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 #include "intel_asm_annotation.h"
  40
  41 #ifdef __cplusplus
  42 }; /* extern "C" */
  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 src_reg : public backend_reg
  94 {
  95 public:
  96    DECLARE_RALLOC_CXX_OPERATORS(src_reg)
  97
  98    void init();
  99
 100    src_reg(register_file file, int reg, const glsl_type *type);
 101    src_reg();
 102    src_reg(float f);
 103    src_reg(uint32_t u);
 104    src_reg(int32_t i);
 105    src_reg(struct brw_reg reg);
 106
 107    bool equals(const src_reg &r) const;
 108
 109    src_reg(class vec4_visitor *v, const struct glsl_type *type);
 110    src_reg(class vec4_visitor *v, const struct glsl_type *type, int size);
 111
 112    explicit src_reg(dst_reg reg);
 113
 114    GLuint swizzle; /**< BRW_SWIZZLE_XYZW macros from brw_reg.h. */
 115
 116    src_reg *reladdr;
 117 };
 118
 119 static inline src_reg
 120 retype(src_reg reg, enum brw_reg_type type)
 121 {
 122    reg.fixed_hw_reg.type = reg.type = type;
 123    return reg;
 124 }
 125
 126 static inline src_reg
 127 offset(src_reg reg, unsigned delta)
 128 {
 129    assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
 130    reg.reg_offset += delta;
 131    return reg;
 132 }
 133
 134 /**
 135  * Reswizzle a given source register.
 136  * \sa brw_swizzle().
 137  */
 138 static inline src_reg
 139 swizzle(src_reg reg, unsigned swizzle)
 140 {
 141    assert(reg.file != HW_REG);
 142    reg.swizzle = BRW_SWIZZLE4(
 143       BRW_GET_SWZ(reg.swizzle, BRW_GET_SWZ(swizzle, 0)),
 144       BRW_GET_SWZ(reg.swizzle, BRW_GET_SWZ(swizzle, 1)),
 145       BRW_GET_SWZ(reg.swizzle, BRW_GET_SWZ(swizzle, 2)),
 146       BRW_GET_SWZ(reg.swizzle, BRW_GET_SWZ(swizzle, 3)));
 147    return reg;
 148 }
 149
 150 static inline src_reg
 151 negate(src_reg reg)
 152 {
 153    assert(reg.file != HW_REG && reg.file != IMM);
 154    reg.negate = !reg.negate;
 155    return reg;
 156 }
 157
 158 class dst_reg : public backend_reg
 159 {
 160 public:
 161    DECLARE_RALLOC_CXX_OPERATORS(dst_reg)
 162
 163    void init();
 164
 165    dst_reg();
 166    dst_reg(register_file file, int reg);
 167    dst_reg(register_file file, int reg, const glsl_type *type, int writemask);
 168    dst_reg(struct brw_reg reg);
 169    dst_reg(class vec4_visitor *v, const struct glsl_type *type);
 170
 171    explicit dst_reg(src_reg reg);
 172
 173    int writemask; /**< Bitfield of WRITEMASK_[XYZW] */
 174
 175    src_reg *reladdr;
 176 };
 177
 178 static inline dst_reg
 179 retype(dst_reg reg, enum brw_reg_type type)
 180 {
 181    reg.fixed_hw_reg.type = reg.type = type;
 182    return reg;
 183 }
 184
 185 static inline dst_reg
 186 offset(dst_reg reg, unsigned delta)
 187 {
 188    assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
 189    reg.reg_offset += delta;
 190    return reg;
 191 }
 192
 193 static inline dst_reg
 194 writemask(dst_reg reg, unsigned mask)
 195 {
 196    assert(reg.file != HW_REG && reg.file != IMM);
 197    assert((reg.writemask & mask) != 0);
 198    reg.writemask &= mask;
 199    return reg;
 200 }
 201
 202 class vec4_instruction : public backend_instruction {
 203 public:
 204    DECLARE_RALLOC_CXX_OPERATORS(vec4_instruction)
 205
 206    vec4_instruction(vec4_visitor *v, enum opcode opcode,
 207                     const dst_reg &dst = dst_reg(),
 208                     const src_reg &src0 = src_reg(),
 209                     const src_reg &src1 = src_reg(),
 210                     const src_reg &src2 = src_reg());
 211
 212    struct brw_reg get_dst(void);
 213    struct brw_reg get_src(const struct brw_vec4_prog_data *prog_data, int i);
 214
 215    dst_reg dst;
 216    src_reg src[3];
 217
 218    bool shadow_compare;
 219
 220    enum brw_urb_write_flags urb_write_flags;
 221    bool header_present;
 222
 223    unsigned sol_binding; /**< gen6: SOL binding table index */
 224    bool sol_final_write; /**< gen6: send commit message */
 225    unsigned sol_vertex; /**< gen6: used for setting dst index in SVB header */
 226
 227    bool is_send_from_grf();
 228    bool can_reswizzle(int dst_writemask, int swizzle, int swizzle_mask);
 229    void reswizzle(int dst_writemask, int swizzle);
 230    bool can_do_source_mods(struct brw_context *brw);
 231
 232    bool reads_flag()
 233    {
 234       return predicate || opcode == VS_OPCODE_UNPACK_FLAGS_SIMD4X2;
 235    }
 236
 237    bool writes_flag()
 238    {
 239       return conditional_mod && opcode != BRW_OPCODE_SEL;
 240    }
 241 };
 242
 243 /**
 244  * The vertex shader front-end.
 245  *
 246  * Translates either GLSL IR or Mesa IR (for ARB_vertex_program and
 247  * fixed-function) into VS IR.
 248  */
 249 class vec4_visitor : public backend_visitor
 250 {
 251 public:
 252    vec4_visitor(struct brw_context *brw,
 253                 struct brw_vec4_compile *c,
 254                 struct gl_program *prog,
 255                 const struct brw_vec4_prog_key *key,
 256                 struct brw_vec4_prog_data *prog_data,
 257                 struct gl_shader_program *shader_prog,
 258                 gl_shader_stage stage,
 259                 void *mem_ctx,
 260                 bool debug_flag,
 261                 bool no_spills,
 262                 shader_time_shader_type st_base,
 263                 shader_time_shader_type st_written,
 264                 shader_time_shader_type st_reset);
 265    ~vec4_visitor();
 266
 267    dst_reg dst_null_f()
 268    {
 269       return dst_reg(brw_null_reg());
 270    }
 271
 272    dst_reg dst_null_d()
 273    {
 274       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
 275    }
 276
 277    dst_reg dst_null_ud()
 278    {
 279       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
 280    }
 281
 282    struct brw_vec4_compile * const c;
 283    const struct brw_vec4_prog_key * const key;
 284    struct brw_vec4_prog_data * const prog_data;
 285    unsigned int sanity_param_count;
 286
 287    char *fail_msg;
 288    bool failed;
 289
 290    /**
 291     * GLSL IR currently being processed, which is associated with our
 292     * driver IR instructions for debugging purposes.
 293     */
 294    const void *base_ir;
 295    const char *current_annotation;
 296
 297    int *virtual_grf_sizes;
 298    int virtual_grf_count;
 299    int virtual_grf_array_size;
 300    int first_non_payload_grf;
 301    unsigned int max_grf;
 302    int *virtual_grf_start;
 303    int *virtual_grf_end;
 304    dst_reg userplane[MAX_CLIP_PLANES];
 305
 306    /**
 307     * This is the size to be used for an array with an element per
 308     * reg_offset
 309     */
 310    int virtual_grf_reg_count;
 311    /** Per-virtual-grf indices into an array of size virtual_grf_reg_count */
 312    int *virtual_grf_reg_map;
 313
 314    bool live_intervals_valid;
 315
 316    dst_reg *variable_storage(ir_variable *var);
 317
 318    void reladdr_to_temp(ir_instruction *ir, src_reg *reg, int *num_reladdr);
 319
 320    bool need_all_constants_in_pull_buffer;
 321
 322    /**
 323     * \name Visit methods
 324     *
 325     * As typical for the visitor pattern, there must be one \c visit method for
 326     * each concrete subclass of \c ir_instruction.  Virtual base classes within
 327     * the hierarchy should not have \c visit methods.
 328     */
 329    /*@{*/
 330    virtual void visit(ir_variable *);
 331    virtual void visit(ir_loop *);
 332    virtual void visit(ir_loop_jump *);
 333    virtual void visit(ir_function_signature *);
 334    virtual void visit(ir_function *);
 335    virtual void visit(ir_expression *);
 336    virtual void visit(ir_swizzle *);
 337    virtual void visit(ir_dereference_variable  *);
 338    virtual void visit(ir_dereference_array *);
 339    virtual void visit(ir_dereference_record *);
 340    virtual void visit(ir_assignment *);
 341    virtual void visit(ir_constant *);
 342    virtual void visit(ir_call *);
 343    virtual void visit(ir_return *);
 344    virtual void visit(ir_discard *);
 345    virtual void visit(ir_texture *);
 346    virtual void visit(ir_if *);
 347    virtual void visit(ir_emit_vertex *);
 348    virtual void visit(ir_end_primitive *);
 349    /*@}*/
 350
 351    src_reg result;
 352
 353    /* Regs for vertex results.  Generated at ir_variable visiting time
 354     * for the ir->location's used.
 355     */
 356    dst_reg output_reg[BRW_VARYING_SLOT_COUNT];
 357    const char *output_reg_annotation[BRW_VARYING_SLOT_COUNT];
 358    int *uniform_size;
 359    int *uniform_vector_size;
 360    int uniform_array_size; /*< Size of uniform_[vector_]size arrays */
 361    int uniforms;
 362
 363    src_reg shader_start_time;
 364
 365    struct hash_table *variable_ht;
 366
 367    bool run(void);
 368    void fail(const char *msg, ...);
 369
 370    int virtual_grf_alloc(int size);
 371    void setup_uniform_clipplane_values();
 372    void setup_uniform_values(ir_variable *ir);
 373    void setup_builtin_uniform_values(ir_variable *ir);
 374    int setup_uniforms(int payload_reg);
 375    bool reg_allocate_trivial();
 376    bool reg_allocate();
 377    void evaluate_spill_costs(float *spill_costs, bool *no_spill);
 378    int choose_spill_reg(struct ra_graph *g);
 379    void spill_reg(int spill_reg);
 380    void move_grf_array_access_to_scratch();
 381    void move_uniform_array_access_to_pull_constants();
 382    void move_push_constants_to_pull_constants();
 383    void split_uniform_registers();
 384    void pack_uniform_registers();
 385    void calculate_live_intervals();
 386    void invalidate_live_intervals(bool invalidate_cfg = true);
 387    void split_virtual_grfs();
 388    bool opt_reduce_swizzle();
 389    bool dead_code_eliminate();
 390    bool virtual_grf_interferes(int a, int b);
 391    bool opt_copy_propagation();
 392    bool opt_cse_local(bblock_t *block);
 393    bool opt_cse();
 394    bool opt_algebraic();
 395    bool opt_register_coalesce();
 396    void opt_set_dependency_control();
 397    void opt_schedule_instructions();
 398
 399    vec4_instruction *emit(vec4_instruction *inst);
 400
 401    vec4_instruction *emit(enum opcode opcode);
 402
 403    vec4_instruction *emit(enum opcode opcode, dst_reg dst);
 404
 405    vec4_instruction *emit(enum opcode opcode, dst_reg dst, src_reg src0);
 406
 407    vec4_instruction *emit(enum opcode opcode, dst_reg dst,
 408                           src_reg src0, src_reg src1);
 409
 410    vec4_instruction *emit(enum opcode opcode, dst_reg dst,
 411                           src_reg src0, src_reg src1, src_reg src2);
 412
 413    vec4_instruction *emit_before(vec4_instruction *inst,
 414                                  vec4_instruction *new_inst);
 415
 416    vec4_instruction *MOV(const dst_reg &dst, const src_reg &src0);
 417    vec4_instruction *NOT(const dst_reg &dst, const src_reg &src0);
 418    vec4_instruction *RNDD(const dst_reg &dst, const src_reg &src0);
 419    vec4_instruction *RNDE(const dst_reg &dst, const src_reg &src0);
 420    vec4_instruction *RNDZ(const dst_reg &dst, const src_reg &src0);
 421    vec4_instruction *FRC(const dst_reg &dst, const src_reg &src0);
 422    vec4_instruction *F32TO16(const dst_reg &dst, const src_reg &src0);
 423    vec4_instruction *F16TO32(const dst_reg &dst, const src_reg &src0);
 424    vec4_instruction *ADD(const dst_reg &dst, const src_reg &src0,
 425                          const src_reg &src1);
 426    vec4_instruction *MUL(const dst_reg &dst, const src_reg &src0,
 427                          const src_reg &src1);
 428    vec4_instruction *MACH(const dst_reg &dst, const src_reg &src0,
 429                           const src_reg &src1);
 430    vec4_instruction *MAC(const dst_reg &dst, const src_reg &src0,
 431                          const src_reg &src1);
 432    vec4_instruction *AND(const dst_reg &dst, const src_reg &src0,
 433                          const src_reg &src1);
 434    vec4_instruction *OR(const dst_reg &dst, const src_reg &src0,
 435                         const src_reg &src1);
 436    vec4_instruction *XOR(const dst_reg &dst, const src_reg &src0,
 437                          const src_reg &src1);
 438    vec4_instruction *DP3(const dst_reg &dst, const src_reg &src0,
 439                          const src_reg &src1);
 440    vec4_instruction *DP4(const dst_reg &dst, const src_reg &src0,
 441                          const src_reg &src1);
 442    vec4_instruction *DPH(const dst_reg &dst, const src_reg &src0,
 443                          const src_reg &src1);
 444    vec4_instruction *SHL(const dst_reg &dst, const src_reg &src0,
 445                          const src_reg &src1);
 446    vec4_instruction *SHR(const dst_reg &dst, const src_reg &src0,
 447                          const src_reg &src1);
 448    vec4_instruction *ASR(const dst_reg &dst, const src_reg &src0,
 449                          const src_reg &src1);
 450    vec4_instruction *CMP(dst_reg dst, src_reg src0, src_reg src1,
 451                          enum brw_conditional_mod condition);
 452    vec4_instruction *IF(src_reg src0, src_reg src1,
 453                         enum brw_conditional_mod condition);
 454    vec4_instruction *IF(enum brw_predicate predicate);
 455    vec4_instruction *PULL_CONSTANT_LOAD(const dst_reg &dst,
 456                                         const src_reg &index);
 457    vec4_instruction *SCRATCH_READ(const dst_reg &dst, const src_reg &index);
 458    vec4_instruction *SCRATCH_WRITE(const dst_reg &dst, const src_reg &src,
 459                                    const src_reg &index);
 460    vec4_instruction *LRP(const dst_reg &dst, const src_reg &a,
 461                          const src_reg &y, const src_reg &x);
 462    vec4_instruction *BFREV(const dst_reg &dst, const src_reg &value);
 463    vec4_instruction *BFE(const dst_reg &dst, const src_reg &bits,
 464                          const src_reg &offset, const src_reg &value);
 465    vec4_instruction *BFI1(const dst_reg &dst, const src_reg &bits,
 466                           const src_reg &offset);
 467    vec4_instruction *BFI2(const dst_reg &dst, const src_reg &bfi1_dst,
 468                           const src_reg &insert, const src_reg &base);
 469    vec4_instruction *FBH(const dst_reg &dst, const src_reg &value);
 470    vec4_instruction *FBL(const dst_reg &dst, const src_reg &value);
 471    vec4_instruction *CBIT(const dst_reg &dst, const src_reg &value);
 472    vec4_instruction *MAD(const dst_reg &dst, const src_reg &c,
 473                          const src_reg &b, const src_reg &a);
 474    vec4_instruction *ADDC(const dst_reg &dst, const src_reg &src0,
 475                           const src_reg &src1);
 476    vec4_instruction *SUBB(const dst_reg &dst, const src_reg &src0,
 477                           const src_reg &src1);
 478
 479    int implied_mrf_writes(vec4_instruction *inst);
 480
 481    bool try_rewrite_rhs_to_dst(ir_assignment *ir,
 482                                dst_reg dst,
 483                                src_reg src,
 484                                vec4_instruction *pre_rhs_inst,
 485                                vec4_instruction *last_rhs_inst);
 486
 487    /** Walks an exec_list of ir_instruction and sends it through this visitor. */
 488    void visit_instructions(const exec_list *list);
 489
 490    void emit_vp_sop(enum brw_conditional_mod condmod, dst_reg dst,
 491                     src_reg src0, src_reg src1, src_reg one);
 492
 493    void emit_bool_to_cond_code(ir_rvalue *ir, enum brw_predicate *predicate);
 494    void emit_if_gen6(ir_if *ir);
 495
 496    void emit_minmax(enum brw_conditional_mod conditionalmod, dst_reg dst,
 497                     src_reg src0, src_reg src1);
 498
 499    void emit_lrp(const dst_reg &dst,
 500                  const src_reg &x, const src_reg &y, const src_reg &a);
 501
 502    void emit_block_move(dst_reg *dst, src_reg *src,
 503                         const struct glsl_type *type, brw_predicate predicate);
 504
 505    void emit_constant_values(dst_reg *dst, ir_constant *value);
 506
 507    /**
 508     * Emit the correct dot-product instruction for the type of arguments
 509     */
 510    void emit_dp(dst_reg dst, src_reg src0, src_reg src1, unsigned elements);
 511
 512    void emit_scalar(ir_instruction *ir, enum prog_opcode op,
 513                     dst_reg dst, src_reg src0);
 514
 515    void emit_scalar(ir_instruction *ir, enum prog_opcode op,
 516                     dst_reg dst, src_reg src0, src_reg src1);
 517
 518    void emit_scs(ir_instruction *ir, enum prog_opcode op,
 519                  dst_reg dst, const src_reg &src);
 520
 521    src_reg fix_3src_operand(src_reg src);
 522
 523    void emit_math1_gen6(enum opcode opcode, dst_reg dst, src_reg src);
 524    void emit_math1_gen4(enum opcode opcode, dst_reg dst, src_reg src);
 525    void emit_math(enum opcode opcode, dst_reg dst, src_reg src);
 526    void emit_math2_gen6(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 527    void emit_math2_gen4(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 528    void emit_math(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 529    src_reg fix_math_operand(src_reg src);
 530
 531    void emit_pack_half_2x16(dst_reg dst, src_reg src0);
 532    void emit_unpack_half_2x16(dst_reg dst, src_reg src0);
 533
 534    uint32_t gather_channel(ir_texture *ir, uint32_t sampler);
 535    src_reg emit_mcs_fetch(ir_texture *ir, src_reg coordinate, src_reg sampler);
 536    void emit_gen6_gather_wa(uint8_t wa, dst_reg dst);
 537    void swizzle_result(ir_texture *ir, src_reg orig_val, uint32_t sampler);
 538
 539    void emit_ndc_computation();
 540    void emit_psiz_and_flags(dst_reg reg);
 541    void emit_clip_distances(dst_reg reg, int offset);
 542    void emit_generic_urb_slot(dst_reg reg, int varying);
 543    void emit_urb_slot(dst_reg reg, int varying);
 544
 545    void emit_shader_time_begin();
 546    void emit_shader_time_end();
 547    void emit_shader_time_write(enum shader_time_shader_type type,
 548                                src_reg value);
 549
 550    void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
 551                             dst_reg dst, src_reg offset, src_reg src0,
 552                             src_reg src1);
 553
 554    void emit_untyped_surface_read(unsigned surf_index, dst_reg dst,
 555                                   src_reg offset);
 556
 557    src_reg get_scratch_offset(vec4_instruction *inst,
 558                               src_reg *reladdr, int reg_offset);
 559    src_reg get_pull_constant_offset(vec4_instruction *inst,
 560                                     src_reg *reladdr, int reg_offset);
 561    void emit_scratch_read(vec4_instruction *inst,
 562                           dst_reg dst,
 563                           src_reg orig_src,
 564                           int base_offset);
 565    void emit_scratch_write(vec4_instruction *inst,
 566                            int base_offset);
 567    void emit_pull_constant_load(vec4_instruction *inst,
 568                                 dst_reg dst,
 569                                 src_reg orig_src,
 570                                 int base_offset);
 571
 572    bool try_emit_mad(ir_expression *ir);
 573    bool try_emit_b2f_of_compare(ir_expression *ir);
 574    void resolve_ud_negate(src_reg *reg);
 575
 576    src_reg get_timestamp();
 577
 578    bool process_move_condition(ir_rvalue *ir);
 579
 580    void dump_instruction(backend_instruction *inst);
 581    void dump_instruction(backend_instruction *inst, FILE *file);
 582
 583    void visit_atomic_counter_intrinsic(ir_call *ir);
 584
 585 protected:
 586    void emit_vertex();
 587    void lower_attributes_to_hw_regs(const int *attribute_map,
 588                                     bool interleaved);
 589    void setup_payload_interference(struct ra_graph *g, int first_payload_node,
 590                                    int reg_node_count);
 591    virtual dst_reg *make_reg_for_system_value(ir_variable *ir) = 0;
 592    virtual void assign_binding_table_offsets();
 593    virtual void setup_payload() = 0;
 594    virtual void emit_prolog() = 0;
 595    virtual void emit_program_code() = 0;
 596    virtual void emit_thread_end() = 0;
 597    virtual void emit_urb_write_header(int mrf) = 0;
 598    virtual vec4_instruction *emit_urb_write_opcode(bool complete) = 0;
 599    virtual int compute_array_stride(ir_dereference_array *ir);
 600
 601    const bool debug_flag;
 602
 603 private:
 604    /**
 605     * If true, then register allocation should fail instead of spilling.
 606     */
 607    const bool no_spills;
 608
 609    const shader_time_shader_type st_base;
 610    const shader_time_shader_type st_written;
 611    const shader_time_shader_type st_reset;
 612 };
 613
 614
 615 /**
 616  * The vertex shader code generator.
 617  *
 618  * Translates VS IR to actual i965 assembly code.
 619  */
 620 class vec4_generator
 621 {
 622 public:
 623    vec4_generator(struct brw_context *brw,
 624                   struct gl_shader_program *shader_prog,
 625                   struct gl_program *prog,
 626                   struct brw_vec4_prog_data *prog_data,
 627                   void *mem_ctx,
 628                   bool debug_flag);
 629    ~vec4_generator();
 630
 631    const unsigned *generate_assembly(const cfg_t *cfg, unsigned *asm_size);
 632
 633 private:
 634    void generate_code(const cfg_t *cfg);
 635
 636    void generate_math1_gen4(vec4_instruction *inst,
 637                             struct brw_reg dst,
 638                             struct brw_reg src);
 639    void generate_math2_gen4(vec4_instruction *inst,
 640                             struct brw_reg dst,
 641                             struct brw_reg src0,
 642                             struct brw_reg src1);
 643    void generate_math_gen6(vec4_instruction *inst,
 644                            struct brw_reg dst,
 645                            struct brw_reg src0,
 646                            struct brw_reg src1);
 647
 648    void generate_tex(vec4_instruction *inst,
 649                      struct brw_reg dst,
 650                      struct brw_reg src,
 651                      struct brw_reg sampler_index);
 652
 653    void generate_vs_urb_write(vec4_instruction *inst);
 654    void generate_gs_urb_write(vec4_instruction *inst);
 655    void generate_gs_urb_write_allocate(vec4_instruction *inst);
 656    void generate_gs_thread_end(vec4_instruction *inst);
 657    void generate_gs_set_write_offset(struct brw_reg dst,
 658                                      struct brw_reg src0,
 659                                      struct brw_reg src1);
 660    void generate_gs_set_vertex_count(struct brw_reg dst,
 661                                      struct brw_reg src);
 662    void generate_gs_svb_write(vec4_instruction *inst,
 663                               struct brw_reg dst,
 664                               struct brw_reg src0,
 665                               struct brw_reg src1);
 666    void generate_gs_svb_set_destination_index(vec4_instruction *inst,
 667                                               struct brw_reg dst,
 668                                               struct brw_reg src);
 669    void generate_gs_set_dword_2(struct brw_reg dst, struct brw_reg src);
 670    void generate_gs_prepare_channel_masks(struct brw_reg dst);
 671    void generate_gs_set_channel_masks(struct brw_reg dst, struct brw_reg src);
 672    void generate_gs_get_instance_id(struct brw_reg dst);
 673    void generate_gs_ff_sync_set_primitives(struct brw_reg dst,
 674                                            struct brw_reg src0,
 675                                            struct brw_reg src1,
 676                                            struct brw_reg src2);
 677    void generate_gs_ff_sync(vec4_instruction *inst,
 678                             struct brw_reg dst,
 679                             struct brw_reg src0,
 680                             struct brw_reg src1);
 681    void generate_gs_set_primitive_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 } /* namespace brw */
 725 #endif /* __cplusplus */
 726
 727 #endif /* BRW_VEC4_H */