src/intel/compiler/brw_fs.h

   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  * Authors:
  24  *    Eric Anholt <eric@anholt.net>
  25  *
  26  */
  27
  28 #ifndef BRW_FS_H
  29 #define BRW_FS_H
  30
  31 #include "brw_shader.h"
  32 #include "brw_ir_fs.h"
  33 #include "brw_fs_builder.h"
  34 #include "compiler/nir/nir.h"
  35
  36 struct bblock_t;
  37 namespace {
  38    struct acp_entry;
  39 }
  40
  41 namespace brw {
  42    class fs_live_variables;
  43 }
  44
  45 struct brw_gs_compile;
  46
  47 static inline fs_reg
  48 offset(const fs_reg &reg, const brw::fs_builder &bld, unsigned delta)
  49 {
  50    return offset(reg, bld.dispatch_width(), delta);
  51 }
  52
  53 #define UBO_START ((1 << 16) - 4)
  54
  55 struct shader_stats {
  56    const char *scheduler_mode;
  57    unsigned promoted_constants;
  58 };
  59
  60 /**
  61  * The fragment shader front-end.
  62  *
  63  * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
  64  */
  65 class fs_visitor : public backend_shader
  66 {
  67 public:
  68    fs_visitor(const struct brw_compiler *compiler, void *log_data,
  69               void *mem_ctx,
  70               const brw_base_prog_key *key,
  71               struct brw_stage_prog_data *prog_data,
  72               struct gl_program *prog,
  73               const nir_shader *shader,
  74               unsigned dispatch_width,
  75               int shader_time_index,
  76               const struct brw_vue_map *input_vue_map = NULL);
  77    fs_visitor(const struct brw_compiler *compiler, void *log_data,
  78               void *mem_ctx,
  79               struct brw_gs_compile *gs_compile,
  80               struct brw_gs_prog_data *prog_data,
  81               const nir_shader *shader,
  82               int shader_time_index);
  83    void init();
  84    ~fs_visitor();
  85
  86    fs_reg vgrf(const glsl_type *const type);
  87    void import_uniforms(fs_visitor *v);
  88
  89    void VARYING_PULL_CONSTANT_LOAD(const brw::fs_builder &bld,
  90                                    const fs_reg &dst,
  91                                    const fs_reg &surf_index,
  92                                    const fs_reg &varying_offset,
  93                                    uint32_t const_offset);
  94    void DEP_RESOLVE_MOV(const brw::fs_builder &bld, int grf);
  95
  96    bool run_fs(bool allow_spilling, bool do_rep_send);
  97    bool run_vs();
  98    bool run_tcs();
  99    bool run_tes();
 100    bool run_gs();
 101    bool run_cs(unsigned min_dispatch_width);
 102    void optimize();
 103    void allocate_registers(unsigned min_dispatch_width, bool allow_spilling);
 104    void setup_fs_payload_gen4();
 105    void setup_fs_payload_gen6();
 106    void setup_vs_payload();
 107    void setup_gs_payload();
 108    void setup_cs_payload();
 109    bool fixup_sends_duplicate_payload();
 110    void fixup_3src_null_dest();
 111    void assign_curb_setup();
 112    void assign_urb_setup();
 113    void convert_attr_sources_to_hw_regs(fs_inst *inst);
 114    void assign_vs_urb_setup();
 115    void assign_tcs_urb_setup();
 116    void assign_tes_urb_setup();
 117    void assign_gs_urb_setup();
 118    bool assign_regs(bool allow_spilling, bool spill_all);
 119    void assign_regs_trivial();
 120    void calculate_payload_ranges(int payload_node_count,
 121                                  int *payload_last_use_ip);
 122    void split_virtual_grfs();
 123    bool compact_virtual_grfs();
 124    void assign_constant_locations();
 125    bool get_pull_locs(const fs_reg &src, unsigned *out_surf_index,
 126                       unsigned *out_pull_index);
 127    void lower_constant_loads();
 128    void invalidate_live_intervals();
 129    void calculate_live_intervals();
 130    void calculate_register_pressure();
 131    void validate();
 132    bool opt_algebraic();
 133    bool opt_redundant_discard_jumps();
 134    bool opt_cse();
 135    bool opt_cse_local(bblock_t *block);
 136    bool opt_copy_propagation();
 137    bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
 138    bool try_constant_propagate(fs_inst *inst, acp_entry *entry);
 139    bool opt_copy_propagation_local(void *mem_ctx, bblock_t *block,
 140                                    exec_list *acp);
 141    bool opt_drop_redundant_mov_to_flags();
 142    bool opt_register_renaming();
 143    bool opt_bank_conflicts();
 144    unsigned bank_conflict_cycles(const fs_inst *inst) const;
 145    bool register_coalesce();
 146    bool compute_to_mrf();
 147    bool eliminate_find_live_channel();
 148    bool dead_code_eliminate();
 149    bool remove_duplicate_mrf_writes();
 150    bool remove_extra_rounding_modes();
 151
 152    bool opt_sampler_eot();
 153    bool virtual_grf_interferes(int a, int b);
 154    void schedule_instructions(instruction_scheduler_mode mode);
 155    void insert_gen4_send_dependency_workarounds();
 156    void insert_gen4_pre_send_dependency_workarounds(bblock_t *block,
 157                                                     fs_inst *inst);
 158    void insert_gen4_post_send_dependency_workarounds(bblock_t *block,
 159                                                      fs_inst *inst);
 160    void vfail(const char *msg, va_list args);
 161    void fail(const char *msg, ...);
 162    void limit_dispatch_width(unsigned n, const char *msg);
 163    void lower_uniform_pull_constant_loads();
 164    bool lower_load_payload();
 165    bool lower_pack();
 166    bool lower_regioning();
 167    bool lower_logical_sends();
 168    bool lower_integer_multiplication();
 169    bool lower_minmax();
 170    bool lower_simd_width();
 171    bool opt_combine_constants();
 172
 173    void emit_dummy_fs();
 174    void emit_repclear_shader();
 175    void emit_fragcoord_interpolation(fs_reg wpos);
 176    fs_reg *emit_frontfacing_interpolation();
 177    fs_reg *emit_samplepos_setup();
 178    fs_reg *emit_sampleid_setup();
 179    fs_reg *emit_samplemaskin_setup();
 180    void emit_interpolation_setup_gen4();
 181    void emit_interpolation_setup_gen6();
 182    void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
 183    fs_reg emit_mcs_fetch(const fs_reg &coordinate, unsigned components,
 184                          const fs_reg &texture,
 185                          const fs_reg &texture_handle);
 186    void emit_gen6_gather_wa(uint8_t wa, fs_reg dst);
 187    fs_reg resolve_source_modifiers(const fs_reg &src);
 188    void emit_discard_jump();
 189    void emit_fsign(const class brw::fs_builder &, const nir_alu_instr *instr,
 190                    fs_reg result, fs_reg *op, unsigned fsign_src);
 191    bool opt_peephole_sel();
 192    bool opt_peephole_csel();
 193    bool opt_peephole_predicated_break();
 194    bool opt_saturate_propagation();
 195    bool opt_cmod_propagation();
 196    bool opt_zero_samples();
 197
 198    void set_tcs_invocation_id();
 199
 200    void emit_nir_code();
 201    void nir_setup_outputs();
 202    void nir_setup_uniforms();
 203    void nir_emit_system_values();
 204    void nir_emit_impl(nir_function_impl *impl);
 205    void nir_emit_cf_list(exec_list *list);
 206    void nir_emit_if(nir_if *if_stmt);
 207    void nir_emit_loop(nir_loop *loop);
 208    void nir_emit_block(nir_block *block);
 209    void nir_emit_instr(nir_instr *instr);
 210    void nir_emit_alu(const brw::fs_builder &bld, nir_alu_instr *instr,
 211                      bool need_dest);
 212    bool try_emit_b2fi_of_inot(const brw::fs_builder &bld, fs_reg result,
 213                               nir_alu_instr *instr);
 214    void nir_emit_load_const(const brw::fs_builder &bld,
 215                             nir_load_const_instr *instr);
 216    void nir_emit_vs_intrinsic(const brw::fs_builder &bld,
 217                               nir_intrinsic_instr *instr);
 218    void nir_emit_tcs_intrinsic(const brw::fs_builder &bld,
 219                                nir_intrinsic_instr *instr);
 220    void nir_emit_gs_intrinsic(const brw::fs_builder &bld,
 221                               nir_intrinsic_instr *instr);
 222    void nir_emit_fs_intrinsic(const brw::fs_builder &bld,
 223                               nir_intrinsic_instr *instr);
 224    void nir_emit_cs_intrinsic(const brw::fs_builder &bld,
 225                               nir_intrinsic_instr *instr);
 226    fs_reg get_nir_image_intrinsic_image(const brw::fs_builder &bld,
 227                                         nir_intrinsic_instr *instr);
 228    fs_reg get_nir_ssbo_intrinsic_index(const brw::fs_builder &bld,
 229                                        nir_intrinsic_instr *instr);
 230    void nir_emit_intrinsic(const brw::fs_builder &bld,
 231                            nir_intrinsic_instr *instr);
 232    void nir_emit_tes_intrinsic(const brw::fs_builder &bld,
 233                                nir_intrinsic_instr *instr);
 234    void nir_emit_ssbo_atomic(const brw::fs_builder &bld,
 235                              int op, nir_intrinsic_instr *instr);
 236    void nir_emit_ssbo_atomic_float(const brw::fs_builder &bld,
 237                                    int op, nir_intrinsic_instr *instr);
 238    void nir_emit_shared_atomic(const brw::fs_builder &bld,
 239                                int op, nir_intrinsic_instr *instr);
 240    void nir_emit_shared_atomic_float(const brw::fs_builder &bld,
 241                                      int op, nir_intrinsic_instr *instr);
 242    void nir_emit_global_atomic(const brw::fs_builder &bld,
 243                                int op, nir_intrinsic_instr *instr);
 244    void nir_emit_global_atomic_float(const brw::fs_builder &bld,
 245                                      int op, nir_intrinsic_instr *instr);
 246    void nir_emit_texture(const brw::fs_builder &bld,
 247                          nir_tex_instr *instr);
 248    void nir_emit_jump(const brw::fs_builder &bld,
 249                       nir_jump_instr *instr);
 250    fs_reg get_nir_src(const nir_src &src);
 251    fs_reg get_nir_src_imm(const nir_src &src);
 252    fs_reg get_nir_dest(const nir_dest &dest);
 253    fs_reg get_indirect_offset(nir_intrinsic_instr *instr);
 254    fs_reg get_tcs_single_patch_icp_handle(const brw::fs_builder &bld,
 255                                           nir_intrinsic_instr *instr);
 256    fs_reg get_tcs_eight_patch_icp_handle(const brw::fs_builder &bld,
 257                                          nir_intrinsic_instr *instr);
 258    struct brw_reg get_tcs_output_urb_handle();
 259
 260    void emit_percomp(const brw::fs_builder &bld, const fs_inst &inst,
 261                      unsigned wr_mask);
 262
 263    bool optimize_extract_to_float(nir_alu_instr *instr,
 264                                   const fs_reg &result);
 265    bool optimize_frontfacing_ternary(nir_alu_instr *instr,
 266                                      const fs_reg &result);
 267
 268    void emit_alpha_test();
 269    fs_inst *emit_single_fb_write(const brw::fs_builder &bld,
 270                                  fs_reg color1, fs_reg color2,
 271                                  fs_reg src0_alpha, unsigned components);
 272    void emit_alpha_to_coverage_workaround(const fs_reg &src0_alpha);
 273    void emit_fb_writes();
 274    fs_inst *emit_non_coherent_fb_read(const brw::fs_builder &bld,
 275                                       const fs_reg &dst, unsigned target);
 276    void emit_urb_writes(const fs_reg &gs_vertex_count = fs_reg());
 277    void set_gs_stream_control_data_bits(const fs_reg &vertex_count,
 278                                         unsigned stream_id);
 279    void emit_gs_control_data_bits(const fs_reg &vertex_count);
 280    void emit_gs_end_primitive(const nir_src &vertex_count_nir_src);
 281    void emit_gs_vertex(const nir_src &vertex_count_nir_src,
 282                        unsigned stream_id);
 283    void emit_gs_thread_end();
 284    void emit_gs_input_load(const fs_reg &dst, const nir_src &vertex_src,
 285                            unsigned base_offset, const nir_src &offset_src,
 286                            unsigned num_components, unsigned first_component);
 287    void emit_cs_terminate();
 288    fs_reg *emit_cs_work_group_id_setup();
 289
 290    void emit_barrier();
 291
 292    void emit_shader_time_begin();
 293    void emit_shader_time_end();
 294    void SHADER_TIME_ADD(const brw::fs_builder &bld,
 295                         int shader_time_subindex,
 296                         fs_reg value);
 297
 298    fs_reg get_timestamp(const brw::fs_builder &bld);
 299
 300    fs_reg interp_reg(int location, int channel);
 301
 302    int implied_mrf_writes(fs_inst *inst) const;
 303
 304    virtual void dump_instructions();
 305    virtual void dump_instructions(const char *name);
 306    void dump_instruction(backend_instruction *inst);
 307    void dump_instruction(backend_instruction *inst, FILE *file);
 308
 309    const brw_base_prog_key *const key;
 310    const struct brw_sampler_prog_key_data *key_tex;
 311
 312    struct brw_gs_compile *gs_compile;
 313
 314    struct brw_stage_prog_data *prog_data;
 315    struct gl_program *prog;
 316
 317    const struct brw_vue_map *input_vue_map;
 318
 319    int *virtual_grf_start;
 320    int *virtual_grf_end;
 321    brw::fs_live_variables *live_intervals;
 322
 323    int *regs_live_at_ip;
 324
 325    /** Number of uniform variable components visited. */
 326    unsigned uniforms;
 327
 328    /** Byte-offset for the next available spot in the scratch space buffer. */
 329    unsigned last_scratch;
 330
 331    /**
 332     * Array mapping UNIFORM register numbers to the pull parameter index,
 333     * or -1 if this uniform register isn't being uploaded as a pull constant.
 334     */
 335    int *pull_constant_loc;
 336
 337    /**
 338     * Array mapping UNIFORM register numbers to the push parameter index,
 339     * or -1 if this uniform register isn't being uploaded as a push constant.
 340     */
 341    int *push_constant_loc;
 342
 343    fs_reg subgroup_id;
 344    fs_reg frag_depth;
 345    fs_reg frag_stencil;
 346    fs_reg sample_mask;
 347    fs_reg outputs[VARYING_SLOT_MAX];
 348    fs_reg dual_src_output;
 349    int first_non_payload_grf;
 350    /** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
 351    unsigned max_grf;
 352
 353    fs_reg *nir_locals;
 354    fs_reg *nir_ssa_values;
 355    fs_reg *nir_system_values;
 356
 357    bool failed;
 358    char *fail_msg;
 359
 360    /** Register numbers for thread payload fields. */
 361    struct thread_payload {
 362       uint8_t subspan_coord_reg[2];
 363       uint8_t source_depth_reg[2];
 364       uint8_t source_w_reg[2];
 365       uint8_t aa_dest_stencil_reg[2];
 366       uint8_t dest_depth_reg[2];
 367       uint8_t sample_pos_reg[2];
 368       uint8_t sample_mask_in_reg[2];
 369       uint8_t barycentric_coord_reg[BRW_BARYCENTRIC_MODE_COUNT][2];
 370       uint8_t local_invocation_id_reg[2];
 371
 372       /** The number of thread payload registers the hardware will supply. */
 373       uint8_t num_regs;
 374    } payload;
 375
 376    bool source_depth_to_render_target;
 377    bool runtime_check_aads_emit;
 378
 379    fs_reg pixel_x;
 380    fs_reg pixel_y;
 381    fs_reg wpos_w;
 382    fs_reg pixel_w;
 383    fs_reg delta_xy[BRW_BARYCENTRIC_MODE_COUNT];
 384    fs_reg shader_start_time;
 385    fs_reg final_gs_vertex_count;
 386    fs_reg control_data_bits;
 387    fs_reg invocation_id;
 388
 389    unsigned grf_used;
 390    bool spilled_any_registers;
 391
 392    const unsigned dispatch_width; /**< 8, 16 or 32 */
 393    unsigned max_dispatch_width;
 394
 395    int shader_time_index;
 396
 397    struct shader_stats shader_stats;
 398
 399    brw::fs_builder bld;
 400
 401 private:
 402    fs_reg prepare_alu_destination_and_sources(const brw::fs_builder &bld,
 403                                               nir_alu_instr *instr,
 404                                               fs_reg *op,
 405                                               bool need_dest);
 406
 407    void resolve_inot_sources(const brw::fs_builder &bld, nir_alu_instr *instr,
 408                              fs_reg *op);
 409    void lower_mul_dword_inst(fs_inst *inst, bblock_t *block);
 410    void lower_mul_qword_inst(fs_inst *inst, bblock_t *block);
 411    void lower_mulh_inst(fs_inst *inst, bblock_t *block);
 412 };
 413
 414 /**
 415  * The fragment shader code generator.
 416  *
 417  * Translates FS IR to actual i965 assembly code.
 418  */
 419 class fs_generator
 420 {
 421 public:
 422    fs_generator(const struct brw_compiler *compiler, void *log_data,
 423                 void *mem_ctx,
 424                 struct brw_stage_prog_data *prog_data,
 425                 struct shader_stats shader_stats,
 426                 bool runtime_check_aads_emit,
 427                 gl_shader_stage stage);
 428    ~fs_generator();
 429
 430    void enable_debug(const char *shader_name);
 431    int generate_code(const cfg_t *cfg, int dispatch_width,
 432                      struct brw_compile_stats *stats);
 433    const unsigned *get_assembly();
 434
 435 private:
 436    void fire_fb_write(fs_inst *inst,
 437                       struct brw_reg payload,
 438                       struct brw_reg implied_header,
 439                       GLuint nr);
 440    void generate_send(fs_inst *inst,
 441                       struct brw_reg dst,
 442                       struct brw_reg desc,
 443                       struct brw_reg ex_desc,
 444                       struct brw_reg payload,
 445                       struct brw_reg payload2);
 446    void generate_fb_write(fs_inst *inst, struct brw_reg payload);
 447    void generate_fb_read(fs_inst *inst, struct brw_reg dst,
 448                          struct brw_reg payload);
 449    void generate_urb_read(fs_inst *inst, struct brw_reg dst, struct brw_reg payload);
 450    void generate_urb_write(fs_inst *inst, struct brw_reg payload);
 451    void generate_cs_terminate(fs_inst *inst, struct brw_reg payload);
 452    void generate_barrier(fs_inst *inst, struct brw_reg src);
 453    bool generate_linterp(fs_inst *inst, struct brw_reg dst,
 454                          struct brw_reg *src);
 455    void generate_tex(fs_inst *inst, struct brw_reg dst,
 456                      struct brw_reg surface_index,
 457                      struct brw_reg sampler_index);
 458    void generate_get_buffer_size(fs_inst *inst, struct brw_reg dst,
 459                                  struct brw_reg src,
 460                                  struct brw_reg surf_index);
 461    void generate_ddx(const fs_inst *inst,
 462                      struct brw_reg dst, struct brw_reg src);
 463    void generate_ddy(const fs_inst *inst,
 464                      struct brw_reg dst, struct brw_reg src);
 465    void generate_scratch_write(fs_inst *inst, struct brw_reg src);
 466    void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
 467    void generate_scratch_read_gen7(fs_inst *inst, struct brw_reg dst);
 468    void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
 469                                             struct brw_reg index,
 470                                             struct brw_reg offset);
 471    void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
 472                                                  struct brw_reg dst,
 473                                                  struct brw_reg surf_index,
 474                                                  struct brw_reg payload);
 475    void generate_varying_pull_constant_load_gen4(fs_inst *inst,
 476                                                  struct brw_reg dst,
 477                                                  struct brw_reg index);
 478    void generate_mov_dispatch_to_flags(fs_inst *inst);
 479
 480    void generate_pixel_interpolator_query(fs_inst *inst,
 481                                           struct brw_reg dst,
 482                                           struct brw_reg src,
 483                                           struct brw_reg msg_data,
 484                                           unsigned msg_type);
 485
 486    void generate_set_sample_id(fs_inst *inst,
 487                                struct brw_reg dst,
 488                                struct brw_reg src0,
 489                                struct brw_reg src1);
 490
 491    void generate_discard_jump(fs_inst *inst);
 492
 493    void generate_pack_half_2x16_split(fs_inst *inst,
 494                                       struct brw_reg dst,
 495                                       struct brw_reg x,
 496                                       struct brw_reg y);
 497
 498    void generate_shader_time_add(fs_inst *inst,
 499                                  struct brw_reg payload,
 500                                  struct brw_reg offset,
 501                                  struct brw_reg value);
 502
 503    void generate_mov_indirect(fs_inst *inst,
 504                               struct brw_reg dst,
 505                               struct brw_reg reg,
 506                               struct brw_reg indirect_byte_offset);
 507
 508    void generate_shuffle(fs_inst *inst,
 509                          struct brw_reg dst,
 510                          struct brw_reg src,
 511                          struct brw_reg idx);
 512
 513    void generate_quad_swizzle(const fs_inst *inst,
 514                               struct brw_reg dst, struct brw_reg src,
 515                               unsigned swiz);
 516
 517    bool patch_discard_jumps_to_fb_writes();
 518
 519    const struct brw_compiler *compiler;
 520    void *log_data; /* Passed to compiler->*_log functions */
 521
 522    const struct gen_device_info *devinfo;
 523
 524    struct brw_codegen *p;
 525    struct brw_stage_prog_data * const prog_data;
 526
 527    unsigned dispatch_width; /**< 8, 16 or 32 */
 528
 529    exec_list discard_halt_patches;
 530    struct shader_stats shader_stats;
 531    bool runtime_check_aads_emit;
 532    bool debug_flag;
 533    const char *shader_name;
 534    gl_shader_stage stage;
 535    void *mem_ctx;
 536 };
 537
 538 namespace brw {
 539    inline fs_reg
 540    fetch_payload_reg(const brw::fs_builder &bld, uint8_t regs[2],
 541                      brw_reg_type type = BRW_REGISTER_TYPE_F, unsigned n = 1)
 542    {
 543       if (!regs[0])
 544          return fs_reg();
 545
 546       if (bld.dispatch_width() > 16) {
 547          const fs_reg tmp = bld.vgrf(type, n);
 548          const brw::fs_builder hbld = bld.exec_all().group(16, 0);
 549          const unsigned m = bld.dispatch_width() / hbld.dispatch_width();
 550          fs_reg *const components = new fs_reg[n * m];
 551
 552          for (unsigned c = 0; c < n; c++) {
 553             for (unsigned g = 0; g < m; g++) {
 554                components[c * m + g] =
 555                   offset(retype(brw_vec8_grf(regs[g], 0), type), hbld, c);
 556             }
 557          }
 558
 559          hbld.LOAD_PAYLOAD(tmp, components, n * m, 0);
 560
 561          delete[] components;
 562          return tmp;
 563
 564       } else {
 565          return fs_reg(retype(brw_vec8_grf(regs[0], 0), type));
 566       }
 567    }
 568
 569    bool
 570    lower_src_modifiers(fs_visitor *v, bblock_t *block, fs_inst *inst, unsigned i);
 571 }
 572
 573 void shuffle_from_32bit_read(const brw::fs_builder &bld,
 574                              const fs_reg &dst,
 575                              const fs_reg &src,
 576                              uint32_t first_component,
 577                              uint32_t components);
 578
 579 fs_reg setup_imm_df(const brw::fs_builder &bld,
 580                     double v);
 581
 582 fs_reg setup_imm_b(const brw::fs_builder &bld,
 583                    int8_t v);
 584
 585 fs_reg setup_imm_ub(const brw::fs_builder &bld,
 586                    uint8_t v);
 587
 588 enum brw_barycentric_mode brw_barycentric_mode(enum glsl_interp_mode mode,
 589                                                nir_intrinsic_op op);
 590
 591 #endif /* BRW_FS_H */