1 /* Machine description for AArch64 architecture.
2 Copyright (C) 2009-2017 Free Software Foundation, Inc.
3 Contributed by ARM Ltd.
5 This file is part of GCC.
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)
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.
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/>. */
25 /* Target CPU builtins. */
26 #define TARGET_CPU_CPP_BUILTINS() \
27 aarch64_cpu_cpp_builtins (pfile)
31 #define REGISTER_TARGET_PRAGMAS() aarch64_register_pragmas ()
33 /* Target machine storage layout. */
35 #define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE) \
36 if (GET_MODE_CLASS (MODE) == MODE_INT \
37 && GET_MODE_SIZE (MODE) < 4) \
39 if (MODE == QImode || MODE == HImode) \
45 /* Bits are always numbered from the LSBit. */
46 #define BITS_BIG_ENDIAN 0
48 /* Big/little-endian flavour. */
49 #define BYTES_BIG_ENDIAN (TARGET_BIG_END != 0)
50 #define WORDS_BIG_ENDIAN (BYTES_BIG_ENDIAN)
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)
57 #define UNITS_PER_WORD 8
59 #define UNITS_PER_VREG 16
61 #define PARM_BOUNDARY 64
63 #define STACK_BOUNDARY 128
65 #define FUNCTION_BOUNDARY 32
67 #define EMPTY_FIELD_BOUNDARY 32
69 #define BIGGEST_ALIGNMENT 128
71 #define SHORT_TYPE_SIZE 16
73 #define INT_TYPE_SIZE 32
75 #define LONG_TYPE_SIZE (TARGET_ILP32 ? 32 : 64)
77 #define POINTER_SIZE (TARGET_ILP32 ? 32 : 64)
79 #define LONG_LONG_TYPE_SIZE 64
81 #define FLOAT_TYPE_SIZE 32
83 #define DOUBLE_TYPE_SIZE 64
85 #define LONG_DOUBLE_TYPE_SIZE 128
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
91 #define TARGET_PTRMEMFUNC_VBIT_LOCATION ptrmemfunc_vbit_in_delta
93 /* Make strings word-aligned so that strcpy from constants will be
95 #define CONSTANT_ALIGNMENT(EXP, ALIGN) \
96 ((TREE_CODE (EXP) == STRING_CST \
98 && (ALIGN) < BITS_PER_WORD) \
99 ? BITS_PER_WORD : ALIGN)
101 /* Align definitions of arrays, unions and structures so that
102 initializations and copies can be made more efficient. This is not
103 ABI-changing, so it only affects places where we can see the
104 definition. Increasing the alignment tends to introduce padding,
105 so don't do this when optimizing for size/conserving stack space. */
106 #define AARCH64_EXPAND_ALIGNMENT(COND, EXP, ALIGN) \
107 (((COND) && ((ALIGN) < BITS_PER_WORD) \
108 && (TREE_CODE (EXP) == ARRAY_TYPE \
109 || TREE_CODE (EXP) == UNION_TYPE \
110 || TREE_CODE (EXP) == RECORD_TYPE)) ? BITS_PER_WORD : (ALIGN))
112 /* Align global data. */
113 #define DATA_ALIGNMENT(EXP, ALIGN) \
114 AARCH64_EXPAND_ALIGNMENT (!optimize_size, EXP, ALIGN)
116 /* Similarly, make sure that objects on the stack are sensibly aligned. */
117 #define LOCAL_ALIGNMENT(EXP, ALIGN) \
118 AARCH64_EXPAND_ALIGNMENT (!flag_conserve_stack, EXP, ALIGN)
120 #define STRUCTURE_SIZE_BOUNDARY 8
122 /* Defined by the ABI */
123 #define WCHAR_TYPE "unsigned int"
124 #define WCHAR_TYPE_SIZE 32
126 /* Using long long breaks -ansi and -std=c90, so these will need to be
127 made conditional for an LLP64 ABI. */
129 #define SIZE_TYPE "long unsigned int"
131 #define PTRDIFF_TYPE "long int"
133 #define PCC_BITFIELD_TYPE_MATTERS 1
135 /* Major revision number of the ARM Architecture implemented by the target. */
136 extern unsigned aarch64_architecture_version
;
138 /* Instruction tuning/selection flags. */
140 /* Bit values used to identify processor capabilities. */
141 #define AARCH64_FL_SIMD (1 << 0) /* Has SIMD instructions. */
142 #define AARCH64_FL_FP (1 << 1) /* Has FP. */
143 #define AARCH64_FL_CRYPTO (1 << 2) /* Has crypto. */
144 #define AARCH64_FL_CRC (1 << 3) /* Has CRC. */
145 /* ARMv8.1-A architecture extensions. */
146 #define AARCH64_FL_LSE (1 << 4) /* Has Large System Extensions. */
147 #define AARCH64_FL_RDMA (1 << 5) /* Has Round Double Multiply Add. */
148 #define AARCH64_FL_V8_1 (1 << 6) /* Has ARMv8.1-A extensions. */
149 /* ARMv8.2-A architecture extensions. */
150 #define AARCH64_FL_V8_2 (1 << 8) /* Has ARMv8.2-A features. */
151 #define AARCH64_FL_F16 (1 << 9) /* Has ARMv8.2-A FP16 extensions. */
152 /* ARMv8.3-A architecture extensions. */
153 #define AARCH64_FL_V8_3 (1 << 10) /* Has ARMv8.3-A features. */
154 #define AARCH64_FL_RCPC (1 << 11) /* Has support for RCpc model. */
156 /* Has FP and SIMD. */
157 #define AARCH64_FL_FPSIMD (AARCH64_FL_FP | AARCH64_FL_SIMD)
159 /* Has FP without SIMD. */
160 #define AARCH64_FL_FPQ16 (AARCH64_FL_FP & ~AARCH64_FL_SIMD)
162 /* Architecture flags that effect instruction selection. */
163 #define AARCH64_FL_FOR_ARCH8 (AARCH64_FL_FPSIMD)
164 #define AARCH64_FL_FOR_ARCH8_1 \
165 (AARCH64_FL_FOR_ARCH8 | AARCH64_FL_LSE | AARCH64_FL_CRC \
166 | AARCH64_FL_RDMA | AARCH64_FL_V8_1)
167 #define AARCH64_FL_FOR_ARCH8_2 \
168 (AARCH64_FL_FOR_ARCH8_1 | AARCH64_FL_V8_2)
169 #define AARCH64_FL_FOR_ARCH8_3 \
170 (AARCH64_FL_FOR_ARCH8_2 | AARCH64_FL_V8_3)
172 /* Macros to test ISA flags. */
174 #define AARCH64_ISA_CRC (aarch64_isa_flags & AARCH64_FL_CRC)
175 #define AARCH64_ISA_CRYPTO (aarch64_isa_flags & AARCH64_FL_CRYPTO)
176 #define AARCH64_ISA_FP (aarch64_isa_flags & AARCH64_FL_FP)
177 #define AARCH64_ISA_SIMD (aarch64_isa_flags & AARCH64_FL_SIMD)
178 #define AARCH64_ISA_LSE (aarch64_isa_flags & AARCH64_FL_LSE)
179 #define AARCH64_ISA_RDMA (aarch64_isa_flags & AARCH64_FL_RDMA)
180 #define AARCH64_ISA_V8_2 (aarch64_isa_flags & AARCH64_FL_V8_2)
181 #define AARCH64_ISA_F16 (aarch64_isa_flags & AARCH64_FL_F16)
182 #define AARCH64_ISA_V8_3 (aarch64_isa_flags & AARCH64_FL_V8_3)
184 /* Crypto is an optional extension to AdvSIMD. */
185 #define TARGET_CRYPTO (TARGET_SIMD && AARCH64_ISA_CRYPTO)
187 /* CRC instructions that can be enabled through +crc arch extension. */
188 #define TARGET_CRC32 (AARCH64_ISA_CRC)
190 /* Atomic instructions that can be enabled through the +lse extension. */
191 #define TARGET_LSE (AARCH64_ISA_LSE)
193 /* ARMv8.2-A FP16 support that can be enabled through the +fp16 extension. */
194 #define TARGET_FP_F16INST (TARGET_FLOAT && AARCH64_ISA_F16)
195 #define TARGET_SIMD_F16INST (TARGET_SIMD && AARCH64_ISA_F16)
197 /* ARMv8.3-A features. */
198 #define TARGET_ARMV8_3 (AARCH64_ISA_V8_3)
200 /* Make sure this is always defined so we don't have to check for ifdefs
201 but rather use normal ifs. */
202 #ifndef TARGET_FIX_ERR_A53_835769_DEFAULT
203 #define TARGET_FIX_ERR_A53_835769_DEFAULT 0
205 #undef TARGET_FIX_ERR_A53_835769_DEFAULT
206 #define TARGET_FIX_ERR_A53_835769_DEFAULT 1
209 /* Apply the workaround for Cortex-A53 erratum 835769. */
210 #define TARGET_FIX_ERR_A53_835769 \
211 ((aarch64_fix_a53_err835769 == 2) \
212 ? TARGET_FIX_ERR_A53_835769_DEFAULT : aarch64_fix_a53_err835769)
214 /* Make sure this is always defined so we don't have to check for ifdefs
215 but rather use normal ifs. */
216 #ifndef TARGET_FIX_ERR_A53_843419_DEFAULT
217 #define TARGET_FIX_ERR_A53_843419_DEFAULT 0
219 #undef TARGET_FIX_ERR_A53_843419_DEFAULT
220 #define TARGET_FIX_ERR_A53_843419_DEFAULT 1
223 /* Apply the workaround for Cortex-A53 erratum 843419. */
224 #define TARGET_FIX_ERR_A53_843419 \
225 ((aarch64_fix_a53_err843419 == 2) \
226 ? TARGET_FIX_ERR_A53_843419_DEFAULT : aarch64_fix_a53_err843419)
228 /* ARMv8.1-A Adv.SIMD support. */
229 #define TARGET_SIMD_RDMA (TARGET_SIMD && AARCH64_ISA_RDMA)
231 /* Standard register usage. */
233 /* 31 64-bit general purpose registers R0-R30:
234 R30 LR (link register)
235 R29 FP (frame pointer)
236 R19-R28 Callee-saved registers
237 R18 The platform register; use as temporary register.
238 R17 IP1 The second intra-procedure-call temporary register
239 (can be used by call veneers and PLT code); otherwise use
240 as a temporary register
241 R16 IP0 The first intra-procedure-call temporary register (can
242 be used by call veneers and PLT code); otherwise use as a
244 R9-R15 Temporary registers
245 R8 Structure value parameter / temporary register
246 R0-R7 Parameter/result registers
248 SP stack pointer, encoded as X/R31 where permitted.
249 ZR zero register, encoded as X/R31 elsewhere
251 32 x 128-bit floating-point/vector registers
252 V16-V31 Caller-saved (temporary) registers
253 V8-V15 Callee-saved registers
254 V0-V7 Parameter/result registers
256 The vector register V0 holds scalar B0, H0, S0 and D0 in its least
257 significant bits. Unlike AArch32 S1 is not packed into D0,
260 /* Note that we don't mark X30 as a call-clobbered register. The idea is
261 that it's really the call instructions themselves which clobber X30.
262 We don't care what the called function does with it afterwards.
264 This approach makes it easier to implement sibcalls. Unlike normal
265 calls, sibcalls don't clobber X30, so the register reaches the
266 called function intact. EPILOGUE_USES says that X30 is useful
267 to the called function. */
269 #define FIXED_REGISTERS \
271 0, 0, 0, 0, 0, 0, 0, 0, /* R0 - R7 */ \
272 0, 0, 0, 0, 0, 0, 0, 0, /* R8 - R15 */ \
273 0, 0, 0, 0, 0, 0, 0, 0, /* R16 - R23 */ \
274 0, 0, 0, 0, 0, 1, 0, 1, /* R24 - R30, SP */ \
275 0, 0, 0, 0, 0, 0, 0, 0, /* V0 - V7 */ \
276 0, 0, 0, 0, 0, 0, 0, 0, /* V8 - V15 */ \
277 0, 0, 0, 0, 0, 0, 0, 0, /* V16 - V23 */ \
278 0, 0, 0, 0, 0, 0, 0, 0, /* V24 - V31 */ \
279 1, 1, 1, /* SFP, AP, CC */ \
282 #define CALL_USED_REGISTERS \
284 1, 1, 1, 1, 1, 1, 1, 1, /* R0 - R7 */ \
285 1, 1, 1, 1, 1, 1, 1, 1, /* R8 - R15 */ \
286 1, 1, 1, 0, 0, 0, 0, 0, /* R16 - R23 */ \
287 0, 0, 0, 0, 0, 1, 1, 1, /* R24 - R30, SP */ \
288 1, 1, 1, 1, 1, 1, 1, 1, /* V0 - V7 */ \
289 0, 0, 0, 0, 0, 0, 0, 0, /* V8 - V15 */ \
290 1, 1, 1, 1, 1, 1, 1, 1, /* V16 - V23 */ \
291 1, 1, 1, 1, 1, 1, 1, 1, /* V24 - V31 */ \
292 1, 1, 1, /* SFP, AP, CC */ \
295 #define REGISTER_NAMES \
297 "x0", "x1", "x2", "x3", "x4", "x5", "x6", "x7", \
298 "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", \
299 "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23", \
300 "x24", "x25", "x26", "x27", "x28", "x29", "x30", "sp", \
301 "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", \
302 "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", \
303 "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", \
304 "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", \
308 /* Generate the register aliases for core register N */
309 #define R_ALIASES(N) {"r" # N, R0_REGNUM + (N)}, \
310 {"w" # N, R0_REGNUM + (N)}
312 #define V_ALIASES(N) {"q" # N, V0_REGNUM + (N)}, \
313 {"d" # N, V0_REGNUM + (N)}, \
314 {"s" # N, V0_REGNUM + (N)}, \
315 {"h" # N, V0_REGNUM + (N)}, \
316 {"b" # N, V0_REGNUM + (N)}
318 /* Provide aliases for all of the ISA defined register name forms.
319 These aliases are convenient for use in the clobber lists of inline
322 #define ADDITIONAL_REGISTER_NAMES \
323 { R_ALIASES(0), R_ALIASES(1), R_ALIASES(2), R_ALIASES(3), \
324 R_ALIASES(4), R_ALIASES(5), R_ALIASES(6), R_ALIASES(7), \
325 R_ALIASES(8), R_ALIASES(9), R_ALIASES(10), R_ALIASES(11), \
326 R_ALIASES(12), R_ALIASES(13), R_ALIASES(14), R_ALIASES(15), \
327 R_ALIASES(16), R_ALIASES(17), R_ALIASES(18), R_ALIASES(19), \
328 R_ALIASES(20), R_ALIASES(21), R_ALIASES(22), R_ALIASES(23), \
329 R_ALIASES(24), R_ALIASES(25), R_ALIASES(26), R_ALIASES(27), \
330 R_ALIASES(28), R_ALIASES(29), R_ALIASES(30), {"wsp", R0_REGNUM + 31}, \
331 V_ALIASES(0), V_ALIASES(1), V_ALIASES(2), V_ALIASES(3), \
332 V_ALIASES(4), V_ALIASES(5), V_ALIASES(6), V_ALIASES(7), \
333 V_ALIASES(8), V_ALIASES(9), V_ALIASES(10), V_ALIASES(11), \
334 V_ALIASES(12), V_ALIASES(13), V_ALIASES(14), V_ALIASES(15), \
335 V_ALIASES(16), V_ALIASES(17), V_ALIASES(18), V_ALIASES(19), \
336 V_ALIASES(20), V_ALIASES(21), V_ALIASES(22), V_ALIASES(23), \
337 V_ALIASES(24), V_ALIASES(25), V_ALIASES(26), V_ALIASES(27), \
338 V_ALIASES(28), V_ALIASES(29), V_ALIASES(30), V_ALIASES(31) \
341 /* Say that the epilogue uses the return address register. Note that
342 in the case of sibcalls, the values "used by the epilogue" are
343 considered live at the start of the called function. */
345 #define EPILOGUE_USES(REGNO) \
346 (epilogue_completed && (REGNO) == LR_REGNUM)
348 /* EXIT_IGNORE_STACK should be nonzero if, when returning from a function,
349 the stack pointer does not matter. The value is tested only in
350 functions that have frame pointers. */
351 #define EXIT_IGNORE_STACK 1
353 #define STATIC_CHAIN_REGNUM R18_REGNUM
354 #define HARD_FRAME_POINTER_REGNUM R29_REGNUM
355 #define FRAME_POINTER_REGNUM SFP_REGNUM
356 #define STACK_POINTER_REGNUM SP_REGNUM
357 #define ARG_POINTER_REGNUM AP_REGNUM
358 #define FIRST_PSEUDO_REGISTER 67
360 /* The number of (integer) argument register available. */
361 #define NUM_ARG_REGS 8
362 #define NUM_FP_ARG_REGS 8
364 /* A Homogeneous Floating-Point or Short-Vector Aggregate may have at most
366 #define HA_MAX_NUM_FLDS 4
368 /* External dwarf register number scheme. These number are used to
369 identify registers in dwarf debug information, the values are
370 defined by the AArch64 ABI. The numbering scheme is independent of
371 GCC's internal register numbering scheme. */
373 #define AARCH64_DWARF_R0 0
375 /* The number of R registers, note 31! not 32. */
376 #define AARCH64_DWARF_NUMBER_R 31
378 #define AARCH64_DWARF_SP 31
379 #define AARCH64_DWARF_V0 64
381 /* The number of V registers. */
382 #define AARCH64_DWARF_NUMBER_V 32
384 /* For signal frames we need to use an alternative return column. This
385 value must not correspond to a hard register and must be out of the
386 range of DWARF_FRAME_REGNUM(). */
387 #define DWARF_ALT_FRAME_RETURN_COLUMN \
388 (AARCH64_DWARF_V0 + AARCH64_DWARF_NUMBER_V)
390 /* We add 1 extra frame register for use as the
391 DWARF_ALT_FRAME_RETURN_COLUMN. */
392 #define DWARF_FRAME_REGISTERS (DWARF_ALT_FRAME_RETURN_COLUMN + 1)
395 #define DBX_REGISTER_NUMBER(REGNO) aarch64_dbx_register_number (REGNO)
396 /* Provide a definition of DWARF_FRAME_REGNUM here so that fallback unwinders
397 can use DWARF_ALT_FRAME_RETURN_COLUMN defined below. This is just the same
398 as the default definition in dwarf2out.c. */
399 #undef DWARF_FRAME_REGNUM
400 #define DWARF_FRAME_REGNUM(REGNO) DBX_REGISTER_NUMBER (REGNO)
402 #define DWARF_FRAME_RETURN_COLUMN DWARF_FRAME_REGNUM (LR_REGNUM)
404 #define HARD_REGNO_NREGS(REGNO, MODE) aarch64_hard_regno_nregs (REGNO, MODE)
406 #define DWARF2_UNWIND_INFO 1
408 /* Use R0 through R3 to pass exception handling information. */
409 #define EH_RETURN_DATA_REGNO(N) \
410 ((N) < 4 ? ((unsigned int) R0_REGNUM + (N)) : INVALID_REGNUM)
412 /* Select a format to encode pointers in exception handling data. */
413 #define ASM_PREFERRED_EH_DATA_FORMAT(CODE, GLOBAL) \
414 aarch64_asm_preferred_eh_data_format ((CODE), (GLOBAL))
416 /* Output the assembly strings we want to add to a function definition. */
417 #define ASM_DECLARE_FUNCTION_NAME(STR, NAME, DECL) \
418 aarch64_declare_function_name (STR, NAME, DECL)
420 /* For EH returns X4 contains the stack adjustment. */
421 #define EH_RETURN_STACKADJ_RTX gen_rtx_REG (Pmode, R4_REGNUM)
422 #define EH_RETURN_HANDLER_RTX aarch64_eh_return_handler_rtx ()
424 /* Don't use __builtin_setjmp until we've defined it. */
425 #undef DONT_USE_BUILTIN_SETJMP
426 #define DONT_USE_BUILTIN_SETJMP 1
428 /* Register in which the structure value is to be returned. */
429 #define AARCH64_STRUCT_VALUE_REGNUM R8_REGNUM
431 /* Non-zero if REGNO is part of the Core register set.
433 The rather unusual way of expressing this check is to avoid
434 warnings when building the compiler when R0_REGNUM is 0 and REGNO
436 #define GP_REGNUM_P(REGNO) \
437 (((unsigned) (REGNO - R0_REGNUM)) <= (R30_REGNUM - R0_REGNUM))
439 #define FP_REGNUM_P(REGNO) \
440 (((unsigned) (REGNO - V0_REGNUM)) <= (V31_REGNUM - V0_REGNUM))
442 #define FP_LO_REGNUM_P(REGNO) \
443 (((unsigned) (REGNO - V0_REGNUM)) <= (V15_REGNUM - V0_REGNUM))
446 /* Register and constant classes. */
458 LIM_REG_CLASSES
/* Last */
461 #define N_REG_CLASSES ((int) LIM_REG_CLASSES)
463 #define REG_CLASS_NAMES \
466 "CALLER_SAVE_REGS", \
475 #define REG_CLASS_CONTENTS \
477 { 0x00000000, 0x00000000, 0x00000000 }, /* NO_REGS */ \
478 { 0x0007ffff, 0x00000000, 0x00000000 }, /* CALLER_SAVE_REGS */ \
479 { 0x7fffffff, 0x00000000, 0x00000003 }, /* GENERAL_REGS */ \
480 { 0x80000000, 0x00000000, 0x00000000 }, /* STACK_REG */ \
481 { 0xffffffff, 0x00000000, 0x00000003 }, /* POINTER_REGS */ \
482 { 0x00000000, 0x0000ffff, 0x00000000 }, /* FP_LO_REGS */ \
483 { 0x00000000, 0xffffffff, 0x00000000 }, /* FP_REGS */ \
484 { 0xffffffff, 0xffffffff, 0x00000007 } /* ALL_REGS */ \
487 #define REGNO_REG_CLASS(REGNO) aarch64_regno_regclass (REGNO)
489 #define INDEX_REG_CLASS GENERAL_REGS
490 #define BASE_REG_CLASS POINTER_REGS
492 /* Register pairs used to eliminate unneeded registers that point into
494 #define ELIMINABLE_REGS \
496 { ARG_POINTER_REGNUM, STACK_POINTER_REGNUM }, \
497 { ARG_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM }, \
498 { FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM }, \
499 { FRAME_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM }, \
502 #define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \
503 (OFFSET) = aarch64_initial_elimination_offset (FROM, TO)
505 /* CPU/ARCH option handling. */
506 #include "config/aarch64/aarch64-opts.h"
510 #define AARCH64_CORE(NAME, INTERNAL_IDENT, SCHED, ARCH, FLAGS, COSTS, IMP, PART, VARIANT) \
511 TARGET_CPU_##INTERNAL_IDENT,
512 #include "aarch64-cores.def"
516 /* If there is no CPU defined at configure, use generic as default. */
517 #ifndef TARGET_CPU_DEFAULT
518 #define TARGET_CPU_DEFAULT \
519 (TARGET_CPU_generic | (AARCH64_CPU_DEFAULT_FLAGS << 6))
522 /* If inserting NOP before a mult-accumulate insn remember to adjust the
523 length so that conditional branching code is updated appropriately. */
524 #define ADJUST_INSN_LENGTH(insn, length) \
527 if (aarch64_madd_needs_nop (insn)) \
531 #define FINAL_PRESCAN_INSN(INSN, OPVEC, NOPERANDS) \
532 aarch64_final_prescan_insn (INSN); \
534 /* The processor for which instructions should be scheduled. */
535 extern enum aarch64_processor aarch64_tune
;
537 /* RTL generation support. */
538 #define INIT_EXPANDERS aarch64_init_expanders ()
541 /* Stack layout; function entry, exit and calling. */
542 #define STACK_GROWS_DOWNWARD 1
544 #define FRAME_GROWS_DOWNWARD 1
546 #define STARTING_FRAME_OFFSET 0
548 #define ACCUMULATE_OUTGOING_ARGS 1
550 #define FIRST_PARM_OFFSET(FNDECL) 0
553 #define LIBCALL_VALUE(MODE) \
554 gen_rtx_REG (MODE, FLOAT_MODE_P (MODE) ? V0_REGNUM : R0_REGNUM)
556 #define DEFAULT_PCC_STRUCT_RETURN 0
559 struct GTY (()) aarch64_frame
561 HOST_WIDE_INT reg_offset
[FIRST_PSEUDO_REGISTER
];
563 /* The number of extra stack bytes taken up by register varargs.
564 This area is allocated by the callee at the very top of the
565 frame. This value is rounded up to a multiple of
567 HOST_WIDE_INT saved_varargs_size
;
569 /* The size of the saved callee-save int/FP registers. */
571 HOST_WIDE_INT saved_regs_size
;
573 /* Offset from the base of the frame (incomming SP) to the
574 top of the locals area. This value is always a multiple of
576 HOST_WIDE_INT locals_offset
;
578 /* Offset from the base of the frame (incomming SP) to the
579 hard_frame_pointer. This value is always a multiple of
581 HOST_WIDE_INT hard_fp_offset
;
583 /* The size of the frame. This value is the offset from base of the
584 * frame (incomming SP) to the stack_pointer. This value is always
585 * a multiple of STACK_BOUNDARY. */
586 HOST_WIDE_INT frame_size
;
588 /* The size of the initial stack adjustment before saving callee-saves. */
589 HOST_WIDE_INT initial_adjust
;
591 /* The writeback value when pushing callee-save registers.
592 It is zero when no push is used. */
593 HOST_WIDE_INT callee_adjust
;
595 /* The offset from SP to the callee-save registers after initial_adjust.
596 It may be non-zero if no push is used (ie. callee_adjust == 0). */
597 HOST_WIDE_INT callee_offset
;
599 /* The size of the stack adjustment after saving callee-saves. */
600 HOST_WIDE_INT final_adjust
;
602 unsigned wb_candidate1
;
603 unsigned wb_candidate2
;
608 typedef struct GTY (()) machine_function
610 struct aarch64_frame frame
;
611 /* One entry for each hard register. */
612 bool reg_is_wrapped_separately
[LAST_SAVED_REGNUM
];
616 /* Which ABI to use. */
617 enum aarch64_abi_type
619 AARCH64_ABI_LP64
= 0,
620 AARCH64_ABI_ILP32
= 1
623 #ifndef AARCH64_ABI_DEFAULT
624 #define AARCH64_ABI_DEFAULT AARCH64_ABI_LP64
627 #define TARGET_ILP32 (aarch64_abi & AARCH64_ABI_ILP32)
631 ARM_PCS_AAPCS64
, /* Base standard AAPCS for 64 bit. */
638 /* We can't use machine_mode inside a generator file because it
639 hasn't been created yet; we shouldn't be using any code that
640 needs the real definition though, so this ought to be safe. */
641 #ifdef GENERATOR_FILE
644 #include "insn-modes.h"
645 #define MACHMODE machine_mode
648 #ifndef USED_FOR_TARGET
649 /* AAPCS related state tracking. */
652 enum arm_pcs pcs_variant
;
653 int aapcs_arg_processed
; /* No need to lay out this argument again. */
654 int aapcs_ncrn
; /* Next Core register number. */
655 int aapcs_nextncrn
; /* Next next core register number. */
656 int aapcs_nvrn
; /* Next Vector register number. */
657 int aapcs_nextnvrn
; /* Next Next Vector register number. */
658 rtx aapcs_reg
; /* Register assigned to this argument. This
659 is NULL_RTX if this parameter goes on
661 MACHMODE aapcs_vfp_rmode
;
662 int aapcs_stack_words
; /* If the argument is passed on the stack, this
663 is the number of words needed, after rounding
664 up. Only meaningful when
665 aapcs_reg == NULL_RTX. */
666 int aapcs_stack_size
; /* The total size (in words, per 8 byte) of the
667 stack arg area so far. */
671 #define BLOCK_REG_PADDING(MODE, TYPE, FIRST) \
672 (aarch64_pad_reg_upward (MODE, TYPE, FIRST) ? PAD_UPWARD : PAD_DOWNWARD)
674 #define PAD_VARARGS_DOWN 0
676 #define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \
677 aarch64_init_cumulative_args (&(CUM), FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS)
679 #define FUNCTION_ARG_REGNO_P(REGNO) \
680 aarch64_function_arg_regno_p(REGNO)
685 /* Addressing modes, etc. */
686 #define HAVE_POST_INCREMENT 1
687 #define HAVE_PRE_INCREMENT 1
688 #define HAVE_POST_DECREMENT 1
689 #define HAVE_PRE_DECREMENT 1
690 #define HAVE_POST_MODIFY_DISP 1
691 #define HAVE_PRE_MODIFY_DISP 1
693 #define MAX_REGS_PER_ADDRESS 2
695 #define CONSTANT_ADDRESS_P(X) aarch64_constant_address_p(X)
697 #define REGNO_OK_FOR_BASE_P(REGNO) \
698 aarch64_regno_ok_for_base_p (REGNO, true)
700 #define REGNO_OK_FOR_INDEX_P(REGNO) \
701 aarch64_regno_ok_for_index_p (REGNO, true)
703 #define LEGITIMATE_PIC_OPERAND_P(X) \
704 aarch64_legitimate_pic_operand_p (X)
706 #define CASE_VECTOR_MODE Pmode
708 #define DEFAULT_SIGNED_CHAR 0
710 /* An integer expression for the size in bits of the largest integer machine
711 mode that should actually be used. We allow pairs of registers. */
712 #define MAX_FIXED_MODE_SIZE GET_MODE_BITSIZE (TImode)
714 /* Maximum bytes moved by a single instruction (load/store pair). */
715 #define MOVE_MAX (UNITS_PER_WORD * 2)
717 /* The base cost overhead of a memcpy call, for MOVE_RATIO and friends. */
718 #define AARCH64_CALL_RATIO 8
720 /* MOVE_RATIO dictates when we will use the move_by_pieces infrastructure.
721 move_by_pieces will continually copy the largest safe chunks. So a
722 7-byte copy is a 4-byte + 2-byte + byte copy. This proves inefficient
723 for both size and speed of copy, so we will instead use the "movmem"
724 standard name to implement the copy. This logic does not apply when
725 targeting -mstrict-align, so keep a sensible default in that case. */
726 #define MOVE_RATIO(speed) \
727 (!STRICT_ALIGNMENT ? 2 : (((speed) ? 15 : AARCH64_CALL_RATIO) / 2))
729 /* For CLEAR_RATIO, when optimizing for size, give a better estimate
730 of the length of a memset call, but use the default otherwise. */
731 #define CLEAR_RATIO(speed) \
732 ((speed) ? 15 : AARCH64_CALL_RATIO)
734 /* SET_RATIO is similar to CLEAR_RATIO, but for a non-zero constant, so when
735 optimizing for size adjust the ratio to account for the overhead of loading
737 #define SET_RATIO(speed) \
738 ((speed) ? 15 : AARCH64_CALL_RATIO - 2)
740 /* Disable auto-increment in move_by_pieces et al. Use of auto-increment is
741 rarely a good idea in straight-line code since it adds an extra address
742 dependency between each instruction. Better to use incrementing offsets. */
743 #define USE_LOAD_POST_INCREMENT(MODE) 0
744 #define USE_LOAD_POST_DECREMENT(MODE) 0
745 #define USE_LOAD_PRE_INCREMENT(MODE) 0
746 #define USE_LOAD_PRE_DECREMENT(MODE) 0
747 #define USE_STORE_POST_INCREMENT(MODE) 0
748 #define USE_STORE_POST_DECREMENT(MODE) 0
749 #define USE_STORE_PRE_INCREMENT(MODE) 0
750 #define USE_STORE_PRE_DECREMENT(MODE) 0
752 /* WORD_REGISTER_OPERATIONS does not hold for AArch64.
753 The assigned word_mode is DImode but operations narrower than SImode
754 behave as 32-bit operations if using the W-form of the registers rather
755 than as word_mode (64-bit) operations as WORD_REGISTER_OPERATIONS
757 #define WORD_REGISTER_OPERATIONS 0
759 /* Define if loading from memory in MODE, an integral mode narrower than
760 BITS_PER_WORD will either zero-extend or sign-extend. The value of this
761 macro should be the code that says which one of the two operations is
762 implicitly done, or UNKNOWN if none. */
763 #define LOAD_EXTEND_OP(MODE) ZERO_EXTEND
765 /* Define this macro to be non-zero if instructions will fail to work
766 if given data not on the nominal alignment. */
767 #define STRICT_ALIGNMENT TARGET_STRICT_ALIGN
769 /* Define this macro to be non-zero if accessing less than a word of
770 memory is no faster than accessing a word of memory, i.e., if such
771 accesses require more than one instruction or if there is no
773 Although there's no difference in instruction count or cycles,
774 in AArch64 we don't want to expand to a sub-word to a 64-bit access
775 if we don't have to, for power-saving reasons. */
776 #define SLOW_BYTE_ACCESS 0
778 #define TRULY_NOOP_TRUNCATION(OUTPREC, INPREC) 1
780 #define NO_FUNCTION_CSE 1
782 /* Specify the machine mode that the hardware addresses have.
783 After generation of rtl, the compiler makes no further distinction
784 between pointers and any other objects of this machine mode. */
787 /* A C expression whose value is zero if pointers that need to be extended
788 from being `POINTER_SIZE' bits wide to `Pmode' are sign-extended and
789 greater then zero if they are zero-extended and less then zero if the
790 ptr_extend instruction should be used. */
791 #define POINTERS_EXTEND_UNSIGNED 1
793 /* Mode of a function address in a call instruction (for indexing purposes). */
794 #define FUNCTION_MODE Pmode
796 #define SELECT_CC_MODE(OP, X, Y) aarch64_select_cc_mode (OP, X, Y)
798 #define REVERSIBLE_CC_MODE(MODE) 1
800 #define REVERSE_CONDITION(CODE, MODE) \
801 (((MODE) == CCFPmode || (MODE) == CCFPEmode) \
802 ? reverse_condition_maybe_unordered (CODE) \
803 : reverse_condition (CODE))
805 #define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
806 ((VALUE) = GET_MODE_UNIT_BITSIZE (MODE), 2)
807 #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
808 ((VALUE) = GET_MODE_UNIT_BITSIZE (MODE), 2)
810 #define INCOMING_RETURN_ADDR_RTX gen_rtx_REG (Pmode, LR_REGNUM)
812 #define RETURN_ADDR_RTX aarch64_return_addr
814 /* 3 insns + padding + 2 pointer-sized entries. */
815 #define TRAMPOLINE_SIZE (TARGET_ILP32 ? 24 : 32)
817 /* Trampolines contain dwords, so must be dword aligned. */
818 #define TRAMPOLINE_ALIGNMENT 64
820 /* Put trampolines in the text section so that mapping symbols work
822 #define TRAMPOLINE_SECTION text_section
825 #define BRANCH_COST(SPEED_P, PREDICTABLE_P) \
826 (aarch64_branch_cost (SPEED_P, PREDICTABLE_P))
829 /* Assembly output. */
831 /* For now we'll make all jump tables pc-relative. */
832 #define CASE_VECTOR_PC_RELATIVE 1
834 #define CASE_VECTOR_SHORTEN_MODE(min, max, body) \
835 ((min < -0x1fff0 || max > 0x1fff0) ? SImode \
836 : (min < -0x1f0 || max > 0x1f0) ? HImode \
839 /* Jump table alignment is explicit in ASM_OUTPUT_CASE_LABEL. */
840 #define ADDR_VEC_ALIGN(JUMPTABLE) 0
842 #define MCOUNT_NAME "_mcount"
844 #define NO_PROFILE_COUNTERS 1
846 /* Emit rtl for profiling. Output assembler code to FILE
847 to call "_mcount" for profiling a function entry. */
848 #define PROFILE_HOOK(LABEL) \
851 lr = get_hard_reg_initial_val (Pmode, LR_REGNUM); \
852 fun = gen_rtx_SYMBOL_REF (Pmode, MCOUNT_NAME); \
853 emit_library_call (fun, LCT_NORMAL, VOIDmode, lr, Pmode); \
856 /* All the work done in PROFILE_HOOK, but still required. */
857 #define FUNCTION_PROFILER(STREAM, LABELNO) do { } while (0)
859 /* For some reason, the Linux headers think they know how to define
860 these macros. They don't!!! */
863 #define ASM_APP_ON "\t" ASM_COMMENT_START " Start of user assembly\n"
864 #define ASM_APP_OFF "\t" ASM_COMMENT_START " End of user assembly\n"
866 #define CONSTANT_POOL_BEFORE_FUNCTION 0
868 /* This definition should be relocated to aarch64-elf-raw.h. This macro
869 should be undefined in aarch64-linux.h and a clear_cache pattern
870 implmented to emit either the call to __aarch64_sync_cache_range()
871 directly or preferably the appropriate sycall or cache clear
872 instructions inline. */
873 #define CLEAR_INSN_CACHE(beg, end) \
874 extern void __aarch64_sync_cache_range (void *, void *); \
875 __aarch64_sync_cache_range (beg, end)
877 #define SHIFT_COUNT_TRUNCATED (!TARGET_SIMD)
879 /* Choose appropriate mode for caller saves, so we do the minimum
880 required size of load/store. */
881 #define HARD_REGNO_CALLER_SAVE_MODE(REGNO, NREGS, MODE) \
882 aarch64_hard_regno_caller_save_mode ((REGNO), (NREGS), (MODE))
884 #undef SWITCHABLE_TARGET
885 #define SWITCHABLE_TARGET 1
887 /* Check TLS Descriptors mechanism is selected. */
888 #define TARGET_TLS_DESC (aarch64_tls_dialect == TLS_DESCRIPTORS)
890 extern enum aarch64_code_model aarch64_cmodel
;
892 /* When using the tiny addressing model conditional and unconditional branches
893 can span the whole of the available address space (1MB). */
894 #define HAS_LONG_COND_BRANCH \
895 (aarch64_cmodel == AARCH64_CMODEL_TINY \
896 || aarch64_cmodel == AARCH64_CMODEL_TINY_PIC)
898 #define HAS_LONG_UNCOND_BRANCH \
899 (aarch64_cmodel == AARCH64_CMODEL_TINY \
900 || aarch64_cmodel == AARCH64_CMODEL_TINY_PIC)
902 #define TARGET_SUPPORTS_WIDE_INT 1
904 /* Modes valid for AdvSIMD D registers, i.e. that fit in half a Q register. */
905 #define AARCH64_VALID_SIMD_DREG_MODE(MODE) \
906 ((MODE) == V2SImode || (MODE) == V4HImode || (MODE) == V8QImode \
907 || (MODE) == V2SFmode || (MODE) == V4HFmode || (MODE) == DImode \
910 /* Modes valid for AdvSIMD Q registers. */
911 #define AARCH64_VALID_SIMD_QREG_MODE(MODE) \
912 ((MODE) == V4SImode || (MODE) == V8HImode || (MODE) == V16QImode \
913 || (MODE) == V4SFmode || (MODE) == V8HFmode || (MODE) == V2DImode \
914 || (MODE) == V2DFmode)
916 #define ENDIAN_LANE_N(mode, n) \
917 (BYTES_BIG_ENDIAN ? GET_MODE_NUNITS (mode) - 1 - n : n)
919 /* Support for a configure-time default CPU, etc. We currently support
920 --with-arch and --with-cpu. Both are ignored if either is specified
921 explicitly on the command line at run time. */
922 #define OPTION_DEFAULT_SPECS \
923 {"arch", "%{!march=*:%{!mcpu=*:-march=%(VALUE)}}" }, \
924 {"cpu", "%{!march=*:%{!mcpu=*:-mcpu=%(VALUE)}}" },
926 #define MCPU_TO_MARCH_SPEC \
927 " %{mcpu=*:-march=%:rewrite_mcpu(%{mcpu=*:%*})}"
929 extern const char *aarch64_rewrite_mcpu (int argc
, const char **argv
);
930 #define MCPU_TO_MARCH_SPEC_FUNCTIONS \
931 { "rewrite_mcpu", aarch64_rewrite_mcpu },
933 #if defined(__aarch64__)
934 extern const char *host_detect_local_cpu (int argc
, const char **argv
);
935 # define EXTRA_SPEC_FUNCTIONS \
936 { "local_cpu_detect", host_detect_local_cpu }, \
937 MCPU_TO_MARCH_SPEC_FUNCTIONS
939 # define MCPU_MTUNE_NATIVE_SPECS \
940 " %{march=native:%<march=native %:local_cpu_detect(arch)}" \
941 " %{mcpu=native:%<mcpu=native %:local_cpu_detect(cpu)}" \
942 " %{mtune=native:%<mtune=native %:local_cpu_detect(tune)}"
944 # define MCPU_MTUNE_NATIVE_SPECS ""
945 # define EXTRA_SPEC_FUNCTIONS MCPU_TO_MARCH_SPEC_FUNCTIONS
948 #define ASM_CPU_SPEC \
951 #define EXTRA_SPECS \
952 { "asm_cpu_spec", ASM_CPU_SPEC }
954 #define ASM_OUTPUT_POOL_EPILOGUE aarch64_asm_output_pool_epilogue
956 /* This type is the user-visible __fp16, and a pointer to that type. We
957 need it in many places in the backend. Defined in aarch64-builtins.c. */
958 extern tree aarch64_fp16_type_node
;
959 extern tree aarch64_fp16_ptr_type_node
;
961 #endif /* GCC_AARCH64_H */