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 #include "brw_ir_vec4.h"
  35
  36 extern "C" {
  37 #endif
  38
  39 #include "brw_context.h"
  40 #include "brw_eu.h"
  41 #include "intel_asm_annotation.h"
  42
  43 #ifdef __cplusplus
  44 }; /* extern "C" */
  45 #endif
  46
  47 #include "glsl/ir.h"
  48
  49
  50 struct brw_vec4_compile {
  51    GLuint last_scratch; /**< measured in 32-byte (register size) units */
  52 };
  53
  54 #ifdef __cplusplus
  55 extern "C" {
  56 #endif
  57
  58 void
  59 brw_vue_setup_prog_key_for_precompile(struct gl_context *ctx,
  60                                       struct brw_vue_prog_key *key,
  61                                       GLuint id, struct gl_program *prog);
  62
  63 #ifdef __cplusplus
  64 } /* extern "C" */
  65
  66 namespace brw {
  67
  68 class vec4_live_variables;
  69
  70 /**
  71  * The vertex shader front-end.
  72  *
  73  * Translates either GLSL IR or Mesa IR (for ARB_vertex_program and
  74  * fixed-function) into VS IR.
  75  */
  76 class vec4_visitor : public backend_visitor
  77 {
  78 public:
  79    vec4_visitor(struct brw_context *brw,
  80                 struct brw_vec4_compile *c,
  81                 struct gl_program *prog,
  82                 const struct brw_vue_prog_key *key,
  83                 struct brw_vue_prog_data *prog_data,
  84                 struct gl_shader_program *shader_prog,
  85                 gl_shader_stage stage,
  86                 void *mem_ctx,
  87                 bool no_spills,
  88                 shader_time_shader_type st_base,
  89                 shader_time_shader_type st_written,
  90                 shader_time_shader_type st_reset);
  91    ~vec4_visitor();
  92
  93    dst_reg dst_null_f()
  94    {
  95       return dst_reg(brw_null_reg());
  96    }
  97
  98    dst_reg dst_null_d()
  99    {
 100       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
 101    }
 102
 103    dst_reg dst_null_ud()
 104    {
 105       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
 106    }
 107
 108    struct brw_vec4_compile * const c;
 109    const struct brw_vue_prog_key * const key;
 110    struct brw_vue_prog_data * const prog_data;
 111    unsigned int sanity_param_count;
 112
 113    char *fail_msg;
 114    bool failed;
 115
 116    /**
 117     * GLSL IR currently being processed, which is associated with our
 118     * driver IR instructions for debugging purposes.
 119     */
 120    const void *base_ir;
 121    const char *current_annotation;
 122
 123    int first_non_payload_grf;
 124    unsigned int max_grf;
 125    int *virtual_grf_start;
 126    int *virtual_grf_end;
 127    brw::vec4_live_variables *live_intervals;
 128    dst_reg userplane[MAX_CLIP_PLANES];
 129
 130    dst_reg *variable_storage(ir_variable *var);
 131
 132    void reladdr_to_temp(ir_instruction *ir, src_reg *reg, int *num_reladdr);
 133
 134    bool need_all_constants_in_pull_buffer;
 135
 136    /**
 137     * \name Visit methods
 138     *
 139     * As typical for the visitor pattern, there must be one \c visit method for
 140     * each concrete subclass of \c ir_instruction.  Virtual base classes within
 141     * the hierarchy should not have \c visit methods.
 142     */
 143    /*@{*/
 144    virtual void visit(ir_variable *);
 145    virtual void visit(ir_loop *);
 146    virtual void visit(ir_loop_jump *);
 147    virtual void visit(ir_function_signature *);
 148    virtual void visit(ir_function *);
 149    virtual void visit(ir_expression *);
 150    virtual void visit(ir_swizzle *);
 151    virtual void visit(ir_dereference_variable  *);
 152    virtual void visit(ir_dereference_array *);
 153    virtual void visit(ir_dereference_record *);
 154    virtual void visit(ir_assignment *);
 155    virtual void visit(ir_constant *);
 156    virtual void visit(ir_call *);
 157    virtual void visit(ir_return *);
 158    virtual void visit(ir_discard *);
 159    virtual void visit(ir_texture *);
 160    virtual void visit(ir_if *);
 161    virtual void visit(ir_emit_vertex *);
 162    virtual void visit(ir_end_primitive *);
 163    /*@}*/
 164
 165    src_reg result;
 166
 167    /* Regs for vertex results.  Generated at ir_variable visiting time
 168     * for the ir->location's used.
 169     */
 170    dst_reg output_reg[BRW_VARYING_SLOT_COUNT];
 171    const char *output_reg_annotation[BRW_VARYING_SLOT_COUNT];
 172    int *uniform_size;
 173    int *uniform_vector_size;
 174    int uniform_array_size; /*< Size of uniform_[vector_]size arrays */
 175    int uniforms;
 176
 177    src_reg shader_start_time;
 178
 179    struct hash_table *variable_ht;
 180
 181    bool run(void);
 182    void fail(const char *msg, ...);
 183
 184    void setup_uniform_clipplane_values();
 185    void setup_uniform_values(ir_variable *ir);
 186    void setup_builtin_uniform_values(ir_variable *ir);
 187    int setup_uniforms(int payload_reg);
 188    bool reg_allocate_trivial();
 189    bool reg_allocate();
 190    void evaluate_spill_costs(float *spill_costs, bool *no_spill);
 191    int choose_spill_reg(struct ra_graph *g);
 192    void spill_reg(int spill_reg);
 193    void move_grf_array_access_to_scratch();
 194    void move_uniform_array_access_to_pull_constants();
 195    void move_push_constants_to_pull_constants();
 196    void split_uniform_registers();
 197    void pack_uniform_registers();
 198    void calculate_live_intervals();
 199    void invalidate_live_intervals();
 200    void split_virtual_grfs();
 201    bool opt_vector_float();
 202    bool opt_reduce_swizzle();
 203    bool dead_code_eliminate();
 204    bool virtual_grf_interferes(int a, int b);
 205    bool opt_copy_propagation(bool do_constant_prop = true);
 206    bool opt_cse_local(bblock_t *block);
 207    bool opt_cse();
 208    bool opt_algebraic();
 209    bool opt_register_coalesce();
 210    bool is_dep_ctrl_unsafe(const vec4_instruction *inst);
 211    void opt_set_dependency_control();
 212    void opt_schedule_instructions();
 213
 214    vec4_instruction *emit(vec4_instruction *inst);
 215
 216    vec4_instruction *emit(enum opcode opcode);
 217    vec4_instruction *emit(enum opcode opcode, const dst_reg &dst);
 218    vec4_instruction *emit(enum opcode opcode, const dst_reg &dst,
 219                           const src_reg &src0);
 220    vec4_instruction *emit(enum opcode opcode, const dst_reg &dst,
 221                           const src_reg &src0, const src_reg &src1);
 222    vec4_instruction *emit(enum opcode opcode, const dst_reg &dst,
 223                           const src_reg &src0, const src_reg &src1,
 224                           const src_reg &src2);
 225
 226    vec4_instruction *emit_before(bblock_t *block,
 227                                  vec4_instruction *inst,
 228                                  vec4_instruction *new_inst);
 229
 230 #define EMIT1(op) vec4_instruction *op(const dst_reg &, const src_reg &);
 231 #define EMIT2(op) vec4_instruction *op(const dst_reg &, const src_reg &, const src_reg &);
 232 #define EMIT3(op) vec4_instruction *op(const dst_reg &, const src_reg &, const src_reg &, const src_reg &);
 233    EMIT1(MOV)
 234    EMIT1(NOT)
 235    EMIT1(RNDD)
 236    EMIT1(RNDE)
 237    EMIT1(RNDZ)
 238    EMIT1(FRC)
 239    EMIT1(F32TO16)
 240    EMIT1(F16TO32)
 241    EMIT2(ADD)
 242    EMIT2(MUL)
 243    EMIT2(MACH)
 244    EMIT2(MAC)
 245    EMIT2(AND)
 246    EMIT2(OR)
 247    EMIT2(XOR)
 248    EMIT2(DP3)
 249    EMIT2(DP4)
 250    EMIT2(DPH)
 251    EMIT2(SHL)
 252    EMIT2(SHR)
 253    EMIT2(ASR)
 254    vec4_instruction *CMP(dst_reg dst, src_reg src0, src_reg src1,
 255                          enum brw_conditional_mod condition);
 256    vec4_instruction *IF(src_reg src0, src_reg src1,
 257                         enum brw_conditional_mod condition);
 258    vec4_instruction *IF(enum brw_predicate predicate);
 259    EMIT1(PULL_CONSTANT_LOAD)
 260    EMIT1(SCRATCH_READ)
 261    EMIT2(SCRATCH_WRITE)
 262    EMIT3(LRP)
 263    EMIT1(BFREV)
 264    EMIT3(BFE)
 265    EMIT2(BFI1)
 266    EMIT3(BFI2)
 267    EMIT1(FBH)
 268    EMIT1(FBL)
 269    EMIT1(CBIT)
 270    EMIT3(MAD)
 271    EMIT2(ADDC)
 272    EMIT2(SUBB)
 273 #undef EMIT1
 274 #undef EMIT2
 275 #undef EMIT3
 276
 277    int implied_mrf_writes(vec4_instruction *inst);
 278
 279    bool try_rewrite_rhs_to_dst(ir_assignment *ir,
 280                                dst_reg dst,
 281                                src_reg src,
 282                                vec4_instruction *pre_rhs_inst,
 283                                vec4_instruction *last_rhs_inst);
 284
 285    /** Walks an exec_list of ir_instruction and sends it through this visitor. */
 286    void visit_instructions(const exec_list *list);
 287
 288    void emit_vp_sop(enum brw_conditional_mod condmod, dst_reg dst,
 289                     src_reg src0, src_reg src1, src_reg one);
 290
 291    void emit_bool_to_cond_code(ir_rvalue *ir, enum brw_predicate *predicate);
 292    void emit_if_gen6(ir_if *ir);
 293
 294    void emit_minmax(enum brw_conditional_mod conditionalmod, dst_reg dst,
 295                     src_reg src0, src_reg src1);
 296
 297    void emit_lrp(const dst_reg &dst,
 298                  const src_reg &x, const src_reg &y, const src_reg &a);
 299
 300    void emit_block_move(dst_reg *dst, src_reg *src,
 301                         const struct glsl_type *type, brw_predicate predicate);
 302
 303    void emit_constant_values(dst_reg *dst, ir_constant *value);
 304
 305    /**
 306     * Emit the correct dot-product instruction for the type of arguments
 307     */
 308    void emit_dp(dst_reg dst, src_reg src0, src_reg src1, unsigned elements);
 309
 310    void emit_scalar(ir_instruction *ir, enum prog_opcode op,
 311                     dst_reg dst, src_reg src0);
 312
 313    void emit_scalar(ir_instruction *ir, enum prog_opcode op,
 314                     dst_reg dst, src_reg src0, src_reg src1);
 315
 316    void emit_scs(ir_instruction *ir, enum prog_opcode op,
 317                  dst_reg dst, const src_reg &src);
 318
 319    src_reg fix_3src_operand(src_reg src);
 320
 321    void emit_math(enum opcode opcode, const dst_reg &dst, const src_reg &src0,
 322                   const src_reg &src1 = src_reg());
 323    src_reg fix_math_operand(src_reg src);
 324
 325    void emit_pack_half_2x16(dst_reg dst, src_reg src0);
 326    void emit_unpack_half_2x16(dst_reg dst, src_reg src0);
 327    void emit_unpack_unorm_4x8(const dst_reg &dst, src_reg src0);
 328    void emit_unpack_snorm_4x8(const dst_reg &dst, src_reg src0);
 329    void emit_pack_unorm_4x8(const dst_reg &dst, const src_reg &src0);
 330    void emit_pack_snorm_4x8(const dst_reg &dst, const src_reg &src0);
 331
 332    uint32_t gather_channel(ir_texture *ir, uint32_t sampler);
 333    src_reg emit_mcs_fetch(ir_texture *ir, src_reg coordinate, src_reg sampler);
 334    void emit_gen6_gather_wa(uint8_t wa, dst_reg dst);
 335    void swizzle_result(ir_texture *ir, src_reg orig_val, uint32_t sampler);
 336
 337    void emit_ndc_computation();
 338    void emit_psiz_and_flags(dst_reg reg);
 339    void emit_clip_distances(dst_reg reg, int offset);
 340    vec4_instruction *emit_generic_urb_slot(dst_reg reg, int varying);
 341    void emit_urb_slot(dst_reg reg, int varying);
 342
 343    void emit_shader_time_begin();
 344    void emit_shader_time_end();
 345    void emit_shader_time_write(enum shader_time_shader_type type,
 346                                src_reg value);
 347
 348    void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
 349                             dst_reg dst, src_reg offset, src_reg src0,
 350                             src_reg src1);
 351
 352    void emit_untyped_surface_read(unsigned surf_index, dst_reg dst,
 353                                   src_reg offset);
 354
 355    src_reg get_scratch_offset(bblock_t *block, vec4_instruction *inst,
 356                               src_reg *reladdr, int reg_offset);
 357    src_reg get_pull_constant_offset(bblock_t *block, vec4_instruction *inst,
 358                                     src_reg *reladdr, int reg_offset);
 359    void emit_scratch_read(bblock_t *block, vec4_instruction *inst,
 360                           dst_reg dst,
 361                           src_reg orig_src,
 362                           int base_offset);
 363    void emit_scratch_write(bblock_t *block, vec4_instruction *inst,
 364                            int base_offset);
 365    void emit_pull_constant_load(bblock_t *block, vec4_instruction *inst,
 366                                 dst_reg dst,
 367                                 src_reg orig_src,
 368                                 int base_offset);
 369
 370    bool try_emit_mad(ir_expression *ir);
 371    bool try_emit_b2f_of_compare(ir_expression *ir);
 372    void resolve_ud_negate(src_reg *reg);
 373    void resolve_bool_comparison(ir_rvalue *rvalue, src_reg *reg);
 374
 375    src_reg get_timestamp();
 376
 377    bool process_move_condition(ir_rvalue *ir);
 378
 379    void dump_instruction(backend_instruction *inst);
 380    void dump_instruction(backend_instruction *inst, FILE *file);
 381
 382    void visit_atomic_counter_intrinsic(ir_call *ir);
 383
 384 protected:
 385    void emit_vertex();
 386    void lower_attributes_to_hw_regs(const int *attribute_map,
 387                                     bool interleaved);
 388    void setup_payload_interference(struct ra_graph *g, int first_payload_node,
 389                                    int reg_node_count);
 390    virtual dst_reg *make_reg_for_system_value(ir_variable *ir) = 0;
 391    virtual void assign_binding_table_offsets();
 392    virtual void setup_payload() = 0;
 393    virtual void emit_prolog() = 0;
 394    virtual void emit_program_code() = 0;
 395    virtual void emit_thread_end() = 0;
 396    virtual void emit_urb_write_header(int mrf) = 0;
 397    virtual vec4_instruction *emit_urb_write_opcode(bool complete) = 0;
 398    virtual int compute_array_stride(ir_dereference_array *ir);
 399
 400 private:
 401    /**
 402     * If true, then register allocation should fail instead of spilling.
 403     */
 404    const bool no_spills;
 405
 406    const shader_time_shader_type st_base;
 407    const shader_time_shader_type st_written;
 408    const shader_time_shader_type st_reset;
 409 };
 410
 411
 412 /**
 413  * The vertex shader code generator.
 414  *
 415  * Translates VS IR to actual i965 assembly code.
 416  */
 417 class vec4_generator
 418 {
 419 public:
 420    vec4_generator(struct brw_context *brw,
 421                   struct gl_shader_program *shader_prog,
 422                   struct gl_program *prog,
 423                   struct brw_vue_prog_data *prog_data,
 424                   void *mem_ctx,
 425                   bool debug_flag,
 426                   const char *stage_name,
 427                   const char *stage_abbrev);
 428    ~vec4_generator();
 429
 430    const unsigned *generate_assembly(const cfg_t *cfg, unsigned *asm_size);
 431
 432 private:
 433    void generate_code(const cfg_t *cfg);
 434
 435    void generate_math1_gen4(vec4_instruction *inst,
 436                             struct brw_reg dst,
 437                             struct brw_reg src);
 438    void generate_math2_gen4(vec4_instruction *inst,
 439                             struct brw_reg dst,
 440                             struct brw_reg src0,
 441                             struct brw_reg src1);
 442    void generate_math_gen6(vec4_instruction *inst,
 443                            struct brw_reg dst,
 444                            struct brw_reg src0,
 445                            struct brw_reg src1);
 446
 447    void generate_tex(vec4_instruction *inst,
 448                      struct brw_reg dst,
 449                      struct brw_reg src,
 450                      struct brw_reg sampler_index);
 451
 452    void generate_vs_urb_write(vec4_instruction *inst);
 453    void generate_gs_urb_write(vec4_instruction *inst);
 454    void generate_gs_urb_write_allocate(vec4_instruction *inst);
 455    void generate_gs_thread_end(vec4_instruction *inst);
 456    void generate_gs_set_write_offset(struct brw_reg dst,
 457                                      struct brw_reg src0,
 458                                      struct brw_reg src1);
 459    void generate_gs_set_vertex_count(struct brw_reg dst,
 460                                      struct brw_reg src);
 461    void generate_gs_svb_write(vec4_instruction *inst,
 462                               struct brw_reg dst,
 463                               struct brw_reg src0,
 464                               struct brw_reg src1);
 465    void generate_gs_svb_set_destination_index(vec4_instruction *inst,
 466                                               struct brw_reg dst,
 467                                               struct brw_reg src);
 468    void generate_gs_set_dword_2(struct brw_reg dst, struct brw_reg src);
 469    void generate_gs_prepare_channel_masks(struct brw_reg dst);
 470    void generate_gs_set_channel_masks(struct brw_reg dst, struct brw_reg src);
 471    void generate_gs_get_instance_id(struct brw_reg dst);
 472    void generate_gs_ff_sync_set_primitives(struct brw_reg dst,
 473                                            struct brw_reg src0,
 474                                            struct brw_reg src1,
 475                                            struct brw_reg src2);
 476    void generate_gs_ff_sync(vec4_instruction *inst,
 477                             struct brw_reg dst,
 478                             struct brw_reg src0,
 479                             struct brw_reg src1);
 480    void generate_gs_set_primitive_id(struct brw_reg dst);
 481    void generate_oword_dual_block_offsets(struct brw_reg m1,
 482                                           struct brw_reg index);
 483    void generate_scratch_write(vec4_instruction *inst,
 484                                struct brw_reg dst,
 485                                struct brw_reg src,
 486                                struct brw_reg index);
 487    void generate_scratch_read(vec4_instruction *inst,
 488                               struct brw_reg dst,
 489                               struct brw_reg index);
 490    void generate_pull_constant_load(vec4_instruction *inst,
 491                                     struct brw_reg dst,
 492                                     struct brw_reg index,
 493                                     struct brw_reg offset);
 494    void generate_pull_constant_load_gen7(vec4_instruction *inst,
 495                                          struct brw_reg dst,
 496                                          struct brw_reg surf_index,
 497                                          struct brw_reg offset);
 498    void generate_unpack_flags(struct brw_reg dst);
 499
 500    void generate_untyped_atomic(vec4_instruction *inst,
 501                                 struct brw_reg dst,
 502                                 struct brw_reg atomic_op,
 503                                 struct brw_reg surf_index);
 504
 505    void generate_untyped_surface_read(vec4_instruction *inst,
 506                                       struct brw_reg dst,
 507                                       struct brw_reg surf_index);
 508
 509    struct brw_context *brw;
 510
 511    struct brw_compile *p;
 512
 513    struct gl_shader_program *shader_prog;
 514    const struct gl_program *prog;
 515
 516    struct brw_vue_prog_data *prog_data;
 517
 518    void *mem_ctx;
 519    const char *stage_name;
 520    const char *stage_abbrev;
 521    const bool debug_flag;
 522 };
 523
 524 } /* namespace brw */
 525 #endif /* __cplusplus */
 526
 527 #endif /* BRW_VEC4_H */