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       lower_noise(shader->ir);
 176       lower_quadop_vector(shader->ir, false);
 177
 178       bool input = true;
 179       bool output = stage == MESA_SHADER_FRAGMENT;
 180       bool temp = stage == MESA_SHADER_FRAGMENT;
 181       bool uniform = false;
 182
 183       bool lowered_variable_indexing =
 184          lower_variable_index_to_cond_assign(shader->ir,
 185                                              input, output, temp, uniform);
 186
 187       if (unlikely(brw->perf_debug && lowered_variable_indexing)) {
 188          perf_debug("Unsupported form of variable indexing in FS; falling "
 189                     "back to very inefficient code generation\n");
 190       }
 191
 192       /* FINISHME: Do this before the variable index lowering. */
 193       lower_ubo_reference(&shader->base, shader->ir);
 194
 195       do {
 196          progress = false;
 197
 198          if (stage == MESA_SHADER_FRAGMENT) {
 199             brw_do_channel_expressions(shader->ir);
 200             brw_do_vector_splitting(shader->ir);
 201          }
 202
 203          progress = do_lower_jumps(shader->ir, true, true,
 204                                    true, /* main return */
 205                                    false, /* continue */
 206                                    false /* loops */
 207                                    ) || progress;
 208
 209          progress = do_common_optimization(shader->ir, true, true, 32,
 210                                            &ctx->ShaderCompilerOptions[stage])
 211            || progress;
 212       } while (progress);
 213
 214       /* Make a pass over the IR to add state references for any built-in
 215        * uniforms that are used.  This has to be done now (during linking).
 216        * Code generation doesn't happen until the first time this shader is
 217        * used for rendering.  Waiting until then to generate the parameters is
 218        * too late.  At that point, the values for the built-in uniforms won't
 219        * get sent to the shader.
 220        */
 221       foreach_list(node, shader->ir) {
 222          ir_variable *var = ((ir_instruction *) node)->as_variable();
 223
 224          if ((var == NULL) || (var->mode != ir_var_uniform)
 225              || (strncmp(var->name, "gl_", 3) != 0))
 226             continue;
 227
 228          const ir_state_slot *const slots = var->state_slots;
 229          assert(var->state_slots != NULL);
 230
 231          for (unsigned int i = 0; i < var->num_state_slots; i++) {
 232             _mesa_add_state_reference(prog->Parameters,
 233                                       (gl_state_index *) slots[i].tokens);
 234          }
 235       }
 236
 237       validate_ir_tree(shader->ir);
 238
 239       reparent_ir(shader->ir, shader->ir);
 240       ralloc_free(mem_ctx);
 241
 242       do_set_program_inouts(shader->ir, prog, shader->base.Type);
 243
 244       prog->SamplersUsed = shader->base.active_samplers;
 245       _mesa_update_shader_textures_used(shProg, prog);
 246
 247       _mesa_reference_program(ctx, &shader->base.Program, prog);
 248
 249       brw_add_texrect_params(prog);
 250
 251       /* This has to be done last.  Any operation that can cause
 252        * prog->ParameterValues to get reallocated (e.g., anything that adds a
 253        * program constant) has to happen before creating this linkage.
 254        */
 255       _mesa_associate_uniform_storage(ctx, shProg, prog->Parameters);
 256
 257       _mesa_reference_program(ctx, &prog, NULL);
 258
 259       if (ctx->Shader.Flags & GLSL_DUMP) {
 260          printf("\n");
 261          printf("GLSL IR for linked %s program %d:\n",
 262                 _mesa_glsl_shader_target_name(shader->base.Type), shProg->Name);
 263          _mesa_print_ir(shader->base.ir, NULL);
 264          printf("\n");
 265       }
 266    }
 267
 268    if (ctx->Shader.Flags & GLSL_DUMP) {
 269       for (unsigned i = 0; i < shProg->NumShaders; i++) {
 270          const struct gl_shader *sh = shProg->Shaders[i];
 271          if (!sh)
 272             continue;
 273
 274          printf("GLSL %s shader %d source for linked program %d:\n",
 275                 _mesa_glsl_shader_target_name(sh->Type),
 276                 i,
 277                 shProg->Name);
 278          printf("%s", sh->Source);
 279          printf("\n");
 280       }
 281    }
 282
 283    if (!brw_shader_precompile(ctx, shProg))
 284       return false;
 285
 286    return true;
 287 }
 288
 289
 290 int
 291 brw_type_for_base_type(const struct glsl_type *type)
 292 {
 293    switch (type->base_type) {
 294    case GLSL_TYPE_FLOAT:
 295       return BRW_REGISTER_TYPE_F;
 296    case GLSL_TYPE_INT:
 297    case GLSL_TYPE_BOOL:
 298       return BRW_REGISTER_TYPE_D;
 299    case GLSL_TYPE_UINT:
 300       return BRW_REGISTER_TYPE_UD;
 301    case GLSL_TYPE_ARRAY:
 302       return brw_type_for_base_type(type->fields.array);
 303    case GLSL_TYPE_STRUCT:
 304    case GLSL_TYPE_SAMPLER:
 305       /* These should be overridden with the type of the member when
 306        * dereferenced into.  BRW_REGISTER_TYPE_UD seems like a likely
 307        * way to trip up if we don't.
 308        */
 309       return BRW_REGISTER_TYPE_UD;
 310    case GLSL_TYPE_VOID:
 311    case GLSL_TYPE_ERROR:
 312    case GLSL_TYPE_INTERFACE:
 313       assert(!"not reached");
 314       break;
 315    }
 316
 317    return BRW_REGISTER_TYPE_F;
 318 }
 319
 320 uint32_t
 321 brw_conditional_for_comparison(unsigned int op)
 322 {
 323    switch (op) {
 324    case ir_binop_less:
 325       return BRW_CONDITIONAL_L;
 326    case ir_binop_greater:
 327       return BRW_CONDITIONAL_G;
 328    case ir_binop_lequal:
 329       return BRW_CONDITIONAL_LE;
 330    case ir_binop_gequal:
 331       return BRW_CONDITIONAL_GE;
 332    case ir_binop_equal:
 333    case ir_binop_all_equal: /* same as equal for scalars */
 334       return BRW_CONDITIONAL_Z;
 335    case ir_binop_nequal:
 336    case ir_binop_any_nequal: /* same as nequal for scalars */
 337       return BRW_CONDITIONAL_NZ;
 338    default:
 339       assert(!"not reached: bad operation for comparison");
 340       return BRW_CONDITIONAL_NZ;
 341    }
 342 }
 343
 344 uint32_t
 345 brw_math_function(enum opcode op)
 346 {
 347    switch (op) {
 348    case SHADER_OPCODE_RCP:
 349       return BRW_MATH_FUNCTION_INV;
 350    case SHADER_OPCODE_RSQ:
 351       return BRW_MATH_FUNCTION_RSQ;
 352    case SHADER_OPCODE_SQRT:
 353       return BRW_MATH_FUNCTION_SQRT;
 354    case SHADER_OPCODE_EXP2:
 355       return BRW_MATH_FUNCTION_EXP;
 356    case SHADER_OPCODE_LOG2:
 357       return BRW_MATH_FUNCTION_LOG;
 358    case SHADER_OPCODE_POW:
 359       return BRW_MATH_FUNCTION_POW;
 360    case SHADER_OPCODE_SIN:
 361       return BRW_MATH_FUNCTION_SIN;
 362    case SHADER_OPCODE_COS:
 363       return BRW_MATH_FUNCTION_COS;
 364    case SHADER_OPCODE_INT_QUOTIENT:
 365       return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
 366    case SHADER_OPCODE_INT_REMAINDER:
 367       return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
 368    default:
 369       assert(!"not reached: unknown math function");
 370       return 0;
 371    }
 372 }
 373
 374 uint32_t
 375 brw_texture_offset(struct gl_context *ctx, ir_constant *offset)
 376 {
 377    /* If the driver does not support GL_ARB_gpu_shader5, the offset
 378     * must be constant.
 379     */
 380    assert(offset != NULL || ctx->Extensions.ARB_gpu_shader5);
 381
 382    if (!offset) return 0;  /* nonconstant offset; caller will handle it. */
 383
 384    signed char offsets[3];
 385    for (unsigned i = 0; i < offset->type->vector_elements; i++)
 386       offsets[i] = (signed char) offset->value.i[i];
 387
 388    /* Combine all three offsets into a single unsigned dword:
 389     *
 390     *    bits 11:8 - U Offset (X component)
 391     *    bits  7:4 - V Offset (Y component)
 392     *    bits  3:0 - R Offset (Z component)
 393     */
 394    unsigned offset_bits = 0;
 395    for (unsigned i = 0; i < offset->type->vector_elements; i++) {
 396       const unsigned shift = 4 * (2 - i);
 397       offset_bits |= (offsets[i] << shift) & (0xF << shift);
 398    }
 399    return offset_bits;
 400 }
 401
 402 const char *
 403 brw_instruction_name(enum opcode op)
 404 {
 405    char *fallback;
 406
 407    if (op < ARRAY_SIZE(opcode_descs) && opcode_descs[op].name)
 408       return opcode_descs[op].name;
 409
 410    switch (op) {
 411    case FS_OPCODE_FB_WRITE:
 412       return "fb_write";
 413
 414    case SHADER_OPCODE_RCP:
 415       return "rcp";
 416    case SHADER_OPCODE_RSQ:
 417       return "rsq";
 418    case SHADER_OPCODE_SQRT:
 419       return "sqrt";
 420    case SHADER_OPCODE_EXP2:
 421       return "exp2";
 422    case SHADER_OPCODE_LOG2:
 423       return "log2";
 424    case SHADER_OPCODE_POW:
 425       return "pow";
 426    case SHADER_OPCODE_INT_QUOTIENT:
 427       return "int_quot";
 428    case SHADER_OPCODE_INT_REMAINDER:
 429       return "int_rem";
 430    case SHADER_OPCODE_SIN:
 431       return "sin";
 432    case SHADER_OPCODE_COS:
 433       return "cos";
 434
 435    case SHADER_OPCODE_TEX:
 436       return "tex";
 437    case SHADER_OPCODE_TXD:
 438       return "txd";
 439    case SHADER_OPCODE_TXF:
 440       return "txf";
 441    case SHADER_OPCODE_TXL:
 442       return "txl";
 443    case SHADER_OPCODE_TXS:
 444       return "txs";
 445    case FS_OPCODE_TXB:
 446       return "txb";
 447    case SHADER_OPCODE_TXF_MS:
 448       return "txf_ms";
 449    case SHADER_OPCODE_TG4:
 450       return "tg4";
 451    case SHADER_OPCODE_TG4_OFFSET:
 452       return "tg4_offset";
 453
 454    case FS_OPCODE_DDX:
 455       return "ddx";
 456    case FS_OPCODE_DDY:
 457       return "ddy";
 458
 459    case FS_OPCODE_PIXEL_X:
 460       return "pixel_x";
 461    case FS_OPCODE_PIXEL_Y:
 462       return "pixel_y";
 463
 464    case FS_OPCODE_CINTERP:
 465       return "cinterp";
 466    case FS_OPCODE_LINTERP:
 467       return "linterp";
 468
 469    case FS_OPCODE_SPILL:
 470       return "spill";
 471    case FS_OPCODE_UNSPILL:
 472       return "unspill";
 473
 474    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
 475       return "uniform_pull_const";
 476    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
 477       return "uniform_pull_const_gen7";
 478    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
 479       return "varying_pull_const";
 480    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
 481       return "varying_pull_const_gen7";
 482
 483    case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
 484       return "mov_dispatch_to_flags";
 485    case FS_OPCODE_DISCARD_JUMP:
 486       return "discard_jump";
 487
 488    case FS_OPCODE_SET_SIMD4X2_OFFSET:
 489       return "set_simd4x2_offset";
 490
 491    case FS_OPCODE_PACK_HALF_2x16_SPLIT:
 492       return "pack_half_2x16_split";
 493    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
 494       return "unpack_half_2x16_split_x";
 495    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
 496       return "unpack_half_2x16_split_y";
 497
 498    case FS_OPCODE_PLACEHOLDER_HALT:
 499       return "placeholder_halt";
 500
 501    case VS_OPCODE_URB_WRITE:
 502       return "vs_urb_write";
 503    case VS_OPCODE_SCRATCH_READ:
 504       return "scratch_read";
 505    case VS_OPCODE_SCRATCH_WRITE:
 506       return "scratch_write";
 507    case VS_OPCODE_PULL_CONSTANT_LOAD:
 508       return "pull_constant_load";
 509    case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
 510       return "pull_constant_load_gen7";
 511    case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
 512       return "unpack_flags_simd4x2";
 513
 514    case GS_OPCODE_URB_WRITE:
 515       return "gs_urb_write";
 516    case GS_OPCODE_THREAD_END:
 517       return "gs_thread_end";
 518    case GS_OPCODE_SET_WRITE_OFFSET:
 519       return "set_write_offset";
 520    case GS_OPCODE_SET_VERTEX_COUNT:
 521       return "set_vertex_count";
 522    case GS_OPCODE_SET_DWORD_2_IMMED:
 523       return "set_dword_2_immed";
 524    case GS_OPCODE_PREPARE_CHANNEL_MASKS:
 525       return "prepare_channel_masks";
 526    case GS_OPCODE_SET_CHANNEL_MASKS:
 527       return "set_channel_masks";
 528
 529    default:
 530       /* Yes, this leaks.  It's in debug code, it should never occur, and if
 531        * it does, you should just add the case to the list above.
 532        */
 533       asprintf(&fallback, "op%d", op);
 534       return fallback;
 535    }
 536 }
 537
 538 bool
 539 backend_instruction::is_tex()
 540 {
 541    return (opcode == SHADER_OPCODE_TEX ||
 542            opcode == FS_OPCODE_TXB ||
 543            opcode == SHADER_OPCODE_TXD ||
 544            opcode == SHADER_OPCODE_TXF ||
 545            opcode == SHADER_OPCODE_TXF_MS ||
 546            opcode == SHADER_OPCODE_TXL ||
 547            opcode == SHADER_OPCODE_TXS ||
 548            opcode == SHADER_OPCODE_LOD ||
 549            opcode == SHADER_OPCODE_TG4 ||
 550            opcode == SHADER_OPCODE_TG4_OFFSET);
 551 }
 552
 553 bool
 554 backend_instruction::is_math()
 555 {
 556    return (opcode == SHADER_OPCODE_RCP ||
 557            opcode == SHADER_OPCODE_RSQ ||
 558            opcode == SHADER_OPCODE_SQRT ||
 559            opcode == SHADER_OPCODE_EXP2 ||
 560            opcode == SHADER_OPCODE_LOG2 ||
 561            opcode == SHADER_OPCODE_SIN ||
 562            opcode == SHADER_OPCODE_COS ||
 563            opcode == SHADER_OPCODE_INT_QUOTIENT ||
 564            opcode == SHADER_OPCODE_INT_REMAINDER ||
 565            opcode == SHADER_OPCODE_POW);
 566 }
 567
 568 bool
 569 backend_instruction::is_control_flow()
 570 {
 571    switch (opcode) {
 572    case BRW_OPCODE_DO:
 573    case BRW_OPCODE_WHILE:
 574    case BRW_OPCODE_IF:
 575    case BRW_OPCODE_ELSE:
 576    case BRW_OPCODE_ENDIF:
 577    case BRW_OPCODE_BREAK:
 578    case BRW_OPCODE_CONTINUE:
 579       return true;
 580    default:
 581       return false;
 582    }
 583 }
 584
 585 bool
 586 backend_instruction::can_do_source_mods()
 587 {
 588    switch (opcode) {
 589    case BRW_OPCODE_ADDC:
 590    case BRW_OPCODE_BFE:
 591    case BRW_OPCODE_BFI1:
 592    case BRW_OPCODE_BFI2:
 593    case BRW_OPCODE_BFREV:
 594    case BRW_OPCODE_CBIT:
 595    case BRW_OPCODE_FBH:
 596    case BRW_OPCODE_FBL:
 597    case BRW_OPCODE_SUBB:
 598       return false;
 599    default:
 600       return true;
 601    }
 602 }
 603
 604 void
 605 backend_visitor::dump_instructions()
 606 {
 607    int ip = 0;
 608    foreach_list(node, &this->instructions) {
 609       backend_instruction *inst = (backend_instruction *)node;
 610       printf("%d: ", ip++);
 611       dump_instruction(inst);
 612    }
 613 }
 614
 615
 616 /**
 617  * Sets up the starting offsets for the groups of binding table entries
 618  * commong to all pipeline stages.
 619  *
 620  * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
 621  * unused but also make sure that addition of small offsets to them will
 622  * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
 623  */
 624 void
 625 backend_visitor::assign_common_binding_table_offsets(uint32_t next_binding_table_offset)
 626 {
 627    int num_textures = _mesa_fls(prog->SamplersUsed);
 628
 629    stage_prog_data->binding_table.texture_start = next_binding_table_offset;
 630    next_binding_table_offset += num_textures;
 631
 632    if (shader) {
 633       stage_prog_data->binding_table.ubo_start = next_binding_table_offset;
 634       next_binding_table_offset += shader->base.NumUniformBlocks;
 635    } else {
 636       stage_prog_data->binding_table.ubo_start = 0xd0d0d0d0;
 637    }
 638
 639    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 640       stage_prog_data->binding_table.shader_time_start = next_binding_table_offset;
 641       next_binding_table_offset++;
 642    } else {
 643       stage_prog_data->binding_table.shader_time_start = 0xd0d0d0d0;
 644    }
 645
 646    if (prog->UsesGather) {
 647       stage_prog_data->binding_table.gather_texture_start = next_binding_table_offset;
 648       next_binding_table_offset += num_textures;
 649    } else {
 650       stage_prog_data->binding_table.gather_texture_start = 0xd0d0d0d0;
 651    }
 652
 653    /* This may or may not be used depending on how the compile goes. */
 654    stage_prog_data->binding_table.pull_constants_start = next_binding_table_offset;
 655    next_binding_table_offset++;
 656
 657    assert(next_binding_table_offset <= BRW_MAX_SURFACES);
 658
 659    /* prog_data->base.binding_table.size will be set by mark_surface_used. */
 660 }