src/mesa/drivers/dri/i965/brw_shader.cpp

   1 /*
   2  * Copyright © 2010 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 extern "C" {
  25 #include "main/macros.h"
  26 #include "brw_context.h"
  27 }
  28 #include "brw_vs.h"
  29 #include "brw_vec4_gs.h"
  30 #include "brw_fs.h"
  31 #include "glsl/ir_optimization.h"
  32 #include "glsl/glsl_parser_extras.h"
  33 #include "main/shaderapi.h"
  34
  35 struct gl_shader *
  36 brw_new_shader(struct gl_context *ctx, GLuint name, GLuint type)
  37 {
  38    struct brw_shader *shader;
  39
  40    shader = rzalloc(NULL, struct brw_shader);
  41    if (shader) {
  42       shader->base.Type = type;
  43       shader->base.Name = name;
  44       _mesa_init_shader(ctx, &shader->base);
  45    }
  46
  47    return &shader->base;
  48 }
  49
  50 struct gl_shader_program *
  51 brw_new_shader_program(struct gl_context *ctx, GLuint name)
  52 {
  53    struct gl_shader_program *prog = rzalloc(NULL, struct gl_shader_program);
  54    if (prog) {
  55       prog->Name = name;
  56       _mesa_init_shader_program(ctx, prog);
  57    }
  58    return prog;
  59 }
  60
  61 /**
  62  * Performs a compile of the shader stages even when we don't know
  63  * what non-orthogonal state will be set, in the hope that it reflects
  64  * the eventual NOS used, and thus allows us to produce link failures.
  65  */
  66 static bool
  67 brw_shader_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
  68 {
  69    struct brw_context *brw = brw_context(ctx);
  70
  71    if (brw->precompile && !brw_fs_precompile(ctx, prog))
  72       return false;
  73
  74    if (brw->precompile && !brw_gs_precompile(ctx, prog))
  75       return false;
  76
  77    if (brw->precompile && !brw_vs_precompile(ctx, prog))
  78       return false;
  79
  80    return true;
  81 }
  82
  83 static void
  84 brw_lower_packing_builtins(struct brw_context *brw,
  85                            gl_shader_type shader_type,
  86                            exec_list *ir)
  87 {
  88    int ops = LOWER_PACK_SNORM_2x16
  89            | LOWER_UNPACK_SNORM_2x16
  90            | LOWER_PACK_UNORM_2x16
  91            | LOWER_UNPACK_UNORM_2x16
  92            | LOWER_PACK_SNORM_4x8
  93            | LOWER_UNPACK_SNORM_4x8
  94            | LOWER_PACK_UNORM_4x8
  95            | LOWER_UNPACK_UNORM_4x8;
  96
  97    if (brw->gen >= 7) {
  98       /* Gen7 introduced the f32to16 and f16to32 instructions, which can be
  99        * used to execute packHalf2x16 and unpackHalf2x16. For AOS code, no
 100        * lowering is needed. For SOA code, the Half2x16 ops must be
 101        * scalarized.
 102        */
 103       if (shader_type == MESA_SHADER_FRAGMENT) {
 104          ops |= LOWER_PACK_HALF_2x16_TO_SPLIT
 105              |  LOWER_UNPACK_HALF_2x16_TO_SPLIT;
 106       }
 107    } else {
 108       ops |= LOWER_PACK_HALF_2x16
 109           |  LOWER_UNPACK_HALF_2x16;
 110    }
 111
 112    lower_packing_builtins(ir, ops);
 113 }
 114
 115 GLboolean
 116 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
 117 {
 118    struct brw_context *brw = brw_context(ctx);
 119    unsigned int stage;
 120
 121    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
 122       struct brw_shader *shader =
 123          (struct brw_shader *)shProg->_LinkedShaders[stage];
 124
 125       if (!shader)
 126          continue;
 127
 128       struct gl_program *prog =
 129          ctx->Driver.NewProgram(ctx, _mesa_program_index_to_target(stage),
 130                                 shader->base.Name);
 131       if (!prog)
 132         return false;
 133       prog->Parameters = _mesa_new_parameter_list();
 134
 135       _mesa_copy_linked_program_data((gl_shader_type) stage, shProg, prog);
 136
 137       void *mem_ctx = ralloc_context(NULL);
 138       bool progress;
 139
 140       if (shader->ir)
 141          ralloc_free(shader->ir);
 142       shader->ir = new(shader) exec_list;
 143       clone_ir_list(mem_ctx, shader->ir, shader->base.ir);
 144
 145       /* lower_packing_builtins() inserts arithmetic instructions, so it
 146        * must precede lower_instructions().
 147        */
 148       brw_lower_packing_builtins(brw, (gl_shader_type) stage, shader->ir);
 149       do_mat_op_to_vec(shader->ir);
 150       const int bitfield_insert = brw->gen >= 7
 151                                   ? BITFIELD_INSERT_TO_BFM_BFI
 152                                   : 0;
 153       const int lrp_to_arith = brw->gen < 6 ? LRP_TO_ARITH : 0;
 154       lower_instructions(shader->ir,
 155                          MOD_TO_FRACT |
 156                          DIV_TO_MUL_RCP |
 157                          SUB_TO_ADD_NEG |
 158                          EXP_TO_EXP2 |
 159                          LOG_TO_LOG2 |
 160                          bitfield_insert |
 161                          lrp_to_arith |
 162                          LDEXP_TO_ARITH);
 163
 164       /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
 165        * if-statements need to be flattened.
 166        */
 167       if (brw->gen < 6)
 168          lower_if_to_cond_assign(shader->ir, 16);
 169
 170       do_lower_texture_projection(shader->ir);
 171       brw_lower_texture_gradients(brw, shader->ir);
 172       do_vec_index_to_cond_assign(shader->ir);
 173       lower_vector_insert(shader->ir, true);
 174       brw_do_cubemap_normalize(shader->ir);
 175       brw_do_lower_offset_arrays(shader->ir);
 176       brw_do_lower_unnormalized_offset(shader->ir);
 177       lower_noise(shader->ir);
 178       lower_quadop_vector(shader->ir, false);
 179
 180       bool input = true;
 181       bool output = stage == MESA_SHADER_FRAGMENT;
 182       bool temp = stage == MESA_SHADER_FRAGMENT;
 183       bool uniform = false;
 184
 185       bool lowered_variable_indexing =
 186          lower_variable_index_to_cond_assign(shader->ir,
 187                                              input, output, temp, uniform);
 188
 189       if (unlikely(brw->perf_debug && lowered_variable_indexing)) {
 190          perf_debug("Unsupported form of variable indexing in FS; falling "
 191                     "back to very inefficient code generation\n");
 192       }
 193
 194       /* FINISHME: Do this before the variable index lowering. */
 195       lower_ubo_reference(&shader->base, shader->ir);
 196
 197       do {
 198          progress = false;
 199
 200          if (stage == MESA_SHADER_FRAGMENT) {
 201             brw_do_channel_expressions(shader->ir);
 202             brw_do_vector_splitting(shader->ir);
 203          }
 204
 205          progress = do_lower_jumps(shader->ir, true, true,
 206                                    true, /* main return */
 207                                    false, /* continue */
 208                                    false /* loops */
 209                                    ) || progress;
 210
 211          progress = do_common_optimization(shader->ir, true, true, 32,
 212                                            &ctx->ShaderCompilerOptions[stage])
 213            || progress;
 214       } while (progress);
 215
 216       /* Make a pass over the IR to add state references for any built-in
 217        * uniforms that are used.  This has to be done now (during linking).
 218        * Code generation doesn't happen until the first time this shader is
 219        * used for rendering.  Waiting until then to generate the parameters is
 220        * too late.  At that point, the values for the built-in uniforms won't
 221        * get sent to the shader.
 222        */
 223       foreach_list(node, shader->ir) {
 224          ir_variable *var = ((ir_instruction *) node)->as_variable();
 225
 226          if ((var == NULL) || (var->mode != ir_var_uniform)
 227              || (strncmp(var->name, "gl_", 3) != 0))
 228             continue;
 229
 230          const ir_state_slot *const slots = var->state_slots;
 231          assert(var->state_slots != NULL);
 232
 233          for (unsigned int i = 0; i < var->num_state_slots; i++) {
 234             _mesa_add_state_reference(prog->Parameters,
 235                                       (gl_state_index *) slots[i].tokens);
 236          }
 237       }
 238
 239       validate_ir_tree(shader->ir);
 240
 241       reparent_ir(shader->ir, shader->ir);
 242       ralloc_free(mem_ctx);
 243
 244       do_set_program_inouts(shader->ir, prog, shader->base.Type);
 245
 246       prog->SamplersUsed = shader->base.active_samplers;
 247       _mesa_update_shader_textures_used(shProg, prog);
 248
 249       _mesa_reference_program(ctx, &shader->base.Program, prog);
 250
 251       brw_add_texrect_params(prog);
 252
 253       /* This has to be done last.  Any operation that can cause
 254        * prog->ParameterValues to get reallocated (e.g., anything that adds a
 255        * program constant) has to happen before creating this linkage.
 256        */
 257       _mesa_associate_uniform_storage(ctx, shProg, prog->Parameters);
 258
 259       _mesa_reference_program(ctx, &prog, NULL);
 260
 261       if (ctx->Shader.Flags & GLSL_DUMP) {
 262          printf("\n");
 263          printf("GLSL IR for linked %s program %d:\n",
 264                 _mesa_glsl_shader_target_name(shader->base.Type), shProg->Name);
 265          _mesa_print_ir(shader->base.ir, NULL);
 266          printf("\n");
 267       }
 268    }
 269
 270    if (ctx->Shader.Flags & GLSL_DUMP) {
 271       for (unsigned i = 0; i < shProg->NumShaders; i++) {
 272          const struct gl_shader *sh = shProg->Shaders[i];
 273          if (!sh)
 274             continue;
 275
 276          printf("GLSL %s shader %d source for linked program %d:\n",
 277                 _mesa_glsl_shader_target_name(sh->Type),
 278                 i,
 279                 shProg->Name);
 280          printf("%s", sh->Source);
 281          printf("\n");
 282       }
 283    }
 284
 285    if (!brw_shader_precompile(ctx, shProg))
 286       return false;
 287
 288    return true;
 289 }
 290
 291
 292 int
 293 brw_type_for_base_type(const struct glsl_type *type)
 294 {
 295    switch (type->base_type) {
 296    case GLSL_TYPE_FLOAT:
 297       return BRW_REGISTER_TYPE_F;
 298    case GLSL_TYPE_INT:
 299    case GLSL_TYPE_BOOL:
 300       return BRW_REGISTER_TYPE_D;
 301    case GLSL_TYPE_UINT:
 302       return BRW_REGISTER_TYPE_UD;
 303    case GLSL_TYPE_ARRAY:
 304       return brw_type_for_base_type(type->fields.array);
 305    case GLSL_TYPE_STRUCT:
 306    case GLSL_TYPE_SAMPLER:
 307    case GLSL_TYPE_ATOMIC_UINT:
 308       /* These should be overridden with the type of the member when
 309        * dereferenced into.  BRW_REGISTER_TYPE_UD seems like a likely
 310        * way to trip up if we don't.
 311        */
 312       return BRW_REGISTER_TYPE_UD;
 313    case GLSL_TYPE_VOID:
 314    case GLSL_TYPE_ERROR:
 315    case GLSL_TYPE_INTERFACE:
 316       assert(!"not reached");
 317       break;
 318    }
 319
 320    return BRW_REGISTER_TYPE_F;
 321 }
 322
 323 uint32_t
 324 brw_conditional_for_comparison(unsigned int op)
 325 {
 326    switch (op) {
 327    case ir_binop_less:
 328       return BRW_CONDITIONAL_L;
 329    case ir_binop_greater:
 330       return BRW_CONDITIONAL_G;
 331    case ir_binop_lequal:
 332       return BRW_CONDITIONAL_LE;
 333    case ir_binop_gequal:
 334       return BRW_CONDITIONAL_GE;
 335    case ir_binop_equal:
 336    case ir_binop_all_equal: /* same as equal for scalars */
 337       return BRW_CONDITIONAL_Z;
 338    case ir_binop_nequal:
 339    case ir_binop_any_nequal: /* same as nequal for scalars */
 340       return BRW_CONDITIONAL_NZ;
 341    default:
 342       assert(!"not reached: bad operation for comparison");
 343       return BRW_CONDITIONAL_NZ;
 344    }
 345 }
 346
 347 uint32_t
 348 brw_math_function(enum opcode op)
 349 {
 350    switch (op) {
 351    case SHADER_OPCODE_RCP:
 352       return BRW_MATH_FUNCTION_INV;
 353    case SHADER_OPCODE_RSQ:
 354       return BRW_MATH_FUNCTION_RSQ;
 355    case SHADER_OPCODE_SQRT:
 356       return BRW_MATH_FUNCTION_SQRT;
 357    case SHADER_OPCODE_EXP2:
 358       return BRW_MATH_FUNCTION_EXP;
 359    case SHADER_OPCODE_LOG2:
 360       return BRW_MATH_FUNCTION_LOG;
 361    case SHADER_OPCODE_POW:
 362       return BRW_MATH_FUNCTION_POW;
 363    case SHADER_OPCODE_SIN:
 364       return BRW_MATH_FUNCTION_SIN;
 365    case SHADER_OPCODE_COS:
 366       return BRW_MATH_FUNCTION_COS;
 367    case SHADER_OPCODE_INT_QUOTIENT:
 368       return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
 369    case SHADER_OPCODE_INT_REMAINDER:
 370       return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
 371    default:
 372       assert(!"not reached: unknown math function");
 373       return 0;
 374    }
 375 }
 376
 377 uint32_t
 378 brw_texture_offset(struct gl_context *ctx, ir_constant *offset)
 379 {
 380    /* If the driver does not support GL_ARB_gpu_shader5, the offset
 381     * must be constant.
 382     */
 383    assert(offset != NULL || ctx->Extensions.ARB_gpu_shader5);
 384
 385    if (!offset) return 0;  /* nonconstant offset; caller will handle it. */
 386
 387    signed char offsets[3];
 388    for (unsigned i = 0; i < offset->type->vector_elements; i++)
 389       offsets[i] = (signed char) offset->value.i[i];
 390
 391    /* Combine all three offsets into a single unsigned dword:
 392     *
 393     *    bits 11:8 - U Offset (X component)
 394     *    bits  7:4 - V Offset (Y component)
 395     *    bits  3:0 - R Offset (Z component)
 396     */
 397    unsigned offset_bits = 0;
 398    for (unsigned i = 0; i < offset->type->vector_elements; i++) {
 399       const unsigned shift = 4 * (2 - i);
 400       offset_bits |= (offsets[i] << shift) & (0xF << shift);
 401    }
 402    return offset_bits;
 403 }
 404
 405 const char *
 406 brw_instruction_name(enum opcode op)
 407 {
 408    char *fallback;
 409
 410    if (op < ARRAY_SIZE(opcode_descs) && opcode_descs[op].name)
 411       return opcode_descs[op].name;
 412
 413    switch (op) {
 414    case FS_OPCODE_FB_WRITE:
 415       return "fb_write";
 416
 417    case SHADER_OPCODE_RCP:
 418       return "rcp";
 419    case SHADER_OPCODE_RSQ:
 420       return "rsq";
 421    case SHADER_OPCODE_SQRT:
 422       return "sqrt";
 423    case SHADER_OPCODE_EXP2:
 424       return "exp2";
 425    case SHADER_OPCODE_LOG2:
 426       return "log2";
 427    case SHADER_OPCODE_POW:
 428       return "pow";
 429    case SHADER_OPCODE_INT_QUOTIENT:
 430       return "int_quot";
 431    case SHADER_OPCODE_INT_REMAINDER:
 432       return "int_rem";
 433    case SHADER_OPCODE_SIN:
 434       return "sin";
 435    case SHADER_OPCODE_COS:
 436       return "cos";
 437
 438    case SHADER_OPCODE_TEX:
 439       return "tex";
 440    case SHADER_OPCODE_TXD:
 441       return "txd";
 442    case SHADER_OPCODE_TXF:
 443       return "txf";
 444    case SHADER_OPCODE_TXL:
 445       return "txl";
 446    case SHADER_OPCODE_TXS:
 447       return "txs";
 448    case FS_OPCODE_TXB:
 449       return "txb";
 450    case SHADER_OPCODE_TXF_MS:
 451       return "txf_ms";
 452    case SHADER_OPCODE_TG4:
 453       return "tg4";
 454    case SHADER_OPCODE_TG4_OFFSET:
 455       return "tg4_offset";
 456
 457    case FS_OPCODE_DDX:
 458       return "ddx";
 459    case FS_OPCODE_DDY:
 460       return "ddy";
 461
 462    case FS_OPCODE_PIXEL_X:
 463       return "pixel_x";
 464    case FS_OPCODE_PIXEL_Y:
 465       return "pixel_y";
 466
 467    case FS_OPCODE_CINTERP:
 468       return "cinterp";
 469    case FS_OPCODE_LINTERP:
 470       return "linterp";
 471
 472    case FS_OPCODE_SPILL:
 473       return "spill";
 474    case FS_OPCODE_UNSPILL:
 475       return "unspill";
 476
 477    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
 478       return "uniform_pull_const";
 479    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
 480       return "uniform_pull_const_gen7";
 481    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
 482       return "varying_pull_const";
 483    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
 484       return "varying_pull_const_gen7";
 485
 486    case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
 487       return "mov_dispatch_to_flags";
 488    case FS_OPCODE_DISCARD_JUMP:
 489       return "discard_jump";
 490
 491    case FS_OPCODE_SET_SIMD4X2_OFFSET:
 492       return "set_simd4x2_offset";
 493
 494    case FS_OPCODE_PACK_HALF_2x16_SPLIT:
 495       return "pack_half_2x16_split";
 496    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
 497       return "unpack_half_2x16_split_x";
 498    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
 499       return "unpack_half_2x16_split_y";
 500
 501    case FS_OPCODE_PLACEHOLDER_HALT:
 502       return "placeholder_halt";
 503
 504    case VS_OPCODE_URB_WRITE:
 505       return "vs_urb_write";
 506    case VS_OPCODE_SCRATCH_READ:
 507       return "scratch_read";
 508    case VS_OPCODE_SCRATCH_WRITE:
 509       return "scratch_write";
 510    case VS_OPCODE_PULL_CONSTANT_LOAD:
 511       return "pull_constant_load";
 512    case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
 513       return "pull_constant_load_gen7";
 514    case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
 515       return "unpack_flags_simd4x2";
 516
 517    case GS_OPCODE_URB_WRITE:
 518       return "gs_urb_write";
 519    case GS_OPCODE_THREAD_END:
 520       return "gs_thread_end";
 521    case GS_OPCODE_SET_WRITE_OFFSET:
 522       return "set_write_offset";
 523    case GS_OPCODE_SET_VERTEX_COUNT:
 524       return "set_vertex_count";
 525    case GS_OPCODE_SET_DWORD_2_IMMED:
 526       return "set_dword_2_immed";
 527    case GS_OPCODE_PREPARE_CHANNEL_MASKS:
 528       return "prepare_channel_masks";
 529    case GS_OPCODE_SET_CHANNEL_MASKS:
 530       return "set_channel_masks";
 531
 532    default:
 533       /* Yes, this leaks.  It's in debug code, it should never occur, and if
 534        * it does, you should just add the case to the list above.
 535        */
 536       asprintf(&fallback, "op%d", op);
 537       return fallback;
 538    }
 539 }
 540
 541 bool
 542 backend_instruction::is_tex()
 543 {
 544    return (opcode == SHADER_OPCODE_TEX ||
 545            opcode == FS_OPCODE_TXB ||
 546            opcode == SHADER_OPCODE_TXD ||
 547            opcode == SHADER_OPCODE_TXF ||
 548            opcode == SHADER_OPCODE_TXF_MS ||
 549            opcode == SHADER_OPCODE_TXL ||
 550            opcode == SHADER_OPCODE_TXS ||
 551            opcode == SHADER_OPCODE_LOD ||
 552            opcode == SHADER_OPCODE_TG4 ||
 553            opcode == SHADER_OPCODE_TG4_OFFSET);
 554 }
 555
 556 bool
 557 backend_instruction::is_math()
 558 {
 559    return (opcode == SHADER_OPCODE_RCP ||
 560            opcode == SHADER_OPCODE_RSQ ||
 561            opcode == SHADER_OPCODE_SQRT ||
 562            opcode == SHADER_OPCODE_EXP2 ||
 563            opcode == SHADER_OPCODE_LOG2 ||
 564            opcode == SHADER_OPCODE_SIN ||
 565            opcode == SHADER_OPCODE_COS ||
 566            opcode == SHADER_OPCODE_INT_QUOTIENT ||
 567            opcode == SHADER_OPCODE_INT_REMAINDER ||
 568            opcode == SHADER_OPCODE_POW);
 569 }
 570
 571 bool
 572 backend_instruction::is_control_flow()
 573 {
 574    switch (opcode) {
 575    case BRW_OPCODE_DO:
 576    case BRW_OPCODE_WHILE:
 577    case BRW_OPCODE_IF:
 578    case BRW_OPCODE_ELSE:
 579    case BRW_OPCODE_ENDIF:
 580    case BRW_OPCODE_BREAK:
 581    case BRW_OPCODE_CONTINUE:
 582       return true;
 583    default:
 584       return false;
 585    }
 586 }
 587
 588 bool
 589 backend_instruction::can_do_source_mods()
 590 {
 591    switch (opcode) {
 592    case BRW_OPCODE_ADDC:
 593    case BRW_OPCODE_BFE:
 594    case BRW_OPCODE_BFI1:
 595    case BRW_OPCODE_BFI2:
 596    case BRW_OPCODE_BFREV:
 597    case BRW_OPCODE_CBIT:
 598    case BRW_OPCODE_FBH:
 599    case BRW_OPCODE_FBL:
 600    case BRW_OPCODE_SUBB:
 601       return false;
 602    default:
 603       return true;
 604    }
 605 }
 606
 607 void
 608 backend_visitor::dump_instructions()
 609 {
 610    int ip = 0;
 611    foreach_list(node, &this->instructions) {
 612       backend_instruction *inst = (backend_instruction *)node;
 613       printf("%d: ", ip++);
 614       dump_instruction(inst);
 615    }
 616 }
 617
 618
 619 /**
 620  * Sets up the starting offsets for the groups of binding table entries
 621  * commong to all pipeline stages.
 622  *
 623  * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
 624  * unused but also make sure that addition of small offsets to them will
 625  * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
 626  */
 627 void
 628 backend_visitor::assign_common_binding_table_offsets(uint32_t next_binding_table_offset)
 629 {
 630    int num_textures = _mesa_fls(prog->SamplersUsed);
 631
 632    stage_prog_data->binding_table.texture_start = next_binding_table_offset;
 633    next_binding_table_offset += num_textures;
 634
 635    if (shader) {
 636       stage_prog_data->binding_table.ubo_start = next_binding_table_offset;
 637       next_binding_table_offset += shader->base.NumUniformBlocks;
 638    } else {
 639       stage_prog_data->binding_table.ubo_start = 0xd0d0d0d0;
 640    }
 641
 642    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 643       stage_prog_data->binding_table.shader_time_start = next_binding_table_offset;
 644       next_binding_table_offset++;
 645    } else {
 646       stage_prog_data->binding_table.shader_time_start = 0xd0d0d0d0;
 647    }
 648
 649    if (prog->UsesGather) {
 650       stage_prog_data->binding_table.gather_texture_start = next_binding_table_offset;
 651       next_binding_table_offset += num_textures;
 652    } else {
 653       stage_prog_data->binding_table.gather_texture_start = 0xd0d0d0d0;
 654    }
 655
 656    /* This may or may not be used depending on how the compile goes. */
 657    stage_prog_data->binding_table.pull_constants_start = next_binding_table_offset;
 658    next_binding_table_offset++;
 659
 660    assert(next_binding_table_offset <= BRW_MAX_SURFACES);
 661
 662    /* prog_data->base.binding_table.size will be set by mark_surface_used. */
 663 }