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 #pragma once
  29
  30 #include "brw_shader.h"
  31
  32 extern "C" {
  33
  34 #include <sys/types.h>
  35
  36 #include "main/macros.h"
  37 #include "main/shaderobj.h"
  38 #include "main/uniforms.h"
  39 #include "program/prog_parameter.h"
  40 #include "program/prog_print.h"
  41 #include "program/prog_optimize.h"
  42 #include "program/register_allocate.h"
  43 #include "program/sampler.h"
  44 #include "program/hash_table.h"
  45 #include "brw_context.h"
  46 #include "brw_eu.h"
  47 #include "brw_wm.h"
  48 }
  49 #include "glsl/glsl_types.h"
  50 #include "glsl/ir.h"
  51
  52 class fs_bblock;
  53 namespace {
  54    class acp_entry;
  55 }
  56
  57 enum register_file {
  58    BAD_FILE,
  59    ARF,
  60    GRF,
  61    MRF,
  62    IMM,
  63    FIXED_HW_REG, /* a struct brw_reg */
  64    UNIFORM, /* prog_data->params[reg] */
  65 };
  66
  67 class fs_reg {
  68 public:
  69    /* Callers of this ralloc-based new need not call delete. It's
  70     * easier to just ralloc_free 'ctx' (or any of its ancestors). */
  71    static void* operator new(size_t size, void *ctx)
  72    {
  73       void *node;
  74
  75       node = ralloc_size(ctx, size);
  76       assert(node != NULL);
  77
  78       return node;
  79    }
  80
  81    void init();
  82
  83    fs_reg();
  84    fs_reg(float f);
  85    fs_reg(int32_t i);
  86    fs_reg(uint32_t u);
  87    fs_reg(struct brw_reg fixed_hw_reg);
  88    fs_reg(enum register_file file, int reg);
  89    fs_reg(enum register_file file, int reg, uint32_t type);
  90    fs_reg(class fs_visitor *v, const struct glsl_type *type);
  91
  92    bool equals(const fs_reg &r) const;
  93
  94    /** Register file: ARF, GRF, MRF, IMM. */
  95    enum register_file file;
  96    /**
  97     * Register number.  For ARF/MRF, it's the hardware register.  For
  98     * GRF, it's a virtual register number until register allocation
  99     */
 100    int reg;
 101    /**
 102     * For virtual registers, this is a hardware register offset from
 103     * the start of the register block (for example, a constant index
 104     * in an array access).
 105     */
 106    int reg_offset;
 107    /** Register type.  BRW_REGISTER_TYPE_* */
 108    int type;
 109    bool negate;
 110    bool abs;
 111    bool sechalf;
 112    struct brw_reg fixed_hw_reg;
 113    int smear; /* -1, or a channel of the reg to smear to all channels. */
 114
 115    /** Value for file == IMM */
 116    union {
 117       int32_t i;
 118       uint32_t u;
 119       float f;
 120    } imm;
 121 };
 122
 123 static const fs_reg reg_undef;
 124 static const fs_reg reg_null_f(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_F);
 125 static const fs_reg reg_null_d(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_D);
 126
 127 class fs_inst : public exec_node {
 128 public:
 129    /* Callers of this ralloc-based new need not call delete. It's
 130     * easier to just ralloc_free 'ctx' (or any of its ancestors). */
 131    static void* operator new(size_t size, void *ctx)
 132    {
 133       void *node;
 134
 135       node = rzalloc_size(ctx, size);
 136       assert(node != NULL);
 137
 138       return node;
 139    }
 140
 141    void init();
 142
 143    fs_inst();
 144    fs_inst(enum opcode opcode);
 145    fs_inst(enum opcode opcode, fs_reg dst);
 146    fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0);
 147    fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
 148    fs_inst(enum opcode opcode, fs_reg dst,
 149            fs_reg src0, fs_reg src1,fs_reg src2);
 150
 151    bool equals(fs_inst *inst);
 152    int regs_written();
 153    bool overwrites_reg(const fs_reg &reg);
 154    bool is_tex();
 155    bool is_math();
 156
 157    enum opcode opcode; /* BRW_OPCODE_* or FS_OPCODE_* */
 158    fs_reg dst;
 159    fs_reg src[3];
 160    bool saturate;
 161    bool predicated;
 162    bool predicate_inverse;
 163    int conditional_mod; /**< BRW_CONDITIONAL_* */
 164
 165    int mlen; /**< SEND message length */
 166    int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
 167    int sampler;
 168    int target; /**< MRT target. */
 169    bool eot;
 170    bool header_present;
 171    bool shadow_compare;
 172    bool force_uncompressed;
 173    bool force_sechalf;
 174    uint32_t offset; /* spill/unspill offset */
 175
 176    /** @{
 177     * Annotation for the generated IR.  One of the two can be set.
 178     */
 179    ir_instruction *ir;
 180    const char *annotation;
 181    /** @} */
 182 };
 183
 184 class fs_visitor : public ir_visitor
 185 {
 186 public:
 187
 188    fs_visitor(struct brw_wm_compile *c, struct gl_shader_program *prog,
 189               struct brw_shader *shader);
 190    ~fs_visitor();
 191
 192    fs_reg *variable_storage(ir_variable *var);
 193    int virtual_grf_alloc(int size);
 194    void import_uniforms(fs_visitor *v);
 195
 196    void visit(ir_variable *ir);
 197    void visit(ir_assignment *ir);
 198    void visit(ir_dereference_variable *ir);
 199    void visit(ir_dereference_record *ir);
 200    void visit(ir_dereference_array *ir);
 201    void visit(ir_expression *ir);
 202    void visit(ir_texture *ir);
 203    void visit(ir_if *ir);
 204    void visit(ir_constant *ir);
 205    void visit(ir_swizzle *ir);
 206    void visit(ir_return *ir);
 207    void visit(ir_loop *ir);
 208    void visit(ir_loop_jump *ir);
 209    void visit(ir_discard *ir);
 210    void visit(ir_call *ir);
 211    void visit(ir_function *ir);
 212    void visit(ir_function_signature *ir);
 213
 214    void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
 215
 216    fs_inst *emit(fs_inst inst);
 217
 218    fs_inst *emit(enum opcode opcode);
 219    fs_inst *emit(enum opcode opcode, fs_reg dst);
 220    fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0);
 221    fs_inst *emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1);
 222    fs_inst *emit(enum opcode opcode, fs_reg dst,
 223                  fs_reg src0, fs_reg src1, fs_reg src2);
 224
 225    int type_size(const struct glsl_type *type);
 226    fs_inst *get_instruction_generating_reg(fs_inst *start,
 227                                            fs_inst *end,
 228                                            fs_reg reg);
 229
 230    bool run();
 231    void setup_paramvalues_refs();
 232    void assign_curb_setup();
 233    void calculate_urb_setup();
 234    void assign_urb_setup();
 235    bool assign_regs();
 236    void assign_regs_trivial();
 237    int choose_spill_reg(struct ra_graph *g);
 238    void spill_reg(int spill_reg);
 239    void split_virtual_grfs();
 240    void setup_pull_constants();
 241    void calculate_live_intervals();
 242    bool propagate_constants();
 243    bool opt_algebraic();
 244    bool opt_cse();
 245    bool opt_cse_local(fs_bblock *block, exec_list *aeb);
 246    bool opt_copy_propagate();
 247    bool try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry);
 248    bool opt_copy_propagate_local(void *mem_ctx, fs_bblock *block,
 249                                  exec_list *acp);
 250    bool register_coalesce();
 251    bool register_coalesce_2();
 252    bool compute_to_mrf();
 253    bool dead_code_eliminate();
 254    bool remove_dead_constants();
 255    bool remove_duplicate_mrf_writes();
 256    bool virtual_grf_interferes(int a, int b);
 257    void schedule_instructions();
 258    void fail(const char *msg, ...);
 259
 260    void push_force_uncompressed();
 261    void pop_force_uncompressed();
 262    void push_force_sechalf();
 263    void pop_force_sechalf();
 264
 265    void generate_code();
 266    void generate_fb_write(fs_inst *inst);
 267    void generate_pixel_xy(struct brw_reg dst, bool is_x);
 268    void generate_linterp(fs_inst *inst, struct brw_reg dst,
 269                          struct brw_reg *src);
 270    void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 271    void generate_math1_gen7(fs_inst *inst,
 272                             struct brw_reg dst,
 273                             struct brw_reg src);
 274    void generate_math2_gen7(fs_inst *inst,
 275                             struct brw_reg dst,
 276                             struct brw_reg src0,
 277                             struct brw_reg src1);
 278    void generate_math1_gen6(fs_inst *inst,
 279                             struct brw_reg dst,
 280                             struct brw_reg src);
 281    void generate_math2_gen6(fs_inst *inst,
 282                             struct brw_reg dst,
 283                             struct brw_reg src0,
 284                             struct brw_reg src1);
 285    void generate_math_gen4(fs_inst *inst,
 286                            struct brw_reg dst,
 287                            struct brw_reg src);
 288    void generate_discard(fs_inst *inst);
 289    void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
 290    void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
 291                      bool negate_value);
 292    void generate_spill(fs_inst *inst, struct brw_reg src);
 293    void generate_unspill(fs_inst *inst, struct brw_reg dst);
 294    void generate_pull_constant_load(fs_inst *inst, struct brw_reg dst);
 295    void generate_mov_dispatch_to_flags();
 296
 297    void emit_dummy_fs();
 298    fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
 299    fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
 300                          glsl_interp_qualifier interpolation_mode,
 301                          bool is_centroid);
 302    fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
 303    fs_reg *emit_general_interpolation(ir_variable *ir);
 304    void emit_interpolation_setup_gen4();
 305    void emit_interpolation_setup_gen6();
 306    fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 307                               int sampler);
 308    fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 309                               int sampler);
 310    fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
 311                               int sampler);
 312    fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
 313    fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
 314    bool try_emit_saturate(ir_expression *ir);
 315    bool try_emit_mad(ir_expression *ir, int mul_arg);
 316    void emit_bool_to_cond_code(ir_rvalue *condition);
 317    void emit_if_gen6(ir_if *ir);
 318    void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset);
 319
 320    void emit_color_write(int target, int index, int first_color_mrf);
 321    void emit_fb_writes();
 322    bool try_rewrite_rhs_to_dst(ir_assignment *ir,
 323                                fs_reg dst,
 324                                fs_reg src,
 325                                fs_inst *pre_rhs_inst,
 326                                fs_inst *last_rhs_inst);
 327    void emit_assignment_writes(fs_reg &l, fs_reg &r,
 328                                const glsl_type *type, bool predicated);
 329    void resolve_ud_negate(fs_reg *reg);
 330    void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
 331
 332    struct brw_reg interp_reg(int location, int channel);
 333    int setup_uniform_values(int loc, const glsl_type *type);
 334    void setup_builtin_uniform_values(ir_variable *ir);
 335    int implied_mrf_writes(fs_inst *inst);
 336
 337    struct brw_context *brw;
 338    const struct gl_fragment_program *fp;
 339    struct intel_context *intel;
 340    struct gl_context *ctx;
 341    struct brw_wm_compile *c;
 342    struct brw_compile *p;
 343    struct brw_shader *shader;
 344    struct gl_shader_program *prog;
 345    void *mem_ctx;
 346    exec_list instructions;
 347
 348    /* Delayed setup of c->prog_data.params[] due to realloc of
 349     * ParamValues[] during compile.
 350     */
 351    int param_index[MAX_UNIFORMS * 4];
 352    int param_offset[MAX_UNIFORMS * 4];
 353
 354    int *virtual_grf_sizes;
 355    int virtual_grf_count;
 356    int virtual_grf_array_size;
 357    int *virtual_grf_def;
 358    int *virtual_grf_use;
 359    bool live_intervals_valid;
 360
 361    /* This is the map from UNIFORM hw_reg + reg_offset as generated by
 362     * the visitor to the packed uniform number after
 363     * remove_dead_constants() that represents the actual uploaded
 364     * uniform index.
 365     */
 366    int *params_remap;
 367
 368    struct hash_table *variable_ht;
 369    ir_variable *frag_depth;
 370    fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
 371    unsigned output_components[BRW_MAX_DRAW_BUFFERS];
 372    fs_reg dual_src_output;
 373    int first_non_payload_grf;
 374    int max_grf;
 375    int urb_setup[FRAG_ATTRIB_MAX];
 376
 377    /** @{ debug annotation info */
 378    const char *current_annotation;
 379    ir_instruction *base_ir;
 380    /** @} */
 381
 382    bool failed;
 383    char *fail_msg;
 384
 385    /* Result of last visit() method. */
 386    fs_reg result;
 387
 388    fs_reg pixel_x;
 389    fs_reg pixel_y;
 390    fs_reg wpos_w;
 391    fs_reg pixel_w;
 392    fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
 393    fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
 394    fs_reg reg_null_cmp;
 395
 396    int grf_used;
 397
 398    int force_uncompressed_stack;
 399    int force_sechalf_stack;
 400
 401    class fs_bblock *bblock;
 402 };
 403
 404 bool brw_do_channel_expressions(struct exec_list *instructions);
 405 bool brw_do_vector_splitting(struct exec_list *instructions);
 406 bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);