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 opt;
  53
  54    struct draw_llvm *llvm;
  55    struct draw_llvm_variant *current_variant;
  56 };
  57
  58
  59 static void
  60 llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
  61                          unsigned in_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 llvm_vertex_shader *shader =
  68       llvm_vertex_shader(draw->vs.vertex_shader);
  69    char store[DRAW_LLVM_MAX_VARIANT_KEY_SIZE];
  70    struct draw_llvm_variant_key *key;
  71    struct draw_llvm_variant *variant = NULL;
  72    struct draw_llvm_variant_list_item *li;
  73    unsigned i;
  74    unsigned instance_id_index = ~0;
  75
  76
  77    unsigned out_prim = (draw->gs.geometry_shader ?
  78                         draw->gs.geometry_shader->output_primitive :
  79                         in_prim);
  80
  81    /* Add one to num_outputs because the pipeline occasionally tags on
  82     * an additional texcoord, eg for AA lines.
  83     */
  84    unsigned nr = MAX2( shader->base.info.num_inputs,
  85                        shader->base.info.num_outputs + 1 );
  86
  87    /* Scan for instanceID system value.
  88     */
  89    for (i = 0; i < shader->base.info.num_inputs; i++) {
  90       if (shader->base.info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
  91          instance_id_index = i;
  92          break;
  93       }
  94    }
  95
  96    fpme->input_prim = in_prim;
  97    fpme->opt = opt;
  98
  99    /* Always leave room for the vertex header whether we need it or
 100     * not.  It's hard to get rid of it in particular because of the
 101     * viewport code in draw_pt_post_vs.c.
 102     */
 103    fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
 104
 105
 106    /* XXX: it's not really gl rasterization rules we care about here,
 107     * but gl vs dx9 clip spaces.
 108     */
 109    draw_pt_post_vs_prepare( fpme->post_vs,
 110                             draw->clip_xy,
 111                             draw->clip_z,
 112                             draw->clip_user,
 113                             draw->identity_viewport,
 114                             (boolean)draw->rasterizer->gl_rasterization_rules,
 115                             (draw->vs.edgeflag_output ? TRUE : FALSE) );
 116
 117    draw_pt_so_emit_prepare( fpme->so_emit );
 118
 119    if (!(opt & PT_PIPELINE)) {
 120       draw_pt_emit_prepare( fpme->emit,
 121                             out_prim,
 122                             max_vertices );
 123
 124       *max_vertices = MAX2( *max_vertices, 4096 );
 125    }
 126    else {
 127       /* limit max fetches by limiting max_vertices */
 128       *max_vertices = 4096;
 129    }
 130
 131    /* return even number */
 132    *max_vertices = *max_vertices & ~1;
 133
 134    key = draw_llvm_make_variant_key(fpme->llvm, store);
 135
 136    li = first_elem(&shader->variants);
 137    while(!at_end(&shader->variants, li)) {
 138       if(memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
 139          variant = li->base;
 140          break;
 141       }
 142       li = next_elem(li);
 143    }
 144
 145    if (variant) {
 146       move_to_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
 147    }
 148    else {
 149       unsigned i;
 150       if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
 151          /*
 152           * XXX: should we flush here ?
 153           */
 154          for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) {
 155             struct draw_llvm_variant_list_item *item =
 156                last_elem(&fpme->llvm->vs_variants_list);
 157             draw_llvm_destroy_variant(item->base);
 158          }
 159       }
 160
 161       variant = draw_llvm_create_variant(fpme->llvm, nr, key);
 162
 163       if (variant) {
 164          insert_at_head(&shader->variants, &variant->list_item_local);
 165          insert_at_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
 166          fpme->llvm->nr_variants++;
 167          shader->variants_cached++;
 168       }
 169    }
 170
 171    fpme->current_variant = variant;
 172
 173    /*XXX we only support one constant buffer */
 174    fpme->llvm->jit_context.vs_constants =
 175       draw->pt.user.vs_constants[0];
 176    fpme->llvm->jit_context.gs_constants =
 177       draw->pt.user.gs_constants[0];
 178    fpme->llvm->jit_context.planes =
 179       (float (*) [12][4]) draw->pt.user.planes[0];
 180
 181 }
 182
 183
 184 static void pipeline(struct llvm_middle_end *llvm,
 185                      const struct draw_vertex_info *vert_info,
 186                      const struct draw_prim_info *prim_info)
 187 {
 188    if (prim_info->linear)
 189       draw_pipeline_run_linear( llvm->draw,
 190                                 vert_info,
 191                                 prim_info);
 192    else
 193       draw_pipeline_run( llvm->draw,
 194                          vert_info,
 195                          prim_info );
 196 }
 197
 198 static void emit(struct pt_emit *emit,
 199                  const struct draw_vertex_info *vert_info,
 200                  const struct draw_prim_info *prim_info)
 201 {
 202    if (prim_info->linear) {
 203       draw_pt_emit_linear(emit, vert_info, prim_info);
 204    }
 205    else {
 206       draw_pt_emit(emit, vert_info, prim_info);
 207    }
 208 }
 209
 210 static void
 211 llvm_pipeline_generic( struct draw_pt_middle_end *middle,
 212                        const struct draw_fetch_info *fetch_info,
 213                        const struct draw_prim_info *prim_info )
 214 {
 215    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 216    struct draw_context *draw = fpme->draw;
 217    struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
 218    struct draw_prim_info gs_prim_info;
 219    struct draw_vertex_info llvm_vert_info;
 220    struct draw_vertex_info gs_vert_info;
 221    struct draw_vertex_info *vert_info;
 222    unsigned opt = fpme->opt;
 223    unsigned clipped = 0;
 224
 225    llvm_vert_info.count = fetch_info->count;
 226    llvm_vert_info.vertex_size = fpme->vertex_size;
 227    llvm_vert_info.stride = fpme->vertex_size;
 228    llvm_vert_info.verts =
 229       (struct vertex_header *)MALLOC(fpme->vertex_size *
 230                                      align(fetch_info->count,  4));
 231    if (!llvm_vert_info.verts) {
 232       assert(0);
 233       return;
 234    }
 235
 236    if (fetch_info->linear)
 237       clipped = fpme->current_variant->jit_func( &fpme->llvm->jit_context,
 238                                        llvm_vert_info.verts,
 239                                        (const char **)draw->pt.user.vbuffer,
 240                                        fetch_info->start,
 241                                        fetch_info->count,
 242                                        fpme->vertex_size,
 243                                        draw->pt.vertex_buffer,
 244                                        draw->instance_id);
 245    else
 246       clipped = fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
 247                                             llvm_vert_info.verts,
 248                                             (const char **)draw->pt.user.vbuffer,
 249                                             fetch_info->elts,
 250                                             fetch_info->count,
 251                                             fpme->vertex_size,
 252                                             draw->pt.vertex_buffer,
 253                                             draw->instance_id);
 254
 255    /* Finished with fetch and vs:
 256     */
 257    fetch_info = NULL;
 258    vert_info = &llvm_vert_info;
 259
 260
 261    if ((opt & PT_SHADE) && gshader) {
 262       draw_geometry_shader_run(gshader,
 263                                draw->pt.user.gs_constants,
 264                                draw->pt.user.gs_constants_size,
 265                                vert_info,
 266                                prim_info,
 267                                &gs_vert_info,
 268                                &gs_prim_info);
 269
 270       FREE(vert_info->verts);
 271       vert_info = &gs_vert_info;
 272       prim_info = &gs_prim_info;
 273
 274       clipped = draw_pt_post_vs_run( fpme->post_vs, vert_info );
 275
 276    }
 277
 278    /* stream output needs to be done before clipping */
 279    draw_pt_so_emit( fpme->so_emit,
 280                     vert_info,
 281                     prim_info );
 282
 283    if (clipped) {
 284       opt |= PT_PIPELINE;
 285    }
 286
 287    /* Do we need to run the pipeline? Now will come here if clipped
 288     */
 289    if (opt & PT_PIPELINE) {
 290       pipeline( fpme,
 291                 vert_info,
 292                 prim_info );
 293    }
 294    else {
 295       emit( fpme->emit,
 296             vert_info,
 297             prim_info );
 298    }
 299    FREE(vert_info->verts);
 300 }
 301
 302
 303 static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 304                                  const unsigned *fetch_elts,
 305                                  unsigned fetch_count,
 306                                  const ushort *draw_elts,
 307                                  unsigned draw_count,
 308                                  unsigned prim_flags )
 309 {
 310    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 311    struct draw_fetch_info fetch_info;
 312    struct draw_prim_info prim_info;
 313
 314    fetch_info.linear = FALSE;
 315    fetch_info.start = 0;
 316    fetch_info.elts = fetch_elts;
 317    fetch_info.count = fetch_count;
 318
 319    prim_info.linear = FALSE;
 320    prim_info.start = 0;
 321    prim_info.count = draw_count;
 322    prim_info.elts = draw_elts;
 323    prim_info.prim = fpme->input_prim;
 324    prim_info.flags = prim_flags;
 325    prim_info.primitive_count = 1;
 326    prim_info.primitive_lengths = &draw_count;
 327
 328    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 329 }
 330
 331
 332 static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle,
 333                                        unsigned start,
 334                                        unsigned count,
 335                                        unsigned prim_flags)
 336 {
 337    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 338    struct draw_fetch_info fetch_info;
 339    struct draw_prim_info prim_info;
 340
 341    fetch_info.linear = TRUE;
 342    fetch_info.start = start;
 343    fetch_info.count = count;
 344    fetch_info.elts = NULL;
 345
 346    prim_info.linear = TRUE;
 347    prim_info.start = 0;
 348    prim_info.count = count;
 349    prim_info.elts = NULL;
 350    prim_info.prim = fpme->input_prim;
 351    prim_info.flags = prim_flags;
 352    prim_info.primitive_count = 1;
 353    prim_info.primitive_lengths = &count;
 354
 355    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 356 }
 357
 358
 359
 360 static boolean
 361 llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle,
 362                                  unsigned start,
 363                                  unsigned count,
 364                                  const ushort *draw_elts,
 365                                  unsigned draw_count,
 366                                  unsigned prim_flags )
 367 {
 368    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 369    struct draw_fetch_info fetch_info;
 370    struct draw_prim_info prim_info;
 371
 372    fetch_info.linear = TRUE;
 373    fetch_info.start = start;
 374    fetch_info.count = count;
 375    fetch_info.elts = NULL;
 376
 377    prim_info.linear = FALSE;
 378    prim_info.start = 0;
 379    prim_info.count = draw_count;
 380    prim_info.elts = draw_elts;
 381    prim_info.prim = fpme->input_prim;
 382    prim_info.flags = prim_flags;
 383    prim_info.primitive_count = 1;
 384    prim_info.primitive_lengths = &draw_count;
 385
 386    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 387
 388    return TRUE;
 389 }
 390
 391
 392
 393 static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
 394 {
 395    /* nothing to do */
 396 }
 397
 398 static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
 399 {
 400    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 401
 402    if (fpme->fetch)
 403       draw_pt_fetch_destroy( fpme->fetch );
 404
 405    if (fpme->emit)
 406       draw_pt_emit_destroy( fpme->emit );
 407
 408    if (fpme->so_emit)
 409       draw_pt_so_emit_destroy( fpme->so_emit );
 410
 411    if (fpme->post_vs)
 412       draw_pt_post_vs_destroy( fpme->post_vs );
 413
 414    FREE(middle);
 415 }
 416
 417
 418 struct draw_pt_middle_end *
 419 draw_pt_fetch_pipeline_or_emit_llvm(struct draw_context *draw)
 420 {
 421    struct llvm_middle_end *fpme = 0;
 422
 423    if (!draw->engine)
 424       return NULL;
 425
 426    fpme = CALLOC_STRUCT( llvm_middle_end );
 427    if (!fpme)
 428       goto fail;
 429
 430    fpme->base.prepare         = llvm_middle_end_prepare;
 431    fpme->base.run             = llvm_middle_end_run;
 432    fpme->base.run_linear      = llvm_middle_end_linear_run;
 433    fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
 434    fpme->base.finish          = llvm_middle_end_finish;
 435    fpme->base.destroy         = llvm_middle_end_destroy;
 436
 437    fpme->draw = draw;
 438
 439    fpme->fetch = draw_pt_fetch_create( draw );
 440    if (!fpme->fetch)
 441       goto fail;
 442
 443    fpme->post_vs = draw_pt_post_vs_create( draw );
 444    if (!fpme->post_vs)
 445       goto fail;
 446
 447    fpme->emit = draw_pt_emit_create( draw );
 448    if (!fpme->emit)
 449       goto fail;
 450
 451    fpme->so_emit = draw_pt_so_emit_create( draw );
 452    if (!fpme->so_emit)
 453       goto fail;
 454
 455    fpme->llvm = draw->llvm;
 456    if (!fpme->llvm)
 457       goto fail;
 458
 459    fpme->current_variant = NULL;
 460
 461    return &fpme->base;
 462
 463  fail:
 464    if (fpme)
 465       llvm_middle_end_destroy( &fpme->base );
 466
 467    return NULL;
 468 }