gcc/config/aarch64/aarch64-protos.h

   1 /* Machine description for AArch64 architecture.
   2    Copyright (C) 2009-2016 Free Software Foundation, Inc.
   3    Contributed by ARM Ltd.
   4
   5    This file is part of GCC.
   6
   7    GCC is free software; you can redistribute it and/or modify it
   8    under the terms of the GNU General Public License as published by
   9    the Free Software Foundation; either version 3, or (at your option)
  10    any later version.
  11
  12    GCC is distributed in the hope that it will be useful, but
  13    WITHOUT ANY WARRANTY; without even the implied warranty of
  14    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15    General Public License for more details.
  16
  17    You should have received a copy of the GNU General Public License
  18    along with GCC; see the file COPYING3.  If not see
  19    <http://www.gnu.org/licenses/>.  */
  20
  21
  22 #ifndef GCC_AARCH64_PROTOS_H
  23 #define GCC_AARCH64_PROTOS_H
  24
  25 #include "input.h"
  26
  27 /* SYMBOL_SMALL_ABSOLUTE: Generate symbol accesses through
  28    high and lo relocs that calculate the base address using a PC
  29    relative reloc.
  30    So to get the address of foo, we generate
  31    adrp x0, foo
  32    add  x0, x0, :lo12:foo
  33
  34    To load or store something to foo, we could use the corresponding
  35    load store variants that generate an
  36    ldr x0, [x0,:lo12:foo]
  37    or
  38    str x1, [x0, :lo12:foo]
  39
  40    This corresponds to the small code model of the compiler.
  41
  42    SYMBOL_SMALL_GOT_4G: Similar to the one above but this
  43    gives us the GOT entry of the symbol being referred to :
  44    Thus calculating the GOT entry for foo is done using the
  45    following sequence of instructions.  The ADRP instruction
  46    gets us to the page containing the GOT entry of the symbol
  47    and the got_lo12 gets us the actual offset in it, together
  48    the base and offset, we can address 4G size GOT table.
  49
  50    adrp  x0, :got:foo
  51    ldr   x0, [x0, :gotoff_lo12:foo]
  52
  53    This corresponds to the small PIC model of the compiler.
  54
  55    SYMBOL_SMALL_GOT_28K: Similar to SYMBOL_SMALL_GOT_4G, but used for symbol
  56    restricted within 28K GOT table size.
  57
  58    ldr reg, [gp, #:gotpage_lo15:sym]
  59
  60    This corresponds to -fpic model for small memory model of the compiler.
  61
  62    SYMBOL_SMALL_TLSGD
  63    SYMBOL_SMALL_TLSDESC
  64    SYMBOL_SMALL_TLSIE
  65    SYMBOL_TINY_TLSIE
  66    SYMBOL_TLSLE12
  67    SYMBOL_TLSLE24
  68    SYMBOL_TLSLE32
  69    SYMBOL_TLSLE48
  70    Each of these represents a thread-local symbol, and corresponds to the
  71    thread local storage relocation operator for the symbol being referred to.
  72
  73    SYMBOL_TINY_ABSOLUTE
  74
  75    Generate symbol accesses as a PC relative address using a single
  76    instruction.  To compute the address of symbol foo, we generate:
  77
  78    ADR x0, foo
  79
  80    SYMBOL_TINY_GOT
  81
  82    Generate symbol accesses via the GOT using a single PC relative
  83    instruction.  To compute the address of symbol foo, we generate:
  84
  85    ldr t0, :got:foo
  86
  87    The value of foo can subsequently read using:
  88
  89    ldrb    t0, [t0]
  90
  91    SYMBOL_FORCE_TO_MEM : Global variables are addressed using
  92    constant pool.  All variable addresses are spilled into constant
  93    pools.  The constant pools themselves are addressed using PC
  94    relative accesses.  This only works for the large code model.
  95  */
  96 enum aarch64_symbol_type
  97 {
  98   SYMBOL_SMALL_ABSOLUTE,
  99   SYMBOL_SMALL_GOT_28K,
 100   SYMBOL_SMALL_GOT_4G,
 101   SYMBOL_SMALL_TLSGD,
 102   SYMBOL_SMALL_TLSDESC,
 103   SYMBOL_SMALL_TLSIE,
 104   SYMBOL_TINY_ABSOLUTE,
 105   SYMBOL_TINY_GOT,
 106   SYMBOL_TINY_TLSIE,
 107   SYMBOL_TLSLE12,
 108   SYMBOL_TLSLE24,
 109   SYMBOL_TLSLE32,
 110   SYMBOL_TLSLE48,
 111   SYMBOL_FORCE_TO_MEM
 112 };
 113
 114 /* A set of tuning parameters contains references to size and time
 115    cost models and vectors for address cost calculations, register
 116    move costs and memory move costs.  */
 117
 118 /* Scaled addressing modes can vary cost depending on the mode of the
 119    value to be loaded/stored.  QImode values cannot use scaled
 120    addressing modes.  */
 121
 122 struct scale_addr_mode_cost
 123 {
 124   const int hi;
 125   const int si;
 126   const int di;
 127   const int ti;
 128 };
 129
 130 /* Additional cost for addresses.  */
 131 struct cpu_addrcost_table
 132 {
 133   const struct scale_addr_mode_cost addr_scale_costs;
 134   const int pre_modify;
 135   const int post_modify;
 136   const int register_offset;
 137   const int register_sextend;
 138   const int register_zextend;
 139   const int imm_offset;
 140 };
 141
 142 /* Additional costs for register copies.  Cost is for one register.  */
 143 struct cpu_regmove_cost
 144 {
 145   const int GP2GP;
 146   const int GP2FP;
 147   const int FP2GP;
 148   const int FP2FP;
 149 };
 150
 151 /* Cost for vector insn classes.  */
 152 struct cpu_vector_cost
 153 {
 154   const int scalar_stmt_cost;            /* Cost of any scalar operation,
 155                                             excluding load and store.  */
 156   const int scalar_load_cost;            /* Cost of scalar load.  */
 157   const int scalar_store_cost;           /* Cost of scalar store.  */
 158   const int vec_stmt_cost;               /* Cost of any vector operation,
 159                                             excluding load, store, permute,
 160                                             vector-to-scalar and
 161                                             scalar-to-vector operation.  */
 162   const int vec_permute_cost;            /* Cost of permute operation.  */
 163   const int vec_to_scalar_cost;          /* Cost of vec-to-scalar operation.  */
 164   const int scalar_to_vec_cost;          /* Cost of scalar-to-vector
 165                                             operation.  */
 166   const int vec_align_load_cost;         /* Cost of aligned vector load.  */
 167   const int vec_unalign_load_cost;       /* Cost of unaligned vector load.  */
 168   const int vec_unalign_store_cost;      /* Cost of unaligned vector store.  */
 169   const int vec_store_cost;              /* Cost of vector store.  */
 170   const int cond_taken_branch_cost;      /* Cost of taken branch.  */
 171   const int cond_not_taken_branch_cost;  /* Cost of not taken branch.  */
 172 };
 173
 174 /* Branch costs.  */
 175 struct cpu_branch_cost
 176 {
 177   const int predictable;    /* Predictable branch or optimizing for size.  */
 178   const int unpredictable;  /* Unpredictable branch or optimizing for speed.  */
 179 };
 180
 181 /* Control approximate alternatives to certain FP operators.  */
 182 #define AARCH64_APPROX_MODE(MODE) \
 183   ((MIN_MODE_FLOAT <= (MODE) && (MODE) <= MAX_MODE_FLOAT) \
 184    ? (1 << ((MODE) - MIN_MODE_FLOAT)) \
 185    : (MIN_MODE_VECTOR_FLOAT <= (MODE) && (MODE) <= MAX_MODE_VECTOR_FLOAT) \
 186      ? (1 << ((MODE) - MIN_MODE_VECTOR_FLOAT \
 187               + MAX_MODE_FLOAT - MIN_MODE_FLOAT + 1)) \
 188      : (0))
 189 #define AARCH64_APPROX_NONE (0)
 190 #define AARCH64_APPROX_ALL (-1)
 191
 192 /* Allowed modes for approximations.  */
 193 struct cpu_approx_modes
 194 {
 195   const unsigned int division;          /* Division.  */
 196   const unsigned int sqrt;              /* Square root.  */
 197   const unsigned int recip_sqrt;        /* Reciprocal square root.  */
 198 };
 199
 200 struct tune_params
 201 {
 202   const struct cpu_cost_table *insn_extra_cost;
 203   const struct cpu_addrcost_table *addr_cost;
 204   const struct cpu_regmove_cost *regmove_cost;
 205   const struct cpu_vector_cost *vec_costs;
 206   const struct cpu_branch_cost *branch_costs;
 207   const struct cpu_approx_modes *approx_modes;
 208   int memmov_cost;
 209   int issue_rate;
 210   unsigned int fusible_ops;
 211   int function_align;
 212   int jump_align;
 213   int loop_align;
 214   int int_reassoc_width;
 215   int fp_reassoc_width;
 216   int vec_reassoc_width;
 217   int min_div_recip_mul_sf;
 218   int min_div_recip_mul_df;
 219   /* Value for aarch64_case_values_threshold; or 0 for the default.  */
 220   unsigned int max_case_values;
 221   /* Value for PARAM_L1_CACHE_LINE_SIZE; or 0 to use the default.  */
 222   unsigned int cache_line_size;
 223
 224 /* An enum specifying how to take into account CPU autoprefetch capabilities
 225    during instruction scheduling:
 226    - AUTOPREFETCHER_OFF: Do not take autoprefetch capabilities into account.
 227    - AUTOPREFETCHER_WEAK: Attempt to sort sequences of loads/store in order of
 228    offsets but allow the pipeline hazard recognizer to alter that order to
 229    maximize multi-issue opportunities.
 230    - AUTOPREFETCHER_STRONG: Attempt to sort sequences of loads/store in order of
 231    offsets and prefer this even if it restricts multi-issue opportunities.  */
 232
 233   enum aarch64_autoprefetch_model
 234   {
 235     AUTOPREFETCHER_OFF,
 236     AUTOPREFETCHER_WEAK,
 237     AUTOPREFETCHER_STRONG
 238   } autoprefetcher_model;
 239
 240   unsigned int extra_tuning_flags;
 241 };
 242
 243 #define AARCH64_FUSION_PAIR(x, name) \
 244   AARCH64_FUSE_##name##_index,
 245 /* Supported fusion operations.  */
 246 enum aarch64_fusion_pairs_index
 247 {
 248 #include "aarch64-fusion-pairs.def"
 249   AARCH64_FUSE_index_END
 250 };
 251
 252 #define AARCH64_FUSION_PAIR(x, name) \
 253   AARCH64_FUSE_##name = (1u << AARCH64_FUSE_##name##_index),
 254 /* Supported fusion operations.  */
 255 enum aarch64_fusion_pairs
 256 {
 257   AARCH64_FUSE_NOTHING = 0,
 258 #include "aarch64-fusion-pairs.def"
 259   AARCH64_FUSE_ALL = (1u << AARCH64_FUSE_index_END) - 1
 260 };
 261
 262 #define AARCH64_EXTRA_TUNING_OPTION(x, name) \
 263   AARCH64_EXTRA_TUNE_##name##_index,
 264 /* Supported tuning flags indexes.  */
 265 enum aarch64_extra_tuning_flags_index
 266 {
 267 #include "aarch64-tuning-flags.def"
 268   AARCH64_EXTRA_TUNE_index_END
 269 };
 270
 271
 272 #define AARCH64_EXTRA_TUNING_OPTION(x, name) \
 273   AARCH64_EXTRA_TUNE_##name = (1u << AARCH64_EXTRA_TUNE_##name##_index),
 274 /* Supported tuning flags.  */
 275 enum aarch64_extra_tuning_flags
 276 {
 277   AARCH64_EXTRA_TUNE_NONE = 0,
 278 #include "aarch64-tuning-flags.def"
 279   AARCH64_EXTRA_TUNE_ALL = (1u << AARCH64_EXTRA_TUNE_index_END) - 1
 280 };
 281
 282 /* Enum describing the various ways that the
 283    aarch64_parse_{arch,tune,cpu,extension} functions can fail.
 284    This way their callers can choose what kind of error to give.  */
 285
 286 enum aarch64_parse_opt_result
 287 {
 288   AARCH64_PARSE_OK,                     /* Parsing was successful.  */
 289   AARCH64_PARSE_MISSING_ARG,            /* Missing argument.  */
 290   AARCH64_PARSE_INVALID_FEATURE,        /* Invalid feature modifier.  */
 291   AARCH64_PARSE_INVALID_ARG             /* Invalid arch, tune, cpu arg.  */
 292 };
 293
 294 extern struct tune_params aarch64_tune_params;
 295
 296 HOST_WIDE_INT aarch64_initial_elimination_offset (unsigned, unsigned);
 297 int aarch64_get_condition_code (rtx);
 298 bool aarch64_bitmask_imm (HOST_WIDE_INT val, machine_mode);
 299 int aarch64_branch_cost (bool, bool);
 300 enum aarch64_symbol_type aarch64_classify_symbolic_expression (rtx);
 301 bool aarch64_const_vec_all_same_int_p (rtx, HOST_WIDE_INT);
 302 bool aarch64_constant_address_p (rtx);
 303 bool aarch64_emit_approx_div (rtx, rtx, rtx);
 304 bool aarch64_emit_approx_sqrt (rtx, rtx, bool);
 305 bool aarch64_expand_movmem (rtx *);
 306 bool aarch64_float_const_zero_rtx_p (rtx);
 307 bool aarch64_function_arg_regno_p (unsigned);
 308 bool aarch64_fusion_enabled_p (enum aarch64_fusion_pairs);
 309 bool aarch64_gen_movmemqi (rtx *);
 310 bool aarch64_gimple_fold_builtin (gimple_stmt_iterator *);
 311 bool aarch64_is_extend_from_extract (machine_mode, rtx, rtx);
 312 bool aarch64_is_long_call_p (rtx);
 313 bool aarch64_is_noplt_call_p (rtx);
 314 bool aarch64_label_mentioned_p (rtx);
 315 void aarch64_declare_function_name (FILE *, const char*, tree);
 316 bool aarch64_legitimate_pic_operand_p (rtx);
 317 bool aarch64_mask_and_shift_for_ubfiz_p (machine_mode, rtx, rtx);
 318 bool aarch64_modes_tieable_p (machine_mode mode1,
 319                               machine_mode mode2);
 320 bool aarch64_zero_extend_const_eq (machine_mode, rtx, machine_mode, rtx);
 321 bool aarch64_move_imm (HOST_WIDE_INT, machine_mode);
 322 bool aarch64_mov_operand_p (rtx, machine_mode);
 323 int aarch64_simd_attr_length_rglist (enum machine_mode);
 324 rtx aarch64_reverse_mask (enum machine_mode);
 325 bool aarch64_offset_7bit_signed_scaled_p (machine_mode, HOST_WIDE_INT);
 326 char *aarch64_output_scalar_simd_mov_immediate (rtx, machine_mode);
 327 char *aarch64_output_simd_mov_immediate (rtx, machine_mode, unsigned);
 328 bool aarch64_pad_arg_upward (machine_mode, const_tree);
 329 bool aarch64_pad_reg_upward (machine_mode, const_tree, bool);
 330 bool aarch64_regno_ok_for_base_p (int, bool);
 331 bool aarch64_regno_ok_for_index_p (int, bool);
 332 bool aarch64_simd_check_vect_par_cnst_half (rtx op, machine_mode mode,
 333                                             bool high);
 334 bool aarch64_simd_imm_scalar_p (rtx x, machine_mode mode);
 335 bool aarch64_simd_imm_zero_p (rtx, machine_mode);
 336 bool aarch64_simd_scalar_immediate_valid_for_move (rtx, machine_mode);
 337 bool aarch64_simd_shift_imm_p (rtx, machine_mode, bool);
 338 bool aarch64_simd_valid_immediate (rtx, machine_mode, bool,
 339                                    struct simd_immediate_info *);
 340 bool aarch64_split_dimode_const_store (rtx, rtx);
 341 bool aarch64_symbolic_address_p (rtx);
 342 bool aarch64_uimm12_shift (HOST_WIDE_INT);
 343 bool aarch64_use_return_insn_p (void);
 344 const char *aarch64_mangle_builtin_type (const_tree);
 345 const char *aarch64_output_casesi (rtx *);
 346
 347 enum aarch64_symbol_type aarch64_classify_symbol (rtx, rtx);
 348 enum aarch64_symbol_type aarch64_classify_tls_symbol (rtx);
 349 enum reg_class aarch64_regno_regclass (unsigned);
 350 int aarch64_asm_preferred_eh_data_format (int, int);
 351 int aarch64_fpconst_pow_of_2 (rtx);
 352 machine_mode aarch64_hard_regno_caller_save_mode (unsigned, unsigned,
 353                                                        machine_mode);
 354 int aarch64_hard_regno_mode_ok (unsigned, machine_mode);
 355 int aarch64_hard_regno_nregs (unsigned, machine_mode);
 356 int aarch64_uxt_size (int, HOST_WIDE_INT);
 357 int aarch64_vec_fpconst_pow_of_2 (rtx);
 358 rtx aarch64_final_eh_return_addr (void);
 359 rtx aarch64_mask_from_zextract_ops (rtx, rtx);
 360 const char *aarch64_output_move_struct (rtx *operands);
 361 rtx aarch64_return_addr (int, rtx);
 362 rtx aarch64_simd_gen_const_vector_dup (machine_mode, int);
 363 bool aarch64_simd_mem_operand_p (rtx);
 364 rtx aarch64_simd_vect_par_cnst_half (machine_mode, bool);
 365 rtx aarch64_tls_get_addr (void);
 366 tree aarch64_fold_builtin (tree, int, tree *, bool);
 367 unsigned aarch64_dbx_register_number (unsigned);
 368 unsigned aarch64_trampoline_size (void);
 369 void aarch64_asm_output_labelref (FILE *, const char *);
 370 void aarch64_cpu_cpp_builtins (cpp_reader *);
 371 const char * aarch64_gen_far_branch (rtx *, int, const char *, const char *);
 372 const char * aarch64_output_probe_stack_range (rtx, rtx);
 373 void aarch64_err_no_fpadvsimd (machine_mode, const char *);
 374 void aarch64_expand_epilogue (bool);
 375 void aarch64_expand_mov_immediate (rtx, rtx);
 376 void aarch64_expand_prologue (void);
 377 void aarch64_expand_vector_init (rtx, rtx);
 378 void aarch64_init_cumulative_args (CUMULATIVE_ARGS *, const_tree, rtx,
 379                                    const_tree, unsigned);
 380 void aarch64_init_expanders (void);
 381 void aarch64_init_simd_builtins (void);
 382 void aarch64_emit_call_insn (rtx);
 383 void aarch64_register_pragmas (void);
 384 void aarch64_relayout_simd_types (void);
 385 void aarch64_reset_previous_fndecl (void);
 386 void aarch64_save_restore_target_globals (tree);
 387
 388 /* Initialize builtins for SIMD intrinsics.  */
 389 void init_aarch64_simd_builtins (void);
 390
 391 void aarch64_simd_emit_reg_reg_move (rtx *, enum machine_mode, unsigned int);
 392
 393 /* Expand builtins for SIMD intrinsics.  */
 394 rtx aarch64_simd_expand_builtin (int, tree, rtx);
 395
 396 void aarch64_simd_lane_bounds (rtx, HOST_WIDE_INT, HOST_WIDE_INT, const_tree);
 397
 398 void aarch64_split_128bit_move (rtx, rtx);
 399
 400 bool aarch64_split_128bit_move_p (rtx, rtx);
 401
 402 void aarch64_split_simd_combine (rtx, rtx, rtx);
 403
 404 void aarch64_split_simd_move (rtx, rtx);
 405
 406 /* Check for a legitimate floating point constant for FMOV.  */
 407 bool aarch64_float_const_representable_p (rtx);
 408
 409 #if defined (RTX_CODE)
 410
 411 bool aarch64_legitimate_address_p (machine_mode, rtx, RTX_CODE, bool);
 412 machine_mode aarch64_select_cc_mode (RTX_CODE, rtx, rtx);
 413 rtx aarch64_gen_compare_reg (RTX_CODE, rtx, rtx);
 414 rtx aarch64_load_tp (rtx);
 415
 416 void aarch64_expand_compare_and_swap (rtx op[]);
 417 void aarch64_split_compare_and_swap (rtx op[]);
 418 void aarch64_gen_atomic_cas (rtx, rtx, rtx, rtx, rtx);
 419
 420 bool aarch64_atomic_ldop_supported_p (enum rtx_code);
 421 void aarch64_gen_atomic_ldop (enum rtx_code, rtx, rtx, rtx, rtx, rtx);
 422 void aarch64_split_atomic_op (enum rtx_code, rtx, rtx, rtx, rtx, rtx, rtx);
 423
 424 bool aarch64_gen_adjusted_ldpstp (rtx *, bool, enum machine_mode, RTX_CODE);
 425 #endif /* RTX_CODE */
 426
 427 void aarch64_init_builtins (void);
 428
 429 bool aarch64_process_target_attr (tree, const char*);
 430 void aarch64_override_options_internal (struct gcc_options *);
 431
 432 rtx aarch64_expand_builtin (tree exp,
 433                             rtx target,
 434                             rtx subtarget ATTRIBUTE_UNUSED,
 435                             machine_mode mode ATTRIBUTE_UNUSED,
 436                             int ignore ATTRIBUTE_UNUSED);
 437 tree aarch64_builtin_decl (unsigned, bool ATTRIBUTE_UNUSED);
 438 tree aarch64_builtin_rsqrt (unsigned int);
 439 tree aarch64_builtin_vectorized_function (unsigned int, tree, tree);
 440
 441 extern void aarch64_split_combinev16qi (rtx operands[3]);
 442 extern void aarch64_expand_vec_perm (rtx target, rtx op0, rtx op1, rtx sel);
 443 extern bool aarch64_madd_needs_nop (rtx_insn *);
 444 extern void aarch64_final_prescan_insn (rtx_insn *);
 445 extern bool
 446 aarch64_expand_vec_perm_const (rtx target, rtx op0, rtx op1, rtx sel);
 447 void aarch64_atomic_assign_expand_fenv (tree *, tree *, tree *);
 448 int aarch64_ccmp_mode_to_code (enum machine_mode mode);
 449
 450 bool extract_base_offset_in_addr (rtx mem, rtx *base, rtx *offset);
 451 bool aarch64_operands_ok_for_ldpstp (rtx *, bool, enum machine_mode);
 452 bool aarch64_operands_adjust_ok_for_ldpstp (rtx *, bool, enum machine_mode);
 453 extern bool aarch64_nopcrelative_literal_loads;
 454
 455 extern void aarch64_asm_output_pool_epilogue (FILE *, const char *,
 456                                               tree, HOST_WIDE_INT);
 457
 458 /* Defined in common/config/aarch64-common.c.  */
 459 bool aarch64_handle_option (struct gcc_options *, struct gcc_options *,
 460                              const struct cl_decoded_option *, location_t);
 461 const char *aarch64_rewrite_selected_cpu (const char *name);
 462 enum aarch64_parse_opt_result aarch64_parse_extension (const char *,
 463                                                        unsigned long *);
 464 std::string aarch64_get_extension_string_for_isa_flags (unsigned long,
 465                                                         unsigned long);
 466
 467 rtl_opt_pass *make_pass_fma_steering (gcc::context *ctxt);
 468
 469 #endif /* GCC_AARCH64_PROTOS_H */