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    const unsigned out_prim = (draw->gs.geometry_shader ?
  75                               draw->gs.geometry_shader->output_primitive :
  76                               in_prim);
  77
  78    /* Add one to num_outputs because the pipeline occasionally tags on
  79     * an additional texcoord, eg for AA lines.
  80     */
  81    const unsigned nr = MAX2( shader->base.info.num_inputs,
  82                              shader->base.info.num_outputs + 1 );
  83
  84    unsigned i;
  85
  86    fpme->input_prim = in_prim;
  87    fpme->opt = opt;
  88
  89    /* Always leave room for the vertex header whether we need it or
  90     * not.  It's hard to get rid of it in particular because of the
  91     * viewport code in draw_pt_post_vs.c.
  92     */
  93    fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
  94
  95
  96    /* XXX: it's not really gl rasterization rules we care about here,
  97     * but gl vs dx9 clip spaces.
  98     */
  99    draw_pt_post_vs_prepare( fpme->post_vs,
 100                             draw->clip_xy,
 101                             draw->clip_z,
 102                             draw->clip_user,
 103                             draw->guard_band_xy,
 104                             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 );
 109
 110    if (!(opt & PT_PIPELINE)) {
 111       draw_pt_emit_prepare( fpme->emit,
 112                             out_prim,
 113                             max_vertices );
 114
 115       *max_vertices = MAX2( *max_vertices, 4096 );
 116    }
 117    else {
 118       /* limit max fetches by limiting max_vertices */
 119       *max_vertices = 4096;
 120    }
 121
 122    /* return even number */
 123    *max_vertices = *max_vertices & ~1;
 124
 125    key = draw_llvm_make_variant_key(fpme->llvm, store);
 126
 127    /* Search shader's list of variants for the key */
 128    li = first_elem(&shader->variants);
 129    while (!at_end(&shader->variants, li)) {
 130       if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
 131          variant = li->base;
 132          break;
 133       }
 134       li = next_elem(li);
 135    }
 136
 137    if (variant) {
 138       /* found the variant, move to head of global list (for LRU) */
 139       move_to_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
 140    }
 141    else {
 142       /* Need to create new variant */
 143
 144       /* First check if we've created too many variants.  If so, free
 145        * 25% of the LRU to avoid using too much memory.
 146        */
 147       if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
 148          /*
 149           * XXX: should we flush here ?
 150           */
 151          for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) {
 152             struct draw_llvm_variant_list_item *item;
 153             if (is_empty_list(&fpme->llvm->vs_variants_list)) {
 154                break;
 155             }
 156             item = last_elem(&fpme->llvm->vs_variants_list);
 157             assert(item);
 158             assert(item->base);
 159             draw_llvm_destroy_variant(item->base);
 160          }
 161       }
 162
 163       variant = draw_llvm_create_variant(fpme->llvm, nr, key);
 164
 165       if (variant) {
 166          insert_at_head(&shader->variants, &variant->list_item_local);
 167          insert_at_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
 168          fpme->llvm->nr_variants++;
 169          shader->variants_cached++;
 170       }
 171    }
 172
 173    fpme->current_variant = variant;
 174
 175    for (i = 0; i < Elements(fpme->llvm->jit_context.vs_constants); ++i) {
 176       fpme->llvm->jit_context.vs_constants[i] =
 177          draw->pt.user.vs_constants[i];
 178    }
 179    for (i = 0; i < Elements(fpme->llvm->jit_context.gs_constants); ++i) {
 180       fpme->llvm->jit_context.gs_constants[i] =
 181          draw->pt.user.gs_constants[i];
 182    }
 183    fpme->llvm->jit_context.planes =
 184       (float (*) [DRAW_TOTAL_CLIP_PLANES][4]) draw->pt.user.planes[0];
 185    fpme->llvm->jit_context.viewport =
 186       (float *)draw->viewport.scale;
 187
 188 }
 189
 190
 191 static void pipeline(struct llvm_middle_end *llvm,
 192                      const struct draw_vertex_info *vert_info,
 193                      const struct draw_prim_info *prim_info)
 194 {
 195    if (prim_info->linear)
 196       draw_pipeline_run_linear( llvm->draw,
 197                                 vert_info,
 198                                 prim_info);
 199    else
 200       draw_pipeline_run( llvm->draw,
 201                          vert_info,
 202                          prim_info );
 203 }
 204
 205 static void emit(struct pt_emit *emit,
 206                  const struct draw_vertex_info *vert_info,
 207                  const struct draw_prim_info *prim_info)
 208 {
 209    if (prim_info->linear) {
 210       draw_pt_emit_linear(emit, vert_info, prim_info);
 211    }
 212    else {
 213       draw_pt_emit(emit, vert_info, prim_info);
 214    }
 215 }
 216
 217 static void
 218 llvm_pipeline_generic( struct draw_pt_middle_end *middle,
 219                        const struct draw_fetch_info *fetch_info,
 220                        const struct draw_prim_info *prim_info )
 221 {
 222    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 223    struct draw_context *draw = fpme->draw;
 224    struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
 225    struct draw_prim_info gs_prim_info;
 226    struct draw_vertex_info llvm_vert_info;
 227    struct draw_vertex_info gs_vert_info;
 228    struct draw_vertex_info *vert_info;
 229    unsigned opt = fpme->opt;
 230    unsigned clipped = 0;
 231
 232    llvm_vert_info.count = fetch_info->count;
 233    llvm_vert_info.vertex_size = fpme->vertex_size;
 234    llvm_vert_info.stride = fpme->vertex_size;
 235    llvm_vert_info.verts =
 236       (struct vertex_header *)MALLOC(fpme->vertex_size *
 237                                      align(fetch_info->count,  lp_native_vector_width / 32));
 238    if (!llvm_vert_info.verts) {
 239       assert(0);
 240       return;
 241    }
 242
 243    if (fetch_info->linear)
 244       clipped = fpme->current_variant->jit_func( &fpme->llvm->jit_context,
 245                                        llvm_vert_info.verts,
 246                                        (const char **)draw->pt.user.vbuffer,
 247                                        fetch_info->start,
 248                                        fetch_info->count,
 249                                        fpme->vertex_size,
 250                                        draw->pt.vertex_buffer,
 251                                        draw->instance_id);
 252    else
 253       clipped = fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
 254                                             llvm_vert_info.verts,
 255                                             (const char **)draw->pt.user.vbuffer,
 256                                             fetch_info->elts,
 257                                             fetch_info->count,
 258                                             fpme->vertex_size,
 259                                             draw->pt.vertex_buffer,
 260                                             draw->instance_id);
 261
 262    /* Finished with fetch and vs:
 263     */
 264    fetch_info = NULL;
 265    vert_info = &llvm_vert_info;
 266
 267
 268    if ((opt & PT_SHADE) && gshader) {
 269       draw_geometry_shader_run(gshader,
 270                                draw->pt.user.gs_constants,
 271                                draw->pt.user.gs_constants_size,
 272                                vert_info,
 273                                prim_info,
 274                                &gs_vert_info,
 275                                &gs_prim_info);
 276
 277       FREE(vert_info->verts);
 278       vert_info = &gs_vert_info;
 279       prim_info = &gs_prim_info;
 280
 281       clipped = draw_pt_post_vs_run( fpme->post_vs, vert_info );
 282
 283    }
 284
 285    /* stream output needs to be done before clipping */
 286    draw_pt_so_emit( fpme->so_emit,
 287                     vert_info,
 288                     prim_info );
 289
 290    if (clipped) {
 291       opt |= PT_PIPELINE;
 292    }
 293
 294    /* Do we need to run the pipeline? Now will come here if clipped
 295     */
 296    if (opt & PT_PIPELINE) {
 297       pipeline( fpme,
 298                 vert_info,
 299                 prim_info );
 300    }
 301    else {
 302       emit( fpme->emit,
 303             vert_info,
 304             prim_info );
 305    }
 306    FREE(vert_info->verts);
 307 }
 308
 309
 310 static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 311                                  const unsigned *fetch_elts,
 312                                  unsigned fetch_count,
 313                                  const ushort *draw_elts,
 314                                  unsigned draw_count,
 315                                  unsigned prim_flags )
 316 {
 317    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 318    struct draw_fetch_info fetch_info;
 319    struct draw_prim_info prim_info;
 320
 321    fetch_info.linear = FALSE;
 322    fetch_info.start = 0;
 323    fetch_info.elts = fetch_elts;
 324    fetch_info.count = fetch_count;
 325
 326    prim_info.linear = FALSE;
 327    prim_info.start = 0;
 328    prim_info.count = draw_count;
 329    prim_info.elts = draw_elts;
 330    prim_info.prim = fpme->input_prim;
 331    prim_info.flags = prim_flags;
 332    prim_info.primitive_count = 1;
 333    prim_info.primitive_lengths = &draw_count;
 334
 335    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 336 }
 337
 338
 339 static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle,
 340                                        unsigned start,
 341                                        unsigned count,
 342                                        unsigned prim_flags)
 343 {
 344    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 345    struct draw_fetch_info fetch_info;
 346    struct draw_prim_info prim_info;
 347
 348    fetch_info.linear = TRUE;
 349    fetch_info.start = start;
 350    fetch_info.count = count;
 351    fetch_info.elts = NULL;
 352
 353    prim_info.linear = TRUE;
 354    prim_info.start = 0;
 355    prim_info.count = count;
 356    prim_info.elts = NULL;
 357    prim_info.prim = fpme->input_prim;
 358    prim_info.flags = prim_flags;
 359    prim_info.primitive_count = 1;
 360    prim_info.primitive_lengths = &count;
 361
 362    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 363 }
 364
 365
 366
 367 static boolean
 368 llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle,
 369                                  unsigned start,
 370                                  unsigned count,
 371                                  const ushort *draw_elts,
 372                                  unsigned draw_count,
 373                                  unsigned prim_flags )
 374 {
 375    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 376    struct draw_fetch_info fetch_info;
 377    struct draw_prim_info prim_info;
 378
 379    fetch_info.linear = TRUE;
 380    fetch_info.start = start;
 381    fetch_info.count = count;
 382    fetch_info.elts = NULL;
 383
 384    prim_info.linear = FALSE;
 385    prim_info.start = 0;
 386    prim_info.count = draw_count;
 387    prim_info.elts = draw_elts;
 388    prim_info.prim = fpme->input_prim;
 389    prim_info.flags = prim_flags;
 390    prim_info.primitive_count = 1;
 391    prim_info.primitive_lengths = &draw_count;
 392
 393    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 394
 395    return TRUE;
 396 }
 397
 398
 399
 400 static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
 401 {
 402    /* nothing to do */
 403 }
 404
 405 static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
 406 {
 407    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 408
 409    if (fpme->fetch)
 410       draw_pt_fetch_destroy( fpme->fetch );
 411
 412    if (fpme->emit)
 413       draw_pt_emit_destroy( fpme->emit );
 414
 415    if (fpme->so_emit)
 416       draw_pt_so_emit_destroy( fpme->so_emit );
 417
 418    if (fpme->post_vs)
 419       draw_pt_post_vs_destroy( fpme->post_vs );
 420
 421    FREE(middle);
 422 }
 423
 424
 425 struct draw_pt_middle_end *
 426 draw_pt_fetch_pipeline_or_emit_llvm(struct draw_context *draw)
 427 {
 428    struct llvm_middle_end *fpme = 0;
 429
 430    if (!draw->llvm)
 431       return NULL;
 432
 433    fpme = CALLOC_STRUCT( llvm_middle_end );
 434    if (!fpme)
 435       goto fail;
 436
 437    fpme->base.prepare         = llvm_middle_end_prepare;
 438    fpme->base.run             = llvm_middle_end_run;
 439    fpme->base.run_linear      = llvm_middle_end_linear_run;
 440    fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
 441    fpme->base.finish          = llvm_middle_end_finish;
 442    fpme->base.destroy         = llvm_middle_end_destroy;
 443
 444    fpme->draw = draw;
 445
 446    fpme->fetch = draw_pt_fetch_create( draw );
 447    if (!fpme->fetch)
 448       goto fail;
 449
 450    fpme->post_vs = draw_pt_post_vs_create( draw );
 451    if (!fpme->post_vs)
 452       goto fail;
 453
 454    fpme->emit = draw_pt_emit_create( draw );
 455    if (!fpme->emit)
 456       goto fail;
 457
 458    fpme->so_emit = draw_pt_so_emit_create( draw );
 459    if (!fpme->so_emit)
 460       goto fail;
 461
 462    fpme->llvm = draw->llvm;
 463    if (!fpme->llvm)
 464       goto fail;
 465
 466    fpme->current_variant = NULL;
 467
 468    return &fpme->base;
 469
 470  fail:
 471    if (fpme)
 472       llvm_middle_end_destroy( &fpme->base );
 473
 474    return NULL;
 475 }