src/mesa/drivers/dri/i965/brw_vec4.h

   1 /*
   2  * Copyright © 2011 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 #ifndef BRW_VEC4_H
  25 #define BRW_VEC4_H
  26
  27 #include <stdint.h>
  28 #include "brw_shader.h"
  29 #include "main/compiler.h"
  30 #include "program/hash_table.h"
  31 #include "brw_program.h"
  32
  33 #ifdef __cplusplus
  34 extern "C" {
  35 #endif
  36
  37 #include "brw_context.h"
  38 #include "brw_eu.h"
  39
  40 #ifdef __cplusplus
  41 }; /* extern "C" */
  42 #include "gen8_generator.h"
  43 #endif
  44
  45 #include "glsl/ir.h"
  46
  47
  48 struct brw_vec4_compile {
  49    GLuint last_scratch; /**< measured in 32-byte (register size) units */
  50 };
  51
  52
  53 struct brw_vec4_prog_key {
  54    GLuint program_string_id;
  55
  56    /**
  57     * True if at least one clip flag is enabled, regardless of whether the
  58     * shader uses clip planes or gl_ClipDistance.
  59     */
  60    GLuint userclip_active:1;
  61
  62    /**
  63     * How many user clipping planes are being uploaded to the vertex shader as
  64     * push constants.
  65     */
  66    GLuint nr_userclip_plane_consts:4;
  67
  68    GLuint clamp_vertex_color:1;
  69
  70    struct brw_sampler_prog_key_data tex;
  71 };
  72
  73
  74 #ifdef __cplusplus
  75 extern "C" {
  76 #endif
  77
  78 void
  79 brw_vec4_setup_prog_key_for_precompile(struct gl_context *ctx,
  80                                        struct brw_vec4_prog_key *key,
  81                                        GLuint id, struct gl_program *prog);
  82
  83 #ifdef __cplusplus
  84 } /* extern "C" */
  85
  86 namespace brw {
  87
  88 class dst_reg;
  89
  90 unsigned
  91 swizzle_for_size(int size);
  92
  93 class reg
  94 {
  95 public:
  96    /** Register file: GRF, MRF, IMM. */
  97    enum register_file file;
  98    /** virtual register number.  0 = fixed hw reg */
  99    int reg;
 100    /** Offset within the virtual register. */
 101    int reg_offset;
 102    /** Register type.  BRW_REGISTER_TYPE_* */
 103    int type;
 104    struct brw_reg fixed_hw_reg;
 105
 106    /** Value for file == BRW_IMMMEDIATE_FILE */
 107    union {
 108       int32_t i;
 109       uint32_t u;
 110       float f;
 111    } imm;
 112 };
 113
 114 class src_reg : public reg
 115 {
 116 public:
 117    DECLARE_RALLOC_CXX_OPERATORS(src_reg)
 118
 119    void init();
 120
 121    src_reg(register_file file, int reg, const glsl_type *type);
 122    src_reg();
 123    src_reg(float f);
 124    src_reg(uint32_t u);
 125    src_reg(int32_t i);
 126    src_reg(struct brw_reg reg);
 127
 128    bool equals(src_reg *r);
 129    bool is_zero() const;
 130    bool is_one() const;
 131
 132    src_reg(class vec4_visitor *v, const struct glsl_type *type);
 133
 134    explicit src_reg(dst_reg reg);
 135
 136    GLuint swizzle; /**< BRW_SWIZZLE_XYZW macros from brw_reg.h. */
 137    bool negate;
 138    bool abs;
 139
 140    src_reg *reladdr;
 141 };
 142
 143 static inline src_reg
 144 retype(src_reg reg, unsigned type)
 145 {
 146    reg.fixed_hw_reg.type = reg.type = type;
 147    return reg;
 148 }
 149
 150 static inline src_reg
 151 offset(src_reg reg, unsigned delta)
 152 {
 153    assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
 154    reg.reg_offset += delta;
 155    return reg;
 156 }
 157
 158 /**
 159  * Reswizzle a given source register.
 160  * \sa brw_swizzle().
 161  */
 162 static inline src_reg
 163 swizzle(src_reg reg, unsigned swizzle)
 164 {
 165    assert(reg.file != HW_REG);
 166    reg.swizzle = BRW_SWIZZLE4(
 167       BRW_GET_SWZ(reg.swizzle, BRW_GET_SWZ(swizzle, 0)),
 168       BRW_GET_SWZ(reg.swizzle, BRW_GET_SWZ(swizzle, 1)),
 169       BRW_GET_SWZ(reg.swizzle, BRW_GET_SWZ(swizzle, 2)),
 170       BRW_GET_SWZ(reg.swizzle, BRW_GET_SWZ(swizzle, 3)));
 171    return reg;
 172 }
 173
 174 static inline src_reg
 175 negate(src_reg reg)
 176 {
 177    assert(reg.file != HW_REG && reg.file != IMM);
 178    reg.negate = !reg.negate;
 179    return reg;
 180 }
 181
 182 class dst_reg : public reg
 183 {
 184 public:
 185    DECLARE_RALLOC_CXX_OPERATORS(dst_reg)
 186
 187    void init();
 188
 189    dst_reg();
 190    dst_reg(register_file file, int reg);
 191    dst_reg(register_file file, int reg, const glsl_type *type, int writemask);
 192    dst_reg(struct brw_reg reg);
 193    dst_reg(class vec4_visitor *v, const struct glsl_type *type);
 194
 195    explicit dst_reg(src_reg reg);
 196
 197    int writemask; /**< Bitfield of WRITEMASK_[XYZW] */
 198
 199    src_reg *reladdr;
 200 };
 201
 202 static inline dst_reg
 203 retype(dst_reg reg, unsigned type)
 204 {
 205    reg.fixed_hw_reg.type = reg.type = type;
 206    return reg;
 207 }
 208
 209 static inline dst_reg
 210 offset(dst_reg reg, unsigned delta)
 211 {
 212    assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
 213    reg.reg_offset += delta;
 214    return reg;
 215 }
 216
 217 static inline dst_reg
 218 writemask(dst_reg reg, unsigned mask)
 219 {
 220    assert(reg.file != HW_REG && reg.file != IMM);
 221    assert((reg.writemask & mask) != 0);
 222    reg.writemask &= mask;
 223    return reg;
 224 }
 225
 226 class vec4_instruction : public backend_instruction {
 227 public:
 228    DECLARE_RALLOC_CXX_OPERATORS(vec4_instruction)
 229
 230    vec4_instruction(vec4_visitor *v, enum opcode opcode,
 231                     dst_reg dst = dst_reg(),
 232                     src_reg src0 = src_reg(),
 233                     src_reg src1 = src_reg(),
 234                     src_reg src2 = src_reg());
 235
 236    struct brw_reg get_dst(void);
 237    struct brw_reg get_src(const struct brw_vec4_prog_data *prog_data, int i);
 238
 239    dst_reg dst;
 240    src_reg src[3];
 241
 242    bool saturate;
 243    bool force_writemask_all;
 244    bool no_dd_clear, no_dd_check;
 245
 246    int conditional_mod; /**< BRW_CONDITIONAL_* */
 247
 248    int sampler;
 249    uint32_t texture_offset; /**< Texture Offset bitfield */
 250    int target; /**< MRT target. */
 251    bool shadow_compare;
 252
 253    enum brw_urb_write_flags urb_write_flags;
 254    bool header_present;
 255    int mlen; /**< SEND message length */
 256    int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
 257
 258    uint32_t offset; /* spill/unspill offset */
 259    /** @{
 260     * Annotation for the generated IR.  One of the two can be set.
 261     */
 262    const void *ir;
 263    const char *annotation;
 264
 265    bool is_send_from_grf();
 266    bool can_reswizzle_dst(int dst_writemask, int swizzle, int swizzle_mask);
 267    void reswizzle_dst(int dst_writemask, int swizzle);
 268
 269    bool depends_on_flags()
 270    {
 271       return predicate || opcode == VS_OPCODE_UNPACK_FLAGS_SIMD4X2;
 272    }
 273 };
 274
 275 /**
 276  * The vertex shader front-end.
 277  *
 278  * Translates either GLSL IR or Mesa IR (for ARB_vertex_program and
 279  * fixed-function) into VS IR.
 280  */
 281 class vec4_visitor : public backend_visitor
 282 {
 283 public:
 284    vec4_visitor(struct brw_context *brw,
 285                 struct brw_vec4_compile *c,
 286                 struct gl_program *prog,
 287                 const struct brw_vec4_prog_key *key,
 288                 struct brw_vec4_prog_data *prog_data,
 289                 struct gl_shader_program *shader_prog,
 290                 struct brw_shader *shader,
 291                 void *mem_ctx,
 292                 bool debug_flag,
 293                 bool no_spills,
 294                 shader_time_shader_type st_base,
 295                 shader_time_shader_type st_written,
 296                 shader_time_shader_type st_reset);
 297    ~vec4_visitor();
 298
 299    dst_reg dst_null_f()
 300    {
 301       return dst_reg(brw_null_reg());
 302    }
 303
 304    dst_reg dst_null_d()
 305    {
 306       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
 307    }
 308
 309    dst_reg dst_null_ud()
 310    {
 311       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
 312    }
 313
 314    struct brw_vec4_compile *c;
 315    const struct brw_vec4_prog_key *key;
 316    struct brw_vec4_prog_data *prog_data;
 317    unsigned int sanity_param_count;
 318
 319    char *fail_msg;
 320    bool failed;
 321
 322    /**
 323     * GLSL IR currently being processed, which is associated with our
 324     * driver IR instructions for debugging purposes.
 325     */
 326    const void *base_ir;
 327    const char *current_annotation;
 328
 329    int *virtual_grf_sizes;
 330    int virtual_grf_count;
 331    int virtual_grf_array_size;
 332    int first_non_payload_grf;
 333    unsigned int max_grf;
 334    int *virtual_grf_start;
 335    int *virtual_grf_end;
 336    dst_reg userplane[MAX_CLIP_PLANES];
 337
 338    /**
 339     * This is the size to be used for an array with an element per
 340     * reg_offset
 341     */
 342    int virtual_grf_reg_count;
 343    /** Per-virtual-grf indices into an array of size virtual_grf_reg_count */
 344    int *virtual_grf_reg_map;
 345
 346    bool live_intervals_valid;
 347
 348    dst_reg *variable_storage(ir_variable *var);
 349
 350    void reladdr_to_temp(ir_instruction *ir, src_reg *reg, int *num_reladdr);
 351
 352    bool need_all_constants_in_pull_buffer;
 353
 354    /**
 355     * \name Visit methods
 356     *
 357     * As typical for the visitor pattern, there must be one \c visit method for
 358     * each concrete subclass of \c ir_instruction.  Virtual base classes within
 359     * the hierarchy should not have \c visit methods.
 360     */
 361    /*@{*/
 362    virtual void visit(ir_variable *);
 363    virtual void visit(ir_loop *);
 364    virtual void visit(ir_loop_jump *);
 365    virtual void visit(ir_function_signature *);
 366    virtual void visit(ir_function *);
 367    virtual void visit(ir_expression *);
 368    virtual void visit(ir_swizzle *);
 369    virtual void visit(ir_dereference_variable  *);
 370    virtual void visit(ir_dereference_array *);
 371    virtual void visit(ir_dereference_record *);
 372    virtual void visit(ir_assignment *);
 373    virtual void visit(ir_constant *);
 374    virtual void visit(ir_call *);
 375    virtual void visit(ir_return *);
 376    virtual void visit(ir_discard *);
 377    virtual void visit(ir_texture *);
 378    virtual void visit(ir_if *);
 379    virtual void visit(ir_emit_vertex *);
 380    virtual void visit(ir_end_primitive *);
 381    /*@}*/
 382
 383    src_reg result;
 384
 385    /* Regs for vertex results.  Generated at ir_variable visiting time
 386     * for the ir->location's used.
 387     */
 388    dst_reg output_reg[BRW_VARYING_SLOT_COUNT];
 389    const char *output_reg_annotation[BRW_VARYING_SLOT_COUNT];
 390    int uniform_size[MAX_UNIFORMS];
 391    int uniform_vector_size[MAX_UNIFORMS];
 392    int uniforms;
 393
 394    src_reg shader_start_time;
 395
 396    struct hash_table *variable_ht;
 397
 398    bool run(void);
 399    void fail(const char *msg, ...);
 400
 401    int virtual_grf_alloc(int size);
 402    void setup_uniform_clipplane_values();
 403    void setup_uniform_values(ir_variable *ir);
 404    void setup_builtin_uniform_values(ir_variable *ir);
 405    int setup_uniforms(int payload_reg);
 406    bool reg_allocate_trivial();
 407    bool reg_allocate();
 408    void evaluate_spill_costs(float *spill_costs, bool *no_spill);
 409    int choose_spill_reg(struct ra_graph *g);
 410    void spill_reg(int spill_reg);
 411    void move_grf_array_access_to_scratch();
 412    void move_uniform_array_access_to_pull_constants();
 413    void move_push_constants_to_pull_constants();
 414    void split_uniform_registers();
 415    void pack_uniform_registers();
 416    void calculate_live_intervals();
 417    void invalidate_live_intervals();
 418    void split_virtual_grfs();
 419    bool dead_code_eliminate();
 420    bool virtual_grf_interferes(int a, int b);
 421    bool opt_copy_propagation();
 422    bool opt_algebraic();
 423    bool opt_register_coalesce();
 424    void opt_set_dependency_control();
 425    void opt_schedule_instructions();
 426
 427    bool can_do_source_mods(vec4_instruction *inst);
 428
 429    vec4_instruction *emit(vec4_instruction *inst);
 430
 431    vec4_instruction *emit(enum opcode opcode);
 432
 433    vec4_instruction *emit(enum opcode opcode, dst_reg dst);
 434
 435    vec4_instruction *emit(enum opcode opcode, dst_reg dst, src_reg src0);
 436
 437    vec4_instruction *emit(enum opcode opcode, dst_reg dst,
 438                           src_reg src0, src_reg src1);
 439
 440    vec4_instruction *emit(enum opcode opcode, dst_reg dst,
 441                           src_reg src0, src_reg src1, src_reg src2);
 442
 443    vec4_instruction *emit_before(vec4_instruction *inst,
 444                                  vec4_instruction *new_inst);
 445
 446    vec4_instruction *MOV(dst_reg dst, src_reg src0);
 447    vec4_instruction *NOT(dst_reg dst, src_reg src0);
 448    vec4_instruction *RNDD(dst_reg dst, src_reg src0);
 449    vec4_instruction *RNDE(dst_reg dst, src_reg src0);
 450    vec4_instruction *RNDZ(dst_reg dst, src_reg src0);
 451    vec4_instruction *FRC(dst_reg dst, src_reg src0);
 452    vec4_instruction *F32TO16(dst_reg dst, src_reg src0);
 453    vec4_instruction *F16TO32(dst_reg dst, src_reg src0);
 454    vec4_instruction *ADD(dst_reg dst, src_reg src0, src_reg src1);
 455    vec4_instruction *MUL(dst_reg dst, src_reg src0, src_reg src1);
 456    vec4_instruction *MACH(dst_reg dst, src_reg src0, src_reg src1);
 457    vec4_instruction *MAC(dst_reg dst, src_reg src0, src_reg src1);
 458    vec4_instruction *AND(dst_reg dst, src_reg src0, src_reg src1);
 459    vec4_instruction *OR(dst_reg dst, src_reg src0, src_reg src1);
 460    vec4_instruction *XOR(dst_reg dst, src_reg src0, src_reg src1);
 461    vec4_instruction *DP3(dst_reg dst, src_reg src0, src_reg src1);
 462    vec4_instruction *DP4(dst_reg dst, src_reg src0, src_reg src1);
 463    vec4_instruction *DPH(dst_reg dst, src_reg src0, src_reg src1);
 464    vec4_instruction *SHL(dst_reg dst, src_reg src0, src_reg src1);
 465    vec4_instruction *SHR(dst_reg dst, src_reg src0, src_reg src1);
 466    vec4_instruction *ASR(dst_reg dst, src_reg src0, src_reg src1);
 467    vec4_instruction *CMP(dst_reg dst, src_reg src0, src_reg src1,
 468                          uint32_t condition);
 469    vec4_instruction *IF(src_reg src0, src_reg src1, uint32_t condition);
 470    vec4_instruction *IF(uint32_t predicate);
 471    vec4_instruction *PULL_CONSTANT_LOAD(dst_reg dst, src_reg index);
 472    vec4_instruction *SCRATCH_READ(dst_reg dst, src_reg index);
 473    vec4_instruction *SCRATCH_WRITE(dst_reg dst, src_reg src, src_reg index);
 474    vec4_instruction *LRP(dst_reg dst, src_reg a, src_reg y, src_reg x);
 475    vec4_instruction *BFREV(dst_reg dst, src_reg value);
 476    vec4_instruction *BFE(dst_reg dst, src_reg bits, src_reg offset, src_reg value);
 477    vec4_instruction *BFI1(dst_reg dst, src_reg bits, src_reg offset);
 478    vec4_instruction *BFI2(dst_reg dst, src_reg bfi1_dst, src_reg insert, src_reg base);
 479    vec4_instruction *FBH(dst_reg dst, src_reg value);
 480    vec4_instruction *FBL(dst_reg dst, src_reg value);
 481    vec4_instruction *CBIT(dst_reg dst, src_reg value);
 482    vec4_instruction *MAD(dst_reg dst, src_reg c, src_reg b, src_reg a);
 483    vec4_instruction *ADDC(dst_reg dst, src_reg src0, src_reg src1);
 484    vec4_instruction *SUBB(dst_reg dst, src_reg src0, src_reg src1);
 485
 486    int implied_mrf_writes(vec4_instruction *inst);
 487
 488    bool try_rewrite_rhs_to_dst(ir_assignment *ir,
 489                                dst_reg dst,
 490                                src_reg src,
 491                                vec4_instruction *pre_rhs_inst,
 492                                vec4_instruction *last_rhs_inst);
 493
 494    bool try_copy_propagation(vec4_instruction *inst, int arg,
 495                              src_reg *values[4]);
 496
 497    /** Walks an exec_list of ir_instruction and sends it through this visitor. */
 498    void visit_instructions(const exec_list *list);
 499
 500    void emit_vp_sop(uint32_t condmod, dst_reg dst,
 501                     src_reg src0, src_reg src1, src_reg one);
 502
 503    void emit_bool_to_cond_code(ir_rvalue *ir, uint32_t *predicate);
 504    void emit_bool_comparison(unsigned int op, dst_reg dst, src_reg src0, src_reg src1);
 505    void emit_if_gen6(ir_if *ir);
 506
 507    void emit_minmax(uint32_t condmod, dst_reg dst, src_reg src0, src_reg src1);
 508
 509    void emit_lrp(const dst_reg &dst,
 510                  const src_reg &x, const src_reg &y, const src_reg &a);
 511
 512    void emit_block_move(dst_reg *dst, src_reg *src,
 513                         const struct glsl_type *type, uint32_t predicate);
 514
 515    void emit_constant_values(dst_reg *dst, ir_constant *value);
 516
 517    /**
 518     * Emit the correct dot-product instruction for the type of arguments
 519     */
 520    void emit_dp(dst_reg dst, src_reg src0, src_reg src1, unsigned elements);
 521
 522    void emit_scalar(ir_instruction *ir, enum prog_opcode op,
 523                     dst_reg dst, src_reg src0);
 524
 525    void emit_scalar(ir_instruction *ir, enum prog_opcode op,
 526                     dst_reg dst, src_reg src0, src_reg src1);
 527
 528    void emit_scs(ir_instruction *ir, enum prog_opcode op,
 529                  dst_reg dst, const src_reg &src);
 530
 531    src_reg fix_3src_operand(src_reg src);
 532
 533    void emit_math1_gen6(enum opcode opcode, dst_reg dst, src_reg src);
 534    void emit_math1_gen4(enum opcode opcode, dst_reg dst, src_reg src);
 535    void emit_math(enum opcode opcode, dst_reg dst, src_reg src);
 536    void emit_math2_gen6(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 537    void emit_math2_gen4(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 538    void emit_math(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 539    src_reg fix_math_operand(src_reg src);
 540
 541    void emit_pack_half_2x16(dst_reg dst, src_reg src0);
 542    void emit_unpack_half_2x16(dst_reg dst, src_reg src0);
 543
 544    uint32_t gather_channel(ir_texture *ir, int sampler);
 545    src_reg emit_mcs_fetch(ir_texture *ir, src_reg coordinate, int sampler);
 546    void emit_gen6_gather_wa(uint8_t wa, dst_reg dst);
 547    void swizzle_result(ir_texture *ir, src_reg orig_val, int sampler);
 548
 549    void emit_ndc_computation();
 550    void emit_psiz_and_flags(struct brw_reg reg);
 551    void emit_clip_distances(dst_reg reg, int offset);
 552    void emit_generic_urb_slot(dst_reg reg, int varying);
 553    void emit_urb_slot(int mrf, int varying);
 554
 555    void emit_shader_time_begin();
 556    void emit_shader_time_end();
 557    void emit_shader_time_write(enum shader_time_shader_type type,
 558                                src_reg value);
 559
 560    void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
 561                             dst_reg dst, src_reg offset, src_reg src0,
 562                             src_reg src1);
 563
 564    void emit_untyped_surface_read(unsigned surf_index, dst_reg dst,
 565                                   src_reg offset);
 566
 567    src_reg get_scratch_offset(vec4_instruction *inst,
 568                               src_reg *reladdr, int reg_offset);
 569    src_reg get_pull_constant_offset(vec4_instruction *inst,
 570                                     src_reg *reladdr, int reg_offset);
 571    void emit_scratch_read(vec4_instruction *inst,
 572                           dst_reg dst,
 573                           src_reg orig_src,
 574                           int base_offset);
 575    void emit_scratch_write(vec4_instruction *inst,
 576                            int base_offset);
 577    void emit_pull_constant_load(vec4_instruction *inst,
 578                                 dst_reg dst,
 579                                 src_reg orig_src,
 580                                 int base_offset);
 581
 582    bool try_emit_sat(ir_expression *ir);
 583    bool try_emit_mad(ir_expression *ir, int mul_arg);
 584    void resolve_ud_negate(src_reg *reg);
 585
 586    src_reg get_timestamp();
 587
 588    bool process_move_condition(ir_rvalue *ir);
 589
 590    void dump_instruction(backend_instruction *inst);
 591
 592    void visit_atomic_counter_intrinsic(ir_call *ir);
 593
 594 protected:
 595    void emit_vertex();
 596    void lower_attributes_to_hw_regs(const int *attribute_map,
 597                                     bool interleaved);
 598    void setup_payload_interference(struct ra_graph *g, int first_payload_node,
 599                                    int reg_node_count);
 600    virtual dst_reg *make_reg_for_system_value(ir_variable *ir) = 0;
 601    virtual void setup_payload() = 0;
 602    virtual void emit_prolog() = 0;
 603    virtual void emit_program_code() = 0;
 604    virtual void emit_thread_end() = 0;
 605    virtual void emit_urb_write_header(int mrf) = 0;
 606    virtual vec4_instruction *emit_urb_write_opcode(bool complete) = 0;
 607    virtual int compute_array_stride(ir_dereference_array *ir);
 608
 609    const bool debug_flag;
 610
 611 private:
 612    /**
 613     * If true, then register allocation should fail instead of spilling.
 614     */
 615    const bool no_spills;
 616
 617    const shader_time_shader_type st_base;
 618    const shader_time_shader_type st_written;
 619    const shader_time_shader_type st_reset;
 620 };
 621
 622
 623 /**
 624  * The vertex shader code generator.
 625  *
 626  * Translates VS IR to actual i965 assembly code.
 627  */
 628 class vec4_generator
 629 {
 630 public:
 631    vec4_generator(struct brw_context *brw,
 632                   struct gl_shader_program *shader_prog,
 633                   struct gl_program *prog,
 634                   struct brw_vec4_prog_data *prog_data,
 635                   void *mem_ctx,
 636                   bool debug_flag);
 637    ~vec4_generator();
 638
 639    const unsigned *generate_assembly(exec_list *insts, unsigned *asm_size);
 640
 641 private:
 642    void generate_code(exec_list *instructions);
 643    void generate_vec4_instruction(vec4_instruction *inst,
 644                                   struct brw_reg dst,
 645                                   struct brw_reg *src);
 646
 647    void generate_math1_gen4(vec4_instruction *inst,
 648                             struct brw_reg dst,
 649                             struct brw_reg src);
 650    void generate_math1_gen6(vec4_instruction *inst,
 651                             struct brw_reg dst,
 652                             struct brw_reg src);
 653    void generate_math2_gen4(vec4_instruction *inst,
 654                             struct brw_reg dst,
 655                             struct brw_reg src0,
 656                             struct brw_reg src1);
 657    void generate_math2_gen6(vec4_instruction *inst,
 658                             struct brw_reg dst,
 659                             struct brw_reg src0,
 660                             struct brw_reg src1);
 661    void generate_math2_gen7(vec4_instruction *inst,
 662                             struct brw_reg dst,
 663                             struct brw_reg src0,
 664                             struct brw_reg src1);
 665
 666    void generate_tex(vec4_instruction *inst,
 667                      struct brw_reg dst,
 668                      struct brw_reg src);
 669
 670    void generate_vs_urb_write(vec4_instruction *inst);
 671    void generate_gs_urb_write(vec4_instruction *inst);
 672    void generate_gs_thread_end(vec4_instruction *inst);
 673    void generate_gs_set_write_offset(struct brw_reg dst,
 674                                      struct brw_reg src0,
 675                                      struct brw_reg src1);
 676    void generate_gs_set_vertex_count(struct brw_reg dst,
 677                                      struct brw_reg src);
 678    void generate_gs_set_dword_2_immed(struct brw_reg dst, struct brw_reg src);
 679    void generate_gs_prepare_channel_masks(struct brw_reg dst);
 680    void generate_gs_set_channel_masks(struct brw_reg dst, struct brw_reg src);
 681    void generate_gs_get_instance_id(struct brw_reg dst);
 682    void generate_oword_dual_block_offsets(struct brw_reg m1,
 683                                           struct brw_reg index);
 684    void generate_scratch_write(vec4_instruction *inst,
 685                                struct brw_reg dst,
 686                                struct brw_reg src,
 687                                struct brw_reg index);
 688    void generate_scratch_read(vec4_instruction *inst,
 689                               struct brw_reg dst,
 690                               struct brw_reg index);
 691    void generate_pull_constant_load(vec4_instruction *inst,
 692                                     struct brw_reg dst,
 693                                     struct brw_reg index,
 694                                     struct brw_reg offset);
 695    void generate_pull_constant_load_gen7(vec4_instruction *inst,
 696                                          struct brw_reg dst,
 697                                          struct brw_reg surf_index,
 698                                          struct brw_reg offset);
 699    void generate_unpack_flags(vec4_instruction *inst,
 700                               struct brw_reg dst);
 701
 702    void generate_untyped_atomic(vec4_instruction *inst,
 703                                 struct brw_reg dst,
 704                                 struct brw_reg atomic_op,
 705                                 struct brw_reg surf_index);
 706
 707    void generate_untyped_surface_read(vec4_instruction *inst,
 708                                       struct brw_reg dst,
 709                                       struct brw_reg surf_index);
 710
 711    struct brw_context *brw;
 712
 713    struct brw_compile *p;
 714
 715    struct gl_shader_program *shader_prog;
 716    const struct gl_program *prog;
 717
 718    struct brw_vec4_prog_data *prog_data;
 719
 720    void *mem_ctx;
 721    const bool debug_flag;
 722 };
 723
 724 /**
 725  * The vertex shader code generator.
 726  *
 727  * Translates VS IR to actual i965 assembly code.
 728  */
 729 class gen8_vec4_generator : public gen8_generator
 730 {
 731 public:
 732    gen8_vec4_generator(struct brw_context *brw,
 733                        struct gl_shader_program *shader_prog,
 734                        struct gl_program *prog,
 735                        struct brw_vec4_prog_data *prog_data,
 736                        void *mem_ctx,
 737                        bool debug_flag);
 738    ~gen8_vec4_generator();
 739
 740    const unsigned *generate_assembly(exec_list *insts, unsigned *asm_size);
 741
 742 private:
 743    void generate_code(exec_list *instructions);
 744    void generate_vec4_instruction(vec4_instruction *inst,
 745                                   struct brw_reg dst,
 746                                   struct brw_reg *src);
 747
 748    void generate_tex(vec4_instruction *inst,
 749                      struct brw_reg dst);
 750
 751    void generate_urb_write(vec4_instruction *ir, bool copy_g0);
 752    void generate_gs_thread_end(vec4_instruction *ir);
 753    void generate_gs_set_write_offset(struct brw_reg dst,
 754                                      struct brw_reg src0,
 755                                      struct brw_reg src1);
 756    void generate_gs_set_vertex_count(struct brw_reg dst,
 757                                      struct brw_reg src);
 758    void generate_gs_set_dword_2_immed(struct brw_reg dst, struct brw_reg src);
 759    void generate_gs_prepare_channel_masks(struct brw_reg dst);
 760    void generate_gs_set_channel_masks(struct brw_reg dst, struct brw_reg src);
 761
 762    void generate_oword_dual_block_offsets(struct brw_reg m1,
 763                                           struct brw_reg index);
 764    void generate_scratch_write(vec4_instruction *inst,
 765                                struct brw_reg dst,
 766                                struct brw_reg src,
 767                                struct brw_reg index);
 768    void generate_scratch_read(vec4_instruction *inst,
 769                               struct brw_reg dst,
 770                               struct brw_reg index);
 771    void generate_pull_constant_load(vec4_instruction *inst,
 772                                     struct brw_reg dst,
 773                                     struct brw_reg index,
 774                                     struct brw_reg offset);
 775
 776    void mark_surface_used(unsigned surf_index);
 777
 778    struct brw_vec4_prog_data *prog_data;
 779
 780    const bool debug_flag;
 781 };
 782
 783
 784 } /* namespace brw */
 785 #endif /* __cplusplus */
 786
 787 #endif /* BRW_VEC4_H */