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