gcc/config/aarch64/aarch64.h

   1 /* Machine description for AArch64 architecture.
   2    Copyright (C) 2009-2015 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_H
  23 #define GCC_AARCH64_H
  24
  25 /* Target CPU builtins.  */
  26 #define TARGET_CPU_CPP_BUILTINS()       \
  27   aarch64_cpu_cpp_builtins (pfile)
  28
  29 \f
  30
  31 #define REGISTER_TARGET_PRAGMAS() aarch64_register_pragmas ()
  32
  33 /* Target machine storage layout.  */
  34
  35 #define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE)     \
  36   if (GET_MODE_CLASS (MODE) == MODE_INT         \
  37       && GET_MODE_SIZE (MODE) < 4)              \
  38     {                                           \
  39       if (MODE == QImode || MODE == HImode)     \
  40         {                                       \
  41           MODE = SImode;                        \
  42         }                                       \
  43     }
  44
  45 /* Bits are always numbered from the LSBit.  */
  46 #define BITS_BIG_ENDIAN 0
  47
  48 /* Big/little-endian flavour.  */
  49 #define BYTES_BIG_ENDIAN (TARGET_BIG_END != 0)
  50 #define WORDS_BIG_ENDIAN (BYTES_BIG_ENDIAN)
  51
  52 /* AdvSIMD is supported in the default configuration, unless disabled by
  53    -mgeneral-regs-only or by the +nosimd extension.  */
  54 #define TARGET_SIMD (!TARGET_GENERAL_REGS_ONLY && AARCH64_ISA_SIMD)
  55 #define TARGET_FLOAT (!TARGET_GENERAL_REGS_ONLY && AARCH64_ISA_FP)
  56
  57 #define UNITS_PER_WORD          8
  58
  59 #define UNITS_PER_VREG          16
  60
  61 #define PARM_BOUNDARY           64
  62
  63 #define STACK_BOUNDARY          128
  64
  65 #define FUNCTION_BOUNDARY       32
  66
  67 #define EMPTY_FIELD_BOUNDARY    32
  68
  69 #define BIGGEST_ALIGNMENT       128
  70
  71 #define SHORT_TYPE_SIZE         16
  72
  73 #define INT_TYPE_SIZE           32
  74
  75 #define LONG_TYPE_SIZE          (TARGET_ILP32 ? 32 : 64)
  76
  77 #define POINTER_SIZE            (TARGET_ILP32 ? 32 : 64)
  78
  79 #define LONG_LONG_TYPE_SIZE     64
  80
  81 #define FLOAT_TYPE_SIZE         32
  82
  83 #define DOUBLE_TYPE_SIZE        64
  84
  85 #define LONG_DOUBLE_TYPE_SIZE   128
  86
  87 /* The architecture reserves all bits of the address for hardware use,
  88    so the vbit must go into the delta field of pointers to member
  89    functions.  This is the same config as that in the AArch32
  90    port.  */
  91 #define TARGET_PTRMEMFUNC_VBIT_LOCATION ptrmemfunc_vbit_in_delta
  92
  93 /* Make strings word-aligned so that strcpy from constants will be
  94    faster.  */
  95 #define CONSTANT_ALIGNMENT(EXP, ALIGN)          \
  96   ((TREE_CODE (EXP) == STRING_CST               \
  97     && !optimize_size                           \
  98     && (ALIGN) < BITS_PER_WORD)                 \
  99    ? BITS_PER_WORD : ALIGN)
 100
 101 #define DATA_ALIGNMENT(EXP, ALIGN)              \
 102   ((((ALIGN) < BITS_PER_WORD)                   \
 103     && (TREE_CODE (EXP) == ARRAY_TYPE           \
 104         || TREE_CODE (EXP) == UNION_TYPE        \
 105         || TREE_CODE (EXP) == RECORD_TYPE))     \
 106    ? BITS_PER_WORD : (ALIGN))
 107
 108 #define LOCAL_ALIGNMENT(EXP, ALIGN) DATA_ALIGNMENT(EXP, ALIGN)
 109
 110 #define STRUCTURE_SIZE_BOUNDARY         8
 111
 112 /* Defined by the ABI */
 113 #define WCHAR_TYPE "unsigned int"
 114 #define WCHAR_TYPE_SIZE                 32
 115
 116 /* Using long long breaks -ansi and -std=c90, so these will need to be
 117    made conditional for an LLP64 ABI.  */
 118
 119 #define SIZE_TYPE       "long unsigned int"
 120
 121 #define PTRDIFF_TYPE    "long int"
 122
 123 #define PCC_BITFIELD_TYPE_MATTERS       1
 124
 125 /* Major revision number of the ARM Architecture implemented by the target.  */
 126 extern unsigned aarch64_architecture_version;
 127
 128 /* Instruction tuning/selection flags.  */
 129
 130 /* Bit values used to identify processor capabilities.  */
 131 #define AARCH64_FL_SIMD       (1 << 0)  /* Has SIMD instructions.  */
 132 #define AARCH64_FL_FP         (1 << 1)  /* Has FP.  */
 133 #define AARCH64_FL_CRYPTO     (1 << 2)  /* Has crypto.  */
 134 #define AARCH64_FL_CRC        (1 << 3)  /* Has CRC.  */
 135 /* ARMv8.1 architecture extensions.  */
 136 #define AARCH64_FL_LSE        (1 << 4)  /* Has Large System Extensions.  */
 137 #define AARCH64_FL_PAN        (1 << 5)  /* Has Privileged Access Never.  */
 138 #define AARCH64_FL_LOR        (1 << 6)  /* Has Limited Ordering regions.  */
 139 #define AARCH64_FL_RDMA       (1 << 7)  /* Has ARMv8.1 Adv.SIMD.  */
 140
 141 /* Has FP and SIMD.  */
 142 #define AARCH64_FL_FPSIMD     (AARCH64_FL_FP | AARCH64_FL_SIMD)
 143
 144 /* Has FP without SIMD.  */
 145 #define AARCH64_FL_FPQ16      (AARCH64_FL_FP & ~AARCH64_FL_SIMD)
 146
 147 /* Architecture flags that effect instruction selection.  */
 148 #define AARCH64_FL_FOR_ARCH8       (AARCH64_FL_FPSIMD)
 149 #define AARCH64_FL_FOR_ARCH8_1                         \
 150   (AARCH64_FL_FOR_ARCH8 | AARCH64_FL_LSE | AARCH64_FL_PAN \
 151    | AARCH64_FL_LOR | AARCH64_FL_RDMA)
 152
 153 /* Macros to test ISA flags.  */
 154
 155 #define AARCH64_ISA_CRC            (aarch64_isa_flags & AARCH64_FL_CRC)
 156 #define AARCH64_ISA_CRYPTO         (aarch64_isa_flags & AARCH64_FL_CRYPTO)
 157 #define AARCH64_ISA_FP             (aarch64_isa_flags & AARCH64_FL_FP)
 158 #define AARCH64_ISA_SIMD           (aarch64_isa_flags & AARCH64_FL_SIMD)
 159 #define AARCH64_ISA_LSE            (aarch64_isa_flags & AARCH64_FL_LSE)
 160
 161 /* Crypto is an optional extension to AdvSIMD.  */
 162 #define TARGET_CRYPTO (TARGET_SIMD && AARCH64_ISA_CRYPTO)
 163
 164 /* CRC instructions that can be enabled through +crc arch extension.  */
 165 #define TARGET_CRC32 (AARCH64_ISA_CRC)
 166
 167 /* Atomic instructions that can be enabled through the +lse extension.  */
 168 #define TARGET_LSE (AARCH64_ISA_LSE)
 169
 170 /* Make sure this is always defined so we don't have to check for ifdefs
 171    but rather use normal ifs.  */
 172 #ifndef TARGET_FIX_ERR_A53_835769_DEFAULT
 173 #define TARGET_FIX_ERR_A53_835769_DEFAULT 0
 174 #else
 175 #undef TARGET_FIX_ERR_A53_835769_DEFAULT
 176 #define TARGET_FIX_ERR_A53_835769_DEFAULT 1
 177 #endif
 178
 179 /* Apply the workaround for Cortex-A53 erratum 835769.  */
 180 #define TARGET_FIX_ERR_A53_835769       \
 181   ((aarch64_fix_a53_err835769 == 2)     \
 182   ? TARGET_FIX_ERR_A53_835769_DEFAULT : aarch64_fix_a53_err835769)
 183
 184 /* Standard register usage.  */
 185
 186 /* 31 64-bit general purpose registers R0-R30:
 187    R30          LR (link register)
 188    R29          FP (frame pointer)
 189    R19-R28      Callee-saved registers
 190    R18          The platform register; use as temporary register.
 191    R17          IP1 The second intra-procedure-call temporary register
 192                 (can be used by call veneers and PLT code); otherwise use
 193                 as a temporary register
 194    R16          IP0 The first intra-procedure-call temporary register (can
 195                 be used by call veneers and PLT code); otherwise use as a
 196                 temporary register
 197    R9-R15       Temporary registers
 198    R8           Structure value parameter / temporary register
 199    R0-R7        Parameter/result registers
 200
 201    SP           stack pointer, encoded as X/R31 where permitted.
 202    ZR           zero register, encoded as X/R31 elsewhere
 203
 204    32 x 128-bit floating-point/vector registers
 205    V16-V31      Caller-saved (temporary) registers
 206    V8-V15       Callee-saved registers
 207    V0-V7        Parameter/result registers
 208
 209    The vector register V0 holds scalar B0, H0, S0 and D0 in its least
 210    significant bits.  Unlike AArch32 S1 is not packed into D0,
 211    etc.  */
 212
 213 /* Note that we don't mark X30 as a call-clobbered register.  The idea is
 214    that it's really the call instructions themselves which clobber X30.
 215    We don't care what the called function does with it afterwards.
 216
 217    This approach makes it easier to implement sibcalls.  Unlike normal
 218    calls, sibcalls don't clobber X30, so the register reaches the
 219    called function intact.  EPILOGUE_USES says that X30 is useful
 220    to the called function.  */
 221
 222 #define FIXED_REGISTERS                                 \
 223   {                                                     \
 224     0, 0, 0, 0,   0, 0, 0, 0,   /* R0 - R7 */           \
 225     0, 0, 0, 0,   0, 0, 0, 0,   /* R8 - R15 */          \
 226     0, 0, 0, 0,   0, 0, 0, 0,   /* R16 - R23 */         \
 227     0, 0, 0, 0,   0, 1, 0, 1,   /* R24 - R30, SP */     \
 228     0, 0, 0, 0,   0, 0, 0, 0,   /* V0 - V7 */           \
 229     0, 0, 0, 0,   0, 0, 0, 0,   /* V8 - V15 */          \
 230     0, 0, 0, 0,   0, 0, 0, 0,   /* V16 - V23 */         \
 231     0, 0, 0, 0,   0, 0, 0, 0,   /* V24 - V31 */         \
 232     1, 1, 1,                    /* SFP, AP, CC */       \
 233   }
 234
 235 #define CALL_USED_REGISTERS                             \
 236   {                                                     \
 237     1, 1, 1, 1,   1, 1, 1, 1,   /* R0 - R7 */           \
 238     1, 1, 1, 1,   1, 1, 1, 1,   /* R8 - R15 */          \
 239     1, 1, 1, 0,   0, 0, 0, 0,   /* R16 - R23 */         \
 240     0, 0, 0, 0,   0, 1, 1, 1,   /* R24 - R30, SP */     \
 241     1, 1, 1, 1,   1, 1, 1, 1,   /* V0 - V7 */           \
 242     0, 0, 0, 0,   0, 0, 0, 0,   /* V8 - V15 */          \
 243     1, 1, 1, 1,   1, 1, 1, 1,   /* V16 - V23 */         \
 244     1, 1, 1, 1,   1, 1, 1, 1,   /* V24 - V31 */         \
 245     1, 1, 1,                    /* SFP, AP, CC */       \
 246   }
 247
 248 #define REGISTER_NAMES                                          \
 249   {                                                             \
 250     "x0",  "x1",  "x2",  "x3",  "x4",  "x5",  "x6",  "x7",      \
 251     "x8",  "x9",  "x10", "x11", "x12", "x13", "x14", "x15",     \
 252     "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23",     \
 253     "x24", "x25", "x26", "x27", "x28", "x29", "x30", "sp",      \
 254     "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",      \
 255     "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",     \
 256     "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",     \
 257     "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",     \
 258     "sfp", "ap",  "cc",                                         \
 259   }
 260
 261 /* Generate the register aliases for core register N */
 262 #define R_ALIASES(N) {"r" # N, R0_REGNUM + (N)}, \
 263                      {"w" # N, R0_REGNUM + (N)}
 264
 265 #define V_ALIASES(N) {"q" # N, V0_REGNUM + (N)}, \
 266                      {"d" # N, V0_REGNUM + (N)}, \
 267                      {"s" # N, V0_REGNUM + (N)}, \
 268                      {"h" # N, V0_REGNUM + (N)}, \
 269                      {"b" # N, V0_REGNUM + (N)}
 270
 271 /* Provide aliases for all of the ISA defined register name forms.
 272    These aliases are convenient for use in the clobber lists of inline
 273    asm statements.  */
 274
 275 #define ADDITIONAL_REGISTER_NAMES \
 276   { R_ALIASES(0),  R_ALIASES(1),  R_ALIASES(2),  R_ALIASES(3),  \
 277     R_ALIASES(4),  R_ALIASES(5),  R_ALIASES(6),  R_ALIASES(7),  \
 278     R_ALIASES(8),  R_ALIASES(9),  R_ALIASES(10), R_ALIASES(11), \
 279     R_ALIASES(12), R_ALIASES(13), R_ALIASES(14), R_ALIASES(15), \
 280     R_ALIASES(16), R_ALIASES(17), R_ALIASES(18), R_ALIASES(19), \
 281     R_ALIASES(20), R_ALIASES(21), R_ALIASES(22), R_ALIASES(23), \
 282     R_ALIASES(24), R_ALIASES(25), R_ALIASES(26), R_ALIASES(27), \
 283     R_ALIASES(28), R_ALIASES(29), R_ALIASES(30), {"wsp", R0_REGNUM + 31}, \
 284     V_ALIASES(0),  V_ALIASES(1),  V_ALIASES(2),  V_ALIASES(3),  \
 285     V_ALIASES(4),  V_ALIASES(5),  V_ALIASES(6),  V_ALIASES(7),  \
 286     V_ALIASES(8),  V_ALIASES(9),  V_ALIASES(10), V_ALIASES(11), \
 287     V_ALIASES(12), V_ALIASES(13), V_ALIASES(14), V_ALIASES(15), \
 288     V_ALIASES(16), V_ALIASES(17), V_ALIASES(18), V_ALIASES(19), \
 289     V_ALIASES(20), V_ALIASES(21), V_ALIASES(22), V_ALIASES(23), \
 290     V_ALIASES(24), V_ALIASES(25), V_ALIASES(26), V_ALIASES(27), \
 291     V_ALIASES(28), V_ALIASES(29), V_ALIASES(30), V_ALIASES(31)  \
 292   }
 293
 294 /* Say that the epilogue uses the return address register.  Note that
 295    in the case of sibcalls, the values "used by the epilogue" are
 296    considered live at the start of the called function.  */
 297
 298 #define EPILOGUE_USES(REGNO) \
 299   (epilogue_completed && (REGNO) == LR_REGNUM)
 300
 301 /* EXIT_IGNORE_STACK should be nonzero if, when returning from a function,
 302    the stack pointer does not matter.  The value is tested only in
 303    functions that have frame pointers.  */
 304 #define EXIT_IGNORE_STACK       1
 305
 306 #define STATIC_CHAIN_REGNUM             R18_REGNUM
 307 #define HARD_FRAME_POINTER_REGNUM       R29_REGNUM
 308 #define FRAME_POINTER_REGNUM            SFP_REGNUM
 309 #define STACK_POINTER_REGNUM            SP_REGNUM
 310 #define ARG_POINTER_REGNUM              AP_REGNUM
 311 #define FIRST_PSEUDO_REGISTER           67
 312
 313 /* The number of (integer) argument register available.  */
 314 #define NUM_ARG_REGS                    8
 315 #define NUM_FP_ARG_REGS                 8
 316
 317 /* A Homogeneous Floating-Point or Short-Vector Aggregate may have at most
 318    four members.  */
 319 #define HA_MAX_NUM_FLDS         4
 320
 321 /* External dwarf register number scheme.  These number are used to
 322    identify registers in dwarf debug information, the values are
 323    defined by the AArch64 ABI.  The numbering scheme is independent of
 324    GCC's internal register numbering scheme.  */
 325
 326 #define AARCH64_DWARF_R0        0
 327
 328 /* The number of R registers, note 31! not 32.  */
 329 #define AARCH64_DWARF_NUMBER_R 31
 330
 331 #define AARCH64_DWARF_SP       31
 332 #define AARCH64_DWARF_V0       64
 333
 334 /* The number of V registers.  */
 335 #define AARCH64_DWARF_NUMBER_V 32
 336
 337 /* For signal frames we need to use an alternative return column.  This
 338    value must not correspond to a hard register and must be out of the
 339    range of DWARF_FRAME_REGNUM().  */
 340 #define DWARF_ALT_FRAME_RETURN_COLUMN   \
 341   (AARCH64_DWARF_V0 + AARCH64_DWARF_NUMBER_V)
 342
 343 /* We add 1 extra frame register for use as the
 344    DWARF_ALT_FRAME_RETURN_COLUMN.  */
 345 #define DWARF_FRAME_REGISTERS           (DWARF_ALT_FRAME_RETURN_COLUMN + 1)
 346
 347
 348 #define DBX_REGISTER_NUMBER(REGNO)      aarch64_dbx_register_number (REGNO)
 349 /* Provide a definition of DWARF_FRAME_REGNUM here so that fallback unwinders
 350    can use DWARF_ALT_FRAME_RETURN_COLUMN defined below.  This is just the same
 351    as the default definition in dwarf2out.c.  */
 352 #undef DWARF_FRAME_REGNUM
 353 #define DWARF_FRAME_REGNUM(REGNO)       DBX_REGISTER_NUMBER (REGNO)
 354
 355 #define DWARF_FRAME_RETURN_COLUMN       DWARF_FRAME_REGNUM (LR_REGNUM)
 356
 357 #define HARD_REGNO_NREGS(REGNO, MODE)   aarch64_hard_regno_nregs (REGNO, MODE)
 358
 359 #define HARD_REGNO_MODE_OK(REGNO, MODE) aarch64_hard_regno_mode_ok (REGNO, MODE)
 360
 361 #define MODES_TIEABLE_P(MODE1, MODE2) aarch64_modes_tieable_p (MODE1, MODE2)
 362
 363 #define DWARF2_UNWIND_INFO 1
 364
 365 /* Use R0 through R3 to pass exception handling information.  */
 366 #define EH_RETURN_DATA_REGNO(N) \
 367   ((N) < 4 ? ((unsigned int) R0_REGNUM + (N)) : INVALID_REGNUM)
 368
 369 /* Select a format to encode pointers in exception handling data.  */
 370 #define ASM_PREFERRED_EH_DATA_FORMAT(CODE, GLOBAL) \
 371   aarch64_asm_preferred_eh_data_format ((CODE), (GLOBAL))
 372
 373 /* Output the assembly strings we want to add to a function definition.  */
 374 #define ASM_DECLARE_FUNCTION_NAME(STR, NAME, DECL)      \
 375   aarch64_declare_function_name (STR, NAME, DECL)
 376
 377 /* The register that holds the return address in exception handlers.  */
 378 #define AARCH64_EH_STACKADJ_REGNUM      (R0_REGNUM + 4)
 379 #define EH_RETURN_STACKADJ_RTX  gen_rtx_REG (Pmode, AARCH64_EH_STACKADJ_REGNUM)
 380
 381 /* Don't use __builtin_setjmp until we've defined it.  */
 382 #undef DONT_USE_BUILTIN_SETJMP
 383 #define DONT_USE_BUILTIN_SETJMP 1
 384
 385 /* Register in which the structure value is to be returned.  */
 386 #define AARCH64_STRUCT_VALUE_REGNUM R8_REGNUM
 387
 388 /* Non-zero if REGNO is part of the Core register set.
 389
 390    The rather unusual way of expressing this check is to avoid
 391    warnings when building the compiler when R0_REGNUM is 0 and REGNO
 392    is unsigned.  */
 393 #define GP_REGNUM_P(REGNO)                                              \
 394   (((unsigned) (REGNO - R0_REGNUM)) <= (R30_REGNUM - R0_REGNUM))
 395
 396 #define FP_REGNUM_P(REGNO)                      \
 397   (((unsigned) (REGNO - V0_REGNUM)) <= (V31_REGNUM - V0_REGNUM))
 398
 399 #define FP_LO_REGNUM_P(REGNO)            \
 400   (((unsigned) (REGNO - V0_REGNUM)) <= (V15_REGNUM - V0_REGNUM))
 401
 402 \f
 403 /* Register and constant classes.  */
 404
 405 enum reg_class
 406 {
 407   NO_REGS,
 408   CALLER_SAVE_REGS,
 409   GENERAL_REGS,
 410   STACK_REG,
 411   POINTER_REGS,
 412   FP_LO_REGS,
 413   FP_REGS,
 414   ALL_REGS,
 415   LIM_REG_CLASSES               /* Last */
 416 };
 417
 418 #define N_REG_CLASSES   ((int) LIM_REG_CLASSES)
 419
 420 #define REG_CLASS_NAMES                         \
 421 {                                               \
 422   "NO_REGS",                                    \
 423   "CALLER_SAVE_REGS",                           \
 424   "GENERAL_REGS",                               \
 425   "STACK_REG",                                  \
 426   "POINTER_REGS",                               \
 427   "FP_LO_REGS",                                 \
 428   "FP_REGS",                                    \
 429   "ALL_REGS"                                    \
 430 }
 431
 432 #define REG_CLASS_CONTENTS                                              \
 433 {                                                                       \
 434   { 0x00000000, 0x00000000, 0x00000000 },       /* NO_REGS */           \
 435   { 0x0007ffff, 0x00000000, 0x00000000 },       /* CALLER_SAVE_REGS */  \
 436   { 0x7fffffff, 0x00000000, 0x00000003 },       /* GENERAL_REGS */      \
 437   { 0x80000000, 0x00000000, 0x00000000 },       /* STACK_REG */         \
 438   { 0xffffffff, 0x00000000, 0x00000003 },       /* POINTER_REGS */      \
 439   { 0x00000000, 0x0000ffff, 0x00000000 },       /* FP_LO_REGS  */       \
 440   { 0x00000000, 0xffffffff, 0x00000000 },       /* FP_REGS  */          \
 441   { 0xffffffff, 0xffffffff, 0x00000007 }        /* ALL_REGS */          \
 442 }
 443
 444 #define REGNO_REG_CLASS(REGNO)  aarch64_regno_regclass (REGNO)
 445
 446 #define INDEX_REG_CLASS GENERAL_REGS
 447 #define BASE_REG_CLASS  POINTER_REGS
 448
 449 /* Register pairs used to eliminate unneeded registers that point into
 450    the stack frame.  */
 451 #define ELIMINABLE_REGS                                                 \
 452 {                                                                       \
 453   { ARG_POINTER_REGNUM,         STACK_POINTER_REGNUM            },      \
 454   { ARG_POINTER_REGNUM,         HARD_FRAME_POINTER_REGNUM       },      \
 455   { FRAME_POINTER_REGNUM,       STACK_POINTER_REGNUM            },      \
 456   { FRAME_POINTER_REGNUM,       HARD_FRAME_POINTER_REGNUM       },      \
 457 }
 458
 459 #define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \
 460   (OFFSET) = aarch64_initial_elimination_offset (FROM, TO)
 461
 462 /* CPU/ARCH option handling.  */
 463 #include "config/aarch64/aarch64-opts.h"
 464
 465 enum target_cpus
 466 {
 467 #define AARCH64_CORE(NAME, INTERNAL_IDENT, SCHED, ARCH, FLAGS, COSTS, IMP, PART) \
 468   TARGET_CPU_##INTERNAL_IDENT,
 469 #include "aarch64-cores.def"
 470 #undef AARCH64_CORE
 471   TARGET_CPU_generic
 472 };
 473
 474 /* If there is no CPU defined at configure, use generic as default.  */
 475 #ifndef TARGET_CPU_DEFAULT
 476 #define TARGET_CPU_DEFAULT \
 477   (TARGET_CPU_generic | (AARCH64_CPU_DEFAULT_FLAGS << 6))
 478 #endif
 479
 480 /* If inserting NOP before a mult-accumulate insn remember to adjust the
 481    length so that conditional branching code is updated appropriately.  */
 482 #define ADJUST_INSN_LENGTH(insn, length)        \
 483   do                                            \
 484     {                                           \
 485        if (aarch64_madd_needs_nop (insn))       \
 486          length += 4;                           \
 487     } while (0)
 488
 489 #define FINAL_PRESCAN_INSN(INSN, OPVEC, NOPERANDS)      \
 490     aarch64_final_prescan_insn (INSN);                  \
 491
 492 /* The processor for which instructions should be scheduled.  */
 493 extern enum aarch64_processor aarch64_tune;
 494
 495 /* RTL generation support.  */
 496 #define INIT_EXPANDERS aarch64_init_expanders ()
 497 \f
 498
 499 /* Stack layout; function entry, exit and calling.  */
 500 #define STACK_GROWS_DOWNWARD    1
 501
 502 #define FRAME_GROWS_DOWNWARD    1
 503
 504 #define STARTING_FRAME_OFFSET   0
 505
 506 #define ACCUMULATE_OUTGOING_ARGS        1
 507
 508 #define FIRST_PARM_OFFSET(FNDECL) 0
 509
 510 /* Fix for VFP */
 511 #define LIBCALL_VALUE(MODE)  \
 512   gen_rtx_REG (MODE, FLOAT_MODE_P (MODE) ? V0_REGNUM : R0_REGNUM)
 513
 514 #define DEFAULT_PCC_STRUCT_RETURN 0
 515
 516 #define AARCH64_ROUND_UP(X, ALIGNMENT) \
 517   (((X) + ((ALIGNMENT) - 1)) & ~((ALIGNMENT) - 1))
 518
 519 #define AARCH64_ROUND_DOWN(X, ALIGNMENT) \
 520   ((X) & ~((ALIGNMENT) - 1))
 521
 522 #ifdef HOST_WIDE_INT
 523 struct GTY (()) aarch64_frame
 524 {
 525   HOST_WIDE_INT reg_offset[FIRST_PSEUDO_REGISTER];
 526
 527   /* The number of extra stack bytes taken up by register varargs.
 528      This area is allocated by the callee at the very top of the
 529      frame.  This value is rounded up to a multiple of
 530      STACK_BOUNDARY.  */
 531   HOST_WIDE_INT saved_varargs_size;
 532
 533   HOST_WIDE_INT saved_regs_size;
 534   /* Padding if needed after the all the callee save registers have
 535      been saved.  */
 536   HOST_WIDE_INT padding0;
 537   HOST_WIDE_INT hardfp_offset;  /* HARD_FRAME_POINTER_REGNUM */
 538
 539   /* Offset from the base of the frame (incomming SP) to the
 540      hard_frame_pointer.  This value is always a multiple of
 541      STACK_BOUNDARY.  */
 542   HOST_WIDE_INT hard_fp_offset;
 543
 544   /* The size of the frame.  This value is the offset from base of the
 545    * frame (incomming SP) to the stack_pointer.  This value is always
 546    * a multiple of STACK_BOUNDARY.  */
 547
 548   unsigned wb_candidate1;
 549   unsigned wb_candidate2;
 550
 551   HOST_WIDE_INT frame_size;
 552
 553   bool laid_out;
 554 };
 555
 556 typedef struct GTY (()) machine_function
 557 {
 558   struct aarch64_frame frame;
 559 } machine_function;
 560 #endif
 561
 562 /* Which ABI to use.  */
 563 enum aarch64_abi_type
 564 {
 565   AARCH64_ABI_LP64 = 0,
 566   AARCH64_ABI_ILP32 = 1
 567 };
 568
 569 #ifndef AARCH64_ABI_DEFAULT
 570 #define AARCH64_ABI_DEFAULT AARCH64_ABI_LP64
 571 #endif
 572
 573 #define TARGET_ILP32    (aarch64_abi & AARCH64_ABI_ILP32)
 574
 575 enum arm_pcs
 576 {
 577   ARM_PCS_AAPCS64,              /* Base standard AAPCS for 64 bit.  */
 578   ARM_PCS_UNKNOWN
 579 };
 580
 581
 582
 583
 584 /* We can't use machine_mode inside a generator file because it
 585    hasn't been created yet; we shouldn't be using any code that
 586    needs the real definition though, so this ought to be safe.  */
 587 #ifdef GENERATOR_FILE
 588 #define MACHMODE int
 589 #else
 590 #include "insn-modes.h"
 591 #define MACHMODE machine_mode
 592 #endif
 593
 594 #ifndef USED_FOR_TARGET
 595 /* AAPCS related state tracking.  */
 596 typedef struct
 597 {
 598   enum arm_pcs pcs_variant;
 599   int aapcs_arg_processed;      /* No need to lay out this argument again.  */
 600   int aapcs_ncrn;               /* Next Core register number.  */
 601   int aapcs_nextncrn;           /* Next next core register number.  */
 602   int aapcs_nvrn;               /* Next Vector register number.  */
 603   int aapcs_nextnvrn;           /* Next Next Vector register number.  */
 604   rtx aapcs_reg;                /* Register assigned to this argument.  This
 605                                    is NULL_RTX if this parameter goes on
 606                                    the stack.  */
 607   MACHMODE aapcs_vfp_rmode;
 608   int aapcs_stack_words;        /* If the argument is passed on the stack, this
 609                                    is the number of words needed, after rounding
 610                                    up.  Only meaningful when
 611                                    aapcs_reg == NULL_RTX.  */
 612   int aapcs_stack_size;         /* The total size (in words, per 8 byte) of the
 613                                    stack arg area so far.  */
 614 } CUMULATIVE_ARGS;
 615 #endif
 616
 617 #define FUNCTION_ARG_PADDING(MODE, TYPE) \
 618   (aarch64_pad_arg_upward (MODE, TYPE) ? upward : downward)
 619
 620 #define BLOCK_REG_PADDING(MODE, TYPE, FIRST) \
 621   (aarch64_pad_reg_upward (MODE, TYPE, FIRST) ? upward : downward)
 622
 623 #define PAD_VARARGS_DOWN        0
 624
 625 #define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \
 626   aarch64_init_cumulative_args (&(CUM), FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS)
 627
 628 #define FUNCTION_ARG_REGNO_P(REGNO) \
 629   aarch64_function_arg_regno_p(REGNO)
 630 \f
 631
 632 /* ISA Features.  */
 633
 634 /* Addressing modes, etc.  */
 635 #define HAVE_POST_INCREMENT     1
 636 #define HAVE_PRE_INCREMENT      1
 637 #define HAVE_POST_DECREMENT     1
 638 #define HAVE_PRE_DECREMENT      1
 639 #define HAVE_POST_MODIFY_DISP   1
 640 #define HAVE_PRE_MODIFY_DISP    1
 641
 642 #define MAX_REGS_PER_ADDRESS    2
 643
 644 #define CONSTANT_ADDRESS_P(X)           aarch64_constant_address_p(X)
 645
 646 /* Try a machine-dependent way of reloading an illegitimate address
 647    operand.  If we find one, push the reload and jump to WIN.  This
 648    macro is used in only one place: `find_reloads_address' in reload.c.  */
 649
 650 #define LEGITIMIZE_RELOAD_ADDRESS(X, MODE, OPNUM, TYPE, IND_L, WIN)          \
 651 do {                                                                         \
 652   rtx new_x = aarch64_legitimize_reload_address (&(X), MODE, OPNUM, TYPE,    \
 653                                                  IND_L);                     \
 654   if (new_x)                                                                 \
 655     {                                                                        \
 656       X = new_x;                                                             \
 657       goto WIN;                                                              \
 658     }                                                                        \
 659 } while (0)
 660
 661 #define REGNO_OK_FOR_BASE_P(REGNO)      \
 662   aarch64_regno_ok_for_base_p (REGNO, true)
 663
 664 #define REGNO_OK_FOR_INDEX_P(REGNO) \
 665   aarch64_regno_ok_for_index_p (REGNO, true)
 666
 667 #define LEGITIMATE_PIC_OPERAND_P(X) \
 668   aarch64_legitimate_pic_operand_p (X)
 669
 670 #define CASE_VECTOR_MODE Pmode
 671
 672 #define DEFAULT_SIGNED_CHAR 0
 673
 674 /* An integer expression for the size in bits of the largest integer machine
 675    mode that should actually be used.  We allow pairs of registers.  */
 676 #define MAX_FIXED_MODE_SIZE GET_MODE_BITSIZE (TImode)
 677
 678 /* Maximum bytes moved by a single instruction (load/store pair).  */
 679 #define MOVE_MAX (UNITS_PER_WORD * 2)
 680
 681 /* The base cost overhead of a memcpy call, for MOVE_RATIO and friends.  */
 682 #define AARCH64_CALL_RATIO 8
 683
 684 /* MOVE_RATIO dictates when we will use the move_by_pieces infrastructure.
 685    move_by_pieces will continually copy the largest safe chunks.  So a
 686    7-byte copy is a 4-byte + 2-byte + byte copy.  This proves inefficient
 687    for both size and speed of copy, so we will instead use the "movmem"
 688    standard name to implement the copy.  This logic does not apply when
 689    targeting -mstrict-align, so keep a sensible default in that case.  */
 690 #define MOVE_RATIO(speed) \
 691   (!STRICT_ALIGNMENT ? 2 : (((speed) ? 15 : AARCH64_CALL_RATIO) / 2))
 692
 693 /* For CLEAR_RATIO, when optimizing for size, give a better estimate
 694    of the length of a memset call, but use the default otherwise.  */
 695 #define CLEAR_RATIO(speed) \
 696   ((speed) ? 15 : AARCH64_CALL_RATIO)
 697
 698 /* SET_RATIO is similar to CLEAR_RATIO, but for a non-zero constant, so when
 699    optimizing for size adjust the ratio to account for the overhead of loading
 700    the constant.  */
 701 #define SET_RATIO(speed) \
 702   ((speed) ? 15 : AARCH64_CALL_RATIO - 2)
 703
 704 /* Disable auto-increment in move_by_pieces et al.  Use of auto-increment is
 705    rarely a good idea in straight-line code since it adds an extra address
 706    dependency between each instruction.  Better to use incrementing offsets.  */
 707 #define USE_LOAD_POST_INCREMENT(MODE)   0
 708 #define USE_LOAD_POST_DECREMENT(MODE)   0
 709 #define USE_LOAD_PRE_INCREMENT(MODE)    0
 710 #define USE_LOAD_PRE_DECREMENT(MODE)    0
 711 #define USE_STORE_POST_INCREMENT(MODE)  0
 712 #define USE_STORE_POST_DECREMENT(MODE)  0
 713 #define USE_STORE_PRE_INCREMENT(MODE)   0
 714 #define USE_STORE_PRE_DECREMENT(MODE)   0
 715
 716 /* ?? #define WORD_REGISTER_OPERATIONS  */
 717
 718 /* Define if loading from memory in MODE, an integral mode narrower than
 719    BITS_PER_WORD will either zero-extend or sign-extend.  The value of this
 720    macro should be the code that says which one of the two operations is
 721    implicitly done, or UNKNOWN if none.  */
 722 #define LOAD_EXTEND_OP(MODE) ZERO_EXTEND
 723
 724 /* Define this macro to be non-zero if instructions will fail to work
 725    if given data not on the nominal alignment.  */
 726 #define STRICT_ALIGNMENT                TARGET_STRICT_ALIGN
 727
 728 /* Define this macro to be non-zero if accessing less than a word of
 729    memory is no faster than accessing a word of memory, i.e., if such
 730    accesses require more than one instruction or if there is no
 731    difference in cost.
 732    Although there's no difference in instruction count or cycles,
 733    in AArch64 we don't want to expand to a sub-word to a 64-bit access
 734    if we don't have to, for power-saving reasons.  */
 735 #define SLOW_BYTE_ACCESS                0
 736
 737 #define TRULY_NOOP_TRUNCATION(OUTPREC, INPREC) 1
 738
 739 #define NO_FUNCTION_CSE 1
 740
 741 /* Specify the machine mode that the hardware addresses have.
 742    After generation of rtl, the compiler makes no further distinction
 743    between pointers and any other objects of this machine mode.  */
 744 #define Pmode           DImode
 745
 746 /* A C expression whose value is zero if pointers that need to be extended
 747    from being `POINTER_SIZE' bits wide to `Pmode' are sign-extended and
 748    greater then zero if they are zero-extended and less then zero if the
 749    ptr_extend instruction should be used.  */
 750 #define POINTERS_EXTEND_UNSIGNED 1
 751
 752 /* Mode of a function address in a call instruction (for indexing purposes).  */
 753 #define FUNCTION_MODE   Pmode
 754
 755 #define SELECT_CC_MODE(OP, X, Y)        aarch64_select_cc_mode (OP, X, Y)
 756
 757 #define REVERSIBLE_CC_MODE(MODE) 1
 758
 759 #define REVERSE_CONDITION(CODE, MODE)           \
 760   (((MODE) == CCFPmode || (MODE) == CCFPEmode)  \
 761    ? reverse_condition_maybe_unordered (CODE)   \
 762    : reverse_condition (CODE))
 763
 764 #define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
 765   ((VALUE) = GET_MODE_UNIT_BITSIZE (MODE), 2)
 766 #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
 767   ((VALUE) = GET_MODE_UNIT_BITSIZE (MODE), 2)
 768
 769 #define INCOMING_RETURN_ADDR_RTX gen_rtx_REG (Pmode, LR_REGNUM)
 770
 771 #define RETURN_ADDR_RTX aarch64_return_addr
 772
 773 /* 3 insns + padding + 2 pointer-sized entries.  */
 774 #define TRAMPOLINE_SIZE (TARGET_ILP32 ? 24 : 32)
 775
 776 /* Trampolines contain dwords, so must be dword aligned.  */
 777 #define TRAMPOLINE_ALIGNMENT 64
 778
 779 /* Put trampolines in the text section so that mapping symbols work
 780    correctly.  */
 781 #define TRAMPOLINE_SECTION text_section
 782
 783 /* To start with.  */
 784 #define BRANCH_COST(SPEED_P, PREDICTABLE_P) \
 785   (aarch64_branch_cost (SPEED_P, PREDICTABLE_P))
 786 \f
 787
 788 /* Assembly output.  */
 789
 790 /* For now we'll make all jump tables pc-relative.  */
 791 #define CASE_VECTOR_PC_RELATIVE 1
 792
 793 #define CASE_VECTOR_SHORTEN_MODE(min, max, body)        \
 794   ((min < -0x1fff0 || max > 0x1fff0) ? SImode           \
 795    : (min < -0x1f0 || max > 0x1f0) ? HImode             \
 796    : QImode)
 797
 798 /* Jump table alignment is explicit in ASM_OUTPUT_CASE_LABEL.  */
 799 #define ADDR_VEC_ALIGN(JUMPTABLE) 0
 800
 801 #define PRINT_OPERAND(STREAM, X, CODE) aarch64_print_operand (STREAM, X, CODE)
 802
 803 #define PRINT_OPERAND_ADDRESS(STREAM, X) \
 804   aarch64_print_operand_address (STREAM, X)
 805
 806 #define MCOUNT_NAME "_mcount"
 807
 808 #define NO_PROFILE_COUNTERS 1
 809
 810 /* Emit rtl for profiling.  Output assembler code to FILE
 811    to call "_mcount" for profiling a function entry.  */
 812 #define PROFILE_HOOK(LABEL)                                             \
 813   {                                                                     \
 814     rtx fun, lr;                                                        \
 815     lr = get_hard_reg_initial_val (Pmode, LR_REGNUM);                   \
 816     fun = gen_rtx_SYMBOL_REF (Pmode, MCOUNT_NAME);                      \
 817     emit_library_call (fun, LCT_NORMAL, VOIDmode, 1, lr, Pmode);        \
 818   }
 819
 820 /* All the work done in PROFILE_HOOK, but still required.  */
 821 #define FUNCTION_PROFILER(STREAM, LABELNO) do { } while (0)
 822
 823 /* For some reason, the Linux headers think they know how to define
 824    these macros.  They don't!!!  */
 825 #undef ASM_APP_ON
 826 #undef ASM_APP_OFF
 827 #define ASM_APP_ON      "\t" ASM_COMMENT_START " Start of user assembly\n"
 828 #define ASM_APP_OFF     "\t" ASM_COMMENT_START " End of user assembly\n"
 829
 830 #define CONSTANT_POOL_BEFORE_FUNCTION 0
 831
 832 /* This definition should be relocated to aarch64-elf-raw.h.  This macro
 833    should be undefined in aarch64-linux.h and a clear_cache pattern
 834    implmented to emit either the call to __aarch64_sync_cache_range()
 835    directly or preferably the appropriate sycall or cache clear
 836    instructions inline.  */
 837 #define CLEAR_INSN_CACHE(beg, end)                              \
 838   extern void  __aarch64_sync_cache_range (void *, void *);     \
 839   __aarch64_sync_cache_range (beg, end)
 840
 841 #define SHIFT_COUNT_TRUNCATED !TARGET_SIMD
 842
 843 /* Choose appropriate mode for caller saves, so we do the minimum
 844    required size of load/store.  */
 845 #define HARD_REGNO_CALLER_SAVE_MODE(REGNO, NREGS, MODE) \
 846   aarch64_hard_regno_caller_save_mode ((REGNO), (NREGS), (MODE))
 847
 848 /* Callee only saves lower 64-bits of a 128-bit register.  Tell the
 849    compiler the callee clobbers the top 64-bits when restoring the
 850    bottom 64-bits.  */
 851 #define HARD_REGNO_CALL_PART_CLOBBERED(REGNO, MODE) \
 852                 (FP_REGNUM_P (REGNO) && GET_MODE_SIZE (MODE) > 8)
 853
 854 #undef SWITCHABLE_TARGET
 855 #define SWITCHABLE_TARGET 1
 856
 857 /* Check TLS Descriptors mechanism is selected.  */
 858 #define TARGET_TLS_DESC (aarch64_tls_dialect == TLS_DESCRIPTORS)
 859
 860 extern enum aarch64_code_model aarch64_cmodel;
 861
 862 /* When using the tiny addressing model conditional and unconditional branches
 863    can span the whole of the available address space (1MB).  */
 864 #define HAS_LONG_COND_BRANCH                            \
 865   (aarch64_cmodel == AARCH64_CMODEL_TINY                \
 866    || aarch64_cmodel == AARCH64_CMODEL_TINY_PIC)
 867
 868 #define HAS_LONG_UNCOND_BRANCH                          \
 869   (aarch64_cmodel == AARCH64_CMODEL_TINY                \
 870    || aarch64_cmodel == AARCH64_CMODEL_TINY_PIC)
 871
 872 /* Modes valid for AdvSIMD D registers, i.e. that fit in half a Q register.  */
 873 #define AARCH64_VALID_SIMD_DREG_MODE(MODE) \
 874   ((MODE) == V2SImode || (MODE) == V4HImode || (MODE) == V8QImode \
 875    || (MODE) == V2SFmode || (MODE) == V4HFmode || (MODE) == DImode \
 876    || (MODE) == DFmode)
 877
 878 /* Modes valid for AdvSIMD Q registers.  */
 879 #define AARCH64_VALID_SIMD_QREG_MODE(MODE) \
 880   ((MODE) == V4SImode || (MODE) == V8HImode || (MODE) == V16QImode \
 881    || (MODE) == V4SFmode || (MODE) == V8HFmode || (MODE) == V2DImode \
 882    || (MODE) == V2DFmode)
 883
 884 #define ENDIAN_LANE_N(mode, n)  \
 885   (BYTES_BIG_ENDIAN ? GET_MODE_NUNITS (mode) - 1 - n : n)
 886
 887 /* Support for a configure-time default CPU, etc.  We currently support
 888    --with-arch and --with-cpu.  Both are ignored if either is specified
 889    explicitly on the command line at run time.  */
 890 #define OPTION_DEFAULT_SPECS                            \
 891   {"arch", "%{!march=*:%{!mcpu=*:-march=%(VALUE)}}" },  \
 892   {"cpu",  "%{!march=*:%{!mcpu=*:-mcpu=%(VALUE)}}" },
 893
 894 #define MCPU_TO_MARCH_SPEC \
 895    " %{mcpu=*:-march=%:rewrite_mcpu(%{mcpu=*:%*})}"
 896
 897 extern const char *aarch64_rewrite_mcpu (int argc, const char **argv);
 898 #define MCPU_TO_MARCH_SPEC_FUNCTIONS \
 899   { "rewrite_mcpu", aarch64_rewrite_mcpu },
 900
 901 #if defined(__aarch64__)
 902 extern const char *host_detect_local_cpu (int argc, const char **argv);
 903 # define EXTRA_SPEC_FUNCTIONS                                           \
 904   { "local_cpu_detect", host_detect_local_cpu },                        \
 905   MCPU_TO_MARCH_SPEC_FUNCTIONS
 906
 907 # define MCPU_MTUNE_NATIVE_SPECS                                        \
 908    " %{march=native:%<march=native %:local_cpu_detect(arch)}"           \
 909    " %{mcpu=native:%<mcpu=native %:local_cpu_detect(cpu)}"              \
 910    " %{mtune=native:%<mtune=native %:local_cpu_detect(tune)}"
 911 #else
 912 # define MCPU_MTUNE_NATIVE_SPECS ""
 913 # define EXTRA_SPEC_FUNCTIONS MCPU_TO_MARCH_SPEC_FUNCTIONS
 914 #endif
 915
 916 #define ASM_CPU_SPEC \
 917    MCPU_TO_MARCH_SPEC
 918
 919 #define EXTRA_SPECS                                             \
 920   { "asm_cpu_spec",             ASM_CPU_SPEC }
 921
 922 #endif /* GCC_AARCH64_H */