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(class fs_visitor *v, const struct glsl_type *type);
 148
 149    /** Register file: ARF, GRF, MRF, IMM. */
 150    enum register_file file;
 151    /** virtual register number.  0 = fixed hw reg */
 152    int reg;
 153    /** Offset within the virtual register. */
 154    int reg_offset;
 155    /** HW register number.  Generally unset until register allocation. */
 156    int hw_reg;
 157    /** Register type.  BRW_REGISTER_TYPE_* */
 158    int type;
 159    bool negate;
 160    bool abs;
 161    struct brw_reg fixed_hw_reg;
 162
 163    /** Value for file == BRW_IMMMEDIATE_FILE */
 164    union {
 165       int32_t i;
 166       uint32_t u;
 167       float f;
 168    } imm;
 169 };
 170
 171 class fs_inst : public exec_node {
 172 public:
 173    /* Callers of this talloc-based new need not call delete. It's
 174     * easier to just talloc_free 'ctx' (or any of its ancestors). */
 175    static void* operator new(size_t size, void *ctx)
 176    {
 177       void *node;
 178
 179       node = talloc_zero_size(ctx, size);
 180       assert(node != NULL);
 181
 182       return node;
 183    }
 184
 185    void init()
 186    {
 187       this->opcode = BRW_OPCODE_NOP;
 188       this->saturate = false;
 189       this->conditional_mod = BRW_CONDITIONAL_NONE;
 190       this->predicated = false;
 191       this->sampler = 0;
 192       this->target = 0;
 193       this->eot = false;
 194       this->header_present = false;
 195       this->shadow_compare = false;
 196       this->mlen = 0;
 197       this->base_mrf = 0;
 198    }
 199
 200    fs_inst()
 201    {
 202       init();
 203    }
 204
 205    fs_inst(int opcode)
 206    {
 207       init();
 208       this->opcode = opcode;
 209    }
 210
 211    fs_inst(int 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(int 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(int 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(int 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    int opcode; /* BRW_OPCODE_* or FS_OPCODE_* */
 270    fs_reg dst;
 271    fs_reg src[3];
 272    bool saturate;
 273    bool predicated;
 274    int conditional_mod; /**< BRW_CONDITIONAL_* */
 275
 276    int mlen; /**< SEND message length */
 277    int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
 278    int sampler;
 279    int target; /**< MRT target. */
 280    bool eot;
 281    bool header_present;
 282    bool shadow_compare;
 283
 284    /** @{
 285     * Annotation for the generated IR.  One of the two can be set.
 286     */
 287    ir_instruction *ir;
 288    const char *annotation;
 289    /** @} */
 290 };
 291
 292 class fs_visitor : public ir_visitor
 293 {
 294 public:
 295
 296    fs_visitor(struct brw_wm_compile *c, struct brw_shader *shader)
 297    {
 298       this->c = c;
 299       this->p = &c->func;
 300       this->brw = p->brw;
 301       this->fp = brw->fragment_program;
 302       this->intel = &brw->intel;
 303       this->ctx = &intel->ctx;
 304       this->mem_ctx = talloc_new(NULL);
 305       this->shader = shader;
 306       this->fail = false;
 307       this->variable_ht = hash_table_ctor(0,
 308                                           hash_table_pointer_hash,
 309                                           hash_table_pointer_compare);
 310
 311       this->frag_color = NULL;
 312       this->frag_data = NULL;
 313       this->frag_depth = NULL;
 314       this->first_non_payload_grf = 0;
 315
 316       this->current_annotation = NULL;
 317       this->annotation_string = NULL;
 318       this->annotation_ir = NULL;
 319       this->base_ir = NULL;
 320
 321       this->virtual_grf_sizes = NULL;
 322       this->virtual_grf_next = 1;
 323       this->virtual_grf_array_size = 0;
 324       this->virtual_grf_def = NULL;
 325       this->virtual_grf_use = NULL;
 326
 327       this->kill_emitted = false;
 328    }
 329
 330    ~fs_visitor()
 331    {
 332       talloc_free(this->mem_ctx);
 333       hash_table_dtor(this->variable_ht);
 334    }
 335
 336    fs_reg *variable_storage(ir_variable *var);
 337    int virtual_grf_alloc(int size);
 338
 339    void visit(ir_variable *ir);
 340    void visit(ir_assignment *ir);
 341    void visit(ir_dereference_variable *ir);
 342    void visit(ir_dereference_record *ir);
 343    void visit(ir_dereference_array *ir);
 344    void visit(ir_expression *ir);
 345    void visit(ir_texture *ir);
 346    void visit(ir_if *ir);
 347    void visit(ir_constant *ir);
 348    void visit(ir_swizzle *ir);
 349    void visit(ir_return *ir);
 350    void visit(ir_loop *ir);
 351    void visit(ir_loop_jump *ir);
 352    void visit(ir_discard *ir);
 353    void visit(ir_call *ir);
 354    void visit(ir_function *ir);
 355    void visit(ir_function_signature *ir);
 356
 357    fs_inst *emit(fs_inst inst);
 358    void assign_curb_setup();
 359    void calculate_urb_setup();
 360    void assign_urb_setup();
 361    void assign_regs();
 362    void assign_regs_trivial();
 363    void split_virtual_grfs();
 364    void calculate_live_intervals();
 365    bool propagate_constants();
 366    bool register_coalesce();
 367    bool compute_to_mrf();
 368    bool dead_code_eliminate();
 369    bool virtual_grf_interferes(int a, int b);
 370    void generate_code();
 371    void generate_fb_write(fs_inst *inst);
 372    void generate_linterp(fs_inst *inst, struct brw_reg dst,
 373                          struct brw_reg *src);
 374    void generate_tex(fs_inst *inst, struct brw_reg dst);
 375    void generate_math(fs_inst *inst, struct brw_reg dst, struct brw_reg *src);
 376    void generate_discard_not(fs_inst *inst, struct brw_reg temp);
 377    void generate_discard_and(fs_inst *inst, struct brw_reg temp);
 378    void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 379    void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 380
 381    void emit_dummy_fs();
 382    fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
 383    fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
 384    fs_reg *emit_general_interpolation(ir_variable *ir);
 385    void emit_interpolation_setup_gen4();
 386    void emit_interpolation_setup_gen6();
 387    fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate);
 388    fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate);
 389    fs_inst *emit_math(fs_opcodes op, fs_reg dst, fs_reg src0);
 390    fs_inst *emit_math(fs_opcodes op, fs_reg dst, fs_reg src0, fs_reg src1);
 391    void emit_bool_to_cond_code(ir_rvalue *condition);
 392
 393    void emit_fb_writes();
 394    void emit_assignment_writes(fs_reg &l, fs_reg &r,
 395                                const glsl_type *type, bool predicated);
 396
 397    struct brw_reg interp_reg(int location, int channel);
 398    int setup_uniform_values(int loc, const glsl_type *type);
 399    void setup_builtin_uniform_values(ir_variable *ir);
 400
 401    struct brw_context *brw;
 402    const struct gl_fragment_program *fp;
 403    struct intel_context *intel;
 404    struct gl_context *ctx;
 405    struct brw_wm_compile *c;
 406    struct brw_compile *p;
 407    struct brw_shader *shader;
 408    void *mem_ctx;
 409    exec_list instructions;
 410
 411    int *virtual_grf_sizes;
 412    int virtual_grf_next;
 413    int virtual_grf_array_size;
 414    int *virtual_grf_def;
 415    int *virtual_grf_use;
 416
 417    struct hash_table *variable_ht;
 418    ir_variable *frag_color, *frag_data, *frag_depth;
 419    int first_non_payload_grf;
 420    int urb_setup[FRAG_ATTRIB_MAX];
 421    bool kill_emitted;
 422
 423    /** @{ debug annotation info */
 424    const char *current_annotation;
 425    ir_instruction *base_ir;
 426    const char **annotation_string;
 427    ir_instruction **annotation_ir;
 428    /** @} */
 429
 430    bool fail;
 431
 432    /* Result of last visit() method. */
 433    fs_reg result;
 434
 435    fs_reg pixel_x;
 436    fs_reg pixel_y;
 437    fs_reg wpos_w;
 438    fs_reg pixel_w;
 439    fs_reg delta_x;
 440    fs_reg delta_y;
 441
 442    int grf_used;
 443 };
 444
 445 GLboolean brw_do_channel_expressions(struct exec_list *instructions);
 446 GLboolean brw_do_vector_splitting(struct exec_list *instructions);