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

   1 /*
   2  * Copyright © 2010 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Eric Anholt <eric@anholt.net>
  25  *
  26  */
  27
  28 #include "brw_shader.h"
  29
  30 extern "C" {
  31
  32 #include <sys/types.h>
  33
  34 #include "main/macros.h"
  35 #include "main/shaderobj.h"
  36 #include "main/uniforms.h"
  37 #include "program/prog_parameter.h"
  38 #include "program/prog_print.h"
  39 #include "program/prog_optimize.h"
  40 #include "program/register_allocate.h"
  41 #include "program/sampler.h"
  42 #include "program/hash_table.h"
  43 #include "brw_context.h"
  44 #include "brw_eu.h"
  45 #include "brw_wm.h"
  46 }
  47 #include "glsl/glsl_types.h"
  48 #include "glsl/ir.h"
  49
  50 enum register_file {
  51    BAD_FILE,
  52    ARF,
  53    GRF,
  54    MRF,
  55    IMM,
  56    FIXED_HW_REG, /* a struct brw_reg */
  57    UNIFORM, /* prog_data->params[reg] */
  58 };
  59
  60 class fs_reg {
  61 public:
  62    /* Callers of this ralloc-based new need not call delete. It's
  63     * easier to just ralloc_free 'ctx' (or any of its ancestors). */
  64    static void* operator new(size_t size, void *ctx)
  65    {
  66       void *node;
  67
  68       node = ralloc_size(ctx, size);
  69       assert(node != NULL);
  70
  71       return node;
  72    }
  73
  74    void init()
  75    {
  76       memset(this, 0, sizeof(*this));
  77       this->smear = -1;
  78    }
  79
  80    /** Generic unset register constructor. */
  81    fs_reg()
  82    {
  83       init();
  84       this->file = BAD_FILE;
  85    }
  86
  87    /** Immediate value constructor. */
  88    fs_reg(float f)
  89    {
  90       init();
  91       this->file = IMM;
  92       this->type = BRW_REGISTER_TYPE_F;
  93       this->imm.f = f;
  94    }
  95
  96    /** Immediate value constructor. */
  97    fs_reg(int32_t i)
  98    {
  99       init();
 100       this->file = IMM;
 101       this->type = BRW_REGISTER_TYPE_D;
 102       this->imm.i = i;
 103    }
 104
 105    /** Immediate value constructor. */
 106    fs_reg(uint32_t u)
 107    {
 108       init();
 109       this->file = IMM;
 110       this->type = BRW_REGISTER_TYPE_UD;
 111       this->imm.u = u;
 112    }
 113
 114    /** Fixed brw_reg Immediate value constructor. */
 115    fs_reg(struct brw_reg fixed_hw_reg)
 116    {
 117       init();
 118       this->file = FIXED_HW_REG;
 119       this->fixed_hw_reg = fixed_hw_reg;
 120       this->type = fixed_hw_reg.type;
 121    }
 122
 123    fs_reg(enum register_file file, int reg);
 124    fs_reg(enum register_file file, int reg, uint32_t type);
 125    fs_reg(class fs_visitor *v, const struct glsl_type *type);
 126
 127    bool equals(fs_reg *r)
 128    {
 129       return (file == r->file &&
 130               reg == r->reg &&
 131               reg_offset == r->reg_offset &&
 132               type == r->type &&
 133               negate == r->negate &&
 134               abs == r->abs &&
 135               memcmp(&fixed_hw_reg, &r->fixed_hw_reg,
 136                      sizeof(fixed_hw_reg)) == 0 &&
 137               smear == r->smear &&
 138               imm.u == r->imm.u);
 139    }
 140
 141    /** Register file: ARF, GRF, MRF, IMM. */
 142    enum register_file file;
 143    /**
 144     * Register number.  For ARF/MRF, it's the hardware register.  For
 145     * GRF, it's a virtual register number until register allocation
 146     */
 147    int reg;
 148    /**
 149     * For virtual registers, this is a hardware register offset from
 150     * the start of the register block (for example, a constant index
 151     * in an array access).
 152     */
 153    int reg_offset;
 154    /** Register type.  BRW_REGISTER_TYPE_* */
 155    int type;
 156    bool negate;
 157    bool abs;
 158    bool sechalf;
 159    struct brw_reg fixed_hw_reg;
 160    int smear; /* -1, or a channel of the reg to smear to all channels. */
 161
 162    /** Value for file == IMM */
 163    union {
 164       int32_t i;
 165       uint32_t u;
 166       float f;
 167    } imm;
 168 };
 169
 170 static const fs_reg reg_undef;
 171 static const fs_reg reg_null_f(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_F);
 172 static const fs_reg reg_null_d(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_D);
 173
 174 class ip_record : public exec_node {
 175 public:
 176    static void* operator new(size_t size, void *ctx)
 177    {
 178       void *node;
 179
 180       node = rzalloc_size(ctx, size);
 181       assert(node != NULL);
 182
 183       return node;
 184    }
 185
 186    ip_record(int ip)
 187    {
 188       this->ip = ip;
 189    }
 190
 191    int ip;
 192 };
 193
 194 class fs_inst : public exec_node {
 195 public:
 196    /* Callers of this ralloc-based new need not call delete. It's
 197     * easier to just ralloc_free 'ctx' (or any of its ancestors). */
 198    static void* operator new(size_t size, void *ctx)
 199    {
 200       void *node;
 201
 202       node = rzalloc_size(ctx, size);
 203       assert(node != NULL);
 204
 205       return node;
 206    }
 207
 208    void init()
 209    {
 210       memset(this, 0, sizeof(*this));
 211       this->opcode = BRW_OPCODE_NOP;
 212       this->conditional_mod = BRW_CONDITIONAL_NONE;
 213
 214       this->dst = reg_undef;
 215       this->src[0] = reg_undef;
 216       this->src[1] = reg_undef;
 217       this->src[2] = reg_undef;
 218    }
 219
 220    fs_inst()
 221    {
 222       init();
 223    }
 224
 225    fs_inst(enum opcode opcode)
 226    {
 227       init();
 228       this->opcode = opcode;
 229    }
 230
 231    fs_inst(enum opcode opcode, fs_reg dst)
 232    {
 233       init();
 234       this->opcode = opcode;
 235       this->dst = dst;
 236
 237       if (dst.file == GRF)
 238          assert(dst.reg_offset >= 0);
 239    }
 240
 241    fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0)
 242    {
 243       init();
 244       this->opcode = opcode;
 245       this->dst = dst;
 246       this->src[0] = src0;
 247
 248       if (dst.file == GRF)
 249          assert(dst.reg_offset >= 0);
 250       if (src[0].file == GRF)
 251          assert(src[0].reg_offset >= 0);
 252    }
 253
 254    fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)
 255    {
 256       init();
 257       this->opcode = opcode;
 258       this->dst = dst;
 259       this->src[0] = src0;
 260       this->src[1] = src1;
 261
 262       if (dst.file == GRF)
 263          assert(dst.reg_offset >= 0);
 264       if (src[0].file == GRF)
 265          assert(src[0].reg_offset >= 0);
 266       if (src[1].file == GRF)
 267          assert(src[1].reg_offset >= 0);
 268    }
 269
 270    fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1, fs_reg src2)
 271    {
 272       init();
 273       this->opcode = opcode;
 274       this->dst = dst;
 275       this->src[0] = src0;
 276       this->src[1] = src1;
 277       this->src[2] = src2;
 278
 279       if (dst.file == GRF)
 280          assert(dst.reg_offset >= 0);
 281       if (src[0].file == GRF)
 282          assert(src[0].reg_offset >= 0);
 283       if (src[1].file == GRF)
 284          assert(src[1].reg_offset >= 0);
 285       if (src[2].file == GRF)
 286          assert(src[2].reg_offset >= 0);
 287    }
 288
 289    bool equals(fs_inst *inst)
 290    {
 291       return (opcode == inst->opcode &&
 292               dst.equals(&inst->dst) &&
 293               src[0].equals(&inst->src[0]) &&
 294               src[1].equals(&inst->src[1]) &&
 295               src[2].equals(&inst->src[2]) &&
 296               saturate == inst->saturate &&
 297               predicated == inst->predicated &&
 298               conditional_mod == inst->conditional_mod &&
 299               mlen == inst->mlen &&
 300               base_mrf == inst->base_mrf &&
 301               sampler == inst->sampler &&
 302               target == inst->target &&
 303               eot == inst->eot &&
 304               header_present == inst->header_present &&
 305               shadow_compare == inst->shadow_compare &&
 306               offset == inst->offset);
 307    }
 308
 309    int regs_written()
 310    {
 311       if (is_tex())
 312          return 4;
 313
 314       /* The SINCOS and INT_DIV_QUOTIENT_AND_REMAINDER math functions return 2,
 315        * but we don't currently use them...nor do we have an opcode for them.
 316        */
 317
 318       return 1;
 319    }
 320
 321    bool is_tex()
 322    {
 323       return (opcode == SHADER_OPCODE_TEX ||
 324               opcode == FS_OPCODE_TXB ||
 325               opcode == SHADER_OPCODE_TXD ||
 326               opcode == SHADER_OPCODE_TXF ||
 327               opcode == SHADER_OPCODE_TXL ||
 328               opcode == SHADER_OPCODE_TXS);
 329    }
 330
 331    bool is_math()
 332    {
 333       return (opcode == SHADER_OPCODE_RCP ||
 334               opcode == SHADER_OPCODE_RSQ ||
 335               opcode == SHADER_OPCODE_SQRT ||
 336               opcode == SHADER_OPCODE_EXP2 ||
 337               opcode == SHADER_OPCODE_LOG2 ||
 338               opcode == SHADER_OPCODE_SIN ||
 339               opcode == SHADER_OPCODE_COS ||
 340               opcode == SHADER_OPCODE_INT_QUOTIENT ||
 341               opcode == SHADER_OPCODE_INT_REMAINDER ||
 342               opcode == SHADER_OPCODE_POW);
 343    }
 344
 345    enum opcode opcode; /* BRW_OPCODE_* or FS_OPCODE_* */
 346    fs_reg dst;
 347    fs_reg src[3];
 348    bool saturate;
 349    bool predicated;
 350    bool predicate_inverse;
 351    int conditional_mod; /**< BRW_CONDITIONAL_* */
 352
 353    int mlen; /**< SEND message length */
 354    int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
 355    int sampler;
 356    int target; /**< MRT target. */
 357    bool eot;
 358    bool header_present;
 359    bool shadow_compare;
 360    bool force_uncompressed;
 361    bool force_sechalf;
 362    uint32_t offset; /* spill/unspill offset */
 363
 364    /** @{
 365     * Annotation for the generated IR.  One of the two can be set.
 366     */
 367    ir_instruction *ir;
 368    const char *annotation;
 369    /** @} */
 370 };
 371
 372 class fs_visitor : public ir_visitor
 373 {
 374 public:
 375
 376    fs_visitor(struct brw_wm_compile *c, struct gl_shader_program *prog,
 377               struct brw_shader *shader)
 378    {
 379       this->c = c;
 380       this->p = &c->func;
 381       this->brw = p->brw;
 382       this->fp = (struct gl_fragment_program *)
 383          prog->_LinkedShaders[MESA_SHADER_FRAGMENT]->Program;
 384       this->prog = prog;
 385       this->intel = &brw->intel;
 386       this->ctx = &intel->ctx;
 387       this->mem_ctx = ralloc_context(NULL);
 388       this->shader = shader;
 389       this->failed = false;
 390       this->variable_ht = hash_table_ctor(0,
 391                                           hash_table_pointer_hash,
 392                                           hash_table_pointer_compare);
 393
 394       /* There's a question that appears to be left open in the spec:
 395        * How do implicit dst conversions interact with the CMP
 396        * instruction or conditional mods?  On gen6, the instruction:
 397        *
 398        * CMP null<d> src0<f> src1<f>
 399        *
 400        * will do src1 - src0 and compare that result as if it was an
 401        * integer.  On gen4, it will do src1 - src0 as float, convert
 402        * the result to int, and compare as int.  In between, it
 403        * appears that it does src1 - src0 and does the compare in the
 404        * execution type so dst type doesn't matter.
 405        */
 406       if (this->intel->gen > 4)
 407          this->reg_null_cmp = reg_null_d;
 408       else
 409          this->reg_null_cmp = reg_null_f;
 410
 411       this->frag_depth = NULL;
 412       memset(this->outputs, 0, sizeof(this->outputs));
 413       this->first_non_payload_grf = 0;
 414       this->max_grf = intel->gen >= 7 ? GEN7_MRF_HACK_START : BRW_MAX_GRF;
 415
 416       this->current_annotation = NULL;
 417       this->base_ir = NULL;
 418
 419       this->virtual_grf_sizes = NULL;
 420       this->virtual_grf_next = 0;
 421       this->virtual_grf_array_size = 0;
 422       this->virtual_grf_def = NULL;
 423       this->virtual_grf_use = NULL;
 424       this->live_intervals_valid = false;
 425
 426       this->kill_emitted = false;
 427       this->force_uncompressed_stack = 0;
 428       this->force_sechalf_stack = 0;
 429    }
 430
 431    ~fs_visitor()
 432    {
 433       ralloc_free(this->mem_ctx);
 434       hash_table_dtor(this->variable_ht);
 435    }
 436
 437    fs_reg *variable_storage(ir_variable *var);
 438    int virtual_grf_alloc(int size);
 439    void import_uniforms(fs_visitor *v);
 440
 441    void visit(ir_variable *ir);
 442    void visit(ir_assignment *ir);
 443    void visit(ir_dereference_variable *ir);
 444    void visit(ir_dereference_record *ir);
 445    void visit(ir_dereference_array *ir);
 446    void visit(ir_expression *ir);
 447    void visit(ir_texture *ir);
 448    void visit(ir_if *ir);
 449    void visit(ir_constant *ir);
 450    void visit(ir_swizzle *ir);
 451    void visit(ir_return *ir);
 452    void visit(ir_loop *ir);
 453    void visit(ir_loop_jump *ir);
 454    void visit(ir_discard *ir);
 455    void visit(ir_call *ir);
 456    void visit(ir_function *ir);
 457    void visit(ir_function_signature *ir);
 458
 459    void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
 460
 461    fs_inst *emit(fs_inst inst);
 462
 463    fs_inst *emit(enum opcode opcode)
 464    {
 465       return emit(fs_inst(opcode));
 466    }
 467
 468    fs_inst *emit(enum opcode opcode, fs_reg dst)
 469    {
 470       return emit(fs_inst(opcode, dst));
 471    }
 472
 473    fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0)
 474    {
 475       return emit(fs_inst(opcode, dst, src0));
 476    }
 477
 478    fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)
 479    {
 480       return emit(fs_inst(opcode, dst, src0, src1));
 481    }
 482
 483    fs_inst *emit(enum opcode opcode, fs_reg dst,
 484                  fs_reg src0, fs_reg src1, fs_reg src2)
 485    {
 486       return emit(fs_inst(opcode, dst, src0, src1, src2));
 487    }
 488
 489    int type_size(const struct glsl_type *type);
 490    fs_inst *get_instruction_generating_reg(fs_inst *start,
 491                                            fs_inst *end,
 492                                            fs_reg reg);
 493
 494    bool run();
 495    void setup_paramvalues_refs();
 496    void assign_curb_setup();
 497    void calculate_urb_setup();
 498    void assign_urb_setup();
 499    bool assign_regs();
 500    void assign_regs_trivial();
 501    int choose_spill_reg(struct ra_graph *g);
 502    void spill_reg(int spill_reg);
 503    void split_virtual_grfs();
 504    void setup_pull_constants();
 505    void calculate_live_intervals();
 506    bool propagate_constants();
 507    bool opt_algebraic();
 508    bool register_coalesce();
 509    bool compute_to_mrf();
 510    bool dead_code_eliminate();
 511    bool remove_dead_constants();
 512    bool remove_duplicate_mrf_writes();
 513    bool virtual_grf_interferes(int a, int b);
 514    void schedule_instructions();
 515    void patch_discard_jumps_to_fb_writes();
 516    void fail(const char *msg, ...);
 517
 518    void push_force_uncompressed();
 519    void pop_force_uncompressed();
 520    void push_force_sechalf();
 521    void pop_force_sechalf();
 522
 523    void generate_code();
 524    void generate_fb_write(fs_inst *inst);
 525    void generate_pixel_xy(struct brw_reg dst, bool is_x);
 526    void generate_linterp(fs_inst *inst, struct brw_reg dst,
 527                          struct brw_reg *src);
 528    void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 529    void generate_math1_gen7(fs_inst *inst,
 530                             struct brw_reg dst,
 531                             struct brw_reg src);
 532    void generate_math2_gen7(fs_inst *inst,
 533                             struct brw_reg dst,
 534                             struct brw_reg src0,
 535                             struct brw_reg src1);
 536    void generate_math1_gen6(fs_inst *inst,
 537                             struct brw_reg dst,
 538                             struct brw_reg src);
 539    void generate_math2_gen6(fs_inst *inst,
 540                             struct brw_reg dst,
 541                             struct brw_reg src0,
 542                             struct brw_reg src1);
 543    void generate_math_gen4(fs_inst *inst,
 544                            struct brw_reg dst,
 545                            struct brw_reg src);
 546    void generate_discard(fs_inst *inst);
 547    void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 548    void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 549    void generate_spill(fs_inst *inst, struct brw_reg src);
 550    void generate_unspill(fs_inst *inst, struct brw_reg dst);
 551    void generate_pull_constant_load(fs_inst *inst, struct brw_reg dst);
 552
 553    void emit_dummy_fs();
 554    fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
 555    fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
 556    fs_reg *emit_general_interpolation(ir_variable *ir);
 557    void emit_interpolation_setup_gen4();
 558    void emit_interpolation_setup_gen6();
 559    fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 560                               int sampler);
 561    fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 562                               int sampler);
 563    fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 564                               int sampler);
 565    fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
 566    fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
 567    bool try_emit_saturate(ir_expression *ir);
 568    bool try_emit_mad(ir_expression *ir, int mul_arg);
 569    void emit_bool_to_cond_code(ir_rvalue *condition);
 570    void emit_if_gen6(ir_if *ir);
 571    void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset);
 572
 573    void emit_color_write(int target, int index, int first_color_mrf);
 574    void emit_fb_writes();
 575    bool try_rewrite_rhs_to_dst(ir_assignment *ir,
 576                                fs_reg dst,
 577                                fs_reg src,
 578                                fs_inst *pre_rhs_inst,
 579                                fs_inst *last_rhs_inst);
 580    void emit_assignment_writes(fs_reg &l, fs_reg &r,
 581                                const glsl_type *type, bool predicated);
 582    void resolve_ud_negate(fs_reg *reg);
 583
 584    struct brw_reg interp_reg(int location, int channel);
 585    int setup_uniform_values(int loc, const glsl_type *type);
 586    void setup_builtin_uniform_values(ir_variable *ir);
 587    int implied_mrf_writes(fs_inst *inst);
 588
 589    struct brw_context *brw;
 590    const struct gl_fragment_program *fp;
 591    struct intel_context *intel;
 592    struct gl_context *ctx;
 593    struct brw_wm_compile *c;
 594    struct brw_compile *p;
 595    struct brw_shader *shader;
 596    struct gl_shader_program *prog;
 597    void *mem_ctx;
 598    exec_list instructions;
 599    exec_list discard_halt_patches;
 600
 601    /* Delayed setup of c->prog_data.params[] due to realloc of
 602     * ParamValues[] during compile.
 603     */
 604    int param_index[MAX_UNIFORMS * 4];
 605    int param_offset[MAX_UNIFORMS * 4];
 606
 607    int *virtual_grf_sizes;
 608    int virtual_grf_next;
 609    int virtual_grf_array_size;
 610    int *virtual_grf_def;
 611    int *virtual_grf_use;
 612    bool live_intervals_valid;
 613
 614    /* This is the map from UNIFORM hw_reg + reg_offset as generated by
 615     * the visitor to the packed uniform number after
 616     * remove_dead_constants() that represents the actual uploaded
 617     * uniform index.
 618     */
 619    int *params_remap;
 620
 621    struct hash_table *variable_ht;
 622    ir_variable *frag_depth;
 623    fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
 624    int first_non_payload_grf;
 625    int max_grf;
 626    int urb_setup[FRAG_ATTRIB_MAX];
 627    bool kill_emitted;
 628
 629    /** @{ debug annotation info */
 630    const char *current_annotation;
 631    ir_instruction *base_ir;
 632    /** @} */
 633
 634    bool failed;
 635    char *fail_msg;
 636
 637    /* Result of last visit() method. */
 638    fs_reg result;
 639
 640    fs_reg pixel_x;
 641    fs_reg pixel_y;
 642    fs_reg wpos_w;
 643    fs_reg pixel_w;
 644    fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
 645    fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
 646    fs_reg reg_null_cmp;
 647
 648    int grf_used;
 649
 650    int force_uncompressed_stack;
 651    int force_sechalf_stack;
 652 };
 653
 654 bool brw_do_channel_expressions(struct exec_list *instructions);
 655 bool brw_do_vector_splitting(struct exec_list *instructions);
 656 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);