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