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    ARF = BRW_ARCHITECTURE_REGISTER_FILE,
  52    GRF = BRW_GENERAL_REGISTER_FILE,
  53    MRF = BRW_MESSAGE_REGISTER_FILE,
  54    IMM = BRW_IMMEDIATE_VALUE,
  55    FIXED_HW_REG, /* a struct brw_reg */
  56    UNIFORM, /* prog_data->params[reg] */
  57    BAD_FILE
  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 == BRW_IMMMEDIATE_FILE */
 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 fs_inst : public exec_node {
 175 public:
 176    /* Callers of this ralloc-based new need not call delete. It's
 177     * easier to just ralloc_free 'ctx' (or any of its ancestors). */
 178    static void* operator new(size_t size, void *ctx)
 179    {
 180       void *node;
 181
 182       node = rzalloc_size(ctx, size);
 183       assert(node != NULL);
 184
 185       return node;
 186    }
 187
 188    void init()
 189    {
 190       memset(this, 0, sizeof(*this));
 191       this->opcode = BRW_OPCODE_NOP;
 192       this->conditional_mod = BRW_CONDITIONAL_NONE;
 193
 194       this->dst = reg_undef;
 195       this->src[0] = reg_undef;
 196       this->src[1] = reg_undef;
 197       this->src[2] = reg_undef;
 198    }
 199
 200    fs_inst()
 201    {
 202       init();
 203    }
 204
 205    fs_inst(enum opcode opcode)
 206    {
 207       init();
 208       this->opcode = opcode;
 209    }
 210
 211    fs_inst(enum opcode opcode, fs_reg dst)
 212    {
 213       init();
 214       this->opcode = opcode;
 215       this->dst = dst;
 216
 217       if (dst.file == GRF)
 218          assert(dst.reg_offset >= 0);
 219    }
 220
 221    fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0)
 222    {
 223       init();
 224       this->opcode = opcode;
 225       this->dst = dst;
 226       this->src[0] = src0;
 227
 228       if (dst.file == GRF)
 229          assert(dst.reg_offset >= 0);
 230       if (src[0].file == GRF)
 231          assert(src[0].reg_offset >= 0);
 232    }
 233
 234    fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)
 235    {
 236       init();
 237       this->opcode = opcode;
 238       this->dst = dst;
 239       this->src[0] = src0;
 240       this->src[1] = src1;
 241
 242       if (dst.file == GRF)
 243          assert(dst.reg_offset >= 0);
 244       if (src[0].file == GRF)
 245          assert(src[0].reg_offset >= 0);
 246       if (src[1].file == GRF)
 247          assert(src[1].reg_offset >= 0);
 248    }
 249
 250    fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1, fs_reg src2)
 251    {
 252       init();
 253       this->opcode = opcode;
 254       this->dst = dst;
 255       this->src[0] = src0;
 256       this->src[1] = src1;
 257       this->src[2] = src2;
 258
 259       if (dst.file == GRF)
 260          assert(dst.reg_offset >= 0);
 261       if (src[0].file == GRF)
 262          assert(src[0].reg_offset >= 0);
 263       if (src[1].file == GRF)
 264          assert(src[1].reg_offset >= 0);
 265       if (src[2].file == GRF)
 266          assert(src[2].reg_offset >= 0);
 267    }
 268
 269    bool equals(fs_inst *inst)
 270    {
 271       return (opcode == inst->opcode &&
 272               dst.equals(&inst->dst) &&
 273               src[0].equals(&inst->src[0]) &&
 274               src[1].equals(&inst->src[1]) &&
 275               src[2].equals(&inst->src[2]) &&
 276               saturate == inst->saturate &&
 277               predicated == inst->predicated &&
 278               conditional_mod == inst->conditional_mod &&
 279               mlen == inst->mlen &&
 280               base_mrf == inst->base_mrf &&
 281               sampler == inst->sampler &&
 282               target == inst->target &&
 283               eot == inst->eot &&
 284               header_present == inst->header_present &&
 285               shadow_compare == inst->shadow_compare &&
 286               offset == inst->offset);
 287    }
 288
 289    bool is_tex()
 290    {
 291       return (opcode == FS_OPCODE_TEX ||
 292               opcode == FS_OPCODE_TXB ||
 293               opcode == FS_OPCODE_TXD ||
 294               opcode == FS_OPCODE_TXF ||
 295               opcode == FS_OPCODE_TXL ||
 296               opcode == FS_OPCODE_TXS);
 297    }
 298
 299    bool is_math()
 300    {
 301       return (opcode == SHADER_OPCODE_RCP ||
 302               opcode == SHADER_OPCODE_RSQ ||
 303               opcode == SHADER_OPCODE_SQRT ||
 304               opcode == SHADER_OPCODE_EXP2 ||
 305               opcode == SHADER_OPCODE_LOG2 ||
 306               opcode == SHADER_OPCODE_SIN ||
 307               opcode == SHADER_OPCODE_COS ||
 308               opcode == SHADER_OPCODE_INT_QUOTIENT ||
 309               opcode == SHADER_OPCODE_INT_REMAINDER ||
 310               opcode == SHADER_OPCODE_POW);
 311    }
 312
 313    enum opcode opcode; /* BRW_OPCODE_* or FS_OPCODE_* */
 314    fs_reg dst;
 315    fs_reg src[3];
 316    bool saturate;
 317    bool predicated;
 318    bool predicate_inverse;
 319    int conditional_mod; /**< BRW_CONDITIONAL_* */
 320
 321    int mlen; /**< SEND message length */
 322    int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
 323    int sampler;
 324    int target; /**< MRT target. */
 325    bool eot;
 326    bool header_present;
 327    bool shadow_compare;
 328    bool force_uncompressed;
 329    bool force_sechalf;
 330    uint32_t offset; /* spill/unspill offset */
 331
 332    /** @{
 333     * Annotation for the generated IR.  One of the two can be set.
 334     */
 335    ir_instruction *ir;
 336    const char *annotation;
 337    /** @} */
 338 };
 339
 340 class fs_visitor : public ir_visitor
 341 {
 342 public:
 343
 344    fs_visitor(struct brw_wm_compile *c, struct gl_shader_program *prog,
 345               struct brw_shader *shader)
 346    {
 347       this->c = c;
 348       this->p = &c->func;
 349       this->brw = p->brw;
 350       this->fp = prog->FragmentProgram;
 351       this->prog = prog;
 352       this->intel = &brw->intel;
 353       this->ctx = &intel->ctx;
 354       this->mem_ctx = ralloc_context(NULL);
 355       this->shader = shader;
 356       this->failed = false;
 357       this->variable_ht = hash_table_ctor(0,
 358                                           hash_table_pointer_hash,
 359                                           hash_table_pointer_compare);
 360
 361       /* There's a question that appears to be left open in the spec:
 362        * How do implicit dst conversions interact with the CMP
 363        * instruction or conditional mods?  On gen6, the instruction:
 364        *
 365        * CMP null<d> src0<f> src1<f>
 366        *
 367        * will do src1 - src0 and compare that result as if it was an
 368        * integer.  On gen4, it will do src1 - src0 as float, convert
 369        * the result to int, and compare as int.  In between, it
 370        * appears that it does src1 - src0 and does the compare in the
 371        * execution type so dst type doesn't matter.
 372        */
 373       if (this->intel->gen > 4)
 374          this->reg_null_cmp = reg_null_d;
 375       else
 376          this->reg_null_cmp = reg_null_f;
 377
 378       this->frag_color = NULL;
 379       this->frag_data = NULL;
 380       this->frag_depth = NULL;
 381       this->first_non_payload_grf = 0;
 382
 383       this->current_annotation = NULL;
 384       this->base_ir = NULL;
 385
 386       this->virtual_grf_sizes = NULL;
 387       this->virtual_grf_next = 0;
 388       this->virtual_grf_array_size = 0;
 389       this->virtual_grf_def = NULL;
 390       this->virtual_grf_use = NULL;
 391       this->live_intervals_valid = false;
 392
 393       this->kill_emitted = false;
 394       this->force_uncompressed_stack = 0;
 395       this->force_sechalf_stack = 0;
 396    }
 397
 398    ~fs_visitor()
 399    {
 400       ralloc_free(this->mem_ctx);
 401       hash_table_dtor(this->variable_ht);
 402    }
 403
 404    fs_reg *variable_storage(ir_variable *var);
 405    int virtual_grf_alloc(int size);
 406    void import_uniforms(fs_visitor *v);
 407
 408    void visit(ir_variable *ir);
 409    void visit(ir_assignment *ir);
 410    void visit(ir_dereference_variable *ir);
 411    void visit(ir_dereference_record *ir);
 412    void visit(ir_dereference_array *ir);
 413    void visit(ir_expression *ir);
 414    void visit(ir_texture *ir);
 415    void visit(ir_if *ir);
 416    void visit(ir_constant *ir);
 417    void visit(ir_swizzle *ir);
 418    void visit(ir_return *ir);
 419    void visit(ir_loop *ir);
 420    void visit(ir_loop_jump *ir);
 421    void visit(ir_discard *ir);
 422    void visit(ir_call *ir);
 423    void visit(ir_function *ir);
 424    void visit(ir_function_signature *ir);
 425
 426    void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
 427
 428    fs_inst *emit(fs_inst inst);
 429
 430    fs_inst *emit(enum opcode opcode)
 431    {
 432       return emit(fs_inst(opcode));
 433    }
 434
 435    fs_inst *emit(enum opcode opcode, fs_reg dst)
 436    {
 437       return emit(fs_inst(opcode, dst));
 438    }
 439
 440    fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0)
 441    {
 442       return emit(fs_inst(opcode, dst, src0));
 443    }
 444
 445    fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)
 446    {
 447       return emit(fs_inst(opcode, dst, src0, src1));
 448    }
 449
 450    fs_inst *emit(enum opcode opcode, fs_reg dst,
 451                  fs_reg src0, fs_reg src1, fs_reg src2)
 452    {
 453       return emit(fs_inst(opcode, dst, src0, src1, src2));
 454    }
 455
 456    int type_size(const struct glsl_type *type);
 457
 458    bool run();
 459    void setup_paramvalues_refs();
 460    void assign_curb_setup();
 461    void calculate_urb_setup();
 462    void assign_urb_setup();
 463    bool assign_regs();
 464    void assign_regs_trivial();
 465    int choose_spill_reg(struct ra_graph *g);
 466    void spill_reg(int spill_reg);
 467    void split_virtual_grfs();
 468    void setup_pull_constants();
 469    void calculate_live_intervals();
 470    bool propagate_constants();
 471    bool opt_algebraic();
 472    bool register_coalesce();
 473    bool compute_to_mrf();
 474    bool dead_code_eliminate();
 475    bool remove_dead_constants();
 476    bool remove_duplicate_mrf_writes();
 477    bool virtual_grf_interferes(int a, int b);
 478    void schedule_instructions();
 479    void fail(const char *msg, ...);
 480
 481    void push_force_uncompressed();
 482    void pop_force_uncompressed();
 483    void push_force_sechalf();
 484    void pop_force_sechalf();
 485
 486    void generate_code();
 487    void generate_fb_write(fs_inst *inst);
 488    void generate_pixel_xy(struct brw_reg dst, bool is_x);
 489    void generate_linterp(fs_inst *inst, struct brw_reg dst,
 490                          struct brw_reg *src);
 491    void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 492    void generate_math1_gen6(fs_inst *inst,
 493                             struct brw_reg dst,
 494                             struct brw_reg src);
 495    void generate_math2_gen6(fs_inst *inst,
 496                             struct brw_reg dst,
 497                             struct brw_reg src0,
 498                             struct brw_reg src1);
 499    void generate_math_gen4(fs_inst *inst,
 500                            struct brw_reg dst,
 501                            struct brw_reg src);
 502    void generate_discard(fs_inst *inst);
 503    void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 504    void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 505    void generate_spill(fs_inst *inst, struct brw_reg src);
 506    void generate_unspill(fs_inst *inst, struct brw_reg dst);
 507    void generate_pull_constant_load(fs_inst *inst, struct brw_reg dst);
 508
 509    void emit_dummy_fs();
 510    fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
 511    fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
 512    fs_reg *emit_general_interpolation(ir_variable *ir);
 513    void emit_interpolation_setup_gen4();
 514    void emit_interpolation_setup_gen6();
 515    fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 516                               int sampler);
 517    fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 518                               int sampler);
 519    fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 520                               int sampler);
 521    fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
 522    fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
 523    bool try_emit_saturate(ir_expression *ir);
 524    void emit_bool_to_cond_code(ir_rvalue *condition);
 525    void emit_if_gen6(ir_if *ir);
 526    void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset);
 527
 528    void emit_color_write(int index, int first_color_mrf, fs_reg color);
 529    void emit_fb_writes();
 530    bool try_rewrite_rhs_to_dst(ir_assignment *ir,
 531                                fs_reg dst,
 532                                fs_reg src,
 533                                fs_inst *pre_rhs_inst,
 534                                fs_inst *last_rhs_inst);
 535    void emit_assignment_writes(fs_reg &l, fs_reg &r,
 536                                const glsl_type *type, bool predicated);
 537
 538    struct brw_reg interp_reg(int location, int channel);
 539    int setup_uniform_values(int loc, const glsl_type *type);
 540    void setup_builtin_uniform_values(ir_variable *ir);
 541    int implied_mrf_writes(fs_inst *inst);
 542
 543    struct brw_context *brw;
 544    const struct gl_fragment_program *fp;
 545    struct intel_context *intel;
 546    struct gl_context *ctx;
 547    struct brw_wm_compile *c;
 548    struct brw_compile *p;
 549    struct brw_shader *shader;
 550    struct gl_shader_program *prog;
 551    void *mem_ctx;
 552    exec_list instructions;
 553
 554    /* Delayed setup of c->prog_data.params[] due to realloc of
 555     * ParamValues[] during compile.
 556     */
 557    int param_index[MAX_UNIFORMS * 4];
 558    int param_offset[MAX_UNIFORMS * 4];
 559
 560    int *virtual_grf_sizes;
 561    int virtual_grf_next;
 562    int virtual_grf_array_size;
 563    int *virtual_grf_def;
 564    int *virtual_grf_use;
 565    bool live_intervals_valid;
 566
 567    /* This is the map from UNIFORM hw_reg + reg_offset as generated by
 568     * the visitor to the packed uniform number after
 569     * remove_dead_constants() that represents the actual uploaded
 570     * uniform index.
 571     */
 572    int *params_remap;
 573
 574    struct hash_table *variable_ht;
 575    ir_variable *frag_color, *frag_data, *frag_depth;
 576    int first_non_payload_grf;
 577    int urb_setup[FRAG_ATTRIB_MAX];
 578    bool kill_emitted;
 579
 580    /** @{ debug annotation info */
 581    const char *current_annotation;
 582    ir_instruction *base_ir;
 583    /** @} */
 584
 585    bool failed;
 586    char *fail_msg;
 587
 588    /* Result of last visit() method. */
 589    fs_reg result;
 590
 591    fs_reg pixel_x;
 592    fs_reg pixel_y;
 593    fs_reg wpos_w;
 594    fs_reg pixel_w;
 595    fs_reg delta_x;
 596    fs_reg delta_y;
 597    fs_reg reg_null_cmp;
 598
 599    int grf_used;
 600
 601    int force_uncompressed_stack;
 602    int force_sechalf_stack;
 603 };
 604
 605 GLboolean brw_do_channel_expressions(struct exec_list *instructions);
 606 GLboolean brw_do_vector_splitting(struct exec_list *instructions);
 607 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);