Home | History | Annotate | Line # | Download | only in aarch64
      1 /* Machine description for AArch64 architecture.
      2    Copyright (C) 2009-2022 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 
     30 
     32 #define REGISTER_TARGET_PRAGMAS() aarch64_register_pragmas ()
     33 
     34 /* Target machine storage layout.  */
     35 
     36 #define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE)	\
     37   if (GET_MODE_CLASS (MODE) == MODE_INT		\
     38       && GET_MODE_SIZE (MODE) < 4)		\
     39     {						\
     40       if (MODE == QImode || MODE == HImode)	\
     41 	{					\
     42 	  MODE = SImode;			\
     43 	}					\
     44     }
     45 
     46 /* Bits are always numbered from the LSBit.  */
     47 #define BITS_BIG_ENDIAN 0
     48 
     49 /* Big/little-endian flavour.  */
     50 #define BYTES_BIG_ENDIAN (TARGET_BIG_END != 0)
     51 #define WORDS_BIG_ENDIAN (BYTES_BIG_ENDIAN)
     52 
     53 /* AdvSIMD is supported in the default configuration, unless disabled by
     54    -mgeneral-regs-only or by the +nosimd extension.  */
     55 #define TARGET_SIMD (!TARGET_GENERAL_REGS_ONLY && AARCH64_ISA_SIMD)
     56 #define TARGET_FLOAT (!TARGET_GENERAL_REGS_ONLY && AARCH64_ISA_FP)
     57 
     58 #define UNITS_PER_WORD		8
     59 
     60 #define UNITS_PER_VREG		16
     61 
     62 #define PARM_BOUNDARY		64
     63 
     64 #define STACK_BOUNDARY		128
     65 
     66 #define FUNCTION_BOUNDARY	32
     67 
     68 #define EMPTY_FIELD_BOUNDARY	32
     69 
     70 #define BIGGEST_ALIGNMENT	128
     71 
     72 #define SHORT_TYPE_SIZE		16
     73 
     74 #define INT_TYPE_SIZE		32
     75 
     76 #define LONG_TYPE_SIZE		(TARGET_ILP32 ? 32 : 64)
     77 
     78 #define POINTER_SIZE		(TARGET_ILP32 ? 32 : 64)
     79 
     80 #define LONG_LONG_TYPE_SIZE	64
     81 
     82 #define FLOAT_TYPE_SIZE		32
     83 
     84 #define DOUBLE_TYPE_SIZE	64
     85 
     86 #define LONG_DOUBLE_TYPE_SIZE	128
     87 
     88 /* This value is the amount of bytes a caller is allowed to drop the stack
     89    before probing has to be done for stack clash protection.  */
     90 #define STACK_CLASH_CALLER_GUARD 1024
     91 
     92 /* This value represents the minimum amount of bytes we expect the function's
     93    outgoing arguments to be when stack-clash is enabled.  */
     94 #define STACK_CLASH_MIN_BYTES_OUTGOING_ARGS 8
     95 
     96 /* This value controls how many pages we manually unroll the loop for when
     97    generating stack clash probes.  */
     98 #define STACK_CLASH_MAX_UNROLL_PAGES 4
     99 
    100 /* The architecture reserves all bits of the address for hardware use,
    101    so the vbit must go into the delta field of pointers to member
    102    functions.  This is the same config as that in the AArch32
    103    port.  */
    104 #define TARGET_PTRMEMFUNC_VBIT_LOCATION ptrmemfunc_vbit_in_delta
    105 
    106 
    107 /* Emit calls to libgcc helpers for atomic operations for runtime detection
    108    of LSE instructions.  */
    109 #define TARGET_OUTLINE_ATOMICS (aarch64_flag_outline_atomics)
    110 
    111 /* Align definitions of arrays, unions and structures so that
    112    initializations and copies can be made more efficient.  This is not
    113    ABI-changing, so it only affects places where we can see the
    114    definition.  Increasing the alignment tends to introduce padding,
    115    so don't do this when optimizing for size/conserving stack space.  */
    116 #define AARCH64_EXPAND_ALIGNMENT(COND, EXP, ALIGN)			\
    117   (((COND) && ((ALIGN) < BITS_PER_WORD)					\
    118     && (TREE_CODE (EXP) == ARRAY_TYPE					\
    119 	|| TREE_CODE (EXP) == UNION_TYPE				\
    120 	|| TREE_CODE (EXP) == RECORD_TYPE)) ? BITS_PER_WORD : (ALIGN))
    121 
    122 /* Align global data.  */
    123 #define DATA_ALIGNMENT(EXP, ALIGN)			\
    124   AARCH64_EXPAND_ALIGNMENT (!optimize_size, EXP, ALIGN)
    125 
    126 /* Similarly, make sure that objects on the stack are sensibly aligned.  */
    127 #define LOCAL_ALIGNMENT(EXP, ALIGN)				\
    128   AARCH64_EXPAND_ALIGNMENT (!flag_conserve_stack, EXP, ALIGN)
    129 
    130 #define STRUCTURE_SIZE_BOUNDARY		8
    131 
    132 /* Heap alignment (same as BIGGEST_ALIGNMENT and STACK_BOUNDARY).  */
    133 #define MALLOC_ABI_ALIGNMENT  128
    134 
    135 /* Defined by the ABI */
    136 #define WCHAR_TYPE "unsigned int"
    137 #define WCHAR_TYPE_SIZE			32
    138 
    139 /* Using long long breaks -ansi and -std=c90, so these will need to be
    140    made conditional for an LLP64 ABI.  */
    141 
    142 #define SIZE_TYPE	"long unsigned int"
    143 
    144 #define PTRDIFF_TYPE	"long int"
    145 
    146 #define PCC_BITFIELD_TYPE_MATTERS	1
    147 
    148 /* Major revision number of the ARM Architecture implemented by the target.  */
    149 extern unsigned aarch64_architecture_version;
    150 
    151 /* Instruction tuning/selection flags.  */
    152 
    153 /* Bit values used to identify processor capabilities.  */
    154 #define AARCH64_FL_SIMD       (1 << 0)	/* Has SIMD instructions.  */
    155 #define AARCH64_FL_FP         (1 << 1)	/* Has FP.  */
    156 #define AARCH64_FL_CRYPTO     (1 << 2)	/* Has crypto.  */
    157 #define AARCH64_FL_CRC        (1 << 3)	/* Has CRC.  */
    158 /* ARMv8.1-A architecture extensions.  */
    159 #define AARCH64_FL_LSE	      (1 << 4)  /* Has Large System Extensions.  */
    160 #define AARCH64_FL_RDMA       (1 << 5)  /* Has Round Double Multiply Add.  */
    161 #define AARCH64_FL_V8_1       (1 << 6)  /* Has ARMv8.1-A extensions.  */
    162 /* Armv8-R.  */
    163 #define AARCH64_FL_V8_R       (1 << 7)  /* Armv8-R AArch64.  */
    164 /* ARMv8.2-A architecture extensions.  */
    165 #define AARCH64_FL_V8_2       (1 << 8)  /* Has ARMv8.2-A features.  */
    166 #define AARCH64_FL_F16	      (1 << 9)  /* Has ARMv8.2-A FP16 extensions.  */
    167 #define AARCH64_FL_SVE        (1 << 10) /* Has Scalable Vector Extensions.  */
    168 /* ARMv8.3-A architecture extensions.  */
    169 #define AARCH64_FL_V8_3       (1 << 11)  /* Has ARMv8.3-A features.  */
    170 #define AARCH64_FL_RCPC       (1 << 12)  /* Has support for RCpc model.  */
    171 #define AARCH64_FL_DOTPROD    (1 << 13)  /* Has ARMv8.2-A Dot Product ins.  */
    172 /* New flags to split crypto into aes and sha2.  */
    173 #define AARCH64_FL_AES	      (1 << 14)  /* Has Crypto AES.  */
    174 #define AARCH64_FL_SHA2	      (1 << 15)  /* Has Crypto SHA2.  */
    175 /* ARMv8.4-A architecture extensions.  */
    176 #define AARCH64_FL_V8_4	      (1 << 16)  /* Has ARMv8.4-A features.  */
    177 #define AARCH64_FL_SM4	      (1 << 17)  /* Has ARMv8.4-A SM3 and SM4.  */
    178 #define AARCH64_FL_SHA3	      (1 << 18)  /* Has ARMv8.4-a SHA3 and SHA512.  */
    179 #define AARCH64_FL_F16FML     (1 << 19)  /* Has ARMv8.4-a FP16 extensions.  */
    180 #define AARCH64_FL_RCPC8_4    (1 << 20)  /* Has ARMv8.4-a RCPC extensions.  */
    181 
    182 /* Statistical Profiling extensions.  */
    183 #define AARCH64_FL_PROFILE    (1 << 21)
    184 
    185 /* ARMv8.5-A architecture extensions.  */
    186 #define AARCH64_FL_V8_5	      (1 << 22)  /* Has ARMv8.5-A features.  */
    187 #define AARCH64_FL_RNG	      (1 << 23)  /* ARMv8.5-A Random Number Insns.  */
    188 #define AARCH64_FL_MEMTAG     (1 << 24)  /* ARMv8.5-A Memory Tagging
    189 					    Extensions.  */
    190 
    191 /* Speculation Barrier instruction supported.  */
    192 #define AARCH64_FL_SB	      (1 << 25)
    193 
    194 /* Speculative Store Bypass Safe instruction supported.  */
    195 #define AARCH64_FL_SSBS	      (1 << 26)
    196 
    197 /* Execution and Data Prediction Restriction instructions supported.  */
    198 #define AARCH64_FL_PREDRES    (1 << 27)
    199 
    200 /* SVE2 instruction supported.  */
    201 #define AARCH64_FL_SVE2		(1 << 28)
    202 #define AARCH64_FL_SVE2_AES	(1 << 29)
    203 #define AARCH64_FL_SVE2_SM4	(1 << 30)
    204 #define AARCH64_FL_SVE2_SHA3	(1ULL << 31)
    205 #define AARCH64_FL_SVE2_BITPERM	(1ULL << 32)
    206 
    207 /* Transactional Memory Extension.  */
    208 #define AARCH64_FL_TME	      (1ULL << 33)  /* Has TME instructions.  */
    209 
    210 /* Armv8.6-A architecture extensions.  */
    211 #define AARCH64_FL_V8_6	      (1ULL << 34)
    212 
    213 /* 8-bit Integer Matrix Multiply (I8MM) extensions.  */
    214 #define AARCH64_FL_I8MM	      (1ULL << 35)
    215 
    216 /* Brain half-precision floating-point (BFloat16) Extension.  */
    217 #define AARCH64_FL_BF16	      (1ULL << 36)
    218 
    219 /* 32-bit Floating-point Matrix Multiply (F32MM) extensions.  */
    220 #define AARCH64_FL_F32MM      (1ULL << 37)
    221 
    222 /* 64-bit Floating-point Matrix Multiply (F64MM) extensions.  */
    223 #define AARCH64_FL_F64MM      (1ULL << 38)
    224 
    225 /* Flag Manipulation Instructions (FLAGM) extension.  */
    226 #define AARCH64_FL_FLAGM      (1ULL << 39)
    227 
    228 /* Pointer Authentication (PAUTH) extension.  */
    229 #define AARCH64_FL_PAUTH      (1ULL << 40)
    230 
    231 /* Armv9.0-A.  */
    232 #define AARCH64_FL_V9         (1ULL << 41)  /* Armv9.0-A Architecture.  */
    233 
    234 /* 64-byte atomic load/store extensions.  */
    235 #define AARCH64_FL_LS64      (1ULL << 42)
    236 
    237 /* Armv8.7-a architecture extensions.  */
    238 #define AARCH64_FL_V8_7       (1ULL << 43)
    239 
    240 /* Hardware memory operation instructions.  */
    241 #define AARCH64_FL_MOPS       (1ULL << 44)
    242 
    243 /* Armv8.8-a architecture extensions.  */
    244 #define AARCH64_FL_V8_8       (1ULL << 45)
    245 
    246 /* Has FP and SIMD.  */
    247 #define AARCH64_FL_FPSIMD     (AARCH64_FL_FP | AARCH64_FL_SIMD)
    248 
    249 /* Has FP without SIMD.  */
    250 #define AARCH64_FL_FPQ16      (AARCH64_FL_FP & ~AARCH64_FL_SIMD)
    251 
    252 /* Architecture flags that effect instruction selection.  */
    253 #define AARCH64_FL_FOR_ARCH8       (AARCH64_FL_FPSIMD)
    254 #define AARCH64_FL_FOR_ARCH8_1			       \
    255   (AARCH64_FL_FOR_ARCH8 | AARCH64_FL_LSE | AARCH64_FL_CRC \
    256    | AARCH64_FL_RDMA | AARCH64_FL_V8_1)
    257 #define AARCH64_FL_FOR_ARCH8_2			\
    258   (AARCH64_FL_FOR_ARCH8_1 | AARCH64_FL_V8_2)
    259 #define AARCH64_FL_FOR_ARCH8_3			\
    260   (AARCH64_FL_FOR_ARCH8_2 | AARCH64_FL_V8_3 | AARCH64_FL_PAUTH \
    261    | AARCH64_FL_RCPC)
    262 #define AARCH64_FL_FOR_ARCH8_4			\
    263   (AARCH64_FL_FOR_ARCH8_3 | AARCH64_FL_V8_4 | AARCH64_FL_F16FML \
    264    | AARCH64_FL_DOTPROD | AARCH64_FL_RCPC8_4 | AARCH64_FL_FLAGM)
    265 #define AARCH64_FL_FOR_ARCH8_5			\
    266   (AARCH64_FL_FOR_ARCH8_4 | AARCH64_FL_V8_5	\
    267    | AARCH64_FL_SB | AARCH64_FL_SSBS | AARCH64_FL_PREDRES)
    268 #define AARCH64_FL_FOR_ARCH8_6			\
    269   (AARCH64_FL_FOR_ARCH8_5 | AARCH64_FL_V8_6 | AARCH64_FL_FPSIMD \
    270    | AARCH64_FL_I8MM | AARCH64_FL_BF16)
    271 #define AARCH64_FL_FOR_ARCH8_7			\
    272   (AARCH64_FL_FOR_ARCH8_6 | AARCH64_FL_V8_7)
    273 #define AARCH64_FL_FOR_ARCH8_8			\
    274   (AARCH64_FL_FOR_ARCH8_7 | AARCH64_FL_V8_8 | AARCH64_FL_MOPS)
    275 
    276 #define AARCH64_FL_FOR_ARCH8_R     \
    277   (AARCH64_FL_FOR_ARCH8_4 | AARCH64_FL_V8_R)
    278 #define AARCH64_FL_FOR_ARCH9       \
    279   (AARCH64_FL_FOR_ARCH8_5 | AARCH64_FL_SVE | AARCH64_FL_SVE2 | AARCH64_FL_V9 \
    280    | AARCH64_FL_F16)
    281 
    282 /* Macros to test ISA flags.  */
    283 
    284 #define AARCH64_ISA_CRC            (aarch64_isa_flags & AARCH64_FL_CRC)
    285 #define AARCH64_ISA_CRYPTO         (aarch64_isa_flags & AARCH64_FL_CRYPTO)
    286 #define AARCH64_ISA_FP             (aarch64_isa_flags & AARCH64_FL_FP)
    287 #define AARCH64_ISA_SIMD           (aarch64_isa_flags & AARCH64_FL_SIMD)
    288 #define AARCH64_ISA_LSE		   (aarch64_isa_flags & AARCH64_FL_LSE)
    289 #define AARCH64_ISA_RDMA	   (aarch64_isa_flags & AARCH64_FL_RDMA)
    290 #define AARCH64_ISA_V8_2	   (aarch64_isa_flags & AARCH64_FL_V8_2)
    291 #define AARCH64_ISA_F16		   (aarch64_isa_flags & AARCH64_FL_F16)
    292 #define AARCH64_ISA_SVE            (aarch64_isa_flags & AARCH64_FL_SVE)
    293 #define AARCH64_ISA_SVE2	   (aarch64_isa_flags & AARCH64_FL_SVE2)
    294 #define AARCH64_ISA_SVE2_AES	   (aarch64_isa_flags & AARCH64_FL_SVE2_AES)
    295 #define AARCH64_ISA_SVE2_BITPERM  (aarch64_isa_flags & AARCH64_FL_SVE2_BITPERM)
    296 #define AARCH64_ISA_SVE2_SHA3	   (aarch64_isa_flags & AARCH64_FL_SVE2_SHA3)
    297 #define AARCH64_ISA_SVE2_SM4	   (aarch64_isa_flags & AARCH64_FL_SVE2_SM4)
    298 #define AARCH64_ISA_V8_3	   (aarch64_isa_flags & AARCH64_FL_V8_3)
    299 #define AARCH64_ISA_DOTPROD	   (aarch64_isa_flags & AARCH64_FL_DOTPROD)
    300 #define AARCH64_ISA_AES	           (aarch64_isa_flags & AARCH64_FL_AES)
    301 #define AARCH64_ISA_SHA2	   (aarch64_isa_flags & AARCH64_FL_SHA2)
    302 #define AARCH64_ISA_V8_4	   (aarch64_isa_flags & AARCH64_FL_V8_4)
    303 #define AARCH64_ISA_SM4	           (aarch64_isa_flags & AARCH64_FL_SM4)
    304 #define AARCH64_ISA_SHA3	   (aarch64_isa_flags & AARCH64_FL_SHA3)
    305 #define AARCH64_ISA_F16FML	   (aarch64_isa_flags & AARCH64_FL_F16FML)
    306 #define AARCH64_ISA_RCPC	   (aarch64_isa_flags & AARCH64_FL_RCPC)
    307 #define AARCH64_ISA_RCPC8_4	   (aarch64_isa_flags & AARCH64_FL_RCPC8_4)
    308 #define AARCH64_ISA_RNG		   (aarch64_isa_flags & AARCH64_FL_RNG)
    309 #define AARCH64_ISA_V8_5	   (aarch64_isa_flags & AARCH64_FL_V8_5)
    310 #define AARCH64_ISA_TME		   (aarch64_isa_flags & AARCH64_FL_TME)
    311 #define AARCH64_ISA_MEMTAG	   (aarch64_isa_flags & AARCH64_FL_MEMTAG)
    312 #define AARCH64_ISA_V8_6	   (aarch64_isa_flags & AARCH64_FL_V8_6)
    313 #define AARCH64_ISA_I8MM	   (aarch64_isa_flags & AARCH64_FL_I8MM)
    314 #define AARCH64_ISA_F32MM	   (aarch64_isa_flags & AARCH64_FL_F32MM)
    315 #define AARCH64_ISA_F64MM	   (aarch64_isa_flags & AARCH64_FL_F64MM)
    316 #define AARCH64_ISA_BF16	   (aarch64_isa_flags & AARCH64_FL_BF16)
    317 #define AARCH64_ISA_SB		   (aarch64_isa_flags & AARCH64_FL_SB)
    318 #define AARCH64_ISA_V8_R	   (aarch64_isa_flags & AARCH64_FL_V8_R)
    319 #define AARCH64_ISA_PAUTH	   (aarch64_isa_flags & AARCH64_FL_PAUTH)
    320 #define AARCH64_ISA_V9		   (aarch64_isa_flags & AARCH64_FL_V9)
    321 #define AARCH64_ISA_MOPS	   (aarch64_isa_flags & AARCH64_FL_MOPS)
    322 #define AARCH64_ISA_LS64	   (aarch64_isa_flags & AARCH64_FL_LS64)
    323 
    324 /* Crypto is an optional extension to AdvSIMD.  */
    325 #define TARGET_CRYPTO (TARGET_SIMD && AARCH64_ISA_CRYPTO)
    326 
    327 /* SHA2 is an optional extension to AdvSIMD.  */
    328 #define TARGET_SHA2 ((TARGET_SIMD && AARCH64_ISA_SHA2) || TARGET_CRYPTO)
    329 
    330 /* SHA3 is an optional extension to AdvSIMD.  */
    331 #define TARGET_SHA3 (TARGET_SIMD && AARCH64_ISA_SHA3)
    332 
    333 /* AES is an optional extension to AdvSIMD.  */
    334 #define TARGET_AES ((TARGET_SIMD && AARCH64_ISA_AES) || TARGET_CRYPTO)
    335 
    336 /* SM is an optional extension to AdvSIMD.  */
    337 #define TARGET_SM4 (TARGET_SIMD && AARCH64_ISA_SM4)
    338 
    339 /* FP16FML is an optional extension to AdvSIMD.  */
    340 #define TARGET_F16FML (TARGET_SIMD && AARCH64_ISA_F16FML && TARGET_FP_F16INST)
    341 
    342 /* CRC instructions that can be enabled through +crc arch extension.  */
    343 #define TARGET_CRC32 (AARCH64_ISA_CRC)
    344 
    345 /* Atomic instructions that can be enabled through the +lse extension.  */
    346 #define TARGET_LSE (AARCH64_ISA_LSE)
    347 
    348 /* ARMv8.2-A FP16 support that can be enabled through the +fp16 extension.  */
    349 #define TARGET_FP_F16INST (TARGET_FLOAT && AARCH64_ISA_F16)
    350 #define TARGET_SIMD_F16INST (TARGET_SIMD && AARCH64_ISA_F16)
    351 
    352 /* Dot Product is an optional extension to AdvSIMD enabled through +dotprod.  */
    353 #define TARGET_DOTPROD (TARGET_SIMD && AARCH64_ISA_DOTPROD)
    354 
    355 /* SVE instructions, enabled through +sve.  */
    356 #define TARGET_SVE (!TARGET_GENERAL_REGS_ONLY && AARCH64_ISA_SVE)
    357 
    358 /* SVE2 instructions, enabled through +sve2.  */
    359 #define TARGET_SVE2 (TARGET_SVE && AARCH64_ISA_SVE2)
    360 
    361 /* SVE2 AES instructions, enabled through +sve2-aes.  */
    362 #define TARGET_SVE2_AES (TARGET_SVE2 && AARCH64_ISA_SVE2_AES)
    363 
    364 /* SVE2 BITPERM instructions, enabled through +sve2-bitperm.  */
    365 #define TARGET_SVE2_BITPERM (TARGET_SVE2 && AARCH64_ISA_SVE2_BITPERM)
    366 
    367 /* SVE2 SHA3 instructions, enabled through +sve2-sha3.  */
    368 #define TARGET_SVE2_SHA3 (TARGET_SVE2 && AARCH64_ISA_SVE2_SHA3)
    369 
    370 /* SVE2 SM4 instructions, enabled through +sve2-sm4.  */
    371 #define TARGET_SVE2_SM4 (TARGET_SVE2 && AARCH64_ISA_SVE2_SM4)
    372 
    373 /* ARMv8.3-A features.  */
    374 #define TARGET_ARMV8_3	(AARCH64_ISA_V8_3)
    375 
    376 /* Javascript conversion instruction from Armv8.3-a.  */
    377 #define TARGET_JSCVT	(TARGET_FLOAT && AARCH64_ISA_V8_3)
    378 
    379 /* Armv8.3-a Complex number extension to AdvSIMD extensions.  */
    380 #define TARGET_COMPLEX (TARGET_SIMD && TARGET_ARMV8_3)
    381 
    382 /* Floating-point rounding instructions from Armv8.5-a.  */
    383 #define TARGET_FRINT (AARCH64_ISA_V8_5 && TARGET_FLOAT)
    384 
    385 /* TME instructions are enabled.  */
    386 #define TARGET_TME (AARCH64_ISA_TME)
    387 
    388 /* Random number instructions from Armv8.5-a.  */
    389 #define TARGET_RNG (AARCH64_ISA_RNG)
    390 
    391 /* Memory Tagging instructions optional to Armv8.5 enabled through +memtag.  */
    392 #define TARGET_MEMTAG (AARCH64_ISA_V8_5 && AARCH64_ISA_MEMTAG)
    393 
    394 /* I8MM instructions are enabled through +i8mm.  */
    395 #define TARGET_I8MM (AARCH64_ISA_I8MM)
    396 #define TARGET_SVE_I8MM (TARGET_SVE && AARCH64_ISA_I8MM)
    397 
    398 /* F32MM instructions are enabled through +f32mm.  */
    399 #define TARGET_F32MM (AARCH64_ISA_F32MM)
    400 #define TARGET_SVE_F32MM (TARGET_SVE && AARCH64_ISA_F32MM)
    401 
    402 /* F64MM instructions are enabled through +f64mm.  */
    403 #define TARGET_F64MM (AARCH64_ISA_F64MM)
    404 #define TARGET_SVE_F64MM (TARGET_SVE && AARCH64_ISA_F64MM)
    405 
    406 /* BF16 instructions are enabled through +bf16.  */
    407 #define TARGET_BF16_FP (AARCH64_ISA_BF16)
    408 #define TARGET_BF16_SIMD (AARCH64_ISA_BF16 && TARGET_SIMD)
    409 #define TARGET_SVE_BF16 (TARGET_SVE && AARCH64_ISA_BF16)
    410 
    411 /* PAUTH instructions are enabled through +pauth.  */
    412 #define TARGET_PAUTH (AARCH64_ISA_PAUTH)
    413 
    414 /* MOPS instructions are enabled through +mops.  */
    415 #define TARGET_MOPS (AARCH64_ISA_MOPS)
    416 
    417 /* LS64 instructions are enabled through +ls64.  */
    418 #define TARGET_LS64 (AARCH64_ISA_LS64)
    419 
    420 /* Make sure this is always defined so we don't have to check for ifdefs
    421    but rather use normal ifs.  */
    422 #ifndef TARGET_FIX_ERR_A53_835769_DEFAULT
    423 #define TARGET_FIX_ERR_A53_835769_DEFAULT 0
    424 #else
    425 #undef TARGET_FIX_ERR_A53_835769_DEFAULT
    426 #define TARGET_FIX_ERR_A53_835769_DEFAULT 1
    427 #endif
    428 
    429 /* SB instruction is enabled through +sb.  */
    430 #define TARGET_SB (AARCH64_ISA_SB)
    431 
    432 /* Apply the workaround for Cortex-A53 erratum 835769.  */
    433 #define TARGET_FIX_ERR_A53_835769	\
    434   ((aarch64_fix_a53_err835769 == 2)	\
    435   ? TARGET_FIX_ERR_A53_835769_DEFAULT : aarch64_fix_a53_err835769)
    436 
    437 /* Make sure this is always defined so we don't have to check for ifdefs
    438    but rather use normal ifs.  */
    439 #ifndef TARGET_FIX_ERR_A53_843419_DEFAULT
    440 #define TARGET_FIX_ERR_A53_843419_DEFAULT 0
    441 #else
    442 #undef TARGET_FIX_ERR_A53_843419_DEFAULT
    443 #define TARGET_FIX_ERR_A53_843419_DEFAULT 1
    444 #endif
    445 
    446 /* Apply the workaround for Cortex-A53 erratum 843419.  */
    447 #define TARGET_FIX_ERR_A53_843419	\
    448   ((aarch64_fix_a53_err843419 == 2)	\
    449   ? TARGET_FIX_ERR_A53_843419_DEFAULT : aarch64_fix_a53_err843419)
    450 
    451 /* ARMv8.1-A Adv.SIMD support.  */
    452 #define TARGET_SIMD_RDMA (TARGET_SIMD && AARCH64_ISA_RDMA)
    453 
    454 /* Standard register usage.  */
    455 
    456 /* 31 64-bit general purpose registers R0-R30:
    457    R30		LR (link register)
    458    R29		FP (frame pointer)
    459    R19-R28	Callee-saved registers
    460    R18		The platform register; use as temporary register.
    461    R17		IP1 The second intra-procedure-call temporary register
    462 		(can be used by call veneers and PLT code); otherwise use
    463 		as a temporary register
    464    R16		IP0 The first intra-procedure-call temporary register (can
    465 		be used by call veneers and PLT code); otherwise use as a
    466 		temporary register
    467    R9-R15	Temporary registers
    468    R8		Structure value parameter / temporary register
    469    R0-R7	Parameter/result registers
    470 
    471    SP		stack pointer, encoded as X/R31 where permitted.
    472    ZR		zero register, encoded as X/R31 elsewhere
    473 
    474    32 x 128-bit floating-point/vector registers
    475    V16-V31	Caller-saved (temporary) registers
    476    V8-V15	Callee-saved registers
    477    V0-V7	Parameter/result registers
    478 
    479    The vector register V0 holds scalar B0, H0, S0 and D0 in its least
    480    significant bits.  Unlike AArch32 S1 is not packed into D0, etc.
    481 
    482    P0-P7        Predicate low registers: valid in all predicate contexts
    483    P8-P15       Predicate high registers: used as scratch space
    484 
    485    FFR		First Fault Register, a fixed-use SVE predicate register
    486    FFRT		FFR token: a fake register used for modelling dependencies
    487 
    488    VG           Pseudo "vector granules" register
    489 
    490    VG is the number of 64-bit elements in an SVE vector.  We define
    491    it as a hard register so that we can easily map it to the DWARF VG
    492    register.  GCC internally uses the poly_int variable aarch64_sve_vg
    493    instead.  */
    494 
    495 #define FIXED_REGISTERS					\
    496   {							\
    497     0, 0, 0, 0,   0, 0, 0, 0,	/* R0 - R7 */		\
    498     0, 0, 0, 0,   0, 0, 0, 0,	/* R8 - R15 */		\
    499     0, 0, 0, 0,   0, 0, 0, 0,	/* R16 - R23 */		\
    500     0, 0, 0, 0,   0, 1, 0, 1,	/* R24 - R30, SP */	\
    501     0, 0, 0, 0,   0, 0, 0, 0,   /* V0 - V7 */           \
    502     0, 0, 0, 0,   0, 0, 0, 0,   /* V8 - V15 */		\
    503     0, 0, 0, 0,   0, 0, 0, 0,   /* V16 - V23 */         \
    504     0, 0, 0, 0,   0, 0, 0, 0,   /* V24 - V31 */         \
    505     1, 1, 1, 1,			/* SFP, AP, CC, VG */	\
    506     0, 0, 0, 0,   0, 0, 0, 0,   /* P0 - P7 */           \
    507     0, 0, 0, 0,   0, 0, 0, 0,   /* P8 - P15 */          \
    508     1, 1			/* FFR and FFRT */	\
    509   }
    510 
    511 /* X30 is marked as caller-saved which is in line with regular function call
    512    behavior since the call instructions clobber it; AARCH64_EXPAND_CALL does
    513    that for regular function calls and avoids it for sibcalls.  X30 is
    514    considered live for sibcalls; EPILOGUE_USES helps achieve that by returning
    515    true but not until function epilogues have been generated.  This ensures
    516    that X30 is available for use in leaf functions if needed.  */
    517 
    518 #define CALL_USED_REGISTERS				\
    519   {							\
    520     1, 1, 1, 1,   1, 1, 1, 1,	/* R0 - R7 */		\
    521     1, 1, 1, 1,   1, 1, 1, 1,	/* R8 - R15 */		\
    522     1, 1, 1, 0,   0, 0, 0, 0,	/* R16 - R23 */		\
    523     0, 0, 0, 0,   0, 1, 1, 1,	/* R24 - R30, SP */	\
    524     1, 1, 1, 1,   1, 1, 1, 1,	/* V0 - V7 */		\
    525     0, 0, 0, 0,   0, 0, 0, 0,	/* V8 - V15 */		\
    526     1, 1, 1, 1,   1, 1, 1, 1,   /* V16 - V23 */         \
    527     1, 1, 1, 1,   1, 1, 1, 1,   /* V24 - V31 */         \
    528     1, 1, 1, 1,			/* SFP, AP, CC, VG */	\
    529     1, 1, 1, 1,   1, 1, 1, 1,	/* P0 - P7 */		\
    530     1, 1, 1, 1,   1, 1, 1, 1,	/* P8 - P15 */		\
    531     1, 1			/* FFR and FFRT */	\
    532   }
    533 
    534 #define REGISTER_NAMES						\
    535   {								\
    536     "x0",  "x1",  "x2",  "x3",  "x4",  "x5",  "x6",  "x7",	\
    537     "x8",  "x9",  "x10", "x11", "x12", "x13", "x14", "x15",	\
    538     "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23",	\
    539     "x24", "x25", "x26", "x27", "x28", "x29", "x30", "sp",	\
    540     "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",	\
    541     "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",	\
    542     "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",	\
    543     "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",	\
    544     "sfp", "ap",  "cc",  "vg",					\
    545     "p0",  "p1",  "p2",  "p3",  "p4",  "p5",  "p6",  "p7",	\
    546     "p8",  "p9",  "p10", "p11", "p12", "p13", "p14", "p15",	\
    547     "ffr", "ffrt"						\
    548   }
    549 
    550 /* Generate the register aliases for core register N */
    551 #define R_ALIASES(N) {"r" # N, R0_REGNUM + (N)}, \
    552                      {"w" # N, R0_REGNUM + (N)}
    553 
    554 #define V_ALIASES(N) {"q" # N, V0_REGNUM + (N)}, \
    555                      {"d" # N, V0_REGNUM + (N)}, \
    556                      {"s" # N, V0_REGNUM + (N)}, \
    557                      {"h" # N, V0_REGNUM + (N)}, \
    558                      {"b" # N, V0_REGNUM + (N)}, \
    559                      {"z" # N, V0_REGNUM + (N)}
    560 
    561 /* Provide aliases for all of the ISA defined register name forms.
    562    These aliases are convenient for use in the clobber lists of inline
    563    asm statements.  */
    564 
    565 #define ADDITIONAL_REGISTER_NAMES \
    566   { R_ALIASES(0),  R_ALIASES(1),  R_ALIASES(2),  R_ALIASES(3),  \
    567     R_ALIASES(4),  R_ALIASES(5),  R_ALIASES(6),  R_ALIASES(7),  \
    568     R_ALIASES(8),  R_ALIASES(9),  R_ALIASES(10), R_ALIASES(11), \
    569     R_ALIASES(12), R_ALIASES(13), R_ALIASES(14), R_ALIASES(15), \
    570     R_ALIASES(16), R_ALIASES(17), R_ALIASES(18), R_ALIASES(19), \
    571     R_ALIASES(20), R_ALIASES(21), R_ALIASES(22), R_ALIASES(23), \
    572     R_ALIASES(24), R_ALIASES(25), R_ALIASES(26), R_ALIASES(27), \
    573     R_ALIASES(28), R_ALIASES(29), R_ALIASES(30), {"wsp", R0_REGNUM + 31}, \
    574     V_ALIASES(0),  V_ALIASES(1),  V_ALIASES(2),  V_ALIASES(3),  \
    575     V_ALIASES(4),  V_ALIASES(5),  V_ALIASES(6),  V_ALIASES(7),  \
    576     V_ALIASES(8),  V_ALIASES(9),  V_ALIASES(10), V_ALIASES(11), \
    577     V_ALIASES(12), V_ALIASES(13), V_ALIASES(14), V_ALIASES(15), \
    578     V_ALIASES(16), V_ALIASES(17), V_ALIASES(18), V_ALIASES(19), \
    579     V_ALIASES(20), V_ALIASES(21), V_ALIASES(22), V_ALIASES(23), \
    580     V_ALIASES(24), V_ALIASES(25), V_ALIASES(26), V_ALIASES(27), \
    581     V_ALIASES(28), V_ALIASES(29), V_ALIASES(30), V_ALIASES(31)  \
    582   }
    583 
    584 #define EPILOGUE_USES(REGNO) (aarch64_epilogue_uses (REGNO))
    585 
    586 /* EXIT_IGNORE_STACK should be nonzero if, when returning from a function,
    587    the stack pointer does not matter.  This is only true if the function
    588    uses alloca.  */
    589 #define EXIT_IGNORE_STACK	(cfun->calls_alloca)
    590 
    591 #define STATIC_CHAIN_REGNUM		R18_REGNUM
    592 #define HARD_FRAME_POINTER_REGNUM	R29_REGNUM
    593 #define FRAME_POINTER_REGNUM		SFP_REGNUM
    594 #define STACK_POINTER_REGNUM		SP_REGNUM
    595 #define ARG_POINTER_REGNUM		AP_REGNUM
    596 #define FIRST_PSEUDO_REGISTER		(FFRT_REGNUM + 1)
    597 
    598 /* The number of argument registers available for each class.  */
    599 #define NUM_ARG_REGS			8
    600 #define NUM_FP_ARG_REGS			8
    601 #define NUM_PR_ARG_REGS			4
    602 
    603 /* A Homogeneous Floating-Point or Short-Vector Aggregate may have at most
    604    four members.  */
    605 #define HA_MAX_NUM_FLDS		4
    606 
    607 /* External dwarf register number scheme.  These number are used to
    608    identify registers in dwarf debug information, the values are
    609    defined by the AArch64 ABI.  The numbering scheme is independent of
    610    GCC's internal register numbering scheme.  */
    611 
    612 #define AARCH64_DWARF_R0        0
    613 
    614 /* The number of R registers, note 31! not 32.  */
    615 #define AARCH64_DWARF_NUMBER_R 31
    616 
    617 #define AARCH64_DWARF_SP       31
    618 #define AARCH64_DWARF_VG       46
    619 #define AARCH64_DWARF_P0       48
    620 #define AARCH64_DWARF_V0       64
    621 
    622 /* The number of V registers.  */
    623 #define AARCH64_DWARF_NUMBER_V 32
    624 
    625 /* For signal frames we need to use an alternative return column.  This
    626    value must not correspond to a hard register and must be out of the
    627    range of DWARF_FRAME_REGNUM().  */
    628 #define DWARF_ALT_FRAME_RETURN_COLUMN   \
    629   (AARCH64_DWARF_V0 + AARCH64_DWARF_NUMBER_V)
    630 
    631 /* We add 1 extra frame register for use as the
    632    DWARF_ALT_FRAME_RETURN_COLUMN.  */
    633 #define DWARF_FRAME_REGISTERS           (DWARF_ALT_FRAME_RETURN_COLUMN + 1)
    634 
    635 
    636 #define DBX_REGISTER_NUMBER(REGNO)	aarch64_dbx_register_number (REGNO)
    637 /* Provide a definition of DWARF_FRAME_REGNUM here so that fallback unwinders
    638    can use DWARF_ALT_FRAME_RETURN_COLUMN defined below.  This is just the same
    639    as the default definition in dwarf2out.cc.  */
    640 #undef DWARF_FRAME_REGNUM
    641 #define DWARF_FRAME_REGNUM(REGNO)	DBX_REGISTER_NUMBER (REGNO)
    642 
    643 #define DWARF_FRAME_RETURN_COLUMN	DWARF_FRAME_REGNUM (LR_REGNUM)
    644 
    645 #define DWARF2_UNWIND_INFO 1
    646 
    647 /* Use R0 through R3 to pass exception handling information.  */
    648 #define EH_RETURN_DATA_REGNO(N) \
    649   ((N) < 4 ? ((unsigned int) R0_REGNUM + (N)) : INVALID_REGNUM)
    650 
    651 /* Select a format to encode pointers in exception handling data.  */
    652 #define ASM_PREFERRED_EH_DATA_FORMAT(CODE, GLOBAL) \
    653   aarch64_asm_preferred_eh_data_format ((CODE), (GLOBAL))
    654 
    655 /* Output the assembly strings we want to add to a function definition.  */
    656 #define ASM_DECLARE_FUNCTION_NAME(STR, NAME, DECL)	\
    657   aarch64_declare_function_name (STR, NAME, DECL)
    658 
    659 /* Output assembly strings for alias definition.  */
    660 #define ASM_OUTPUT_DEF_FROM_DECLS(STR, DECL, TARGET) \
    661   aarch64_asm_output_alias (STR, DECL, TARGET)
    662 
    663 /* Output assembly strings for undefined extern symbols.  */
    664 #undef ASM_OUTPUT_EXTERNAL
    665 #define ASM_OUTPUT_EXTERNAL(STR, DECL, NAME) \
    666   aarch64_asm_output_external (STR, DECL, NAME)
    667 
    668 /* Output assembly strings after .cfi_startproc is emitted.  */
    669 #define ASM_POST_CFI_STARTPROC  aarch64_post_cfi_startproc
    670 
    671 /* For EH returns X4 contains the stack adjustment.  */
    672 #define EH_RETURN_STACKADJ_RTX	gen_rtx_REG (Pmode, R4_REGNUM)
    673 #define EH_RETURN_HANDLER_RTX  aarch64_eh_return_handler_rtx ()
    674 
    675 /* Don't use __builtin_setjmp until we've defined it.  */
    676 #undef DONT_USE_BUILTIN_SETJMP
    677 #define DONT_USE_BUILTIN_SETJMP 1
    678 
    679 #undef TARGET_COMPUTE_FRAME_LAYOUT
    680 #define TARGET_COMPUTE_FRAME_LAYOUT aarch64_layout_frame
    681 
    682 /* Register in which the structure value is to be returned.  */
    683 #define AARCH64_STRUCT_VALUE_REGNUM R8_REGNUM
    684 
    685 /* Non-zero if REGNO is part of the Core register set.
    686 
    687    The rather unusual way of expressing this check is to avoid
    688    warnings when building the compiler when R0_REGNUM is 0 and REGNO
    689    is unsigned.  */
    690 #define GP_REGNUM_P(REGNO)						\
    691   (((unsigned) (REGNO - R0_REGNUM)) <= (R30_REGNUM - R0_REGNUM))
    692 
    693 /* Registers known to be preserved over a BL instruction.  This consists of the
    694    GENERAL_REGS without x16, x17, and x30.  The x30 register is changed by the
    695    BL instruction itself, while the x16 and x17 registers may be used by
    696    veneers which can be inserted by the linker.  */
    697 #define STUB_REGNUM_P(REGNO) \
    698   (GP_REGNUM_P (REGNO) \
    699    && (REGNO) != R16_REGNUM \
    700    && (REGNO) != R17_REGNUM \
    701    && (REGNO) != R30_REGNUM) \
    702 
    703 #define FP_REGNUM_P(REGNO)			\
    704   (((unsigned) (REGNO - V0_REGNUM)) <= (V31_REGNUM - V0_REGNUM))
    705 
    706 #define FP_LO_REGNUM_P(REGNO)            \
    707   (((unsigned) (REGNO - V0_REGNUM)) <= (V15_REGNUM - V0_REGNUM))
    708 
    709 #define FP_LO8_REGNUM_P(REGNO)            \
    710   (((unsigned) (REGNO - V0_REGNUM)) <= (V7_REGNUM - V0_REGNUM))
    711 
    712 #define PR_REGNUM_P(REGNO)\
    713   (((unsigned) (REGNO - P0_REGNUM)) <= (P15_REGNUM - P0_REGNUM))
    714 
    715 #define PR_LO_REGNUM_P(REGNO)\
    716   (((unsigned) (REGNO - P0_REGNUM)) <= (P7_REGNUM - P0_REGNUM))
    717 
    718 #define FP_SIMD_SAVED_REGNUM_P(REGNO)			\
    719   (((unsigned) (REGNO - V8_REGNUM)) <= (V23_REGNUM - V8_REGNUM))
    720 
    721 /* Register and constant classes.  */
    723 
    724 enum reg_class
    725 {
    726   NO_REGS,
    727   TAILCALL_ADDR_REGS,
    728   STUB_REGS,
    729   GENERAL_REGS,
    730   STACK_REG,
    731   POINTER_REGS,
    732   FP_LO8_REGS,
    733   FP_LO_REGS,
    734   FP_REGS,
    735   POINTER_AND_FP_REGS,
    736   PR_LO_REGS,
    737   PR_HI_REGS,
    738   PR_REGS,
    739   FFR_REGS,
    740   PR_AND_FFR_REGS,
    741   ALL_REGS,
    742   LIM_REG_CLASSES		/* Last */
    743 };
    744 
    745 #define N_REG_CLASSES	((int) LIM_REG_CLASSES)
    746 
    747 #define REG_CLASS_NAMES				\
    748 {						\
    749   "NO_REGS",					\
    750   "TAILCALL_ADDR_REGS",				\
    751   "STUB_REGS",					\
    752   "GENERAL_REGS",				\
    753   "STACK_REG",					\
    754   "POINTER_REGS",				\
    755   "FP_LO8_REGS",				\
    756   "FP_LO_REGS",					\
    757   "FP_REGS",					\
    758   "POINTER_AND_FP_REGS",			\
    759   "PR_LO_REGS",					\
    760   "PR_HI_REGS",					\
    761   "PR_REGS",					\
    762   "FFR_REGS",					\
    763   "PR_AND_FFR_REGS",				\
    764   "ALL_REGS"					\
    765 }
    766 
    767 #define REG_CLASS_CONTENTS						\
    768 {									\
    769   { 0x00000000, 0x00000000, 0x00000000 },	/* NO_REGS */		\
    770   { 0x00030000, 0x00000000, 0x00000000 },	/* TAILCALL_ADDR_REGS */\
    771   { 0x3ffcffff, 0x00000000, 0x00000000 },	/* STUB_REGS */		\
    772   { 0x7fffffff, 0x00000000, 0x00000003 },	/* GENERAL_REGS */	\
    773   { 0x80000000, 0x00000000, 0x00000000 },	/* STACK_REG */		\
    774   { 0xffffffff, 0x00000000, 0x00000003 },	/* POINTER_REGS */	\
    775   { 0x00000000, 0x000000ff, 0x00000000 },       /* FP_LO8_REGS  */	\
    776   { 0x00000000, 0x0000ffff, 0x00000000 },       /* FP_LO_REGS  */	\
    777   { 0x00000000, 0xffffffff, 0x00000000 },       /* FP_REGS  */		\
    778   { 0xffffffff, 0xffffffff, 0x00000003 },	/* POINTER_AND_FP_REGS */\
    779   { 0x00000000, 0x00000000, 0x00000ff0 },	/* PR_LO_REGS */	\
    780   { 0x00000000, 0x00000000, 0x000ff000 },	/* PR_HI_REGS */	\
    781   { 0x00000000, 0x00000000, 0x000ffff0 },	/* PR_REGS */		\
    782   { 0x00000000, 0x00000000, 0x00300000 },	/* FFR_REGS */		\
    783   { 0x00000000, 0x00000000, 0x003ffff0 },	/* PR_AND_FFR_REGS */	\
    784   { 0xffffffff, 0xffffffff, 0x000fffff }	/* ALL_REGS */		\
    785 }
    786 
    787 #define REGNO_REG_CLASS(REGNO)	aarch64_regno_regclass (REGNO)
    788 
    789 #define INDEX_REG_CLASS	GENERAL_REGS
    790 #define BASE_REG_CLASS  POINTER_REGS
    791 
    792 /* Register pairs used to eliminate unneeded registers that point into
    793    the stack frame.  */
    794 #define ELIMINABLE_REGS							\
    795 {									\
    796   { ARG_POINTER_REGNUM,		STACK_POINTER_REGNUM		},	\
    797   { ARG_POINTER_REGNUM,		HARD_FRAME_POINTER_REGNUM	},	\
    798   { FRAME_POINTER_REGNUM,	STACK_POINTER_REGNUM		},	\
    799   { FRAME_POINTER_REGNUM,	HARD_FRAME_POINTER_REGNUM	},	\
    800 }
    801 
    802 #define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \
    803   (OFFSET) = aarch64_initial_elimination_offset (FROM, TO)
    804 
    805 /* CPU/ARCH option handling.  */
    806 #include "config/aarch64/aarch64-opts.h"
    807 
    808 enum target_cpus
    809 {
    810 #define AARCH64_CORE(NAME, INTERNAL_IDENT, SCHED, ARCH, FLAGS, COSTS, IMP, PART, VARIANT) \
    811   TARGET_CPU_##INTERNAL_IDENT,
    812 #include "aarch64-cores.def"
    813   TARGET_CPU_generic
    814 };
    815 
    816 /* Define how many bits are used to represent the CPU in TARGET_CPU_DEFAULT.
    817    This needs to be big enough to fit the value of TARGET_CPU_generic.
    818    All bits after this are used to represent the AARCH64_CPU_DEFAULT_FLAGS.  */
    819 #define TARGET_CPU_NBITS 8
    820 #define TARGET_CPU_MASK ((1 << TARGET_CPU_NBITS) - 1)
    821 
    822 /* If there is no CPU defined at configure, use generic as default.  */
    823 #ifndef TARGET_CPU_DEFAULT
    824 #define TARGET_CPU_DEFAULT \
    825   (TARGET_CPU_generic | (AARCH64_CPU_DEFAULT_FLAGS << TARGET_CPU_NBITS))
    826 #endif
    827 
    828 /* If inserting NOP before a mult-accumulate insn remember to adjust the
    829    length so that conditional branching code is updated appropriately.  */
    830 #define ADJUST_INSN_LENGTH(insn, length)	\
    831   do						\
    832     {						\
    833        if (aarch64_madd_needs_nop (insn))	\
    834          length += 4;				\
    835     } while (0)
    836 
    837 #define FINAL_PRESCAN_INSN(INSN, OPVEC, NOPERANDS)	\
    838     aarch64_final_prescan_insn (INSN);			\
    839 
    840 /* The processor for which instructions should be scheduled.  */
    841 extern enum aarch64_processor aarch64_tune;
    842 
    843 /* RTL generation support.  */
    844 #define INIT_EXPANDERS aarch64_init_expanders ()
    845 
    846 
    848 /* Stack layout; function entry, exit and calling.  */
    849 #define STACK_GROWS_DOWNWARD	1
    850 
    851 #define FRAME_GROWS_DOWNWARD	1
    852 
    853 #define ACCUMULATE_OUTGOING_ARGS	1
    854 
    855 #define FIRST_PARM_OFFSET(FNDECL) 0
    856 
    857 /* Fix for VFP */
    858 #define LIBCALL_VALUE(MODE)  \
    859   gen_rtx_REG (MODE, FLOAT_MODE_P (MODE) ? V0_REGNUM : R0_REGNUM)
    860 
    861 #define DEFAULT_PCC_STRUCT_RETURN 0
    862 
    863 #ifdef HAVE_POLY_INT_H
    864 struct GTY (()) aarch64_frame
    865 {
    866   /* The offset from the bottom of the static frame (the bottom of the
    867      outgoing arguments) of each register save slot, or -2 if no save is
    868      needed.  */
    869   poly_int64 reg_offset[LAST_SAVED_REGNUM + 1];
    870 
    871   /* The number of extra stack bytes taken up by register varargs.
    872      This area is allocated by the callee at the very top of the
    873      frame.  This value is rounded up to a multiple of
    874      STACK_BOUNDARY.  */
    875   HOST_WIDE_INT saved_varargs_size;
    876 
    877   /* The number of bytes between the bottom of the static frame (the bottom
    878      of the outgoing arguments) and the bottom of the register save area.
    879      This value is always a multiple of STACK_BOUNDARY.  */
    880   poly_int64 bytes_below_saved_regs;
    881 
    882   /* The number of bytes between the bottom of the static frame (the bottom
    883      of the outgoing arguments) and the hard frame pointer.  This value is
    884      always a multiple of STACK_BOUNDARY.  */
    885   poly_int64 bytes_below_hard_fp;
    886 
    887   /* The number of bytes between the top of the locals area and the top
    888      of the frame (the incomming SP).  This value is always a multiple of
    889      STACK_BOUNDARY.  */
    890   poly_int64 bytes_above_locals;
    891 
    892   /* The number of bytes between the hard_frame_pointer and the top of
    893      the frame (the incomming SP).  This value is always a multiple of
    894      STACK_BOUNDARY.  */
    895   poly_int64 bytes_above_hard_fp;
    896 
    897   /* The size of the frame, i.e. the number of bytes between the bottom
    898      of the outgoing arguments and the incoming SP.  This value is always
    899      a multiple of STACK_BOUNDARY.  */
    900   poly_int64 frame_size;
    901 
    902   /* The size of the initial stack adjustment before saving callee-saves.  */
    903   poly_int64 initial_adjust;
    904 
    905   /* The writeback value when pushing callee-save registers.
    906      It is zero when no push is used.  */
    907   HOST_WIDE_INT callee_adjust;
    908 
    909   /* The size of the stack adjustment before saving or after restoring
    910      SVE registers.  */
    911   poly_int64 sve_callee_adjust;
    912 
    913   /* The size of the stack adjustment after saving callee-saves.  */
    914   poly_int64 final_adjust;
    915 
    916   /* Store FP,LR and setup a frame pointer.  */
    917   bool emit_frame_chain;
    918 
    919   /* In each frame, we can associate up to two register saves with the
    920      initial stack allocation.  This happens in one of two ways:
    921 
    922      (1) Using an STR or STP with writeback to perform the initial
    923 	 stack allocation.  When EMIT_FRAME_CHAIN, the registers will
    924 	 be those needed to create a frame chain.
    925 
    926 	 Indicated by CALLEE_ADJUST != 0.
    927 
    928      (2) Using a separate STP to set up the frame record, after the
    929 	 initial stack allocation but before setting up the frame pointer.
    930 	 This is used if the offset is too large to use writeback.
    931 
    932 	 Indicated by CALLEE_ADJUST == 0 && EMIT_FRAME_CHAIN.
    933 
    934      These fields indicate which registers we've decided to handle using
    935      (1) or (2), or INVALID_REGNUM if none.
    936 
    937      In some cases we don't always need to pop all registers in the push
    938      candidates, pop candidates record which registers need to be popped
    939      eventually.  The initial value of a pop candidate is copied from its
    940      corresponding push candidate.
    941 
    942      Currently, different pop candidates are only used for shadow call
    943      stack.  When "-fsanitize=shadow-call-stack" is specified, we replace
    944      x30 in the pop candidate with INVALID_REGNUM to ensure that x30 is
    945      not popped twice.  */
    946   unsigned wb_push_candidate1;
    947   unsigned wb_push_candidate2;
    948   unsigned wb_pop_candidate1;
    949   unsigned wb_pop_candidate2;
    950 
    951   /* Big-endian SVE frames need a spare predicate register in order
    952      to save vector registers in the correct layout for unwinding.
    953      This is the register they should use.  */
    954   unsigned spare_pred_reg;
    955 
    956   /* An SVE register that is saved below the hard frame pointer and that acts
    957      as a probe for later allocations, or INVALID_REGNUM if none.  */
    958   unsigned sve_save_and_probe;
    959 
    960   /* A register that is saved at the hard frame pointer and that acts
    961      as a probe for later allocations, or INVALID_REGNUM if none.  */
    962   unsigned hard_fp_save_and_probe;
    963 
    964   bool laid_out;
    965 
    966   /* True if shadow call stack should be enabled for the current function.  */
    967   bool is_scs_enabled;
    968 };
    969 
    970 #ifdef hash_set_h
    971 typedef struct GTY (()) machine_function
    972 {
    973   struct aarch64_frame frame;
    974   /* One entry for each hard register.  */
    975   bool reg_is_wrapped_separately[LAST_SAVED_REGNUM];
    976   /* One entry for each general purpose register.  */
    977   rtx call_via[SP_REGNUM];
    978   bool label_is_assembled;
    979   /* A set of all decls that have been passed to a vld1 intrinsic in the
    980      current function.  This is used to help guide the vector cost model.  */
    981   hash_set<tree> *vector_load_decls;
    982 } machine_function;
    983 #endif
    984 #endif
    985 
    986 /* Which ABI to use.  */
    987 enum aarch64_abi_type
    988 {
    989   AARCH64_ABI_LP64 = 0,
    990   AARCH64_ABI_ILP32 = 1
    991 };
    992 
    993 #ifndef AARCH64_ABI_DEFAULT
    994 #define AARCH64_ABI_DEFAULT AARCH64_ABI_LP64
    995 #endif
    996 
    997 #define TARGET_ILP32	(aarch64_abi & AARCH64_ABI_ILP32)
    998 
    999 enum arm_pcs
   1000 {
   1001   ARM_PCS_AAPCS64,		/* Base standard AAPCS for 64 bit.  */
   1002   ARM_PCS_SIMD,			/* For aarch64_vector_pcs functions.  */
   1003   ARM_PCS_SVE,			/* For functions that pass or return
   1004 				   values in SVE registers.  */
   1005   ARM_PCS_TLSDESC,		/* For targets of tlsdesc calls.  */
   1006   ARM_PCS_UNKNOWN
   1007 };
   1008 
   1009 
   1010 
   1011 
   1012 /* We can't use machine_mode inside a generator file because it
   1013    hasn't been created yet; we shouldn't be using any code that
   1014    needs the real definition though, so this ought to be safe.  */
   1015 #ifdef GENERATOR_FILE
   1016 #define MACHMODE int
   1017 #else
   1018 #include "insn-modes.h"
   1019 #define MACHMODE machine_mode
   1020 #endif
   1021 
   1022 #ifndef USED_FOR_TARGET
   1023 /* AAPCS related state tracking.  */
   1024 typedef struct
   1025 {
   1026   enum arm_pcs pcs_variant;
   1027   int aapcs_arg_processed;	/* No need to lay out this argument again.  */
   1028   int aapcs_ncrn;		/* Next Core register number.  */
   1029   int aapcs_nextncrn;		/* Next next core register number.  */
   1030   int aapcs_nvrn;		/* Next Vector register number.  */
   1031   int aapcs_nextnvrn;		/* Next Next Vector register number.  */
   1032   int aapcs_nprn;		/* Next Predicate register number.  */
   1033   int aapcs_nextnprn;		/* Next Next Predicate register number.  */
   1034   rtx aapcs_reg;		/* Register assigned to this argument.  This
   1035 				   is NULL_RTX if this parameter goes on
   1036 				   the stack.  */
   1037   MACHMODE aapcs_vfp_rmode;
   1038   int aapcs_stack_words;	/* If the argument is passed on the stack, this
   1039 				   is the number of words needed, after rounding
   1040 				   up.  Only meaningful when
   1041 				   aapcs_reg == NULL_RTX.  */
   1042   int aapcs_stack_size;		/* The total size (in words, per 8 byte) of the
   1043 				   stack arg area so far.  */
   1044   bool silent_p;		/* True if we should act silently, rather than
   1045 				   raise an error for invalid calls.  */
   1046 } CUMULATIVE_ARGS;
   1047 #endif
   1048 
   1049 #define BLOCK_REG_PADDING(MODE, TYPE, FIRST) \
   1050   (aarch64_pad_reg_upward (MODE, TYPE, FIRST) ? PAD_UPWARD : PAD_DOWNWARD)
   1051 
   1052 #define PAD_VARARGS_DOWN	0
   1053 
   1054 #define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \
   1055   aarch64_init_cumulative_args (&(CUM), FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS)
   1056 
   1057 #define FUNCTION_ARG_REGNO_P(REGNO) \
   1058   aarch64_function_arg_regno_p(REGNO)
   1059 
   1060 
   1062 /* ISA Features.  */
   1063 
   1064 /* Addressing modes, etc.  */
   1065 #define HAVE_POST_INCREMENT	1
   1066 #define HAVE_PRE_INCREMENT	1
   1067 #define HAVE_POST_DECREMENT	1
   1068 #define HAVE_PRE_DECREMENT	1
   1069 #define HAVE_POST_MODIFY_DISP	1
   1070 #define HAVE_PRE_MODIFY_DISP	1
   1071 
   1072 #define MAX_REGS_PER_ADDRESS	2
   1073 
   1074 #define CONSTANT_ADDRESS_P(X)		aarch64_constant_address_p(X)
   1075 
   1076 #define REGNO_OK_FOR_BASE_P(REGNO)	\
   1077   aarch64_regno_ok_for_base_p (REGNO, true)
   1078 
   1079 #define REGNO_OK_FOR_INDEX_P(REGNO) \
   1080   aarch64_regno_ok_for_index_p (REGNO, true)
   1081 
   1082 #define LEGITIMATE_PIC_OPERAND_P(X) \
   1083   aarch64_legitimate_pic_operand_p (X)
   1084 
   1085 #define CASE_VECTOR_MODE Pmode
   1086 
   1087 #define DEFAULT_SIGNED_CHAR 0
   1088 
   1089 /* An integer expression for the size in bits of the largest integer machine
   1090    mode that should actually be used.  We allow pairs of registers.  */
   1091 #define MAX_FIXED_MODE_SIZE GET_MODE_BITSIZE (TImode)
   1092 
   1093 /* Maximum bytes moved by a single instruction (load/store pair).  */
   1094 #define MOVE_MAX (UNITS_PER_WORD * 2)
   1095 
   1096 /* The base cost overhead of a memcpy call, for MOVE_RATIO and friends.  */
   1097 #define AARCH64_CALL_RATIO 8
   1098 
   1099 /* MOVE_RATIO dictates when we will use the move_by_pieces infrastructure.
   1100    move_by_pieces will continually copy the largest safe chunks.  So a
   1101    7-byte copy is a 4-byte + 2-byte + byte copy.  This proves inefficient
   1102    for both size and speed of copy, so we will instead use the "cpymem"
   1103    standard name to implement the copy.  This logic does not apply when
   1104    targeting -mstrict-align or TARGET_MOPS, so keep a sensible default in
   1105    that case.  */
   1106 #define MOVE_RATIO(speed) \
   1107   ((!STRICT_ALIGNMENT || TARGET_MOPS) ? 2 : (((speed) ? 15 : AARCH64_CALL_RATIO) / 2))
   1108 
   1109 /* Like MOVE_RATIO, without -mstrict-align, make decisions in "setmem" when
   1110    we would use more than 3 scalar instructions.
   1111    Otherwise follow a sensible default: when optimizing for size, give a better
   1112    estimate of the length of a memset call, but use the default otherwise.  */
   1113 #define CLEAR_RATIO(speed) \
   1114   (!STRICT_ALIGNMENT ? (TARGET_MOPS ? 0 : 4) : (speed) ? 15 : AARCH64_CALL_RATIO)
   1115 
   1116 /* SET_RATIO is similar to CLEAR_RATIO, but for a non-zero constant.  Without
   1117    -mstrict-align, make decisions in "setmem".  Otherwise follow a sensible
   1118    default: when optimizing for size adjust the ratio to account for the
   1119    overhead of loading the constant.  */
   1120 #define SET_RATIO(speed) \
   1121   ((!STRICT_ALIGNMENT || TARGET_MOPS) ? 0 : (speed) ? 15 : AARCH64_CALL_RATIO - 2)
   1122 
   1123 /* Disable auto-increment in move_by_pieces et al.  Use of auto-increment is
   1124    rarely a good idea in straight-line code since it adds an extra address
   1125    dependency between each instruction.  Better to use incrementing offsets.  */
   1126 #define USE_LOAD_POST_INCREMENT(MODE)   0
   1127 #define USE_LOAD_POST_DECREMENT(MODE)   0
   1128 #define USE_LOAD_PRE_INCREMENT(MODE)    0
   1129 #define USE_LOAD_PRE_DECREMENT(MODE)    0
   1130 #define USE_STORE_POST_INCREMENT(MODE)  0
   1131 #define USE_STORE_POST_DECREMENT(MODE)  0
   1132 #define USE_STORE_PRE_INCREMENT(MODE)   0
   1133 #define USE_STORE_PRE_DECREMENT(MODE)   0
   1134 
   1135 /* WORD_REGISTER_OPERATIONS does not hold for AArch64.
   1136    The assigned word_mode is DImode but operations narrower than SImode
   1137    behave as 32-bit operations if using the W-form of the registers rather
   1138    than as word_mode (64-bit) operations as WORD_REGISTER_OPERATIONS
   1139    expects.  */
   1140 #define WORD_REGISTER_OPERATIONS 0
   1141 
   1142 /* Define if loading from memory in MODE, an integral mode narrower than
   1143    BITS_PER_WORD will either zero-extend or sign-extend.  The value of this
   1144    macro should be the code that says which one of the two operations is
   1145    implicitly done, or UNKNOWN if none.  */
   1146 #define LOAD_EXTEND_OP(MODE) ZERO_EXTEND
   1147 
   1148 /* Define this macro to be non-zero if instructions will fail to work
   1149    if given data not on the nominal alignment.  */
   1150 #define STRICT_ALIGNMENT		TARGET_STRICT_ALIGN
   1151 
   1152 /* Enable wide bitfield accesses for more efficient bitfield code.  */
   1153 #define SLOW_BYTE_ACCESS 1
   1154 
   1155 #define NO_FUNCTION_CSE	1
   1156 
   1157 /* Specify the machine mode that the hardware addresses have.
   1158    After generation of rtl, the compiler makes no further distinction
   1159    between pointers and any other objects of this machine mode.  */
   1160 #define Pmode		DImode
   1161 
   1162 /* A C expression whose value is zero if pointers that need to be extended
   1163    from being `POINTER_SIZE' bits wide to `Pmode' are sign-extended and
   1164    greater then zero if they are zero-extended and less then zero if the
   1165    ptr_extend instruction should be used.  */
   1166 #define POINTERS_EXTEND_UNSIGNED 1
   1167 
   1168 /* Mode of a function address in a call instruction (for indexing purposes).  */
   1169 #define FUNCTION_MODE	Pmode
   1170 
   1171 #define SELECT_CC_MODE(OP, X, Y)	aarch64_select_cc_mode (OP, X, Y)
   1172 
   1173 /* Having an integer comparison mode guarantees that we can use
   1174    reverse_condition, but the usual restrictions apply to floating-point
   1175    comparisons.  */
   1176 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPmode && (MODE) != CCFPEmode)
   1177 
   1178 #define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
   1179   ((VALUE) = GET_MODE_UNIT_BITSIZE (MODE), 2)
   1180 #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
   1181   ((VALUE) = GET_MODE_UNIT_BITSIZE (MODE), 2)
   1182 
   1183 #define INCOMING_RETURN_ADDR_RTX gen_rtx_REG (Pmode, LR_REGNUM)
   1184 
   1185 #define RETURN_ADDR_RTX aarch64_return_addr
   1186 
   1187 /* BTI c + 3 insns
   1188    + sls barrier of DSB + ISB.
   1189    + 2 pointer-sized entries.  */
   1190 #define TRAMPOLINE_SIZE	(24 + (TARGET_ILP32 ? 8 : 16))
   1191 
   1192 /* Trampolines contain dwords, so must be dword aligned.  */
   1193 #define TRAMPOLINE_ALIGNMENT 64
   1194 
   1195 /* Put trampolines in the text section so that mapping symbols work
   1196    correctly.  */
   1197 #define TRAMPOLINE_SECTION text_section
   1198 
   1199 /* To start with.  */
   1200 #define BRANCH_COST(SPEED_P, PREDICTABLE_P) \
   1201   (aarch64_branch_cost (SPEED_P, PREDICTABLE_P))
   1202 
   1203 
   1205 /* Assembly output.  */
   1206 
   1207 /* For now we'll make all jump tables pc-relative.  */
   1208 #define CASE_VECTOR_PC_RELATIVE	1
   1209 
   1210 #define CASE_VECTOR_SHORTEN_MODE(min, max, body)	\
   1211   ((min < -0x1fff0 || max > 0x1fff0) ? SImode		\
   1212    : (min < -0x1f0 || max > 0x1f0) ? HImode		\
   1213    : QImode)
   1214 
   1215 /* Jump table alignment is explicit in ASM_OUTPUT_CASE_LABEL.  */
   1216 #define ADDR_VEC_ALIGN(JUMPTABLE) 0
   1217 
   1218 #define MCOUNT_NAME "_mcount"
   1219 
   1220 #define NO_PROFILE_COUNTERS 1
   1221 
   1222 /* Emit rtl for profiling.  Output assembler code to FILE
   1223    to call "_mcount" for profiling a function entry.  */
   1224 #define PROFILE_HOOK(LABEL)						\
   1225   {									\
   1226     rtx fun, lr;							\
   1227     lr = aarch64_return_addr_rtx ();					\
   1228     fun = gen_rtx_SYMBOL_REF (Pmode, MCOUNT_NAME);			\
   1229     emit_library_call (fun, LCT_NORMAL, VOIDmode, lr, Pmode);		\
   1230   }
   1231 
   1232 /* All the work done in PROFILE_HOOK, but still required.  */
   1233 #define FUNCTION_PROFILER(STREAM, LABELNO) do { } while (0)
   1234 
   1235 /* For some reason, the Linux headers think they know how to define
   1236    these macros.  They don't!!!  */
   1237 #undef ASM_APP_ON
   1238 #undef ASM_APP_OFF
   1239 #define ASM_APP_ON	"\t" ASM_COMMENT_START " Start of user assembly\n"
   1240 #define ASM_APP_OFF	"\t" ASM_COMMENT_START " End of user assembly\n"
   1241 
   1242 #define CONSTANT_POOL_BEFORE_FUNCTION 0
   1243 
   1244 /* This definition should be relocated to aarch64-elf-raw.h.  This macro
   1245    should be undefined in aarch64-linux.h and a clear_cache pattern
   1246    implmented to emit either the call to __aarch64_sync_cache_range()
   1247    directly or preferably the appropriate sycall or cache clear
   1248    instructions inline.  */
   1249 #define CLEAR_INSN_CACHE(beg, end)				\
   1250   extern void  __aarch64_sync_cache_range (void *, void *);	\
   1251   __aarch64_sync_cache_range (beg, end)
   1252 
   1253 #define SHIFT_COUNT_TRUNCATED (!TARGET_SIMD)
   1254 
   1255 /* Choose appropriate mode for caller saves, so we do the minimum
   1256    required size of load/store.  */
   1257 #define HARD_REGNO_CALLER_SAVE_MODE(REGNO, NREGS, MODE) \
   1258   aarch64_hard_regno_caller_save_mode ((REGNO), (NREGS), (MODE))
   1259 
   1260 #undef SWITCHABLE_TARGET
   1261 #define SWITCHABLE_TARGET 1
   1262 
   1263 /* Check TLS Descriptors mechanism is selected.  */
   1264 #define TARGET_TLS_DESC (aarch64_tls_dialect == TLS_DESCRIPTORS)
   1265 
   1266 extern enum aarch64_code_model aarch64_cmodel;
   1267 
   1268 /* When using the tiny addressing model conditional and unconditional branches
   1269    can span the whole of the available address space (1MB).  */
   1270 #define HAS_LONG_COND_BRANCH				\
   1271   (aarch64_cmodel == AARCH64_CMODEL_TINY		\
   1272    || aarch64_cmodel == AARCH64_CMODEL_TINY_PIC)
   1273 
   1274 #define HAS_LONG_UNCOND_BRANCH				\
   1275   (aarch64_cmodel == AARCH64_CMODEL_TINY		\
   1276    || aarch64_cmodel == AARCH64_CMODEL_TINY_PIC)
   1277 
   1278 #define TARGET_SUPPORTS_WIDE_INT 1
   1279 
   1280 /* Modes valid for AdvSIMD D registers, i.e. that fit in half a Q register.  */
   1281 #define AARCH64_VALID_SIMD_DREG_MODE(MODE) \
   1282   ((MODE) == V2SImode || (MODE) == V4HImode || (MODE) == V8QImode \
   1283    || (MODE) == V2SFmode || (MODE) == V4HFmode || (MODE) == DImode \
   1284    || (MODE) == DFmode || (MODE) == V4BFmode)
   1285 
   1286 /* Modes valid for AdvSIMD Q registers.  */
   1287 #define AARCH64_VALID_SIMD_QREG_MODE(MODE) \
   1288   ((MODE) == V4SImode || (MODE) == V8HImode || (MODE) == V16QImode \
   1289    || (MODE) == V4SFmode || (MODE) == V8HFmode || (MODE) == V2DImode \
   1290    || (MODE) == V2DFmode || (MODE) == V8BFmode)
   1291 
   1292 #define ENDIAN_LANE_N(NUNITS, N) \
   1293   (BYTES_BIG_ENDIAN ? NUNITS - 1 - N : N)
   1294 
   1295 /* Support for configure-time --with-arch, --with-cpu and --with-tune.
   1296    --with-arch and --with-cpu are ignored if either -mcpu or -march is used.
   1297    --with-tune is ignored if either -mtune or -mcpu is used (but is not
   1298    affected by -march).  */
   1299 #define OPTION_DEFAULT_SPECS				\
   1300   {"arch", "%{!march=*:%{!mcpu=*:-march=%(VALUE)}}" },	\
   1301   {"cpu",  "%{!march=*:%{!mcpu=*:-mcpu=%(VALUE)}}" },   \
   1302   {"tune", "%{!mcpu=*:%{!mtune=*:-mtune=%(VALUE)}}"},
   1303 
   1304 #define MCPU_TO_MARCH_SPEC \
   1305    " %{mcpu=*:-march=%:rewrite_mcpu(%{mcpu=*:%*})}"
   1306 
   1307 extern const char *aarch64_rewrite_mcpu (int argc, const char **argv);
   1308 #define MCPU_TO_MARCH_SPEC_FUNCTIONS \
   1309   { "rewrite_mcpu", aarch64_rewrite_mcpu },
   1310 
   1311 #if defined(__aarch64__) && (defined(__linux__) || defined(__NetBSD__))
   1312 extern const char *host_detect_local_cpu (int argc, const char **argv);
   1313 #define HAVE_LOCAL_CPU_DETECT
   1314 # define EXTRA_SPEC_FUNCTIONS						\
   1315   { "local_cpu_detect", host_detect_local_cpu },			\
   1316   MCPU_TO_MARCH_SPEC_FUNCTIONS
   1317 
   1318 # define MCPU_MTUNE_NATIVE_SPECS					\
   1319    " %{march=native:%<march=native %:local_cpu_detect(arch)}"		\
   1320    " %{mcpu=native:%<mcpu=native %:local_cpu_detect(cpu)}"		\
   1321    " %{mtune=native:%<mtune=native %:local_cpu_detect(tune)}"
   1322 #else
   1323 # define MCPU_MTUNE_NATIVE_SPECS ""
   1324 # define EXTRA_SPEC_FUNCTIONS MCPU_TO_MARCH_SPEC_FUNCTIONS
   1325 #endif
   1326 
   1327 #define ASM_CPU_SPEC \
   1328    MCPU_TO_MARCH_SPEC
   1329 
   1330 #define EXTRA_SPECS						\
   1331   { "asm_cpu_spec",		ASM_CPU_SPEC }
   1332 
   1333 #define ASM_OUTPUT_POOL_EPILOGUE  aarch64_asm_output_pool_epilogue
   1334 
   1335 /* This type is the user-visible __fp16, and a pointer to that type.  We
   1336    need it in many places in the backend.  Defined in aarch64-builtins.cc.  */
   1337 extern GTY(()) tree aarch64_fp16_type_node;
   1338 extern GTY(()) tree aarch64_fp16_ptr_type_node;
   1339 
   1340 /* This type is the user-visible __bf16, and a pointer to that type.  Defined
   1341    in aarch64-builtins.cc.  */
   1342 extern GTY(()) tree aarch64_bf16_type_node;
   1343 extern GTY(()) tree aarch64_bf16_ptr_type_node;
   1344 
   1345 /* The generic unwind code in libgcc does not initialize the frame pointer.
   1346    So in order to unwind a function using a frame pointer, the very first
   1347    function that is unwound must save the frame pointer.  That way the frame
   1348    pointer is restored and its value is now valid - otherwise _Unwind_GetGR
   1349    crashes.  Libgcc can now be safely built with -fomit-frame-pointer.  */
   1350 #define LIBGCC2_UNWIND_ATTRIBUTE \
   1351   __attribute__((optimize ("no-omit-frame-pointer")))
   1352 
   1353 #ifndef USED_FOR_TARGET
   1354 extern poly_uint16 aarch64_sve_vg;
   1355 
   1356 /* The number of bits and bytes in an SVE vector.  */
   1357 #define BITS_PER_SVE_VECTOR (poly_uint16 (aarch64_sve_vg * 64))
   1358 #define BYTES_PER_SVE_VECTOR (poly_uint16 (aarch64_sve_vg * 8))
   1359 
   1360 /* The number of bits and bytes in an SVE predicate.  */
   1361 #define BITS_PER_SVE_PRED BYTES_PER_SVE_VECTOR
   1362 #define BYTES_PER_SVE_PRED aarch64_sve_vg
   1363 
   1364 /* The SVE mode for a vector of bytes.  */
   1365 #define SVE_BYTE_MODE VNx16QImode
   1366 
   1367 /* The maximum number of bytes in a fixed-size vector.  This is 256 bytes
   1368    (for -msve-vector-bits=2048) multiplied by the maximum number of
   1369    vectors in a structure mode (4).
   1370 
   1371    This limit must not be used for variable-size vectors, since
   1372    VL-agnostic code must work with arbitary vector lengths.  */
   1373 #define MAX_COMPILE_TIME_VEC_BYTES (256 * 4)
   1374 #endif
   1375 
   1376 #define REGMODE_NATURAL_SIZE(MODE) aarch64_regmode_natural_size (MODE)
   1377 
   1378 /* Allocate a minimum of STACK_CLASH_MIN_BYTES_OUTGOING_ARGS bytes for the
   1379    outgoing arguments if stack clash protection is enabled.  This is essential
   1380    as the extra arg space allows us to skip a check in alloca.  */
   1381 #undef STACK_DYNAMIC_OFFSET
   1382 #define STACK_DYNAMIC_OFFSET(FUNDECL)			   \
   1383    ((flag_stack_clash_protection			   \
   1384      && cfun->calls_alloca				   \
   1385      && known_lt (crtl->outgoing_args_size,		   \
   1386 		  STACK_CLASH_MIN_BYTES_OUTGOING_ARGS))    \
   1387     ? ROUND_UP (STACK_CLASH_MIN_BYTES_OUTGOING_ARGS,       \
   1388 		STACK_BOUNDARY / BITS_PER_UNIT)		   \
   1389     : (crtl->outgoing_args_size + STACK_POINTER_OFFSET))
   1390 
   1391 #endif /* GCC_AARCH64_H */
   1392