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 extern "C" {
  29
  30 #include <sys/types.h>
  31
  32 #include "main/macros.h"
  33 #include "main/shaderobj.h"
  34 #include "main/uniforms.h"
  35 #include "program/prog_parameter.h"
  36 #include "program/prog_print.h"
  37 #include "program/prog_optimize.h"
  38 #include "program/register_allocate.h"
  39 #include "program/sampler.h"
  40 #include "program/hash_table.h"
  41 #include "brw_context.h"
  42 #include "brw_eu.h"
  43 #include "brw_wm.h"
  44 #include "talloc.h"
  45 }
  46 #include "../glsl/glsl_types.h"
  47 #include "../glsl/ir.h"
  48
  49 enum register_file {
  50    ARF = BRW_ARCHITECTURE_REGISTER_FILE,
  51    GRF = BRW_GENERAL_REGISTER_FILE,
  52    MRF = BRW_MESSAGE_REGISTER_FILE,
  53    IMM = BRW_IMMEDIATE_VALUE,
  54    FIXED_HW_REG, /* a struct brw_reg */
  55    UNIFORM, /* prog_data->params[hw_reg] */
  56    BAD_FILE
  57 };
  58
  59 enum fs_opcodes {
  60    FS_OPCODE_FB_WRITE = 256,
  61    FS_OPCODE_RCP,
  62    FS_OPCODE_RSQ,
  63    FS_OPCODE_SQRT,
  64    FS_OPCODE_EXP2,
  65    FS_OPCODE_LOG2,
  66    FS_OPCODE_POW,
  67    FS_OPCODE_SIN,
  68    FS_OPCODE_COS,
  69    FS_OPCODE_DDX,
  70    FS_OPCODE_DDY,
  71    FS_OPCODE_CINTERP,
  72    FS_OPCODE_LINTERP,
  73    FS_OPCODE_TEX,
  74    FS_OPCODE_TXB,
  75    FS_OPCODE_TXL,
  76    FS_OPCODE_DISCARD_NOT,
  77    FS_OPCODE_DISCARD_AND,
  78    FS_OPCODE_SPILL,
  79    FS_OPCODE_UNSPILL,
  80    FS_OPCODE_PULL_CONSTANT_LOAD,
  81 };
  82
  83
  84 class fs_reg {
  85 public:
  86    /* Callers of this talloc-based new need not call delete. It's
  87     * easier to just talloc_free 'ctx' (or any of its ancestors). */
  88    static void* operator new(size_t size, void *ctx)
  89    {
  90       void *node;
  91
  92       node = talloc_size(ctx, size);
  93       assert(node != NULL);
  94
  95       return node;
  96    }
  97
  98    void init()
  99    {
 100       memset(this, 0, sizeof(*this));
 101       this->hw_reg = -1;
 102       this->smear = -1;
 103    }
 104
 105    /** Generic unset register constructor. */
 106    fs_reg()
 107    {
 108       init();
 109       this->file = BAD_FILE;
 110    }
 111
 112    /** Immediate value constructor. */
 113    fs_reg(float f)
 114    {
 115       init();
 116       this->file = IMM;
 117       this->type = BRW_REGISTER_TYPE_F;
 118       this->imm.f = f;
 119    }
 120
 121    /** Immediate value constructor. */
 122    fs_reg(int32_t i)
 123    {
 124       init();
 125       this->file = IMM;
 126       this->type = BRW_REGISTER_TYPE_D;
 127       this->imm.i = i;
 128    }
 129
 130    /** Immediate value constructor. */
 131    fs_reg(uint32_t u)
 132    {
 133       init();
 134       this->file = IMM;
 135       this->type = BRW_REGISTER_TYPE_UD;
 136       this->imm.u = u;
 137    }
 138
 139    /** Fixed brw_reg Immediate value constructor. */
 140    fs_reg(struct brw_reg fixed_hw_reg)
 141    {
 142       init();
 143       this->file = FIXED_HW_REG;
 144       this->fixed_hw_reg = fixed_hw_reg;
 145       this->type = fixed_hw_reg.type;
 146    }
 147
 148    fs_reg(enum register_file file, int hw_reg);
 149    fs_reg(enum register_file file, int hw_reg, uint32_t type);
 150    fs_reg(class fs_visitor *v, const struct glsl_type *type);
 151
 152    bool equals(fs_reg *r)
 153    {
 154       return (file == r->file &&
 155               reg == r->reg &&
 156               reg_offset == r->reg_offset &&
 157               hw_reg == r->hw_reg &&
 158               type == r->type &&
 159               negate == r->negate &&
 160               abs == r->abs &&
 161               memcmp(&fixed_hw_reg, &r->fixed_hw_reg,
 162                      sizeof(fixed_hw_reg)) == 0 &&
 163               smear == r->smear &&
 164               imm.u == r->imm.u);
 165    }
 166
 167    /** Register file: ARF, GRF, MRF, IMM. */
 168    enum register_file file;
 169    /** virtual register number.  0 = fixed hw reg */
 170    int reg;
 171    /** Offset within the virtual register. */
 172    int reg_offset;
 173    /** HW register number.  Generally unset until register allocation. */
 174    int hw_reg;
 175    /** Register type.  BRW_REGISTER_TYPE_* */
 176    int type;
 177    bool negate;
 178    bool abs;
 179    struct brw_reg fixed_hw_reg;
 180    int smear; /* -1, or a channel of the reg to smear to all channels. */
 181
 182    /** Value for file == BRW_IMMMEDIATE_FILE */
 183    union {
 184       int32_t i;
 185       uint32_t u;
 186       float f;
 187    } imm;
 188 };
 189
 190 static const fs_reg reg_undef;
 191 static const fs_reg reg_null_f(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_F);
 192 static const fs_reg reg_null_d(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_D);
 193
 194 class fs_inst : public exec_node {
 195 public:
 196    /* Callers of this talloc-based new need not call delete. It's
 197     * easier to just talloc_free 'ctx' (or any of its ancestors). */
 198    static void* operator new(size_t size, void *ctx)
 199    {
 200       void *node;
 201
 202       node = talloc_zero_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(int opcode)
 226    {
 227       init();
 228       this->opcode = opcode;
 229    }
 230
 231    fs_inst(int 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(int 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(int 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(int 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    bool is_tex()
 310    {
 311       return (opcode == FS_OPCODE_TEX ||
 312               opcode == FS_OPCODE_TXB ||
 313               opcode == FS_OPCODE_TXL);
 314    }
 315
 316    bool is_math()
 317    {
 318       return (opcode == FS_OPCODE_RCP ||
 319               opcode == FS_OPCODE_RSQ ||
 320               opcode == FS_OPCODE_SQRT ||
 321               opcode == FS_OPCODE_EXP2 ||
 322               opcode == FS_OPCODE_LOG2 ||
 323               opcode == FS_OPCODE_SIN ||
 324               opcode == FS_OPCODE_COS ||
 325               opcode == FS_OPCODE_POW);
 326    }
 327
 328    int opcode; /* BRW_OPCODE_* or FS_OPCODE_* */
 329    fs_reg dst;
 330    fs_reg src[3];
 331    bool saturate;
 332    bool predicated;
 333    int conditional_mod; /**< BRW_CONDITIONAL_* */
 334
 335    int mlen; /**< SEND message length */
 336    int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
 337    int sampler;
 338    int target; /**< MRT target. */
 339    bool eot;
 340    bool header_present;
 341    bool shadow_compare;
 342    uint32_t offset; /* spill/unspill offset */
 343
 344    /** @{
 345     * Annotation for the generated IR.  One of the two can be set.
 346     */
 347    ir_instruction *ir;
 348    const char *annotation;
 349    /** @} */
 350 };
 351
 352 class fs_visitor : public ir_visitor
 353 {
 354 public:
 355
 356    fs_visitor(struct brw_wm_compile *c, struct brw_shader *shader)
 357    {
 358       this->c = c;
 359       this->p = &c->func;
 360       this->brw = p->brw;
 361       this->fp = brw->fragment_program;
 362       this->intel = &brw->intel;
 363       this->ctx = &intel->ctx;
 364       this->mem_ctx = talloc_new(NULL);
 365       this->shader = shader;
 366       this->fail = false;
 367       this->variable_ht = hash_table_ctor(0,
 368                                           hash_table_pointer_hash,
 369                                           hash_table_pointer_compare);
 370
 371       /* There's a question that appears to be left open in the spec:
 372        * How do implicit dst conversions interact with the CMP
 373        * instruction or conditional mods?  On gen6, the instruction:
 374        *
 375        * CMP null<d> src0<f> src1<f>
 376        *
 377        * will do src1 - src0 and compare that result as if it was an
 378        * integer.  On gen4, it will do src1 - src0 as float, convert
 379        * the result to int, and compare as int.  In between, it
 380        * appears that it does src1 - src0 and does the compare in the
 381        * execution type so dst type doesn't matter.
 382        */
 383       if (this->intel->gen > 4)
 384          this->reg_null_cmp = reg_null_d;
 385       else
 386          this->reg_null_cmp = reg_null_f;
 387
 388       this->frag_color = NULL;
 389       this->frag_data = NULL;
 390       this->frag_depth = NULL;
 391       this->first_non_payload_grf = 0;
 392
 393       this->current_annotation = NULL;
 394       this->base_ir = NULL;
 395
 396       this->virtual_grf_sizes = NULL;
 397       this->virtual_grf_next = 1;
 398       this->virtual_grf_array_size = 0;
 399       this->virtual_grf_def = NULL;
 400       this->virtual_grf_use = NULL;
 401       this->live_intervals_valid = false;
 402
 403       this->kill_emitted = false;
 404    }
 405
 406    ~fs_visitor()
 407    {
 408       talloc_free(this->mem_ctx);
 409       hash_table_dtor(this->variable_ht);
 410    }
 411
 412    fs_reg *variable_storage(ir_variable *var);
 413    int virtual_grf_alloc(int size);
 414
 415    void visit(ir_variable *ir);
 416    void visit(ir_assignment *ir);
 417    void visit(ir_dereference_variable *ir);
 418    void visit(ir_dereference_record *ir);
 419    void visit(ir_dereference_array *ir);
 420    void visit(ir_expression *ir);
 421    void visit(ir_texture *ir);
 422    void visit(ir_if *ir);
 423    void visit(ir_constant *ir);
 424    void visit(ir_swizzle *ir);
 425    void visit(ir_return *ir);
 426    void visit(ir_loop *ir);
 427    void visit(ir_loop_jump *ir);
 428    void visit(ir_discard *ir);
 429    void visit(ir_call *ir);
 430    void visit(ir_function *ir);
 431    void visit(ir_function_signature *ir);
 432
 433    fs_inst *emit(fs_inst inst);
 434    void setup_paramvalues_refs();
 435    void assign_curb_setup();
 436    void calculate_urb_setup();
 437    void assign_urb_setup();
 438    bool assign_regs();
 439    void assign_regs_trivial();
 440    int choose_spill_reg(struct ra_graph *g);
 441    void spill_reg(int spill_reg);
 442    void split_virtual_grfs();
 443    void setup_pull_constants();
 444    void calculate_live_intervals();
 445    bool propagate_constants();
 446    bool register_coalesce();
 447    bool compute_to_mrf();
 448    bool dead_code_eliminate();
 449    bool remove_duplicate_mrf_writes();
 450    bool virtual_grf_interferes(int a, int b);
 451    void schedule_instructions();
 452
 453    void generate_code();
 454    void generate_fb_write(fs_inst *inst);
 455    void generate_linterp(fs_inst *inst, struct brw_reg dst,
 456                          struct brw_reg *src);
 457    void generate_tex(fs_inst *inst, struct brw_reg dst);
 458    void generate_math(fs_inst *inst, struct brw_reg dst, struct brw_reg *src);
 459    void generate_discard_not(fs_inst *inst, struct brw_reg temp);
 460    void generate_discard_and(fs_inst *inst, struct brw_reg temp);
 461    void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 462    void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 463    void generate_spill(fs_inst *inst, struct brw_reg src);
 464    void generate_unspill(fs_inst *inst, struct brw_reg dst);
 465    void generate_pull_constant_load(fs_inst *inst, struct brw_reg dst);
 466
 467    void emit_dummy_fs();
 468    fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
 469    fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
 470    fs_reg *emit_general_interpolation(ir_variable *ir);
 471    void emit_interpolation_setup_gen4();
 472    void emit_interpolation_setup_gen6();
 473    fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate);
 474    fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate);
 475    fs_inst *emit_math(fs_opcodes op, fs_reg dst, fs_reg src0);
 476    fs_inst *emit_math(fs_opcodes op, fs_reg dst, fs_reg src0, fs_reg src1);
 477    bool try_emit_saturate(ir_expression *ir);
 478    void emit_bool_to_cond_code(ir_rvalue *condition);
 479    void emit_if_gen6(ir_if *ir);
 480    void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset);
 481
 482    void emit_fb_writes();
 483    void emit_assignment_writes(fs_reg &l, fs_reg &r,
 484                                const glsl_type *type, bool predicated);
 485
 486    struct brw_reg interp_reg(int location, int channel);
 487    int setup_uniform_values(int loc, const glsl_type *type);
 488    void setup_builtin_uniform_values(ir_variable *ir);
 489    int implied_mrf_writes(fs_inst *inst);
 490
 491    struct brw_context *brw;
 492    const struct gl_fragment_program *fp;
 493    struct intel_context *intel;
 494    struct gl_context *ctx;
 495    struct brw_wm_compile *c;
 496    struct brw_compile *p;
 497    struct brw_shader *shader;
 498    void *mem_ctx;
 499    exec_list instructions;
 500
 501    /* Delayed setup of c->prog_data.params[] due to realloc of
 502     * ParamValues[] during compile.
 503     */
 504    int param_index[MAX_UNIFORMS * 4];
 505    int param_offset[MAX_UNIFORMS * 4];
 506
 507    int *virtual_grf_sizes;
 508    int virtual_grf_next;
 509    int virtual_grf_array_size;
 510    int *virtual_grf_def;
 511    int *virtual_grf_use;
 512    bool live_intervals_valid;
 513
 514    struct hash_table *variable_ht;
 515    ir_variable *frag_color, *frag_data, *frag_depth;
 516    int first_non_payload_grf;
 517    int urb_setup[FRAG_ATTRIB_MAX];
 518    bool kill_emitted;
 519
 520    /** @{ debug annotation info */
 521    const char *current_annotation;
 522    ir_instruction *base_ir;
 523    /** @} */
 524
 525    bool fail;
 526
 527    /* Result of last visit() method. */
 528    fs_reg result;
 529
 530    fs_reg pixel_x;
 531    fs_reg pixel_y;
 532    fs_reg wpos_w;
 533    fs_reg pixel_w;
 534    fs_reg delta_x;
 535    fs_reg delta_y;
 536    fs_reg reg_null_cmp;
 537
 538    int grf_used;
 539 };
 540
 541 GLboolean brw_do_channel_expressions(struct exec_list *instructions);
 542 GLboolean brw_do_vector_splitting(struct exec_list *instructions);