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 #include "gallivm/lp_bld_init.h"
  38
  39
  40 struct llvm_middle_end {
  41    struct draw_pt_middle_end base;
  42    struct draw_context *draw;
  43
  44    struct pt_emit *emit;
  45    struct pt_so_emit *so_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 input_prim;
  53    unsigned opt;
  54
  55    struct draw_llvm *llvm;
  56    struct draw_llvm_variant *current_variant;
  57 };
  58
  59
  60 static void
  61 llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
  62                          unsigned in_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 llvm_vertex_shader *shader =
  69       llvm_vertex_shader(draw->vs.vertex_shader);
  70    char store[DRAW_LLVM_MAX_VARIANT_KEY_SIZE];
  71    struct draw_llvm_variant_key *key;
  72    struct draw_llvm_variant *variant = NULL;
  73    struct draw_llvm_variant_list_item *li;
  74    unsigned i;
  75    unsigned instance_id_index = ~0;
  76    const unsigned out_prim = (draw->gs.geometry_shader ?
  77                               draw->gs.geometry_shader->output_primitive :
  78                               in_prim);
  79
  80    /* Add one to num_outputs because the pipeline occasionally tags on
  81     * an additional texcoord, eg for AA lines.
  82     */
  83    const unsigned nr = MAX2( shader->base.info.num_inputs,
  84                              shader->base.info.num_outputs + 1 );
  85
  86    /* Scan for instanceID system value.
  87     * XXX but we never use instance_id_index?!
  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->guard_band_xy,
 114                             draw->identity_viewport,
 115                             (boolean)draw->rasterizer->gl_rasterization_rules,
 116                             (draw->vs.edgeflag_output ? TRUE : FALSE) );
 117
 118    draw_pt_so_emit_prepare( fpme->so_emit );
 119
 120    if (!(opt & PT_PIPELINE)) {
 121       draw_pt_emit_prepare( fpme->emit,
 122                             out_prim,
 123                             max_vertices );
 124
 125       *max_vertices = MAX2( *max_vertices, 4096 );
 126    }
 127    else {
 128       /* limit max fetches by limiting max_vertices */
 129       *max_vertices = 4096;
 130    }
 131
 132    /* return even number */
 133    *max_vertices = *max_vertices & ~1;
 134
 135    key = draw_llvm_make_variant_key(fpme->llvm, store);
 136
 137    /* Search shader's list of variants for the key */
 138    li = first_elem(&shader->variants);
 139    while (!at_end(&shader->variants, li)) {
 140       if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
 141          variant = li->base;
 142          break;
 143       }
 144       li = next_elem(li);
 145    }
 146
 147    if (variant) {
 148       /* found the variant, move to head of global list (for LRU) */
 149       move_to_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
 150    }
 151    else {
 152       /* Need to create new variant */
 153       unsigned i;
 154
 155       /* First check if we've created too many variants.  If so, free
 156        * 25% of the LRU to avoid using too much memory.
 157        */
 158       if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
 159          /*
 160           * XXX: should we flush here ?
 161           */
 162          for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) {
 163             struct draw_llvm_variant_list_item *item =
 164                last_elem(&fpme->llvm->vs_variants_list);
 165             draw_llvm_destroy_variant(item->base);
 166          }
 167       }
 168
 169       variant = draw_llvm_create_variant(fpme->llvm, nr, key);
 170
 171       if (variant) {
 172          insert_at_head(&shader->variants, &variant->list_item_local);
 173          insert_at_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
 174          fpme->llvm->nr_variants++;
 175          shader->variants_cached++;
 176       }
 177    }
 178
 179    fpme->current_variant = variant;
 180
 181    /*XXX we only support one constant buffer */
 182    fpme->llvm->jit_context.vs_constants =
 183       draw->pt.user.vs_constants[0];
 184    fpme->llvm->jit_context.gs_constants =
 185       draw->pt.user.gs_constants[0];
 186    fpme->llvm->jit_context.planes =
 187       (float (*) [12][4]) draw->pt.user.planes[0];
 188    fpme->llvm->jit_context.viewport =
 189       (float *)draw->viewport.scale;
 190
 191 }
 192
 193
 194 static void pipeline(struct llvm_middle_end *llvm,
 195                      const struct draw_vertex_info *vert_info,
 196                      const struct draw_prim_info *prim_info)
 197 {
 198    if (prim_info->linear)
 199       draw_pipeline_run_linear( llvm->draw,
 200                                 vert_info,
 201                                 prim_info);
 202    else
 203       draw_pipeline_run( llvm->draw,
 204                          vert_info,
 205                          prim_info );
 206 }
 207
 208 static void emit(struct pt_emit *emit,
 209                  const struct draw_vertex_info *vert_info,
 210                  const struct draw_prim_info *prim_info)
 211 {
 212    if (prim_info->linear) {
 213       draw_pt_emit_linear(emit, vert_info, prim_info);
 214    }
 215    else {
 216       draw_pt_emit(emit, vert_info, prim_info);
 217    }
 218 }
 219
 220 static void
 221 llvm_pipeline_generic( struct draw_pt_middle_end *middle,
 222                        const struct draw_fetch_info *fetch_info,
 223                        const struct draw_prim_info *prim_info )
 224 {
 225    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 226    struct draw_context *draw = fpme->draw;
 227    struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
 228    struct draw_prim_info gs_prim_info;
 229    struct draw_vertex_info llvm_vert_info;
 230    struct draw_vertex_info gs_vert_info;
 231    struct draw_vertex_info *vert_info;
 232    unsigned opt = fpme->opt;
 233    unsigned clipped = 0;
 234
 235    llvm_vert_info.count = fetch_info->count;
 236    llvm_vert_info.vertex_size = fpme->vertex_size;
 237    llvm_vert_info.stride = fpme->vertex_size;
 238    llvm_vert_info.verts =
 239       (struct vertex_header *)MALLOC(fpme->vertex_size *
 240                                      align(fetch_info->count,  4));
 241    if (!llvm_vert_info.verts) {
 242       assert(0);
 243       return;
 244    }
 245
 246    if (fetch_info->linear)
 247       clipped = fpme->current_variant->jit_func( &fpme->llvm->jit_context,
 248                                        llvm_vert_info.verts,
 249                                        (const char **)draw->pt.user.vbuffer,
 250                                        fetch_info->start,
 251                                        fetch_info->count,
 252                                        fpme->vertex_size,
 253                                        draw->pt.vertex_buffer,
 254                                        draw->instance_id);
 255    else
 256       clipped = fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
 257                                             llvm_vert_info.verts,
 258                                             (const char **)draw->pt.user.vbuffer,
 259                                             fetch_info->elts,
 260                                             fetch_info->count,
 261                                             fpme->vertex_size,
 262                                             draw->pt.vertex_buffer,
 263                                             draw->instance_id);
 264
 265    /* Finished with fetch and vs:
 266     */
 267    fetch_info = NULL;
 268    vert_info = &llvm_vert_info;
 269
 270
 271    if ((opt & PT_SHADE) && gshader) {
 272       draw_geometry_shader_run(gshader,
 273                                draw->pt.user.gs_constants,
 274                                draw->pt.user.gs_constants_size,
 275                                vert_info,
 276                                prim_info,
 277                                &gs_vert_info,
 278                                &gs_prim_info);
 279
 280       FREE(vert_info->verts);
 281       vert_info = &gs_vert_info;
 282       prim_info = &gs_prim_info;
 283
 284       clipped = draw_pt_post_vs_run( fpme->post_vs, vert_info );
 285
 286    }
 287
 288    /* stream output needs to be done before clipping */
 289    draw_pt_so_emit( fpme->so_emit,
 290                     vert_info,
 291                     prim_info );
 292
 293    if (clipped) {
 294       opt |= PT_PIPELINE;
 295    }
 296
 297    /* Do we need to run the pipeline? Now will come here if clipped
 298     */
 299    if (opt & PT_PIPELINE) {
 300       pipeline( fpme,
 301                 vert_info,
 302                 prim_info );
 303    }
 304    else {
 305       emit( fpme->emit,
 306             vert_info,
 307             prim_info );
 308    }
 309    FREE(vert_info->verts);
 310 }
 311
 312
 313 static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 314                                  const unsigned *fetch_elts,
 315                                  unsigned fetch_count,
 316                                  const ushort *draw_elts,
 317                                  unsigned draw_count,
 318                                  unsigned prim_flags )
 319 {
 320    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 321    struct draw_fetch_info fetch_info;
 322    struct draw_prim_info prim_info;
 323
 324    fetch_info.linear = FALSE;
 325    fetch_info.start = 0;
 326    fetch_info.elts = fetch_elts;
 327    fetch_info.count = fetch_count;
 328
 329    prim_info.linear = FALSE;
 330    prim_info.start = 0;
 331    prim_info.count = draw_count;
 332    prim_info.elts = draw_elts;
 333    prim_info.prim = fpme->input_prim;
 334    prim_info.flags = prim_flags;
 335    prim_info.primitive_count = 1;
 336    prim_info.primitive_lengths = &draw_count;
 337
 338    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 339 }
 340
 341
 342 static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle,
 343                                        unsigned start,
 344                                        unsigned count,
 345                                        unsigned prim_flags)
 346 {
 347    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 348    struct draw_fetch_info fetch_info;
 349    struct draw_prim_info prim_info;
 350
 351    fetch_info.linear = TRUE;
 352    fetch_info.start = start;
 353    fetch_info.count = count;
 354    fetch_info.elts = NULL;
 355
 356    prim_info.linear = TRUE;
 357    prim_info.start = 0;
 358    prim_info.count = count;
 359    prim_info.elts = NULL;
 360    prim_info.prim = fpme->input_prim;
 361    prim_info.flags = prim_flags;
 362    prim_info.primitive_count = 1;
 363    prim_info.primitive_lengths = &count;
 364
 365    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 366 }
 367
 368
 369
 370 static boolean
 371 llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle,
 372                                  unsigned start,
 373                                  unsigned count,
 374                                  const ushort *draw_elts,
 375                                  unsigned draw_count,
 376                                  unsigned prim_flags )
 377 {
 378    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 379    struct draw_fetch_info fetch_info;
 380    struct draw_prim_info prim_info;
 381
 382    fetch_info.linear = TRUE;
 383    fetch_info.start = start;
 384    fetch_info.count = count;
 385    fetch_info.elts = NULL;
 386
 387    prim_info.linear = FALSE;
 388    prim_info.start = 0;
 389    prim_info.count = draw_count;
 390    prim_info.elts = draw_elts;
 391    prim_info.prim = fpme->input_prim;
 392    prim_info.flags = prim_flags;
 393    prim_info.primitive_count = 1;
 394    prim_info.primitive_lengths = &draw_count;
 395
 396    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 397
 398    return TRUE;
 399 }
 400
 401
 402
 403 static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
 404 {
 405    /* nothing to do */
 406 }
 407
 408 static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
 409 {
 410    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 411
 412    if (fpme->fetch)
 413       draw_pt_fetch_destroy( fpme->fetch );
 414
 415    if (fpme->emit)
 416       draw_pt_emit_destroy( fpme->emit );
 417
 418    if (fpme->so_emit)
 419       draw_pt_so_emit_destroy( fpme->so_emit );
 420
 421    if (fpme->post_vs)
 422       draw_pt_post_vs_destroy( fpme->post_vs );
 423
 424    FREE(middle);
 425 }
 426
 427
 428 struct draw_pt_middle_end *
 429 draw_pt_fetch_pipeline_or_emit_llvm(struct draw_context *draw)
 430 {
 431    struct llvm_middle_end *fpme = 0;
 432
 433    if (!draw->llvm->gallivm->engine)
 434       return NULL;
 435
 436    fpme = CALLOC_STRUCT( llvm_middle_end );
 437    if (!fpme)
 438       goto fail;
 439
 440    fpme->base.prepare         = llvm_middle_end_prepare;
 441    fpme->base.run             = llvm_middle_end_run;
 442    fpme->base.run_linear      = llvm_middle_end_linear_run;
 443    fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
 444    fpme->base.finish          = llvm_middle_end_finish;
 445    fpme->base.destroy         = llvm_middle_end_destroy;
 446
 447    fpme->draw = draw;
 448
 449    fpme->fetch = draw_pt_fetch_create( draw );
 450    if (!fpme->fetch)
 451       goto fail;
 452
 453    fpme->post_vs = draw_pt_post_vs_create( draw );
 454    if (!fpme->post_vs)
 455       goto fail;
 456
 457    fpme->emit = draw_pt_emit_create( draw );
 458    if (!fpme->emit)
 459       goto fail;
 460
 461    fpme->so_emit = draw_pt_so_emit_create( draw );
 462    if (!fpme->so_emit)
 463       goto fail;
 464
 465    fpme->llvm = draw->llvm;
 466    if (!fpme->llvm)
 467       goto fail;
 468
 469    fpme->current_variant = NULL;
 470
 471    return &fpme->base;
 472
 473  fail:
 474    if (fpme)
 475       llvm_middle_end_destroy( &fpme->base );
 476
 477    return NULL;
 478 }