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_printer.h"
  40
  41 #ifdef __cplusplus
  42 }; /* extern "C" */
  43 #include "gen8_generator.h"
  44 #endif
  45
  46 #include "glsl/ir.h"
  47
  48
  49 struct brw_vec4_compile {
  50    GLuint last_scratch; /**< measured in 32-byte (register size) units */
  51 };
  52
  53
  54 struct brw_vec4_prog_key {
  55    GLuint program_string_id;
  56
  57    /**
  58     * True if at least one clip flag is enabled, regardless of whether the
  59     * shader uses clip planes or gl_ClipDistance.
  60     */
  61    GLuint userclip_active:1;
  62
  63    /**
  64     * How many user clipping planes are being uploaded to the vertex shader as
  65     * push constants.
  66     */
  67    GLuint nr_userclip_plane_consts:4;
  68
  69    GLuint clamp_vertex_color:1;
  70
  71    struct brw_sampler_prog_key_data tex;
  72 };
  73
  74
  75 #ifdef __cplusplus
  76 extern "C" {
  77 #endif
  78
  79 void
  80 brw_vec4_setup_prog_key_for_precompile(struct gl_context *ctx,
  81                                        struct brw_vec4_prog_key *key,
  82                                        GLuint id, struct gl_program *prog);
  83
  84 #ifdef __cplusplus
  85 } /* extern "C" */
  86
  87 namespace brw {
  88
  89 class dst_reg;
  90
  91 unsigned
  92 swizzle_for_size(int size);
  93
  94 class src_reg : public backend_reg
  95 {
  96 public:
  97    DECLARE_RALLOC_CXX_OPERATORS(src_reg)
  98
  99    void init();
 100
 101    src_reg(register_file file, int reg, const glsl_type *type);
 102    src_reg();
 103    src_reg(float f);
 104    src_reg(uint32_t u);
 105    src_reg(int32_t i);
 106    src_reg(struct brw_reg reg);
 107
 108    bool equals(const src_reg &r) const;
 109
 110    src_reg(class vec4_visitor *v, const struct glsl_type *type);
 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    bool is_send_from_grf();
 224    bool can_reswizzle_dst(int dst_writemask, int swizzle, int swizzle_mask);
 225    void reswizzle_dst(int dst_writemask, int swizzle);
 226    bool can_do_source_mods(struct brw_context *brw);
 227
 228    bool reads_flag()
 229    {
 230       return predicate || opcode == VS_OPCODE_UNPACK_FLAGS_SIMD4X2;
 231    }
 232
 233    bool writes_flag()
 234    {
 235       return conditional_mod && opcode != BRW_OPCODE_SEL;
 236    }
 237 };
 238
 239 /**
 240  * The vertex shader front-end.
 241  *
 242  * Translates either GLSL IR or Mesa IR (for ARB_vertex_program and
 243  * fixed-function) into VS IR.
 244  */
 245 class vec4_visitor : public backend_visitor
 246 {
 247 public:
 248    vec4_visitor(struct brw_context *brw,
 249                 struct brw_vec4_compile *c,
 250                 struct gl_program *prog,
 251                 const struct brw_vec4_prog_key *key,
 252                 struct brw_vec4_prog_data *prog_data,
 253                 struct gl_shader_program *shader_prog,
 254                 gl_shader_stage stage,
 255                 void *mem_ctx,
 256                 bool debug_flag,
 257                 bool no_spills,
 258                 shader_time_shader_type st_base,
 259                 shader_time_shader_type st_written,
 260                 shader_time_shader_type st_reset);
 261    ~vec4_visitor();
 262
 263    dst_reg dst_null_f()
 264    {
 265       return dst_reg(brw_null_reg());
 266    }
 267
 268    dst_reg dst_null_d()
 269    {
 270       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
 271    }
 272
 273    dst_reg dst_null_ud()
 274    {
 275       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
 276    }
 277
 278    struct brw_vec4_compile * const c;
 279    const struct brw_vec4_prog_key * const key;
 280    struct brw_vec4_prog_data * const prog_data;
 281    unsigned int sanity_param_count;
 282
 283    char *fail_msg;
 284    bool failed;
 285
 286    /**
 287     * GLSL IR currently being processed, which is associated with our
 288     * driver IR instructions for debugging purposes.
 289     */
 290    const void *base_ir;
 291    const char *current_annotation;
 292
 293    int *virtual_grf_sizes;
 294    int virtual_grf_count;
 295    int virtual_grf_array_size;
 296    int first_non_payload_grf;
 297    unsigned int max_grf;
 298    int *virtual_grf_start;
 299    int *virtual_grf_end;
 300    dst_reg userplane[MAX_CLIP_PLANES];
 301
 302    /**
 303     * This is the size to be used for an array with an element per
 304     * reg_offset
 305     */
 306    int virtual_grf_reg_count;
 307    /** Per-virtual-grf indices into an array of size virtual_grf_reg_count */
 308    int *virtual_grf_reg_map;
 309
 310    bool live_intervals_valid;
 311
 312    dst_reg *variable_storage(ir_variable *var);
 313
 314    void reladdr_to_temp(ir_instruction *ir, src_reg *reg, int *num_reladdr);
 315
 316    bool need_all_constants_in_pull_buffer;
 317
 318    /**
 319     * \name Visit methods
 320     *
 321     * As typical for the visitor pattern, there must be one \c visit method for
 322     * each concrete subclass of \c ir_instruction.  Virtual base classes within
 323     * the hierarchy should not have \c visit methods.
 324     */
 325    /*@{*/
 326    virtual void visit(ir_variable *);
 327    virtual void visit(ir_loop *);
 328    virtual void visit(ir_loop_jump *);
 329    virtual void visit(ir_function_signature *);
 330    virtual void visit(ir_function *);
 331    virtual void visit(ir_expression *);
 332    virtual void visit(ir_swizzle *);
 333    virtual void visit(ir_dereference_variable  *);
 334    virtual void visit(ir_dereference_array *);
 335    virtual void visit(ir_dereference_record *);
 336    virtual void visit(ir_assignment *);
 337    virtual void visit(ir_constant *);
 338    virtual void visit(ir_call *);
 339    virtual void visit(ir_return *);
 340    virtual void visit(ir_discard *);
 341    virtual void visit(ir_texture *);
 342    virtual void visit(ir_if *);
 343    virtual void visit(ir_emit_vertex *);
 344    virtual void visit(ir_end_primitive *);
 345    /*@}*/
 346
 347    src_reg result;
 348
 349    /* Regs for vertex results.  Generated at ir_variable visiting time
 350     * for the ir->location's used.
 351     */
 352    dst_reg output_reg[BRW_VARYING_SLOT_COUNT];
 353    const char *output_reg_annotation[BRW_VARYING_SLOT_COUNT];
 354    int *uniform_size;
 355    int *uniform_vector_size;
 356    int uniform_array_size; /*< Size of uniform_[vector_]size arrays */
 357    int uniforms;
 358
 359    src_reg shader_start_time;
 360
 361    struct hash_table *variable_ht;
 362
 363    bool run(void);
 364    void fail(const char *msg, ...);
 365
 366    int virtual_grf_alloc(int size);
 367    void setup_uniform_clipplane_values();
 368    void setup_uniform_values(ir_variable *ir);
 369    void setup_builtin_uniform_values(ir_variable *ir);
 370    int setup_uniforms(int payload_reg);
 371    bool reg_allocate_trivial();
 372    bool reg_allocate();
 373    void evaluate_spill_costs(float *spill_costs, bool *no_spill);
 374    int choose_spill_reg(struct ra_graph *g);
 375    void spill_reg(int spill_reg);
 376    void move_grf_array_access_to_scratch();
 377    void move_uniform_array_access_to_pull_constants();
 378    void move_push_constants_to_pull_constants();
 379    void split_uniform_registers();
 380    void pack_uniform_registers();
 381    void calculate_live_intervals();
 382    void invalidate_live_intervals();
 383    void split_virtual_grfs();
 384    bool dead_code_eliminate();
 385    bool virtual_grf_interferes(int a, int b);
 386    bool opt_copy_propagation();
 387    bool opt_algebraic();
 388    bool opt_register_coalesce();
 389    void opt_set_dependency_control();
 390    void opt_schedule_instructions();
 391
 392    vec4_instruction *emit(vec4_instruction *inst);
 393
 394    vec4_instruction *emit(enum opcode opcode);
 395
 396    vec4_instruction *emit(enum opcode opcode, dst_reg dst);
 397
 398    vec4_instruction *emit(enum opcode opcode, dst_reg dst, src_reg src0);
 399
 400    vec4_instruction *emit(enum opcode opcode, dst_reg dst,
 401                           src_reg src0, src_reg src1);
 402
 403    vec4_instruction *emit(enum opcode opcode, dst_reg dst,
 404                           src_reg src0, src_reg src1, src_reg src2);
 405
 406    vec4_instruction *emit_before(vec4_instruction *inst,
 407                                  vec4_instruction *new_inst);
 408
 409    vec4_instruction *MOV(const dst_reg &dst, const src_reg &src0);
 410    vec4_instruction *NOT(const dst_reg &dst, const src_reg &src0);
 411    vec4_instruction *RNDD(const dst_reg &dst, const src_reg &src0);
 412    vec4_instruction *RNDE(const dst_reg &dst, const src_reg &src0);
 413    vec4_instruction *RNDZ(const dst_reg &dst, const src_reg &src0);
 414    vec4_instruction *FRC(const dst_reg &dst, const src_reg &src0);
 415    vec4_instruction *F32TO16(const dst_reg &dst, const src_reg &src0);
 416    vec4_instruction *F16TO32(const dst_reg &dst, const src_reg &src0);
 417    vec4_instruction *ADD(const dst_reg &dst, const src_reg &src0,
 418                          const src_reg &src1);
 419    vec4_instruction *MUL(const dst_reg &dst, const src_reg &src0,
 420                          const src_reg &src1);
 421    vec4_instruction *MACH(const dst_reg &dst, const src_reg &src0,
 422                           const src_reg &src1);
 423    vec4_instruction *MAC(const dst_reg &dst, const src_reg &src0,
 424                          const src_reg &src1);
 425    vec4_instruction *AND(const dst_reg &dst, const src_reg &src0,
 426                          const src_reg &src1);
 427    vec4_instruction *OR(const dst_reg &dst, const src_reg &src0,
 428                         const src_reg &src1);
 429    vec4_instruction *XOR(const dst_reg &dst, const src_reg &src0,
 430                          const src_reg &src1);
 431    vec4_instruction *DP3(const dst_reg &dst, const src_reg &src0,
 432                          const src_reg &src1);
 433    vec4_instruction *DP4(const dst_reg &dst, const src_reg &src0,
 434                          const src_reg &src1);
 435    vec4_instruction *DPH(const dst_reg &dst, const src_reg &src0,
 436                          const src_reg &src1);
 437    vec4_instruction *SHL(const dst_reg &dst, const src_reg &src0,
 438                          const src_reg &src1);
 439    vec4_instruction *SHR(const dst_reg &dst, const src_reg &src0,
 440                          const src_reg &src1);
 441    vec4_instruction *ASR(const dst_reg &dst, const src_reg &src0,
 442                          const src_reg &src1);
 443    vec4_instruction *CMP(dst_reg dst, src_reg src0, src_reg src1,
 444                          enum brw_conditional_mod condition);
 445    vec4_instruction *IF(src_reg src0, src_reg src1,
 446                         enum brw_conditional_mod condition);
 447    vec4_instruction *IF(enum brw_predicate predicate);
 448    vec4_instruction *PULL_CONSTANT_LOAD(const dst_reg &dst,
 449                                         const src_reg &index);
 450    vec4_instruction *SCRATCH_READ(const dst_reg &dst, const src_reg &index);
 451    vec4_instruction *SCRATCH_WRITE(const dst_reg &dst, const src_reg &src,
 452                                    const src_reg &index);
 453    vec4_instruction *LRP(const dst_reg &dst, const src_reg &a,
 454                          const src_reg &y, const src_reg &x);
 455    vec4_instruction *BFREV(const dst_reg &dst, const src_reg &value);
 456    vec4_instruction *BFE(const dst_reg &dst, const src_reg &bits,
 457                          const src_reg &offset, const src_reg &value);
 458    vec4_instruction *BFI1(const dst_reg &dst, const src_reg &bits,
 459                           const src_reg &offset);
 460    vec4_instruction *BFI2(const dst_reg &dst, const src_reg &bfi1_dst,
 461                           const src_reg &insert, const src_reg &base);
 462    vec4_instruction *FBH(const dst_reg &dst, const src_reg &value);
 463    vec4_instruction *FBL(const dst_reg &dst, const src_reg &value);
 464    vec4_instruction *CBIT(const dst_reg &dst, const src_reg &value);
 465    vec4_instruction *MAD(const dst_reg &dst, const src_reg &c,
 466                          const src_reg &b, const src_reg &a);
 467    vec4_instruction *ADDC(const dst_reg &dst, const src_reg &src0,
 468                           const src_reg &src1);
 469    vec4_instruction *SUBB(const dst_reg &dst, const src_reg &src0,
 470                           const src_reg &src1);
 471
 472    int implied_mrf_writes(vec4_instruction *inst);
 473
 474    bool try_rewrite_rhs_to_dst(ir_assignment *ir,
 475                                dst_reg dst,
 476                                src_reg src,
 477                                vec4_instruction *pre_rhs_inst,
 478                                vec4_instruction *last_rhs_inst);
 479
 480    /** Walks an exec_list of ir_instruction and sends it through this visitor. */
 481    void visit_instructions(const exec_list *list);
 482
 483    void emit_vp_sop(enum brw_conditional_mod condmod, dst_reg dst,
 484                     src_reg src0, src_reg src1, src_reg one);
 485
 486    void emit_bool_to_cond_code(ir_rvalue *ir, enum brw_predicate *predicate);
 487    void emit_bool_comparison(unsigned int op, dst_reg dst, src_reg src0, src_reg src1);
 488    void emit_if_gen6(ir_if *ir);
 489
 490    void emit_minmax(enum brw_conditional_mod conditionalmod, dst_reg dst,
 491                     src_reg src0, src_reg src1);
 492
 493    void emit_lrp(const dst_reg &dst,
 494                  const src_reg &x, const src_reg &y, const src_reg &a);
 495
 496    void emit_block_move(dst_reg *dst, src_reg *src,
 497                         const struct glsl_type *type, brw_predicate predicate);
 498
 499    void emit_constant_values(dst_reg *dst, ir_constant *value);
 500
 501    /**
 502     * Emit the correct dot-product instruction for the type of arguments
 503     */
 504    void emit_dp(dst_reg dst, src_reg src0, src_reg src1, unsigned elements);
 505
 506    void emit_scalar(ir_instruction *ir, enum prog_opcode op,
 507                     dst_reg dst, src_reg src0);
 508
 509    void emit_scalar(ir_instruction *ir, enum prog_opcode op,
 510                     dst_reg dst, src_reg src0, src_reg src1);
 511
 512    void emit_scs(ir_instruction *ir, enum prog_opcode op,
 513                  dst_reg dst, const src_reg &src);
 514
 515    src_reg fix_3src_operand(src_reg src);
 516
 517    void emit_math1_gen6(enum opcode opcode, dst_reg dst, src_reg src);
 518    void emit_math1_gen4(enum opcode opcode, dst_reg dst, src_reg src);
 519    void emit_math(enum opcode opcode, dst_reg dst, src_reg src);
 520    void emit_math2_gen6(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 521    void emit_math2_gen4(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 522    void emit_math(enum opcode opcode, dst_reg dst, src_reg src0, src_reg src1);
 523    src_reg fix_math_operand(src_reg src);
 524
 525    void emit_pack_half_2x16(dst_reg dst, src_reg src0);
 526    void emit_unpack_half_2x16(dst_reg dst, src_reg src0);
 527
 528    uint32_t gather_channel(ir_texture *ir, int sampler);
 529    src_reg emit_mcs_fetch(ir_texture *ir, src_reg coordinate, int sampler);
 530    void emit_gen6_gather_wa(uint8_t wa, dst_reg dst);
 531    void swizzle_result(ir_texture *ir, src_reg orig_val, int sampler);
 532
 533    void emit_ndc_computation();
 534    void emit_psiz_and_flags(struct brw_reg reg);
 535    void emit_clip_distances(dst_reg reg, int offset);
 536    void emit_generic_urb_slot(dst_reg reg, int varying);
 537    void emit_urb_slot(int mrf, int varying);
 538
 539    void emit_shader_time_begin();
 540    void emit_shader_time_end();
 541    void emit_shader_time_write(enum shader_time_shader_type type,
 542                                src_reg value);
 543
 544    void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
 545                             dst_reg dst, src_reg offset, src_reg src0,
 546                             src_reg src1);
 547
 548    void emit_untyped_surface_read(unsigned surf_index, dst_reg dst,
 549                                   src_reg offset);
 550
 551    src_reg get_scratch_offset(vec4_instruction *inst,
 552                               src_reg *reladdr, int reg_offset);
 553    src_reg get_pull_constant_offset(vec4_instruction *inst,
 554                                     src_reg *reladdr, int reg_offset);
 555    void emit_scratch_read(vec4_instruction *inst,
 556                           dst_reg dst,
 557                           src_reg orig_src,
 558                           int base_offset);
 559    void emit_scratch_write(vec4_instruction *inst,
 560                            int base_offset);
 561    void emit_pull_constant_load(vec4_instruction *inst,
 562                                 dst_reg dst,
 563                                 src_reg orig_src,
 564                                 int base_offset);
 565
 566    bool try_emit_sat(ir_expression *ir);
 567    bool try_emit_mad(ir_expression *ir);
 568    bool try_emit_b2f_of_compare(ir_expression *ir);
 569    void resolve_ud_negate(src_reg *reg);
 570
 571    src_reg get_timestamp();
 572
 573    bool process_move_condition(ir_rvalue *ir);
 574
 575    void dump_instruction(backend_instruction *inst);
 576    void dump_instruction(backend_instruction *inst, FILE *file);
 577
 578    void visit_atomic_counter_intrinsic(ir_call *ir);
 579
 580 protected:
 581    void emit_vertex();
 582    void lower_attributes_to_hw_regs(const int *attribute_map,
 583                                     bool interleaved);
 584    void setup_payload_interference(struct ra_graph *g, int first_payload_node,
 585                                    int reg_node_count);
 586    virtual dst_reg *make_reg_for_system_value(ir_variable *ir) = 0;
 587    virtual void setup_payload() = 0;
 588    virtual void emit_prolog() = 0;
 589    virtual void emit_program_code() = 0;
 590    virtual void emit_thread_end() = 0;
 591    virtual void emit_urb_write_header(int mrf) = 0;
 592    virtual vec4_instruction *emit_urb_write_opcode(bool complete) = 0;
 593    virtual int compute_array_stride(ir_dereference_array *ir);
 594
 595    const bool debug_flag;
 596
 597 private:
 598    /**
 599     * If true, then register allocation should fail instead of spilling.
 600     */
 601    const bool no_spills;
 602
 603    const shader_time_shader_type st_base;
 604    const shader_time_shader_type st_written;
 605    const shader_time_shader_type st_reset;
 606 };
 607
 608
 609 /**
 610  * The vertex shader code generator.
 611  *
 612  * Translates VS IR to actual i965 assembly code.
 613  */
 614 class vec4_generator
 615 {
 616 public:
 617    vec4_generator(struct brw_context *brw,
 618                   struct gl_shader_program *shader_prog,
 619                   struct gl_program *prog,
 620                   struct brw_vec4_prog_data *prog_data,
 621                   void *mem_ctx,
 622                   bool debug_flag);
 623    ~vec4_generator();
 624
 625    const unsigned *generate_assembly(exec_list *insts, unsigned *asm_size);
 626
 627 private:
 628    void generate_code(exec_list *instructions);
 629    void generate_vec4_instruction(vec4_instruction *inst,
 630                                   struct brw_reg dst,
 631                                   struct brw_reg *src);
 632
 633    void generate_math1_gen4(vec4_instruction *inst,
 634                             struct brw_reg dst,
 635                             struct brw_reg src);
 636    void generate_math2_gen4(vec4_instruction *inst,
 637                             struct brw_reg dst,
 638                             struct brw_reg src0,
 639                             struct brw_reg src1);
 640    void generate_math_gen6(vec4_instruction *inst,
 641                            struct brw_reg dst,
 642                            struct brw_reg src0,
 643                            struct brw_reg src1);
 644
 645    void generate_tex(vec4_instruction *inst,
 646                      struct brw_reg dst,
 647                      struct brw_reg src);
 648
 649    void generate_vs_urb_write(vec4_instruction *inst);
 650    void generate_gs_urb_write(vec4_instruction *inst);
 651    void generate_gs_thread_end(vec4_instruction *inst);
 652    void generate_gs_set_write_offset(struct brw_reg dst,
 653                                      struct brw_reg src0,
 654                                      struct brw_reg src1);
 655    void generate_gs_set_vertex_count(struct brw_reg dst,
 656                                      struct brw_reg src);
 657    void generate_gs_set_dword_2_immed(struct brw_reg dst, struct brw_reg src);
 658    void generate_gs_prepare_channel_masks(struct brw_reg dst);
 659    void generate_gs_set_channel_masks(struct brw_reg dst, struct brw_reg src);
 660    void generate_gs_get_instance_id(struct brw_reg dst);
 661    void generate_oword_dual_block_offsets(struct brw_reg m1,
 662                                           struct brw_reg index);
 663    void generate_scratch_write(vec4_instruction *inst,
 664                                struct brw_reg dst,
 665                                struct brw_reg src,
 666                                struct brw_reg index);
 667    void generate_scratch_read(vec4_instruction *inst,
 668                               struct brw_reg dst,
 669                               struct brw_reg index);
 670    void generate_pull_constant_load(vec4_instruction *inst,
 671                                     struct brw_reg dst,
 672                                     struct brw_reg index,
 673                                     struct brw_reg offset);
 674    void generate_pull_constant_load_gen7(vec4_instruction *inst,
 675                                          struct brw_reg dst,
 676                                          struct brw_reg surf_index,
 677                                          struct brw_reg offset);
 678    void generate_unpack_flags(vec4_instruction *inst,
 679                               struct brw_reg dst);
 680
 681    void generate_untyped_atomic(vec4_instruction *inst,
 682                                 struct brw_reg dst,
 683                                 struct brw_reg atomic_op,
 684                                 struct brw_reg surf_index);
 685
 686    void generate_untyped_surface_read(vec4_instruction *inst,
 687                                       struct brw_reg dst,
 688                                       struct brw_reg surf_index);
 689
 690    struct brw_context *brw;
 691
 692    struct brw_compile *p;
 693
 694    struct gl_shader_program *shader_prog;
 695    const struct gl_program *prog;
 696
 697    struct brw_vec4_prog_data *prog_data;
 698
 699    void *mem_ctx;
 700    const bool debug_flag;
 701 };
 702
 703 /**
 704  * The vertex shader code generator.
 705  *
 706  * Translates VS IR to actual i965 assembly code.
 707  */
 708 class gen8_vec4_generator : public gen8_generator
 709 {
 710 public:
 711    gen8_vec4_generator(struct brw_context *brw,
 712                        struct gl_shader_program *shader_prog,
 713                        struct gl_program *prog,
 714                        struct brw_vec4_prog_data *prog_data,
 715                        void *mem_ctx,
 716                        bool debug_flag);
 717    ~gen8_vec4_generator();
 718
 719    const unsigned *generate_assembly(exec_list *insts, unsigned *asm_size);
 720
 721 private:
 722    void generate_code(exec_list *instructions);
 723    void generate_vec4_instruction(vec4_instruction *inst,
 724                                   struct brw_reg dst,
 725                                   struct brw_reg *src);
 726
 727    void generate_tex(vec4_instruction *inst,
 728                      struct brw_reg dst);
 729
 730    void generate_urb_write(vec4_instruction *ir, bool copy_g0);
 731    void generate_gs_thread_end(vec4_instruction *ir);
 732    void generate_gs_set_write_offset(struct brw_reg dst,
 733                                      struct brw_reg src0,
 734                                      struct brw_reg src1);
 735    void generate_gs_set_vertex_count(struct brw_reg dst,
 736                                      struct brw_reg src);
 737    void generate_gs_set_dword_2_immed(struct brw_reg dst, struct brw_reg src);
 738    void generate_gs_prepare_channel_masks(struct brw_reg dst);
 739    void generate_gs_set_channel_masks(struct brw_reg dst, struct brw_reg src);
 740
 741    void generate_oword_dual_block_offsets(struct brw_reg m1,
 742                                           struct brw_reg index);
 743    void generate_scratch_write(vec4_instruction *inst,
 744                                struct brw_reg dst,
 745                                struct brw_reg src,
 746                                struct brw_reg index);
 747    void generate_scratch_read(vec4_instruction *inst,
 748                               struct brw_reg dst,
 749                               struct brw_reg index);
 750    void generate_pull_constant_load(vec4_instruction *inst,
 751                                     struct brw_reg dst,
 752                                     struct brw_reg index,
 753                                     struct brw_reg offset);
 754    void generate_untyped_atomic(vec4_instruction *ir,
 755                                 struct brw_reg dst,
 756                                 struct brw_reg atomic_op,
 757                                 struct brw_reg surf_index);
 758    void generate_untyped_surface_read(vec4_instruction *ir,
 759                                       struct brw_reg dst,
 760                                       struct brw_reg surf_index);
 761
 762    struct brw_vec4_prog_data *prog_data;
 763
 764    const bool debug_flag;
 765 };
 766
 767
 768 } /* namespace brw */
 769 #endif /* __cplusplus */
 770
 771 #endif /* BRW_VEC4_H */