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_LINTERP,
  72    FS_OPCODE_TEX,
  73    FS_OPCODE_TXB,
  74    FS_OPCODE_TXL,
  75    FS_OPCODE_DISCARD_NOT,
  76    FS_OPCODE_DISCARD_AND,
  77 };
  78
  79
  80 class fs_reg {
  81 public:
  82    /* Callers of this talloc-based new need not call delete. It's
  83     * easier to just talloc_free 'ctx' (or any of its ancestors). */
  84    static void* operator new(size_t size, void *ctx)
  85    {
  86       void *node;
  87
  88       node = talloc_size(ctx, size);
  89       assert(node != NULL);
  90
  91       return node;
  92    }
  93
  94    void init()
  95    {
  96       this->reg = 0;
  97       this->reg_offset = 0;
  98       this->negate = 0;
  99       this->abs = 0;
 100       this->hw_reg = -1;
 101    }
 102
 103    /** Generic unset register constructor. */
 104    fs_reg()
 105    {
 106       init();
 107       this->file = BAD_FILE;
 108    }
 109
 110    /** Immediate value constructor. */
 111    fs_reg(float f)
 112    {
 113       init();
 114       this->file = IMM;
 115       this->type = BRW_REGISTER_TYPE_F;
 116       this->imm.f = f;
 117    }
 118
 119    /** Immediate value constructor. */
 120    fs_reg(int32_t i)
 121    {
 122       init();
 123       this->file = IMM;
 124       this->type = BRW_REGISTER_TYPE_D;
 125       this->imm.i = i;
 126    }
 127
 128    /** Immediate value constructor. */
 129    fs_reg(uint32_t u)
 130    {
 131       init();
 132       this->file = IMM;
 133       this->type = BRW_REGISTER_TYPE_UD;
 134       this->imm.u = u;
 135    }
 136
 137    /** Fixed brw_reg Immediate value constructor. */
 138    fs_reg(struct brw_reg fixed_hw_reg)
 139    {
 140       init();
 141       this->file = FIXED_HW_REG;
 142       this->fixed_hw_reg = fixed_hw_reg;
 143       this->type = fixed_hw_reg.type;
 144    }
 145
 146    fs_reg(enum register_file file, int hw_reg);
 147    fs_reg(enum register_file file, int hw_reg, uint32_t type);
 148    fs_reg(class fs_visitor *v, const struct glsl_type *type);
 149
 150    /** Register file: ARF, GRF, MRF, IMM. */
 151    enum register_file file;
 152    /** virtual register number.  0 = fixed hw reg */
 153    int reg;
 154    /** Offset within the virtual register. */
 155    int reg_offset;
 156    /** HW register number.  Generally unset until register allocation. */
 157    int hw_reg;
 158    /** Register type.  BRW_REGISTER_TYPE_* */
 159    int type;
 160    bool negate;
 161    bool abs;
 162    struct brw_reg fixed_hw_reg;
 163
 164    /** Value for file == BRW_IMMMEDIATE_FILE */
 165    union {
 166       int32_t i;
 167       uint32_t u;
 168       float f;
 169    } imm;
 170 };
 171
 172 class fs_inst : public exec_node {
 173 public:
 174    /* Callers of this talloc-based new need not call delete. It's
 175     * easier to just talloc_free 'ctx' (or any of its ancestors). */
 176    static void* operator new(size_t size, void *ctx)
 177    {
 178       void *node;
 179
 180       node = talloc_zero_size(ctx, size);
 181       assert(node != NULL);
 182
 183       return node;
 184    }
 185
 186    void init()
 187    {
 188       this->opcode = BRW_OPCODE_NOP;
 189       this->saturate = false;
 190       this->conditional_mod = BRW_CONDITIONAL_NONE;
 191       this->predicated = false;
 192       this->sampler = 0;
 193       this->target = 0;
 194       this->eot = false;
 195       this->header_present = false;
 196       this->shadow_compare = false;
 197       this->mlen = 0;
 198       this->base_mrf = 0;
 199    }
 200
 201    fs_inst()
 202    {
 203       init();
 204    }
 205
 206    fs_inst(int opcode)
 207    {
 208       init();
 209       this->opcode = opcode;
 210    }
 211
 212    fs_inst(int opcode, fs_reg dst)
 213    {
 214       init();
 215       this->opcode = opcode;
 216       this->dst = dst;
 217
 218       if (dst.file == GRF)
 219          assert(dst.reg_offset >= 0);
 220    }
 221
 222    fs_inst(int opcode, fs_reg dst, fs_reg src0)
 223    {
 224       init();
 225       this->opcode = opcode;
 226       this->dst = dst;
 227       this->src[0] = src0;
 228
 229       if (dst.file == GRF)
 230          assert(dst.reg_offset >= 0);
 231       if (src[0].file == GRF)
 232          assert(src[0].reg_offset >= 0);
 233    }
 234
 235    fs_inst(int opcode, fs_reg dst, fs_reg src0, fs_reg src1)
 236    {
 237       init();
 238       this->opcode = opcode;
 239       this->dst = dst;
 240       this->src[0] = src0;
 241       this->src[1] = src1;
 242
 243       if (dst.file == GRF)
 244          assert(dst.reg_offset >= 0);
 245       if (src[0].file == GRF)
 246          assert(src[0].reg_offset >= 0);
 247       if (src[1].file == GRF)
 248          assert(src[1].reg_offset >= 0);
 249    }
 250
 251    fs_inst(int opcode, fs_reg dst, fs_reg src0, fs_reg src1, fs_reg src2)
 252    {
 253       init();
 254       this->opcode = opcode;
 255       this->dst = dst;
 256       this->src[0] = src0;
 257       this->src[1] = src1;
 258       this->src[2] = src2;
 259
 260       if (dst.file == GRF)
 261          assert(dst.reg_offset >= 0);
 262       if (src[0].file == GRF)
 263          assert(src[0].reg_offset >= 0);
 264       if (src[1].file == GRF)
 265          assert(src[1].reg_offset >= 0);
 266       if (src[2].file == GRF)
 267          assert(src[2].reg_offset >= 0);
 268    }
 269
 270    int opcode; /* BRW_OPCODE_* or FS_OPCODE_* */
 271    fs_reg dst;
 272    fs_reg src[3];
 273    bool saturate;
 274    bool predicated;
 275    int conditional_mod; /**< BRW_CONDITIONAL_* */
 276
 277    int mlen; /**< SEND message length */
 278    int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
 279    int sampler;
 280    int target; /**< MRT target. */
 281    bool eot;
 282    bool header_present;
 283    bool shadow_compare;
 284
 285    /** @{
 286     * Annotation for the generated IR.  One of the two can be set.
 287     */
 288    ir_instruction *ir;
 289    const char *annotation;
 290    /** @} */
 291 };
 292
 293 class fs_visitor : public ir_visitor
 294 {
 295 public:
 296
 297    fs_visitor(struct brw_wm_compile *c, struct brw_shader *shader)
 298    {
 299       this->c = c;
 300       this->p = &c->func;
 301       this->brw = p->brw;
 302       this->fp = brw->fragment_program;
 303       this->intel = &brw->intel;
 304       this->ctx = &intel->ctx;
 305       this->mem_ctx = talloc_new(NULL);
 306       this->shader = shader;
 307       this->fail = false;
 308       this->variable_ht = hash_table_ctor(0,
 309                                           hash_table_pointer_hash,
 310                                           hash_table_pointer_compare);
 311
 312       this->frag_color = NULL;
 313       this->frag_data = NULL;
 314       this->frag_depth = NULL;
 315       this->first_non_payload_grf = 0;
 316
 317       this->current_annotation = NULL;
 318       this->annotation_string = NULL;
 319       this->annotation_ir = NULL;
 320       this->base_ir = NULL;
 321
 322       this->virtual_grf_sizes = NULL;
 323       this->virtual_grf_next = 1;
 324       this->virtual_grf_array_size = 0;
 325       this->virtual_grf_def = NULL;
 326       this->virtual_grf_use = NULL;
 327
 328       this->kill_emitted = false;
 329    }
 330
 331    ~fs_visitor()
 332    {
 333       talloc_free(this->mem_ctx);
 334       hash_table_dtor(this->variable_ht);
 335    }
 336
 337    fs_reg *variable_storage(ir_variable *var);
 338    int virtual_grf_alloc(int size);
 339
 340    void visit(ir_variable *ir);
 341    void visit(ir_assignment *ir);
 342    void visit(ir_dereference_variable *ir);
 343    void visit(ir_dereference_record *ir);
 344    void visit(ir_dereference_array *ir);
 345    void visit(ir_expression *ir);
 346    void visit(ir_texture *ir);
 347    void visit(ir_if *ir);
 348    void visit(ir_constant *ir);
 349    void visit(ir_swizzle *ir);
 350    void visit(ir_return *ir);
 351    void visit(ir_loop *ir);
 352    void visit(ir_loop_jump *ir);
 353    void visit(ir_discard *ir);
 354    void visit(ir_call *ir);
 355    void visit(ir_function *ir);
 356    void visit(ir_function_signature *ir);
 357
 358    fs_inst *emit(fs_inst inst);
 359    void assign_curb_setup();
 360    void calculate_urb_setup();
 361    void assign_urb_setup();
 362    void assign_regs();
 363    void assign_regs_trivial();
 364    void split_virtual_grfs();
 365    void calculate_live_intervals();
 366    bool propagate_constants();
 367    bool register_coalesce();
 368    bool compute_to_mrf();
 369    bool dead_code_eliminate();
 370    bool virtual_grf_interferes(int a, int b);
 371    void generate_code();
 372    void generate_fb_write(fs_inst *inst);
 373    void generate_linterp(fs_inst *inst, struct brw_reg dst,
 374                          struct brw_reg *src);
 375    void generate_tex(fs_inst *inst, struct brw_reg dst);
 376    void generate_math(fs_inst *inst, struct brw_reg dst, struct brw_reg *src);
 377    void generate_discard_not(fs_inst *inst, struct brw_reg temp);
 378    void generate_discard_and(fs_inst *inst, struct brw_reg temp);
 379    void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 380    void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 381
 382    void emit_dummy_fs();
 383    fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
 384    fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
 385    fs_reg *emit_general_interpolation(ir_variable *ir);
 386    void emit_interpolation_setup_gen4();
 387    void emit_interpolation_setup_gen6();
 388    fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate);
 389    fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate);
 390    fs_inst *emit_math(fs_opcodes op, fs_reg dst, fs_reg src0);
 391    fs_inst *emit_math(fs_opcodes op, fs_reg dst, fs_reg src0, fs_reg src1);
 392    void emit_bool_to_cond_code(ir_rvalue *condition);
 393    void emit_if_gen6(ir_if *ir);
 394
 395    void emit_fb_writes();
 396    void emit_assignment_writes(fs_reg &l, fs_reg &r,
 397                                const glsl_type *type, bool predicated);
 398
 399    struct brw_reg interp_reg(int location, int channel);
 400    int setup_uniform_values(int loc, const glsl_type *type);
 401    void setup_builtin_uniform_values(ir_variable *ir);
 402
 403    struct brw_context *brw;
 404    const struct gl_fragment_program *fp;
 405    struct intel_context *intel;
 406    struct gl_context *ctx;
 407    struct brw_wm_compile *c;
 408    struct brw_compile *p;
 409    struct brw_shader *shader;
 410    void *mem_ctx;
 411    exec_list instructions;
 412
 413    int *virtual_grf_sizes;
 414    int virtual_grf_next;
 415    int virtual_grf_array_size;
 416    int *virtual_grf_def;
 417    int *virtual_grf_use;
 418
 419    struct hash_table *variable_ht;
 420    ir_variable *frag_color, *frag_data, *frag_depth;
 421    int first_non_payload_grf;
 422    int urb_setup[FRAG_ATTRIB_MAX];
 423    bool kill_emitted;
 424
 425    /** @{ debug annotation info */
 426    const char *current_annotation;
 427    ir_instruction *base_ir;
 428    const char **annotation_string;
 429    ir_instruction **annotation_ir;
 430    /** @} */
 431
 432    bool fail;
 433
 434    /* Result of last visit() method. */
 435    fs_reg result;
 436
 437    fs_reg pixel_x;
 438    fs_reg pixel_y;
 439    fs_reg wpos_w;
 440    fs_reg pixel_w;
 441    fs_reg delta_x;
 442    fs_reg delta_y;
 443
 444    int grf_used;
 445 };
 446
 447 GLboolean brw_do_channel_expressions(struct exec_list *instructions);
 448 GLboolean brw_do_vector_splitting(struct exec_list *instructions);