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 /**
  61  * Prepare/validate middle part of the vertex pipeline.
  62  * NOTE: if you change this function, also look at the non-LLVM
  63  * function fetch_pipeline_prepare() for similar changes.
  64  */
  65 static void
  66 llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
  67                          unsigned in_prim,
  68                          unsigned opt,
  69                          unsigned *max_vertices )
  70 {
  71    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
  72    struct draw_context *draw = fpme->draw;
  73    struct draw_vertex_shader *vs = draw->vs.vertex_shader;
  74    struct draw_geometry_shader *gs = draw->gs.geometry_shader;
  75    const unsigned out_prim = gs ? gs->output_primitive : in_prim;
  76
  77    /* Add one to num_outputs because the pipeline occasionally tags on
  78     * an additional texcoord, eg for AA lines.
  79     */
  80    const unsigned nr = MAX2( vs->info.num_inputs,
  81                              vs->info.num_outputs + 1 );
  82
  83    fpme->input_prim = in_prim;
  84    fpme->opt = opt;
  85
  86    /* Always leave room for the vertex header whether we need it or
  87     * not.  It's hard to get rid of it in particular because of the
  88     * viewport code in draw_pt_post_vs.c.
  89     */
  90    fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
  91
  92
  93    /* XXX: it's not really gl rasterization rules we care about here,
  94     * but gl vs dx9 clip spaces.
  95     */
  96    draw_pt_post_vs_prepare( fpme->post_vs,
  97                             draw->clip_xy,
  98                             draw->clip_z,
  99                             draw->clip_user,
 100                             draw->guard_band_xy,
 101                             draw->identity_viewport,
 102                             (boolean)draw->rasterizer->gl_rasterization_rules,
 103                             (draw->vs.edgeflag_output ? TRUE : FALSE) );
 104
 105    draw_pt_so_emit_prepare( fpme->so_emit );
 106
 107    if (!(opt & PT_PIPELINE)) {
 108       draw_pt_emit_prepare( fpme->emit,
 109                             out_prim,
 110                             max_vertices );
 111
 112       *max_vertices = MAX2( *max_vertices, 4096 );
 113    }
 114    else {
 115       /* limit max fetches by limiting max_vertices */
 116       *max_vertices = 4096;
 117    }
 118
 119    /* return even number */
 120    *max_vertices = *max_vertices & ~1;
 121
 122    /* Find/create the vertex shader variant */
 123    {
 124       struct draw_llvm_variant_key *key;
 125       struct draw_llvm_variant *variant = NULL;
 126       struct draw_llvm_variant_list_item *li;
 127       struct llvm_vertex_shader *shader = llvm_vertex_shader(vs);
 128       char store[DRAW_LLVM_MAX_VARIANT_KEY_SIZE];
 129       unsigned i;
 130
 131       key = draw_llvm_make_variant_key(fpme->llvm, store);
 132
 133       /* Search shader's list of variants for the key */
 134       li = first_elem(&shader->variants);
 135       while (!at_end(&shader->variants, li)) {
 136          if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
 137             variant = li->base;
 138             break;
 139          }
 140          li = next_elem(li);
 141       }
 142
 143       if (variant) {
 144          /* found the variant, move to head of global list (for LRU) */
 145          move_to_head(&fpme->llvm->vs_variants_list,
 146                       &variant->list_item_global);
 147       }
 148       else {
 149          /* Need to create new variant */
 150
 151          /* First check if we've created too many variants.  If so, free
 152           * 25% of the LRU to avoid using too much memory.
 153           */
 154          if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
 155             /*
 156              * XXX: should we flush here ?
 157              */
 158             for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) {
 159                struct draw_llvm_variant_list_item *item;
 160                if (is_empty_list(&fpme->llvm->vs_variants_list)) {
 161                   break;
 162                }
 163                item = last_elem(&fpme->llvm->vs_variants_list);
 164                assert(item);
 165                assert(item->base);
 166                draw_llvm_destroy_variant(item->base);
 167             }
 168          }
 169
 170          variant = draw_llvm_create_variant(fpme->llvm, nr, key);
 171
 172          if (variant) {
 173             insert_at_head(&shader->variants, &variant->list_item_local);
 174             insert_at_head(&fpme->llvm->vs_variants_list,
 175                            &variant->list_item_global);
 176             fpme->llvm->nr_variants++;
 177             shader->variants_cached++;
 178          }
 179       }
 180
 181       fpme->current_variant = variant;
 182    }
 183 }
 184
 185
 186 /**
 187  * Bind/update constant buffer pointers, clip planes and viewport dims.
 188  * These are "light weight" parameters which aren't baked into the
 189  * generated code.  Updating these items is much cheaper than revalidating
 190  * and rebuilding the generated pipeline code.
 191  */
 192 static void
 193 llvm_middle_end_bind_parameters(struct draw_pt_middle_end *middle)
 194 {
 195    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 196    struct draw_context *draw = fpme->draw;
 197    unsigned i;
 198
 199    for (i = 0; i < Elements(fpme->llvm->jit_context.vs_constants); ++i) {
 200       fpme->llvm->jit_context.vs_constants[i] = draw->pt.user.vs_constants[i];
 201    }
 202
 203    for (i = 0; i < Elements(fpme->llvm->jit_context.gs_constants); ++i) {
 204       fpme->llvm->jit_context.gs_constants[i] = draw->pt.user.gs_constants[i];
 205    }
 206
 207    fpme->llvm->jit_context.planes =
 208       (float (*)[DRAW_TOTAL_CLIP_PLANES][4]) draw->pt.user.planes[0];
 209
 210    fpme->llvm->jit_context.viewport = (float *) draw->viewport.scale;
 211 }
 212
 213
 214 static void pipeline(struct llvm_middle_end *llvm,
 215                      const struct draw_vertex_info *vert_info,
 216                      const struct draw_prim_info *prim_info)
 217 {
 218    if (prim_info->linear)
 219       draw_pipeline_run_linear( llvm->draw,
 220                                 vert_info,
 221                                 prim_info);
 222    else
 223       draw_pipeline_run( llvm->draw,
 224                          vert_info,
 225                          prim_info );
 226 }
 227
 228 static void emit(struct pt_emit *emit,
 229                  const struct draw_vertex_info *vert_info,
 230                  const struct draw_prim_info *prim_info)
 231 {
 232    if (prim_info->linear) {
 233       draw_pt_emit_linear(emit, vert_info, prim_info);
 234    }
 235    else {
 236       draw_pt_emit(emit, vert_info, prim_info);
 237    }
 238 }
 239
 240 static void
 241 llvm_pipeline_generic( struct draw_pt_middle_end *middle,
 242                        const struct draw_fetch_info *fetch_info,
 243                        const struct draw_prim_info *prim_info )
 244 {
 245    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 246    struct draw_context *draw = fpme->draw;
 247    struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
 248    struct draw_prim_info gs_prim_info;
 249    struct draw_vertex_info llvm_vert_info;
 250    struct draw_vertex_info gs_vert_info;
 251    struct draw_vertex_info *vert_info;
 252    unsigned opt = fpme->opt;
 253    unsigned clipped = 0;
 254
 255    llvm_vert_info.count = fetch_info->count;
 256    llvm_vert_info.vertex_size = fpme->vertex_size;
 257    llvm_vert_info.stride = fpme->vertex_size;
 258    llvm_vert_info.verts =
 259       (struct vertex_header *)MALLOC(fpme->vertex_size *
 260                                      align(fetch_info->count,  lp_native_vector_width / 32));
 261    if (!llvm_vert_info.verts) {
 262       assert(0);
 263       return;
 264    }
 265
 266    if (fetch_info->linear)
 267       clipped = fpme->current_variant->jit_func( &fpme->llvm->jit_context,
 268                                        llvm_vert_info.verts,
 269                                        (const char **)draw->pt.user.vbuffer,
 270                                        fetch_info->start,
 271                                        fetch_info->count,
 272                                        fpme->vertex_size,
 273                                        draw->pt.vertex_buffer,
 274                                        draw->instance_id);
 275    else
 276       clipped = fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
 277                                             llvm_vert_info.verts,
 278                                             (const char **)draw->pt.user.vbuffer,
 279                                             fetch_info->elts,
 280                                             fetch_info->count,
 281                                             fpme->vertex_size,
 282                                             draw->pt.vertex_buffer,
 283                                             draw->instance_id);
 284
 285    /* Finished with fetch and vs:
 286     */
 287    fetch_info = NULL;
 288    vert_info = &llvm_vert_info;
 289
 290
 291    if ((opt & PT_SHADE) && gshader) {
 292       draw_geometry_shader_run(gshader,
 293                                draw->pt.user.gs_constants,
 294                                draw->pt.user.gs_constants_size,
 295                                vert_info,
 296                                prim_info,
 297                                &gs_vert_info,
 298                                &gs_prim_info);
 299
 300       FREE(vert_info->verts);
 301       vert_info = &gs_vert_info;
 302       prim_info = &gs_prim_info;
 303
 304       clipped = draw_pt_post_vs_run( fpme->post_vs, vert_info );
 305
 306    }
 307
 308    /* stream output needs to be done before clipping */
 309    draw_pt_so_emit( fpme->so_emit,
 310                     vert_info,
 311                     prim_info );
 312
 313    if (clipped) {
 314       opt |= PT_PIPELINE;
 315    }
 316
 317    /* Do we need to run the pipeline? Now will come here if clipped
 318     */
 319    if (opt & PT_PIPELINE) {
 320       pipeline( fpme,
 321                 vert_info,
 322                 prim_info );
 323    }
 324    else {
 325       emit( fpme->emit,
 326             vert_info,
 327             prim_info );
 328    }
 329    FREE(vert_info->verts);
 330 }
 331
 332
 333 static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 334                                  const unsigned *fetch_elts,
 335                                  unsigned fetch_count,
 336                                  const ushort *draw_elts,
 337                                  unsigned draw_count,
 338                                  unsigned prim_flags )
 339 {
 340    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 341    struct draw_fetch_info fetch_info;
 342    struct draw_prim_info prim_info;
 343
 344    fetch_info.linear = FALSE;
 345    fetch_info.start = 0;
 346    fetch_info.elts = fetch_elts;
 347    fetch_info.count = fetch_count;
 348
 349    prim_info.linear = FALSE;
 350    prim_info.start = 0;
 351    prim_info.count = draw_count;
 352    prim_info.elts = draw_elts;
 353    prim_info.prim = fpme->input_prim;
 354    prim_info.flags = prim_flags;
 355    prim_info.primitive_count = 1;
 356    prim_info.primitive_lengths = &draw_count;
 357
 358    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 359 }
 360
 361
 362 static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle,
 363                                        unsigned start,
 364                                        unsigned count,
 365                                        unsigned prim_flags)
 366 {
 367    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 368    struct draw_fetch_info fetch_info;
 369    struct draw_prim_info prim_info;
 370
 371    fetch_info.linear = TRUE;
 372    fetch_info.start = start;
 373    fetch_info.count = count;
 374    fetch_info.elts = NULL;
 375
 376    prim_info.linear = TRUE;
 377    prim_info.start = 0;
 378    prim_info.count = count;
 379    prim_info.elts = NULL;
 380    prim_info.prim = fpme->input_prim;
 381    prim_info.flags = prim_flags;
 382    prim_info.primitive_count = 1;
 383    prim_info.primitive_lengths = &count;
 384
 385    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 386 }
 387
 388
 389
 390 static boolean
 391 llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle,
 392                                  unsigned start,
 393                                  unsigned count,
 394                                  const ushort *draw_elts,
 395                                  unsigned draw_count,
 396                                  unsigned prim_flags )
 397 {
 398    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 399    struct draw_fetch_info fetch_info;
 400    struct draw_prim_info prim_info;
 401
 402    fetch_info.linear = TRUE;
 403    fetch_info.start = start;
 404    fetch_info.count = count;
 405    fetch_info.elts = NULL;
 406
 407    prim_info.linear = FALSE;
 408    prim_info.start = 0;
 409    prim_info.count = draw_count;
 410    prim_info.elts = draw_elts;
 411    prim_info.prim = fpme->input_prim;
 412    prim_info.flags = prim_flags;
 413    prim_info.primitive_count = 1;
 414    prim_info.primitive_lengths = &draw_count;
 415
 416    llvm_pipeline_generic( middle, &fetch_info, &prim_info );
 417
 418    return TRUE;
 419 }
 420
 421
 422
 423 static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
 424 {
 425    /* nothing to do */
 426 }
 427
 428 static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
 429 {
 430    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
 431
 432    if (fpme->fetch)
 433       draw_pt_fetch_destroy( fpme->fetch );
 434
 435    if (fpme->emit)
 436       draw_pt_emit_destroy( fpme->emit );
 437
 438    if (fpme->so_emit)
 439       draw_pt_so_emit_destroy( fpme->so_emit );
 440
 441    if (fpme->post_vs)
 442       draw_pt_post_vs_destroy( fpme->post_vs );
 443
 444    FREE(middle);
 445 }
 446
 447
 448 struct draw_pt_middle_end *
 449 draw_pt_fetch_pipeline_or_emit_llvm(struct draw_context *draw)
 450 {
 451    struct llvm_middle_end *fpme = 0;
 452
 453    if (!draw->llvm)
 454       return NULL;
 455
 456    fpme = CALLOC_STRUCT( llvm_middle_end );
 457    if (!fpme)
 458       goto fail;
 459
 460    fpme->base.prepare         = llvm_middle_end_prepare;
 461    fpme->base.bind_parameters = llvm_middle_end_bind_parameters;
 462    fpme->base.run             = llvm_middle_end_run;
 463    fpme->base.run_linear      = llvm_middle_end_linear_run;
 464    fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
 465    fpme->base.finish          = llvm_middle_end_finish;
 466    fpme->base.destroy         = llvm_middle_end_destroy;
 467
 468    fpme->draw = draw;
 469
 470    fpme->fetch = draw_pt_fetch_create( draw );
 471    if (!fpme->fetch)
 472       goto fail;
 473
 474    fpme->post_vs = draw_pt_post_vs_create( draw );
 475    if (!fpme->post_vs)
 476       goto fail;
 477
 478    fpme->emit = draw_pt_emit_create( draw );
 479    if (!fpme->emit)
 480       goto fail;
 481
 482    fpme->so_emit = draw_pt_so_emit_create( draw );
 483    if (!fpme->so_emit)
 484       goto fail;
 485
 486    fpme->llvm = draw->llvm;
 487    if (!fpme->llvm)
 488       goto fail;
 489
 490    fpme->current_variant = NULL;
 491
 492    return &fpme->base;
 493
 494  fail:
 495    if (fpme)
 496       llvm_middle_end_destroy( &fpme->base );
 497
 498    return NULL;
 499 }