src/mesa/drivers/dri/i965/brw_vec4.h

   1 /*
   2  * Copyright © 2011 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
  24 #ifndef BRW_VEC4_H
  25 #define BRW_VEC4_H
  26
  27 #include <stdint.h>
  28 #include "brw_shader.h"
  29 #include "main/compiler.h"
  30 #include "program/hash_table.h"
  31 #include "brw_program.h"
  32
  33 #ifdef __cplusplus
  34 #include "brw_ir_vec4.h"
  35
  36 extern "C" {
  37 #endif
  38
  39 #include "brw_context.h"
  40 #include "brw_eu.h"
  41 #include "intel_asm_annotation.h"
  42
  43 #ifdef __cplusplus
  44 }; /* extern "C" */
  45 #endif
  46
  47 #include "glsl/ir.h"
  48 #include "glsl/nir/nir.h"
  49
  50
  51 #ifdef __cplusplus
  52 extern "C" {
  53 #endif
  54
  55 const unsigned *
  56 brw_vec4_generate_assembly(const struct brw_compiler *compiler,
  57                            void *log_data,
  58                            void *mem_ctx,
  59                            const nir_shader *nir,
  60                            struct brw_vue_prog_data *prog_data,
  61                            const struct cfg_t *cfg,
  62                            unsigned *out_assembly_size);
  63
  64 #ifdef __cplusplus
  65 } /* extern "C" */
  66
  67 namespace brw {
  68
  69 class vec4_live_variables;
  70
  71 /**
  72  * The vertex shader front-end.
  73  *
  74  * Translates either GLSL IR or Mesa IR (for ARB_vertex_program and
  75  * fixed-function) into VS IR.
  76  */
  77 class vec4_visitor : public backend_shader
  78 {
  79 public:
  80    vec4_visitor(const struct brw_compiler *compiler,
  81                 void *log_data,
  82                 const struct brw_sampler_prog_key_data *key,
  83                 struct brw_vue_prog_data *prog_data,
  84                 const nir_shader *shader,
  85                 void *mem_ctx,
  86                 bool no_spills,
  87                 int shader_time_index);
  88    virtual ~vec4_visitor();
  89
  90    dst_reg dst_null_f()
  91    {
  92       return dst_reg(brw_null_reg());
  93    }
  94
  95    dst_reg dst_null_d()
  96    {
  97       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
  98    }
  99
 100    dst_reg dst_null_ud()
 101    {
 102       return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
 103    }
 104
 105    const struct brw_sampler_prog_key_data * const key_tex;
 106    struct brw_vue_prog_data * const prog_data;
 107    char *fail_msg;
 108    bool failed;
 109
 110    /**
 111     * GLSL IR currently being processed, which is associated with our
 112     * driver IR instructions for debugging purposes.
 113     */
 114    const void *base_ir;
 115    const char *current_annotation;
 116
 117    int first_non_payload_grf;
 118    unsigned int max_grf;
 119    int *virtual_grf_start;
 120    int *virtual_grf_end;
 121    brw::vec4_live_variables *live_intervals;
 122    dst_reg userplane[MAX_CLIP_PLANES];
 123
 124    bool need_all_constants_in_pull_buffer;
 125
 126    /* Regs for vertex results.  Generated at ir_variable visiting time
 127     * for the ir->location's used.
 128     */
 129    dst_reg output_reg[BRW_VARYING_SLOT_COUNT];
 130    const char *output_reg_annotation[BRW_VARYING_SLOT_COUNT];
 131    int *uniform_size;
 132    int uniform_array_size; /*< Size of the uniform_size array */
 133    int uniforms;
 134
 135    src_reg shader_start_time;
 136
 137    bool run();
 138    void fail(const char *msg, ...);
 139
 140    int setup_uniforms(int payload_reg);
 141
 142    bool reg_allocate_trivial();
 143    bool reg_allocate();
 144    void evaluate_spill_costs(float *spill_costs, bool *no_spill);
 145    int choose_spill_reg(struct ra_graph *g);
 146    void spill_reg(int spill_reg);
 147    void move_grf_array_access_to_scratch();
 148    void move_uniform_array_access_to_pull_constants();
 149    void move_push_constants_to_pull_constants();
 150    void split_uniform_registers();
 151    void pack_uniform_registers();
 152    void calculate_live_intervals();
 153    void invalidate_live_intervals();
 154    void split_virtual_grfs();
 155    bool opt_vector_float();
 156    bool opt_reduce_swizzle();
 157    bool dead_code_eliminate();
 158    int var_range_start(unsigned v, unsigned n) const;
 159    int var_range_end(unsigned v, unsigned n) const;
 160    bool virtual_grf_interferes(int a, int b);
 161    bool opt_cmod_propagation();
 162    bool opt_copy_propagation(bool do_constant_prop = true);
 163    bool opt_cse_local(bblock_t *block);
 164    bool opt_cse();
 165    bool opt_algebraic();
 166    bool opt_register_coalesce();
 167    bool eliminate_find_live_channel();
 168    bool is_dep_ctrl_unsafe(const vec4_instruction *inst);
 169    void opt_set_dependency_control();
 170    void opt_schedule_instructions();
 171    void convert_to_hw_regs();
 172
 173    vec4_instruction *emit(vec4_instruction *inst);
 174
 175    vec4_instruction *emit(enum opcode opcode);
 176    vec4_instruction *emit(enum opcode opcode, const dst_reg &dst);
 177    vec4_instruction *emit(enum opcode opcode, const dst_reg &dst,
 178                           const src_reg &src0);
 179    vec4_instruction *emit(enum opcode opcode, const dst_reg &dst,
 180                           const src_reg &src0, const src_reg &src1);
 181    vec4_instruction *emit(enum opcode opcode, const dst_reg &dst,
 182                           const src_reg &src0, const src_reg &src1,
 183                           const src_reg &src2);
 184
 185    vec4_instruction *emit_before(bblock_t *block,
 186                                  vec4_instruction *inst,
 187                                  vec4_instruction *new_inst);
 188
 189 #define EMIT1(op) vec4_instruction *op(const dst_reg &, const src_reg &);
 190 #define EMIT2(op) vec4_instruction *op(const dst_reg &, const src_reg &, const src_reg &);
 191 #define EMIT3(op) vec4_instruction *op(const dst_reg &, const src_reg &, const src_reg &, const src_reg &);
 192    EMIT1(MOV)
 193    EMIT1(NOT)
 194    EMIT1(RNDD)
 195    EMIT1(RNDE)
 196    EMIT1(RNDZ)
 197    EMIT1(FRC)
 198    EMIT1(F32TO16)
 199    EMIT1(F16TO32)
 200    EMIT2(ADD)
 201    EMIT2(MUL)
 202    EMIT2(MACH)
 203    EMIT2(MAC)
 204    EMIT2(AND)
 205    EMIT2(OR)
 206    EMIT2(XOR)
 207    EMIT2(DP3)
 208    EMIT2(DP4)
 209    EMIT2(DPH)
 210    EMIT2(SHL)
 211    EMIT2(SHR)
 212    EMIT2(ASR)
 213    vec4_instruction *CMP(dst_reg dst, src_reg src0, src_reg src1,
 214                          enum brw_conditional_mod condition);
 215    vec4_instruction *IF(src_reg src0, src_reg src1,
 216                         enum brw_conditional_mod condition);
 217    vec4_instruction *IF(enum brw_predicate predicate);
 218    EMIT1(SCRATCH_READ)
 219    EMIT2(SCRATCH_WRITE)
 220    EMIT3(LRP)
 221    EMIT1(BFREV)
 222    EMIT3(BFE)
 223    EMIT2(BFI1)
 224    EMIT3(BFI2)
 225    EMIT1(FBH)
 226    EMIT1(FBL)
 227    EMIT1(CBIT)
 228    EMIT3(MAD)
 229    EMIT2(ADDC)
 230    EMIT2(SUBB)
 231 #undef EMIT1
 232 #undef EMIT2
 233 #undef EMIT3
 234
 235    int implied_mrf_writes(vec4_instruction *inst);
 236
 237    vec4_instruction *emit_minmax(enum brw_conditional_mod conditionalmod, dst_reg dst,
 238                                  src_reg src0, src_reg src1);
 239
 240    vec4_instruction *emit_lrp(const dst_reg &dst, const src_reg &x,
 241                               const src_reg &y, const src_reg &a);
 242
 243    /**
 244     * Copy any live channel from \p src to the first channel of the
 245     * result.
 246     */
 247    src_reg emit_uniformize(const src_reg &src);
 248
 249    src_reg fix_3src_operand(const src_reg &src);
 250    src_reg resolve_source_modifiers(const src_reg &src);
 251
 252    vec4_instruction *emit_math(enum opcode opcode, const dst_reg &dst, const src_reg &src0,
 253                                const src_reg &src1 = src_reg());
 254
 255    src_reg fix_math_operand(const src_reg &src);
 256
 257    void emit_pack_half_2x16(dst_reg dst, src_reg src0);
 258    void emit_unpack_half_2x16(dst_reg dst, src_reg src0);
 259    void emit_unpack_unorm_4x8(const dst_reg &dst, src_reg src0);
 260    void emit_unpack_snorm_4x8(const dst_reg &dst, src_reg src0);
 261    void emit_pack_unorm_4x8(const dst_reg &dst, const src_reg &src0);
 262    void emit_pack_snorm_4x8(const dst_reg &dst, const src_reg &src0);
 263
 264    void emit_texture(ir_texture_opcode op,
 265                      dst_reg dest,
 266                      const glsl_type *dest_type,
 267                      src_reg coordinate,
 268                      int coord_components,
 269                      src_reg shadow_comparitor,
 270                      src_reg lod, src_reg lod2,
 271                      src_reg sample_index,
 272                      uint32_t constant_offset,
 273                      src_reg offset_value,
 274                      src_reg mcs,
 275                      bool is_cube_array,
 276                      uint32_t sampler, src_reg sampler_reg);
 277
 278    src_reg emit_mcs_fetch(const glsl_type *coordinate_type, src_reg coordinate,
 279                           src_reg sampler);
 280    void emit_gen6_gather_wa(uint8_t wa, dst_reg dst);
 281
 282    void emit_ndc_computation();
 283    void emit_psiz_and_flags(dst_reg reg);
 284    vec4_instruction *emit_generic_urb_slot(dst_reg reg, int varying);
 285    virtual void emit_urb_slot(dst_reg reg, int varying);
 286
 287    void emit_shader_time_begin();
 288    void emit_shader_time_end();
 289    void emit_shader_time_write(int shader_time_subindex, src_reg value);
 290
 291    void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
 292                             dst_reg dst, src_reg offset, src_reg src0,
 293                             src_reg src1);
 294
 295    void emit_untyped_surface_read(unsigned surf_index, dst_reg dst,
 296                                   src_reg offset);
 297
 298    src_reg get_scratch_offset(bblock_t *block, vec4_instruction *inst,
 299                               src_reg *reladdr, int reg_offset);
 300    src_reg get_pull_constant_offset(bblock_t *block, vec4_instruction *inst,
 301                                     src_reg *reladdr, int reg_offset);
 302    void emit_scratch_read(bblock_t *block, vec4_instruction *inst,
 303                           dst_reg dst,
 304                           src_reg orig_src,
 305                           int base_offset);
 306    void emit_scratch_write(bblock_t *block, vec4_instruction *inst,
 307                            int base_offset);
 308    void emit_pull_constant_load(bblock_t *block, vec4_instruction *inst,
 309                                 dst_reg dst,
 310                                 src_reg orig_src,
 311                                 int base_offset);
 312    void emit_pull_constant_load_reg(dst_reg dst,
 313                                     src_reg surf_index,
 314                                     src_reg offset,
 315                                     bblock_t *before_block,
 316                                     vec4_instruction *before_inst);
 317    src_reg emit_resolve_reladdr(int scratch_loc[], bblock_t *block,
 318                                 vec4_instruction *inst, src_reg src);
 319
 320    void resolve_ud_negate(src_reg *reg);
 321
 322    src_reg get_timestamp();
 323
 324    void dump_instruction(backend_instruction *inst);
 325    void dump_instruction(backend_instruction *inst, FILE *file);
 326
 327    bool is_high_sampler(src_reg sampler);
 328
 329    virtual void emit_nir_code();
 330    virtual void nir_setup_inputs();
 331    virtual void nir_setup_uniforms();
 332    virtual void nir_setup_system_value_intrinsic(nir_intrinsic_instr *instr);
 333    virtual void nir_setup_system_values();
 334    virtual void nir_emit_impl(nir_function_impl *impl);
 335    virtual void nir_emit_cf_list(exec_list *list);
 336    virtual void nir_emit_if(nir_if *if_stmt);
 337    virtual void nir_emit_loop(nir_loop *loop);
 338    virtual void nir_emit_block(nir_block *block);
 339    virtual void nir_emit_instr(nir_instr *instr);
 340    virtual void nir_emit_load_const(nir_load_const_instr *instr);
 341    virtual void nir_emit_intrinsic(nir_intrinsic_instr *instr);
 342    virtual void nir_emit_alu(nir_alu_instr *instr);
 343    virtual void nir_emit_jump(nir_jump_instr *instr);
 344    virtual void nir_emit_texture(nir_tex_instr *instr);
 345    virtual void nir_emit_undef(nir_ssa_undef_instr *instr);
 346    virtual void nir_emit_ssbo_atomic(int op, nir_intrinsic_instr *instr);
 347
 348    dst_reg get_nir_dest(nir_dest dest, enum brw_reg_type type);
 349    dst_reg get_nir_dest(nir_dest dest, nir_alu_type type);
 350    dst_reg get_nir_dest(nir_dest dest);
 351    src_reg get_nir_src(nir_src src, enum brw_reg_type type,
 352                        unsigned num_components = 4);
 353    src_reg get_nir_src(nir_src src, nir_alu_type type,
 354                        unsigned num_components = 4);
 355    src_reg get_nir_src(nir_src src,
 356                        unsigned num_components = 4);
 357
 358    virtual dst_reg *make_reg_for_system_value(int location,
 359                                               const glsl_type *type) = 0;
 360
 361    dst_reg *nir_locals;
 362    dst_reg *nir_ssa_values;
 363    src_reg *nir_inputs;
 364    dst_reg *nir_system_values;
 365
 366 protected:
 367    void emit_vertex();
 368    void lower_attributes_to_hw_regs(const int *attribute_map,
 369                                     bool interleaved);
 370    void setup_payload_interference(struct ra_graph *g, int first_payload_node,
 371                                    int reg_node_count);
 372    virtual void setup_payload() = 0;
 373    virtual void emit_prolog() = 0;
 374    virtual void emit_thread_end() = 0;
 375    virtual void emit_urb_write_header(int mrf) = 0;
 376    virtual vec4_instruction *emit_urb_write_opcode(bool complete) = 0;
 377    virtual void gs_emit_vertex(int stream_id);
 378    virtual void gs_end_primitive();
 379
 380 private:
 381    /**
 382     * If true, then register allocation should fail instead of spilling.
 383     */
 384    const bool no_spills;
 385
 386    int shader_time_index;
 387
 388    unsigned last_scratch; /**< measured in 32-byte (register size) units */
 389 };
 390
 391 } /* namespace brw */
 392 #endif /* __cplusplus */
 393
 394 #endif /* BRW_VEC4_H */