70    struct draw_llvm_variant_key key;
  71    struct draw_llvm_variant *variant = NULL;
  72    unsigned i;
  73    unsigned instance_id_index = ~0;
  74
  75    /* Add one to num_outputs because the pipeline occasionally tags on
  76     * an additional texcoord, eg for AA lines.
  77     */
  78    unsigned nr = MAX2( vs->info.num_inputs,
  79                        vs->info.num_outputs + 1 );
  80
  81    /* Scan for instanceID system value.
  82     */
  83    for (i = 0; i < vs->info.num_inputs; i++) {
  84       if (vs->info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
  85          instance_id_index = i;
  86          break;
  87       }
  88    }
  89
  90    fpme->prim = prim;
  91    fpme->opt = opt;
  92
  93    /* Always leave room for the vertex header whether we need it or
  94     * not.  It's hard to get rid of it in particular because of the
  95     * viewport code in draw_pt_post_vs.c.
  96     */
  97    fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
  98
  99
 100
 101    draw_pt_fetch_prepare( fpme->fetch,
 102                           vs->info.num_inputs,
 103                           fpme->vertex_size,
 104                           instance_id_index );
 105    if (opt & PT_SHADE) {
 106       vs->prepare(vs, draw);
 107       draw_geometry_shader_prepare(gs, draw);
 108    }
 109
 110
 111    /* XXX: it's not really gl rasterization rules we care about here,
 112     * but gl vs dx9 clip spaces.
 113     */
 114    draw_pt_post_vs_prepare( fpme->post_vs,
 115                             (boolean)draw->bypass_clipping,
 116                             (boolean)(draw->identity_viewport),
 117                             (boolean)draw->rasterizer->gl_rasterization_rules,
 118                             (draw->vs.edgeflag_output ? true : false) );
 119
 120    if (!(opt & PT_PIPELINE)) {
 121       draw_pt_emit_prepare( fpme->emit,
 122                             prim,
 123                             max_vertices );
 124
 125       *max_vertices = MAX2( *max_vertices,
 126                             DRAW_PIPE_MAX_VERTICES );
 127    }
 128    else {
 129       *max_vertices = DRAW_PIPE_MAX_VERTICES;
 130    }
 131
 132    /* return even number */
 133    *max_vertices = *max_vertices & ~1;
 134
 135    draw_llvm_make_variant_key(fpme->llvm, &key);
 136
 137    variant = fpme->variants;
 138    while(variant) {
 139       if(memcmp(&variant->key, &key, sizeof key) == 0)
 140          break;
 141
 142       variant = variant->next;
 143    }
 144
 145    if (!variant) {
 146       variant = draw_llvm_prepare(fpme->llvm, nr);
 147       variant->next = fpme->variants;
 148       fpme->variants = variant;
 149       ++fpme->nr_variants;
 150    }
 151    fpme->current_variant = variant;
 152
 153    /*XXX we only support one constant buffer */
 154    fpme->llvm->jit_context.vs_constants =
 155       draw->pt.user.vs_constants[0];
 156    fpme->llvm->jit_context.gs_constants =
 157       draw->pt.user.gs_constants[0];
 158 }
 159
 160
 161
 162 static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 163                                  const unsigned *fetch_elts,
 164                                  unsigned fetch_count,
 165                                  const ushort *draw_elts,
 166                                  unsigned draw_count )
 167 {
 168    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 169    struct draw_context *draw = fpme->draw;
 170    unsigned opt = fpme->opt;
 171    unsigned alloc_count = align( fetch_count, 4 );
 172
 173    struct vertex_header *pipeline_verts =
 174       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 175
 176    if (!pipeline_verts) {
 177       /* Not much we can do here - just skip the rendering.
 178        */
 179       assert(0);
 180       return;
 181    }
 182
 183    fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
 184                                          pipeline_verts,
 185                                          (const char **)draw->pt.user.vbuffer,
 186                                          fetch_elts,
 187                                          fetch_count,
 188                                          fpme->vertex_size,
 189                                          draw->pt.vertex_buffer );
 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 #if 0
 244    debug_printf("#### Pipeline = %p (data = %p)\n",
 245                 pipeline_verts, pipeline_verts->data);
 246 #endif
 247    fpme->current_variant->jit_func( &fpme->llvm->jit_context,
 248                                     pipeline_verts,
 249                                     (const char **)draw->pt.user.vbuffer,
 250                                     start,
 251                                     count,
 252                                     fpme->vertex_size,
 253                                     draw->pt.vertex_buffer );
 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    unsigned opt = fpme->opt;
 294    unsigned alloc_count = align( count, 4 );
 295
 296    struct vertex_header *pipeline_verts =
 297       (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
 298
 299    if (!pipeline_verts)
 300       return FALSE;
 301
 302    fpme->current_variant->jit_func( &fpme->llvm->jit_context,
 303                                     pipeline_verts,
 304                                     (const char **)draw->pt.user.vbuffer,
 305                                     start,
 306                                     count,
 307                                     fpme->vertex_size,
 308                                     draw->pt.vertex_buffer );
 309
 310    if (draw_pt_post_vs_run( fpme->post_vs,
 311                             pipeline_verts,
 312                             count,
 313                             fpme->vertex_size ))
 314    {
 315       opt |= PT_PIPELINE;
 316    }
 317
 318    /* Do we need to run the pipeline?
 319     */
 320    if (opt & PT_PIPELINE) {
 321       draw_pipeline_run( fpme->draw,
 322                          fpme->prim,
 323                          pipeline_verts,
 324                          count,
 325                          fpme->vertex_size,
 326                          draw_elts,
 327                          draw_count );
 328    }
 329    else {
 330       draw_pt_emit( fpme->emit,
 331                     (const float (*)[4])pipeline_verts->data,
 332                     count,
 333                     fpme->vertex_size,
 334                     draw_elts,
 335                     draw_count );
 336    }
 337
 338    FREE(pipeline_verts);
 339    return TRUE;
 340 }
 341
 342
 343
 344 static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
 345 {
 346    /* nothing to do */
 347 }
 348
 349 static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
 350 {
 351    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 352
 353    if (fpme->fetch)
 354       draw_pt_fetch_destroy( fpme->fetch );
 355
 356    if (fpme->emit)
 357       draw_pt_emit_destroy( fpme->emit );
 358
 359    if (fpme->post_vs)
 360       draw_pt_post_vs_destroy( fpme->post_vs );
 361
 362    if (fpme->llvm)
 363       draw_llvm_destroy( fpme->llvm );
 364
 365    FREE(middle);
 366 }
 367
 368
 369 struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit_llvm( struct draw_context *draw )
 370 {
 371    struct llvm_middle_end *fpme = 0;
 372
 373    if (!draw->engine)
 374       return NULL;
 375
 376    fpme = CALLOC_STRUCT( llvm_middle_end );
 377    if (!fpme)
 378       goto fail;
 379
 380    fpme->base.prepare         = llvm_middle_end_prepare;
 381    fpme->base.run             = llvm_middle_end_run;
 382    fpme->base.run_linear      = llvm_middle_end_linear_run;
 383    fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
 384    fpme->base.finish          = llvm_middle_end_finish;
 385    fpme->base.destroy         = llvm_middle_end_destroy;
 386
 387    fpme->draw = draw;
 388
 389    fpme->fetch = draw_pt_fetch_create( draw );
 390    if (!fpme->fetch)
 391       goto fail;
 392
 393    fpme->post_vs = draw_pt_post_vs_create( draw );
 394    if (!fpme->post_vs)
 395       goto fail;
 396
 397    fpme->emit = draw_pt_emit_create( draw );
 398    if (!fpme->emit)
 399       goto fail;
 400
 401    fpme->llvm = draw_llvm_create(draw);
 402    if (!fpme->llvm)
 403       goto fail;
 404
 405    fpme->variants = NULL;
 406    fpme->current_variant = NULL;
 407    fpme->nr_variants = 0;
 408
 409    return &fpme->base;
 410
 411  fail:
 412    if (fpme)
 413       llvm_middle_end_destroy( &fpme->base );
 414
 415    return NULL;
 416 }