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