[committed] [RISC-V] Fix wrong patch application
[official-gcc.git] / gcc / config / aarch64 / aarch64.h
blob2b89f6f88ef00b4691eb3e7b27593ace8d872fb5
1 /* Machine description for AArch64 architecture.
2 Copyright (C) 2009-2024 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)
10 any later version.
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/>. */
22 #ifndef GCC_AARCH64_H
23 #define GCC_AARCH64_H
25 /* Make these flags read-only so that all uses go via
26 aarch64_set_asm_isa_flags. */
27 #ifndef GENERATOR_FILE
28 #undef aarch64_asm_isa_flags
29 #define aarch64_asm_isa_flags \
30 ((aarch64_feature_flags) global_options.x_aarch64_asm_isa_flags)
31 #undef aarch64_isa_flags
32 #define aarch64_isa_flags \
33 ((aarch64_feature_flags) global_options.x_aarch64_isa_flags)
34 #endif
36 /* Target CPU builtins. */
37 #define TARGET_CPU_CPP_BUILTINS() \
38 aarch64_cpu_cpp_builtins (pfile)
42 #define REGISTER_TARGET_PRAGMAS() aarch64_register_pragmas ()
44 /* Target machine storage layout. */
46 #define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE) \
47 if (GET_MODE_CLASS (MODE) == MODE_INT \
48 && GET_MODE_SIZE (MODE) < 4) \
49 { \
50 if (MODE == QImode || MODE == HImode) \
51 { \
52 MODE = SImode; \
53 } \
56 /* Bits are always numbered from the LSBit. */
57 #define BITS_BIG_ENDIAN 0
59 /* Big/little-endian flavour. */
60 #define BYTES_BIG_ENDIAN (TARGET_BIG_END != 0)
61 #define WORDS_BIG_ENDIAN (BYTES_BIG_ENDIAN)
63 /* AdvSIMD is supported in the default configuration, unless disabled by
64 -mgeneral-regs-only or by the +nosimd extension. The set of available
65 instructions is then subdivided into:
67 - the "base" set, available both in SME streaming mode and in
68 non-streaming mode
70 - the full set, available only in non-streaming mode. */
71 #define TARGET_BASE_SIMD (AARCH64_ISA_SIMD)
72 #define TARGET_SIMD (AARCH64_ISA_SIMD && AARCH64_ISA_SM_OFF)
73 #define TARGET_FLOAT (AARCH64_ISA_FP)
75 #define UNITS_PER_WORD 8
77 #define UNITS_PER_VREG 16
79 #define PARM_BOUNDARY 64
81 #define STACK_BOUNDARY 128
83 #define FUNCTION_BOUNDARY 32
85 #define EMPTY_FIELD_BOUNDARY 32
87 #define BIGGEST_ALIGNMENT 128
89 #define SHORT_TYPE_SIZE 16
91 #define INT_TYPE_SIZE 32
93 #define LONG_TYPE_SIZE (TARGET_ILP32 ? 32 : 64)
95 #define POINTER_SIZE (TARGET_ILP32 ? 32 : 64)
97 #define LONG_LONG_TYPE_SIZE 64
99 #define FLOAT_TYPE_SIZE 32
101 #define DOUBLE_TYPE_SIZE 64
103 #define LONG_DOUBLE_TYPE_SIZE 128
105 /* This value is the amount of bytes a caller is allowed to drop the stack
106 before probing has to be done for stack clash protection. */
107 #define STACK_CLASH_CALLER_GUARD 1024
109 /* This value represents the minimum amount of bytes we expect the function's
110 outgoing arguments to be when stack-clash is enabled. */
111 #define STACK_CLASH_MIN_BYTES_OUTGOING_ARGS 8
113 /* This value controls how many pages we manually unroll the loop for when
114 generating stack clash probes. */
115 #define STACK_CLASH_MAX_UNROLL_PAGES 4
117 /* The architecture reserves all bits of the address for hardware use,
118 so the vbit must go into the delta field of pointers to member
119 functions. This is the same config as that in the AArch32
120 port. */
121 #define TARGET_PTRMEMFUNC_VBIT_LOCATION ptrmemfunc_vbit_in_delta
124 /* Emit calls to libgcc helpers for atomic operations for runtime detection
125 of LSE instructions. */
126 #define TARGET_OUTLINE_ATOMICS (aarch64_flag_outline_atomics)
128 /* Align definitions of arrays, unions and structures so that
129 initializations and copies can be made more efficient. This is not
130 ABI-changing, so it only affects places where we can see the
131 definition. Increasing the alignment tends to introduce padding,
132 so don't do this when optimizing for size/conserving stack space. */
133 #define AARCH64_EXPAND_ALIGNMENT(COND, EXP, ALIGN) \
134 (((COND) && ((ALIGN) < BITS_PER_WORD) \
135 && (TREE_CODE (EXP) == ARRAY_TYPE \
136 || TREE_CODE (EXP) == UNION_TYPE \
137 || TREE_CODE (EXP) == RECORD_TYPE)) ? BITS_PER_WORD : (ALIGN))
139 /* Align global data. */
140 #define DATA_ALIGNMENT(EXP, ALIGN) \
141 AARCH64_EXPAND_ALIGNMENT (!optimize_size, EXP, ALIGN)
143 /* Similarly, make sure that objects on the stack are sensibly aligned. */
144 #define LOCAL_ALIGNMENT(EXP, ALIGN) \
145 AARCH64_EXPAND_ALIGNMENT (!flag_conserve_stack, EXP, ALIGN)
147 #define STRUCTURE_SIZE_BOUNDARY 8
149 /* Heap alignment (same as BIGGEST_ALIGNMENT and STACK_BOUNDARY). */
150 #define MALLOC_ABI_ALIGNMENT 128
152 /* Defined by the ABI */
153 #define WCHAR_TYPE "unsigned int"
154 #define WCHAR_TYPE_SIZE 32
156 /* Using long long breaks -ansi and -std=c90, so these will need to be
157 made conditional for an LLP64 ABI. */
159 #define SIZE_TYPE "long unsigned int"
161 #define PTRDIFF_TYPE "long int"
163 #define PCC_BITFIELD_TYPE_MATTERS 1
165 #ifndef USED_FOR_TARGET
167 /* Define an enum of all features (ISA modes, architectures and extensions).
168 The ISA modes must come first. */
169 enum class aarch64_feature : unsigned char {
170 #define DEF_AARCH64_ISA_MODE(IDENT) IDENT,
171 #define AARCH64_OPT_EXTENSION(A, IDENT, C, D, E, F) IDENT,
172 #define AARCH64_ARCH(A, B, IDENT, D, E) IDENT,
173 #include "aarch64-isa-modes.def"
174 #include "aarch64-option-extensions.def"
175 #include "aarch64-arches.def"
178 /* Define unique flags for each of the above. */
179 #define HANDLE(IDENT) \
180 constexpr auto AARCH64_FL_##IDENT \
181 = aarch64_feature_flags (1) << int (aarch64_feature::IDENT);
182 #define DEF_AARCH64_ISA_MODE(IDENT) HANDLE (IDENT)
183 #define AARCH64_OPT_EXTENSION(A, IDENT, C, D, E, F) HANDLE (IDENT)
184 #define AARCH64_ARCH(A, B, IDENT, D, E) HANDLE (IDENT)
185 #include "aarch64-isa-modes.def"
186 #include "aarch64-option-extensions.def"
187 #include "aarch64-arches.def"
188 #undef HANDLE
190 constexpr auto AARCH64_FL_SM_STATE = AARCH64_FL_SM_ON | AARCH64_FL_SM_OFF;
192 constexpr unsigned int AARCH64_NUM_ISA_MODES = (0
193 #define DEF_AARCH64_ISA_MODE(IDENT) + 1
194 #include "aarch64-isa-modes.def"
197 /* The mask of all ISA modes. */
198 constexpr auto AARCH64_FL_ISA_MODES
199 = (aarch64_feature_flags (1) << AARCH64_NUM_ISA_MODES) - 1;
201 /* The default ISA mode, for functions with no attributes that specify
202 something to the contrary. */
203 constexpr auto AARCH64_FL_DEFAULT_ISA_MODE = AARCH64_FL_SM_OFF;
205 #endif
207 /* Macros to test ISA flags.
209 There is intentionally no macro for AARCH64_FL_CRYPTO, since this flag bit
210 is not always set when its constituent features are present.
211 Check (TARGET_AES && TARGET_SHA2) instead. */
213 #define AARCH64_ISA_SM_OFF (aarch64_isa_flags & AARCH64_FL_SM_OFF)
214 #define AARCH64_ISA_SM_ON (aarch64_isa_flags & AARCH64_FL_SM_ON)
215 #define AARCH64_ISA_ZA_ON (aarch64_isa_flags & AARCH64_FL_ZA_ON)
216 #define AARCH64_ISA_MODE (aarch64_isa_flags & AARCH64_FL_ISA_MODES)
217 #define AARCH64_ISA_V8A (aarch64_isa_flags & AARCH64_FL_V8A)
218 #define AARCH64_ISA_V8_1A (aarch64_isa_flags & AARCH64_FL_V8_1A)
219 #define AARCH64_ISA_CRC (aarch64_isa_flags & AARCH64_FL_CRC)
220 #define AARCH64_ISA_FP (aarch64_isa_flags & AARCH64_FL_FP)
221 #define AARCH64_ISA_SIMD (aarch64_isa_flags & AARCH64_FL_SIMD)
222 #define AARCH64_ISA_LSE (aarch64_isa_flags & AARCH64_FL_LSE)
223 #define AARCH64_ISA_RDMA (aarch64_isa_flags & AARCH64_FL_RDMA)
224 #define AARCH64_ISA_V8_2A (aarch64_isa_flags & AARCH64_FL_V8_2A)
225 #define AARCH64_ISA_F16 (aarch64_isa_flags & AARCH64_FL_F16)
226 #define AARCH64_ISA_SVE (aarch64_isa_flags & AARCH64_FL_SVE)
227 #define AARCH64_ISA_SVE2 (aarch64_isa_flags & AARCH64_FL_SVE2)
228 #define AARCH64_ISA_SVE2_AES (aarch64_isa_flags & AARCH64_FL_SVE2_AES)
229 #define AARCH64_ISA_SVE2_BITPERM (aarch64_isa_flags & AARCH64_FL_SVE2_BITPERM)
230 #define AARCH64_ISA_SVE2_SHA3 (aarch64_isa_flags & AARCH64_FL_SVE2_SHA3)
231 #define AARCH64_ISA_SVE2_SM4 (aarch64_isa_flags & AARCH64_FL_SVE2_SM4)
232 #define AARCH64_ISA_SME (aarch64_isa_flags & AARCH64_FL_SME)
233 #define AARCH64_ISA_SME_I16I64 (aarch64_isa_flags & AARCH64_FL_SME_I16I64)
234 #define AARCH64_ISA_SME_F64F64 (aarch64_isa_flags & AARCH64_FL_SME_F64F64)
235 #define AARCH64_ISA_SME2 (aarch64_isa_flags & AARCH64_FL_SME2)
236 #define AARCH64_ISA_V8_3A (aarch64_isa_flags & AARCH64_FL_V8_3A)
237 #define AARCH64_ISA_DOTPROD (aarch64_isa_flags & AARCH64_FL_DOTPROD)
238 #define AARCH64_ISA_AES (aarch64_isa_flags & AARCH64_FL_AES)
239 #define AARCH64_ISA_SHA2 (aarch64_isa_flags & AARCH64_FL_SHA2)
240 #define AARCH64_ISA_V8_4A (aarch64_isa_flags & AARCH64_FL_V8_4A)
241 #define AARCH64_ISA_SM4 (aarch64_isa_flags & AARCH64_FL_SM4)
242 #define AARCH64_ISA_SHA3 (aarch64_isa_flags & AARCH64_FL_SHA3)
243 #define AARCH64_ISA_F16FML (aarch64_isa_flags & AARCH64_FL_F16FML)
244 #define AARCH64_ISA_RCPC (aarch64_isa_flags & AARCH64_FL_RCPC)
245 #define AARCH64_ISA_RCPC8_4 ((AARCH64_ISA_RCPC && AARCH64_ISA_V8_4A) \
246 || (aarch64_isa_flags & AARCH64_FL_RCPC3))
247 #define AARCH64_ISA_RNG (aarch64_isa_flags & AARCH64_FL_RNG)
248 #define AARCH64_ISA_V8_5A (aarch64_isa_flags & AARCH64_FL_V8_5A)
249 #define AARCH64_ISA_TME (aarch64_isa_flags & AARCH64_FL_TME)
250 #define AARCH64_ISA_MEMTAG (aarch64_isa_flags & AARCH64_FL_MEMTAG)
251 #define AARCH64_ISA_V8_6A (aarch64_isa_flags & AARCH64_FL_V8_6A)
252 #define AARCH64_ISA_I8MM (aarch64_isa_flags & AARCH64_FL_I8MM)
253 #define AARCH64_ISA_F32MM (aarch64_isa_flags & AARCH64_FL_F32MM)
254 #define AARCH64_ISA_F64MM (aarch64_isa_flags & AARCH64_FL_F64MM)
255 #define AARCH64_ISA_BF16 (aarch64_isa_flags & AARCH64_FL_BF16)
256 #define AARCH64_ISA_SB (aarch64_isa_flags & AARCH64_FL_SB)
257 #define AARCH64_ISA_RCPC3 (aarch64_isa_flags & AARCH64_FL_RCPC3)
258 #define AARCH64_ISA_V8R (aarch64_isa_flags & AARCH64_FL_V8R)
259 #define AARCH64_ISA_PAUTH (aarch64_isa_flags & AARCH64_FL_PAUTH)
260 #define AARCH64_ISA_V8_7A (aarch64_isa_flags & AARCH64_FL_V8_7A)
261 #define AARCH64_ISA_V8_8A (aarch64_isa_flags & AARCH64_FL_V8_8A)
262 #define AARCH64_ISA_V8_9A (aarch64_isa_flags & AARCH64_FL_V8_9A)
263 #define AARCH64_ISA_V9A (aarch64_isa_flags & AARCH64_FL_V9A)
264 #define AARCH64_ISA_V9_1A (aarch64_isa_flags & AARCH64_FL_V9_1A)
265 #define AARCH64_ISA_V9_2A (aarch64_isa_flags & AARCH64_FL_V9_2A)
266 #define AARCH64_ISA_V9_3A (aarch64_isa_flags & AARCH64_FL_V9_3A)
267 #define AARCH64_ISA_V9_4A (aarch64_isa_flags & AARCH64_FL_V9_4A)
268 #define AARCH64_ISA_MOPS (aarch64_isa_flags & AARCH64_FL_MOPS)
269 #define AARCH64_ISA_LS64 (aarch64_isa_flags & AARCH64_FL_LS64)
270 #define AARCH64_ISA_CSSC (aarch64_isa_flags & AARCH64_FL_CSSC)
271 #define AARCH64_ISA_D128 (aarch64_isa_flags & AARCH64_FL_D128)
272 #define AARCH64_ISA_THE (aarch64_isa_flags & AARCH64_FL_THE)
273 #define AARCH64_ISA_GCS (aarch64_isa_flags & AARCH64_FL_GCS)
275 /* The current function is a normal non-streaming function. */
276 #define TARGET_NON_STREAMING (AARCH64_ISA_SM_OFF)
278 /* The current function has a streaming body. */
279 #define TARGET_STREAMING (AARCH64_ISA_SM_ON)
281 /* The current function has a streaming-compatible body. */
282 #define TARGET_STREAMING_COMPATIBLE \
283 ((aarch64_isa_flags & AARCH64_FL_SM_STATE) == 0)
285 /* PSTATE.ZA is enabled in the current function body. */
286 #define TARGET_ZA (AARCH64_ISA_ZA_ON)
287 /* AARCH64_FL options necessary for system register implementation. */
289 /* Define AARCH64_FL aliases for architectural features which are protected
290 by -march flags in binutils but which receive no special treatment by GCC.
292 Such flags are inherited from the Binutils definition of system registers
293 and are mapped to the architecture in which the feature is implemented. */
294 #define AARCH64_FL_RAS AARCH64_FL_V8A
295 #define AARCH64_FL_LOR AARCH64_FL_V8_1A
296 #define AARCH64_FL_PAN AARCH64_FL_V8_1A
297 #define AARCH64_FL_AMU AARCH64_FL_V8_4A
298 #define AARCH64_FL_SCXTNUM AARCH64_FL_V8_5A
299 #define AARCH64_FL_ID_PFR2 AARCH64_FL_V8_5A
301 /* Armv8.9-A extension feature bits defined in Binutils but absent from GCC,
302 aliased to their base architecture. */
303 #define AARCH64_FL_AIE AARCH64_FL_V8_9A
304 #define AARCH64_FL_DEBUGv8p9 AARCH64_FL_V8_9A
305 #define AARCH64_FL_FGT2 AARCH64_FL_V8_9A
306 #define AARCH64_FL_ITE AARCH64_FL_V8_9A
307 #define AARCH64_FL_PFAR AARCH64_FL_V8_9A
308 #define AARCH64_FL_PMUv3_ICNTR AARCH64_FL_V8_9A
309 #define AARCH64_FL_PMUv3_SS AARCH64_FL_V8_9A
310 #define AARCH64_FL_PMUv3p9 AARCH64_FL_V8_9A
311 #define AARCH64_FL_RASv2 AARCH64_FL_V8_9A
312 #define AARCH64_FL_S1PIE AARCH64_FL_V8_9A
313 #define AARCH64_FL_S1POE AARCH64_FL_V8_9A
314 #define AARCH64_FL_S2PIE AARCH64_FL_V8_9A
315 #define AARCH64_FL_S2POE AARCH64_FL_V8_9A
316 #define AARCH64_FL_SCTLR2 AARCH64_FL_V8_9A
317 #define AARCH64_FL_SEBEP AARCH64_FL_V8_9A
318 #define AARCH64_FL_SPE_FDS AARCH64_FL_V8_9A
319 #define AARCH64_FL_TCR2 AARCH64_FL_V8_9A
321 /* SHA2 is an optional extension to AdvSIMD. */
322 #define TARGET_SHA2 (AARCH64_ISA_SHA2)
324 /* SHA3 is an optional extension to AdvSIMD. */
325 #define TARGET_SHA3 (AARCH64_ISA_SHA3)
327 /* AES is an optional extension to AdvSIMD. */
328 #define TARGET_AES (AARCH64_ISA_AES)
330 /* SM is an optional extension to AdvSIMD. */
331 #define TARGET_SM4 (AARCH64_ISA_SM4)
333 /* FP16FML is an optional extension to AdvSIMD. */
334 #define TARGET_F16FML (TARGET_SIMD && AARCH64_ISA_F16FML && TARGET_FP_F16INST)
336 /* CRC instructions that can be enabled through +crc arch extension. */
337 #define TARGET_CRC32 (AARCH64_ISA_CRC)
339 /* Atomic instructions that can be enabled through the +lse extension. */
340 #define TARGET_LSE (AARCH64_ISA_LSE)
342 /* ARMv8.2-A FP16 support that can be enabled through the +fp16 extension. */
343 #define TARGET_FP_F16INST (AARCH64_ISA_F16)
344 #define TARGET_SIMD_F16INST (TARGET_SIMD && AARCH64_ISA_F16)
346 /* Dot Product is an optional extension to AdvSIMD enabled through +dotprod. */
347 #define TARGET_DOTPROD (AARCH64_ISA_DOTPROD)
349 /* SVE instructions, enabled through +sve. */
350 #define TARGET_SVE (AARCH64_ISA_SVE)
352 /* SVE2 instructions, enabled through +sve2. */
353 #define TARGET_SVE2 (AARCH64_ISA_SVE2)
355 /* SVE2 AES instructions, enabled through +sve2-aes. */
356 #define TARGET_SVE2_AES (AARCH64_ISA_SVE2_AES && TARGET_NON_STREAMING)
358 /* SVE2 BITPERM instructions, enabled through +sve2-bitperm. */
359 #define TARGET_SVE2_BITPERM (AARCH64_ISA_SVE2_BITPERM && TARGET_NON_STREAMING)
361 /* SVE2 SHA3 instructions, enabled through +sve2-sha3. */
362 #define TARGET_SVE2_SHA3 (AARCH64_ISA_SVE2_SHA3 && TARGET_NON_STREAMING)
364 /* SVE2 SM4 instructions, enabled through +sve2-sm4. */
365 #define TARGET_SVE2_SM4 (AARCH64_ISA_SVE2_SM4 && TARGET_NON_STREAMING)
367 /* SME instructions, enabled through +sme. Note that this does not
368 imply anything about the state of PSTATE.SM. */
369 #define TARGET_SME (AARCH64_ISA_SME)
371 /* Same with streaming mode enabled. */
372 #define TARGET_STREAMING_SME (TARGET_STREAMING && TARGET_SME)
374 /* The FEAT_SME_I16I64 extension to SME, enabled through +sme-i16i64. */
375 #define TARGET_SME_I16I64 (AARCH64_ISA_SME_I16I64)
377 /* The FEAT_SME_F64F64 extension to SME, enabled through +sme-f64f64. */
378 #define TARGET_SME_F64F64 (AARCH64_ISA_SME_F64F64)
380 /* SME2 instructions, enabled through +sme2. */
381 #define TARGET_SME2 (AARCH64_ISA_SME2)
383 /* Same with streaming mode enabled. */
384 #define TARGET_STREAMING_SME2 (TARGET_STREAMING && TARGET_SME2)
386 /* ARMv8.3-A features. */
387 #define TARGET_ARMV8_3 (AARCH64_ISA_V8_3A)
389 /* Javascript conversion instruction from Armv8.3-a. */
390 #define TARGET_JSCVT (TARGET_FLOAT && AARCH64_ISA_V8_3A)
392 /* Armv8.3-a Complex number extension to AdvSIMD extensions. */
393 #define TARGET_COMPLEX (TARGET_SIMD && TARGET_ARMV8_3)
395 /* Floating-point rounding instructions from Armv8.5-a. */
396 #define TARGET_FRINT (AARCH64_ISA_V8_5A && TARGET_FLOAT)
398 /* TME instructions are enabled. */
399 #define TARGET_TME (AARCH64_ISA_TME)
401 /* Random number instructions from Armv8.5-a. */
402 #define TARGET_RNG (AARCH64_ISA_RNG)
404 /* Memory Tagging instructions optional to Armv8.5 enabled through +memtag. */
405 #define TARGET_MEMTAG (AARCH64_ISA_MEMTAG)
407 /* I8MM instructions are enabled through +i8mm. */
408 #define TARGET_I8MM (AARCH64_ISA_I8MM)
409 #define TARGET_SVE_I8MM (TARGET_SVE && AARCH64_ISA_I8MM)
411 /* F32MM instructions are enabled through +f32mm. */
412 #define TARGET_SVE_F32MM (AARCH64_ISA_F32MM)
414 /* F64MM instructions are enabled through +f64mm. */
415 #define TARGET_SVE_F64MM (AARCH64_ISA_F64MM)
417 /* BF16 instructions are enabled through +bf16. */
418 #define TARGET_BF16_FP (AARCH64_ISA_BF16)
419 #define TARGET_BF16_SIMD (AARCH64_ISA_BF16 && TARGET_SIMD)
420 #define TARGET_SVE_BF16 (TARGET_SVE && AARCH64_ISA_BF16)
422 /* PAUTH instructions are enabled through +pauth. */
423 #define TARGET_PAUTH (AARCH64_ISA_PAUTH)
425 /* BTI instructions exist from Armv8.5-a onwards. Their automatic use is
426 enabled through -mbranch-protection by using NOP-space instructions,
427 but this TARGET_ is used for defining BTI-related ACLE things. */
428 #define TARGET_BTI (AARCH64_ISA_V8_5A)
430 /* MOPS instructions are enabled through +mops. */
431 #define TARGET_MOPS (AARCH64_ISA_MOPS)
433 /* LS64 instructions are enabled through +ls64. */
434 #define TARGET_LS64 (AARCH64_ISA_LS64)
436 /* CSSC instructions are enabled through +cssc. */
437 #define TARGET_CSSC (AARCH64_ISA_CSSC)
439 /* Make sure this is always defined so we don't have to check for ifdefs
440 but rather use normal ifs. */
441 #ifndef TARGET_FIX_ERR_A53_835769_DEFAULT
442 #define TARGET_FIX_ERR_A53_835769_DEFAULT 0
443 #else
444 #undef TARGET_FIX_ERR_A53_835769_DEFAULT
445 #define TARGET_FIX_ERR_A53_835769_DEFAULT 1
446 #endif
448 /* SB instruction is enabled through +sb. */
449 #define TARGET_SB (AARCH64_ISA_SB)
451 /* RCPC loads from Armv8.3-a. */
452 #define TARGET_RCPC (AARCH64_ISA_RCPC)
454 /* The RCPC2 extensions from Armv8.4-a that allow immediate offsets to LDAPR
455 and sign-extending versions.*/
456 #define TARGET_RCPC2 (AARCH64_ISA_RCPC8_4)
458 /* RCPC3 (Release Consistency) extensions, optional from Armv8.2-a. */
459 #define TARGET_RCPC3 (AARCH64_ISA_RCPC3)
461 /* Apply the workaround for Cortex-A53 erratum 835769. */
462 #define TARGET_FIX_ERR_A53_835769 \
463 ((aarch64_fix_a53_err835769 == 2) \
464 ? TARGET_FIX_ERR_A53_835769_DEFAULT : aarch64_fix_a53_err835769)
466 /* Make sure this is always defined so we don't have to check for ifdefs
467 but rather use normal ifs. */
468 #ifndef TARGET_FIX_ERR_A53_843419_DEFAULT
469 #define TARGET_FIX_ERR_A53_843419_DEFAULT 0
470 #else
471 #undef TARGET_FIX_ERR_A53_843419_DEFAULT
472 #define TARGET_FIX_ERR_A53_843419_DEFAULT 1
473 #endif
475 /* Apply the workaround for Cortex-A53 erratum 843419. */
476 #define TARGET_FIX_ERR_A53_843419 \
477 ((aarch64_fix_a53_err843419 == 2) \
478 ? TARGET_FIX_ERR_A53_843419_DEFAULT : aarch64_fix_a53_err843419)
480 /* ARMv8.1-A Adv.SIMD support. */
481 #define TARGET_SIMD_RDMA (TARGET_SIMD && AARCH64_ISA_RDMA)
483 /* Armv9.4-A features. */
484 #define TARGET_ARMV9_4 (AARCH64_ISA_V9_4A)
486 /* 128-bit System Registers and Instructions from Armv9.4-a are enabled
487 through +d128. */
488 #define TARGET_D128 (AARCH64_ISA_D128)
490 /* Armv8.9-A/9.4-A Translation Hardening Extension system registers are
491 enabled through +the. */
492 #define TARGET_THE (AARCH64_ISA_THE)
494 /* Armv9.4-A Guarded Control Stack extension system registers are
495 enabled through +gcs. */
496 #define TARGET_GCS (AARCH64_ISA_GCS)
498 /* Prefer different predicate registers for the output of a predicated
499 operation over re-using an existing input predicate. */
500 #define TARGET_SVE_PRED_CLOBBER (TARGET_SVE \
501 && (aarch64_tune_params.extra_tuning_flags \
502 & AARCH64_EXTRA_TUNE_AVOID_PRED_RMW))
504 /* Standard register usage. */
506 /* 31 64-bit general purpose registers R0-R30:
507 R30 LR (link register)
508 R29 FP (frame pointer)
509 R19-R28 Callee-saved registers
510 R18 The platform register; use as temporary register.
511 R17 IP1 The second intra-procedure-call temporary register
512 (can be used by call veneers and PLT code); otherwise use
513 as a temporary register
514 R16 IP0 The first intra-procedure-call temporary register (can
515 be used by call veneers and PLT code); otherwise use as a
516 temporary register
517 R9-R15 Temporary registers
518 R8 Structure value parameter / temporary register
519 R0-R7 Parameter/result registers
521 SP stack pointer, encoded as X/R31 where permitted.
522 ZR zero register, encoded as X/R31 elsewhere
524 32 x 128-bit floating-point/vector registers
525 V16-V31 Caller-saved (temporary) registers
526 V8-V15 Callee-saved registers
527 V0-V7 Parameter/result registers
529 The vector register V0 holds scalar B0, H0, S0 and D0 in its least
530 significant bits. Unlike AArch32 S1 is not packed into D0, etc.
532 P0-P7 Predicate low registers: valid in all predicate contexts
533 P8-P15 Predicate high registers: used as scratch space
535 FFR First Fault Register, a fixed-use SVE predicate register
536 FFRT FFR token: a fake register used for modelling dependencies
538 VG Pseudo "vector granules" register
540 VG is the number of 64-bit elements in an SVE vector. We define
541 it as a hard register so that we can easily map it to the DWARF VG
542 register. GCC internally uses the poly_int variable aarch64_sve_vg
543 instead. */
545 #define FIXED_X18 0
546 #define CALL_USED_X18 1
548 #define FIXED_REGISTERS \
550 0, 0, 0, 0, 0, 0, 0, 0, /* R0 - R7 */ \
551 0, 0, 0, 0, 0, 0, 0, 0, /* R8 - R15 */ \
552 0, 0, FIXED_X18, 0, 0, 0, 0, 0, /* R16 - R23. */ \
553 0, 0, 0, 0, 0, 1, 0, 1, /* R24 - R30, SP */ \
554 0, 0, 0, 0, 0, 0, 0, 0, /* V0 - V7 */ \
555 0, 0, 0, 0, 0, 0, 0, 0, /* V8 - V15 */ \
556 0, 0, 0, 0, 0, 0, 0, 0, /* V16 - V23 */ \
557 0, 0, 0, 0, 0, 0, 0, 0, /* V24 - V31 */ \
558 1, 1, 1, 1, /* SFP, AP, CC, VG */ \
559 0, 0, 0, 0, 0, 0, 0, 0, /* P0 - P7 */ \
560 0, 0, 0, 0, 0, 0, 0, 0, /* P8 - P15 */ \
561 1, 1, /* FFR and FFRT */ \
562 1, 1, 1, 1, 1, 1, 1, 1 /* Fake registers */ \
565 /* X30 is marked as caller-saved which is in line with regular function call
566 behavior since the call instructions clobber it; AARCH64_EXPAND_CALL does
567 that for regular function calls and avoids it for sibcalls. X30 is
568 considered live for sibcalls; EPILOGUE_USES helps achieve that by returning
569 true but not until function epilogues have been generated. This ensures
570 that X30 is available for use in leaf functions if needed. */
572 #define CALL_REALLY_USED_REGISTERS \
574 1, 1, 1, 1, 1, 1, 1, 1, /* R0 - R7 */ \
575 1, 1, 1, 1, 1, 1, 1, 1, /* R8 - R15 */ \
576 1, 1, CALL_USED_X18, 0, 0, 0, 0, 0, /* R16 - R23. */ \
577 0, 0, 0, 0, 0, 1, 1, 1, /* R24 - R30, SP */ \
578 1, 1, 1, 1, 1, 1, 1, 1, /* V0 - V7 */ \
579 0, 0, 0, 0, 0, 0, 0, 0, /* V8 - V15 */ \
580 1, 1, 1, 1, 1, 1, 1, 1, /* V16 - V23 */ \
581 1, 1, 1, 1, 1, 1, 1, 1, /* V24 - V31 */ \
582 1, 1, 1, 0, /* SFP, AP, CC, VG */ \
583 1, 1, 1, 1, 1, 1, 1, 1, /* P0 - P7 */ \
584 1, 1, 1, 1, 1, 1, 1, 1, /* P8 - P15 */ \
585 1, 1, /* FFR and FFRT */ \
586 0, 0, 0, 0, 0, 0, 0, 0 /* Fake registers */ \
589 #define REGISTER_NAMES \
591 "x0", "x1", "x2", "x3", "x4", "x5", "x6", "x7", \
592 "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", \
593 "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23", \
594 "x24", "x25", "x26", "x27", "x28", "x29", "x30", "sp", \
595 "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", \
596 "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", \
597 "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", \
598 "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", \
599 "sfp", "ap", "cc", "vg", \
600 "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", \
601 "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", \
602 "ffr", "ffrt", \
603 "lowering", "tpidr2_block", "sme_state", "tpidr2_setup", \
604 "za_free", "za_saved", "za", "zt0" \
607 /* Generate the register aliases for core register N */
608 #define R_ALIASES(N) {"r" # N, R0_REGNUM + (N)}, \
609 {"w" # N, R0_REGNUM + (N)}
611 #define V_ALIASES(N) {"q" # N, V0_REGNUM + (N)}, \
612 {"d" # N, V0_REGNUM + (N)}, \
613 {"s" # N, V0_REGNUM + (N)}, \
614 {"h" # N, V0_REGNUM + (N)}, \
615 {"b" # N, V0_REGNUM + (N)}, \
616 {"z" # N, V0_REGNUM + (N)}
618 #define P_ALIASES(N) {"pn" # N, P0_REGNUM + (N)}
620 /* Provide aliases for all of the ISA defined register name forms.
621 These aliases are convenient for use in the clobber lists of inline
622 asm statements. */
624 #define ADDITIONAL_REGISTER_NAMES \
625 { R_ALIASES(0), R_ALIASES(1), R_ALIASES(2), R_ALIASES(3), \
626 R_ALIASES(4), R_ALIASES(5), R_ALIASES(6), R_ALIASES(7), \
627 R_ALIASES(8), R_ALIASES(9), R_ALIASES(10), R_ALIASES(11), \
628 R_ALIASES(12), R_ALIASES(13), R_ALIASES(14), R_ALIASES(15), \
629 R_ALIASES(16), R_ALIASES(17), R_ALIASES(18), R_ALIASES(19), \
630 R_ALIASES(20), R_ALIASES(21), R_ALIASES(22), R_ALIASES(23), \
631 R_ALIASES(24), R_ALIASES(25), R_ALIASES(26), R_ALIASES(27), \
632 R_ALIASES(28), R_ALIASES(29), R_ALIASES(30), {"wsp", R0_REGNUM + 31}, \
633 V_ALIASES(0), V_ALIASES(1), V_ALIASES(2), V_ALIASES(3), \
634 V_ALIASES(4), V_ALIASES(5), V_ALIASES(6), V_ALIASES(7), \
635 V_ALIASES(8), V_ALIASES(9), V_ALIASES(10), V_ALIASES(11), \
636 V_ALIASES(12), V_ALIASES(13), V_ALIASES(14), V_ALIASES(15), \
637 V_ALIASES(16), V_ALIASES(17), V_ALIASES(18), V_ALIASES(19), \
638 V_ALIASES(20), V_ALIASES(21), V_ALIASES(22), V_ALIASES(23), \
639 V_ALIASES(24), V_ALIASES(25), V_ALIASES(26), V_ALIASES(27), \
640 V_ALIASES(28), V_ALIASES(29), V_ALIASES(30), V_ALIASES(31), \
641 P_ALIASES(0), P_ALIASES(1), P_ALIASES(2), P_ALIASES(3), \
642 P_ALIASES(4), P_ALIASES(5), P_ALIASES(6), P_ALIASES(7), \
643 P_ALIASES(8), P_ALIASES(9), P_ALIASES(10), P_ALIASES(11), \
644 P_ALIASES(12), P_ALIASES(13), P_ALIASES(14), P_ALIASES(15) \
647 #define EPILOGUE_USES(REGNO) (aarch64_epilogue_uses (REGNO))
649 /* EXIT_IGNORE_STACK should be nonzero if, when returning from a function,
650 the stack pointer does not matter. This is only true if the function
651 uses alloca. */
652 #define EXIT_IGNORE_STACK (cfun->calls_alloca)
654 #define STATIC_CHAIN_REGNUM R18_REGNUM
655 #define HARD_FRAME_POINTER_REGNUM R29_REGNUM
656 #define FRAME_POINTER_REGNUM SFP_REGNUM
657 #define STACK_POINTER_REGNUM SP_REGNUM
658 #define ARG_POINTER_REGNUM AP_REGNUM
659 #define FIRST_PSEUDO_REGISTER (LAST_FAKE_REGNUM + 1)
661 /* The number of argument registers available for each class. */
662 #define NUM_ARG_REGS 8
663 #define NUM_FP_ARG_REGS 8
664 #define NUM_PR_ARG_REGS 4
666 /* A Homogeneous Floating-Point or Short-Vector Aggregate may have at most
667 four members. */
668 #define HA_MAX_NUM_FLDS 4
670 /* External dwarf register number scheme. These number are used to
671 identify registers in dwarf debug information, the values are
672 defined by the AArch64 ABI. The numbering scheme is independent of
673 GCC's internal register numbering scheme. */
675 #define AARCH64_DWARF_R0 0
677 /* The number of R registers, note 31! not 32. */
678 #define AARCH64_DWARF_NUMBER_R 31
680 #define AARCH64_DWARF_SP 31
681 #define AARCH64_DWARF_VG 46
682 #define AARCH64_DWARF_P0 48
683 #define AARCH64_DWARF_V0 64
685 /* The number of V registers. */
686 #define AARCH64_DWARF_NUMBER_V 32
688 /* For signal frames we need to use an alternative return column. This
689 value must not correspond to a hard register and must be out of the
690 range of DWARF_FRAME_REGNUM(). */
691 #define DWARF_ALT_FRAME_RETURN_COLUMN \
692 (AARCH64_DWARF_V0 + AARCH64_DWARF_NUMBER_V)
694 /* We add 1 extra frame register for use as the
695 DWARF_ALT_FRAME_RETURN_COLUMN. */
696 #define DWARF_FRAME_REGISTERS (DWARF_ALT_FRAME_RETURN_COLUMN + 1)
699 #define DEBUGGER_REGNO(REGNO) aarch64_debugger_regno (REGNO)
700 /* Provide a definition of DWARF_FRAME_REGNUM here so that fallback unwinders
701 can use DWARF_ALT_FRAME_RETURN_COLUMN defined below. This is just the same
702 as the default definition in dwarf2out.cc. */
703 #undef DWARF_FRAME_REGNUM
704 #define DWARF_FRAME_REGNUM(REGNO) DEBUGGER_REGNO (REGNO)
706 #define DWARF_FRAME_RETURN_COLUMN DWARF_FRAME_REGNUM (LR_REGNUM)
708 #define DWARF2_UNWIND_INFO 1
710 /* Use R0 through R3 to pass exception handling information. */
711 #define EH_RETURN_DATA_REGNO(N) \
712 ((N) < 4 ? ((unsigned int) R0_REGNUM + (N)) : INVALID_REGNUM)
714 /* Select a format to encode pointers in exception handling data. */
715 #define ASM_PREFERRED_EH_DATA_FORMAT(CODE, GLOBAL) \
716 aarch64_asm_preferred_eh_data_format ((CODE), (GLOBAL))
718 /* Output the assembly strings we want to add to a function definition. */
719 #define ASM_DECLARE_FUNCTION_NAME(STR, NAME, DECL) \
720 aarch64_declare_function_name (STR, NAME, DECL)
722 /* Output assembly strings for alias definition. */
723 #define ASM_OUTPUT_DEF_FROM_DECLS(STR, DECL, TARGET) \
724 aarch64_asm_output_alias (STR, DECL, TARGET)
726 /* Output assembly strings for undefined extern symbols. */
727 #undef ASM_OUTPUT_EXTERNAL
728 #define ASM_OUTPUT_EXTERNAL(STR, DECL, NAME) \
729 aarch64_asm_output_external (STR, DECL, NAME)
731 /* Output assembly strings after .cfi_startproc is emitted. */
732 #define ASM_POST_CFI_STARTPROC aarch64_post_cfi_startproc
734 /* For EH returns X4 is a flag that is set in the EH return
735 code paths and then X5 and X6 contain the stack adjustment
736 and return address respectively. */
737 #define EH_RETURN_TAKEN_RTX gen_rtx_REG (Pmode, R4_REGNUM)
738 #define EH_RETURN_STACKADJ_RTX gen_rtx_REG (Pmode, R5_REGNUM)
739 #define EH_RETURN_HANDLER_RTX gen_rtx_REG (Pmode, R6_REGNUM)
741 #undef TARGET_COMPUTE_FRAME_LAYOUT
742 #define TARGET_COMPUTE_FRAME_LAYOUT aarch64_layout_frame
744 /* Register in which the structure value is to be returned. */
745 #define AARCH64_STRUCT_VALUE_REGNUM R8_REGNUM
747 /* Non-zero if REGNO is part of the Core register set.
749 The rather unusual way of expressing this check is to avoid
750 warnings when building the compiler when R0_REGNUM is 0 and REGNO
751 is unsigned. */
752 #define GP_REGNUM_P(REGNO) \
753 (((unsigned) (REGNO - R0_REGNUM)) <= (R30_REGNUM - R0_REGNUM))
755 /* Registers known to be preserved over a BL instruction. This consists of the
756 GENERAL_REGS without x16, x17, and x30. The x30 register is changed by the
757 BL instruction itself, while the x16 and x17 registers may be used by
758 veneers which can be inserted by the linker. */
759 #define STUB_REGNUM_P(REGNO) \
760 (GP_REGNUM_P (REGNO) \
761 && (REGNO) != R16_REGNUM \
762 && (REGNO) != R17_REGNUM \
763 && (REGNO) != R30_REGNUM) \
765 #define W8_W11_REGNUM_P(REGNO) \
766 IN_RANGE (REGNO, R8_REGNUM, R11_REGNUM)
768 #define W12_W15_REGNUM_P(REGNO) \
769 IN_RANGE (REGNO, R12_REGNUM, R15_REGNUM)
771 #define FP_REGNUM_P(REGNO) \
772 (((unsigned) (REGNO - V0_REGNUM)) <= (V31_REGNUM - V0_REGNUM))
774 #define FP_LO_REGNUM_P(REGNO) \
775 (((unsigned) (REGNO - V0_REGNUM)) <= (V15_REGNUM - V0_REGNUM))
777 #define FP_LO8_REGNUM_P(REGNO) \
778 (((unsigned) (REGNO - V0_REGNUM)) <= (V7_REGNUM - V0_REGNUM))
780 #define PR_REGNUM_P(REGNO)\
781 (((unsigned) (REGNO - P0_REGNUM)) <= (P15_REGNUM - P0_REGNUM))
783 #define PR_LO_REGNUM_P(REGNO)\
784 (((unsigned) (REGNO - P0_REGNUM)) <= (P7_REGNUM - P0_REGNUM))
786 #define FP_SIMD_SAVED_REGNUM_P(REGNO) \
787 (((unsigned) (REGNO - V8_REGNUM)) <= (V23_REGNUM - V8_REGNUM))
789 #define FAKE_REGNUM_P(REGNO) \
790 IN_RANGE (REGNO, FIRST_FAKE_REGNUM, LAST_FAKE_REGNUM)
792 /* Register and constant classes. */
794 enum reg_class
796 NO_REGS,
797 W8_W11_REGS,
798 W12_W15_REGS,
799 TAILCALL_ADDR_REGS,
800 STUB_REGS,
801 GENERAL_REGS,
802 STACK_REG,
803 POINTER_REGS,
804 FP_LO8_REGS,
805 FP_LO_REGS,
806 FP_REGS,
807 POINTER_AND_FP_REGS,
808 PR_LO_REGS,
809 PR_HI_REGS,
810 PR_REGS,
811 FFR_REGS,
812 PR_AND_FFR_REGS,
813 FAKE_REGS,
814 ALL_REGS,
815 LIM_REG_CLASSES /* Last */
818 #define N_REG_CLASSES ((int) LIM_REG_CLASSES)
820 #define REG_CLASS_NAMES \
822 "NO_REGS", \
823 "W8_W11_REGS", \
824 "W12_W15_REGS", \
825 "TAILCALL_ADDR_REGS", \
826 "STUB_REGS", \
827 "GENERAL_REGS", \
828 "STACK_REG", \
829 "POINTER_REGS", \
830 "FP_LO8_REGS", \
831 "FP_LO_REGS", \
832 "FP_REGS", \
833 "POINTER_AND_FP_REGS", \
834 "PR_LO_REGS", \
835 "PR_HI_REGS", \
836 "PR_REGS", \
837 "FFR_REGS", \
838 "PR_AND_FFR_REGS", \
839 "FAKE_REGS", \
840 "ALL_REGS" \
843 #define REG_CLASS_CONTENTS \
845 { 0x00000000, 0x00000000, 0x00000000 }, /* NO_REGS */ \
846 { 0x00000f00, 0x00000000, 0x00000000 }, /* W8_W11_REGS */ \
847 { 0x0000f000, 0x00000000, 0x00000000 }, /* W12_W15_REGS */ \
848 { 0x00030000, 0x00000000, 0x00000000 }, /* TAILCALL_ADDR_REGS */\
849 { 0x3ffcffff, 0x00000000, 0x00000000 }, /* STUB_REGS */ \
850 { 0x7fffffff, 0x00000000, 0x00000003 }, /* GENERAL_REGS */ \
851 { 0x80000000, 0x00000000, 0x00000000 }, /* STACK_REG */ \
852 { 0xffffffff, 0x00000000, 0x00000003 }, /* POINTER_REGS */ \
853 { 0x00000000, 0x000000ff, 0x00000000 }, /* FP_LO8_REGS */ \
854 { 0x00000000, 0x0000ffff, 0x00000000 }, /* FP_LO_REGS */ \
855 { 0x00000000, 0xffffffff, 0x00000000 }, /* FP_REGS */ \
856 { 0xffffffff, 0xffffffff, 0x00000003 }, /* POINTER_AND_FP_REGS */\
857 { 0x00000000, 0x00000000, 0x00000ff0 }, /* PR_LO_REGS */ \
858 { 0x00000000, 0x00000000, 0x000ff000 }, /* PR_HI_REGS */ \
859 { 0x00000000, 0x00000000, 0x000ffff0 }, /* PR_REGS */ \
860 { 0x00000000, 0x00000000, 0x00300000 }, /* FFR_REGS */ \
861 { 0x00000000, 0x00000000, 0x003ffff0 }, /* PR_AND_FFR_REGS */ \
862 { 0x00000000, 0x00000000, 0x3fc00000 }, /* FAKE_REGS */ \
863 { 0xffffffff, 0xffffffff, 0x000fffff } /* ALL_REGS */ \
866 #define REGNO_REG_CLASS(REGNO) aarch64_regno_regclass (REGNO)
868 #define INDEX_REG_CLASS GENERAL_REGS
869 #define BASE_REG_CLASS POINTER_REGS
871 /* Register pairs used to eliminate unneeded registers that point into
872 the stack frame. */
873 #define ELIMINABLE_REGS \
875 { ARG_POINTER_REGNUM, STACK_POINTER_REGNUM }, \
876 { ARG_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM }, \
877 { FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM }, \
878 { FRAME_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM }, \
881 #define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \
882 (OFFSET) = aarch64_initial_elimination_offset (FROM, TO)
884 /* CPU/ARCH option handling. */
885 #include "config/aarch64/aarch64-opts.h"
887 enum target_cpus
889 #define AARCH64_CORE(NAME, INTERNAL_IDENT, SCHED, ARCH, FLAGS, COSTS, IMP, PART, VARIANT) \
890 TARGET_CPU_##INTERNAL_IDENT,
891 #include "aarch64-cores.def"
894 /* If there is no CPU defined at configure, use generic as default. */
895 #ifndef TARGET_CPU_DEFAULT
896 # define TARGET_CPU_DEFAULT TARGET_CPU_generic_armv8_a
897 #endif
899 /* If inserting NOP before a mult-accumulate insn remember to adjust the
900 length so that conditional branching code is updated appropriately. */
901 #define ADJUST_INSN_LENGTH(insn, length) \
902 do \
904 if (aarch64_madd_needs_nop (insn)) \
905 length += 4; \
906 } while (0)
908 #define FINAL_PRESCAN_INSN(INSN, OPVEC, NOPERANDS) \
909 aarch64_final_prescan_insn (INSN); \
911 /* The processor for which instructions should be scheduled. */
912 extern enum aarch64_processor aarch64_tune;
914 /* RTL generation support. */
915 #define INIT_EXPANDERS aarch64_init_expanders ()
918 /* Stack layout; function entry, exit and calling. */
919 #define STACK_GROWS_DOWNWARD 1
921 #define FRAME_GROWS_DOWNWARD 1
923 #define ACCUMULATE_OUTGOING_ARGS 1
925 #define FIRST_PARM_OFFSET(FNDECL) 0
927 /* Fix for VFP */
928 #define LIBCALL_VALUE(MODE) \
929 gen_rtx_REG (MODE, FLOAT_MODE_P (MODE) ? V0_REGNUM : R0_REGNUM)
931 #define DEFAULT_PCC_STRUCT_RETURN 0
933 #if defined(HAVE_POLY_INT_H) && defined(GCC_VEC_H)
934 struct GTY (()) aarch64_frame
936 /* The offset from the bottom of the static frame (the bottom of the
937 outgoing arguments) of each register save slot, or -2 if no save is
938 needed. */
939 poly_int64 reg_offset[LAST_SAVED_REGNUM + 1];
941 /* The list of GPRs, FPRs and predicate registers that have nonnegative
942 entries in reg_offset. The registers are listed in order of
943 increasing offset (rather than increasing register number). */
944 vec<unsigned, va_gc_atomic> *saved_gprs;
945 vec<unsigned, va_gc_atomic> *saved_fprs;
946 vec<unsigned, va_gc_atomic> *saved_prs;
948 /* The offset from the base of the frame of a 64-bit slot whose low
949 bit contains the incoming value of PSTATE.SM. This slot must be
950 within reach of the hard frame pointer.
952 The offset is -1 if such a slot isn't needed. */
953 poly_int64 old_svcr_offset;
955 /* The number of extra stack bytes taken up by register varargs.
956 This area is allocated by the callee at the very top of the
957 frame. This value is rounded up to a multiple of
958 STACK_BOUNDARY. */
959 HOST_WIDE_INT saved_varargs_size;
961 /* The number of bytes between the bottom of the static frame (the bottom
962 of the outgoing arguments) and the bottom of the register save area.
963 This value is always a multiple of STACK_BOUNDARY. */
964 poly_int64 bytes_below_saved_regs;
966 /* The number of bytes between the bottom of the static frame (the bottom
967 of the outgoing arguments) and the hard frame pointer. This value is
968 always a multiple of STACK_BOUNDARY. */
969 poly_int64 bytes_below_hard_fp;
971 /* The number of bytes between the top of the locals area and the top
972 of the frame (the incomming SP). This value is always a multiple of
973 STACK_BOUNDARY. */
974 poly_int64 bytes_above_locals;
976 /* The number of bytes between the hard_frame_pointer and the top of
977 the frame (the incomming SP). This value is always a multiple of
978 STACK_BOUNDARY. */
979 poly_int64 bytes_above_hard_fp;
981 /* The size of the frame, i.e. the number of bytes between the bottom
982 of the outgoing arguments and the incoming SP. This value is always
983 a multiple of STACK_BOUNDARY. */
984 poly_int64 frame_size;
986 /* The size of the initial stack adjustment before saving callee-saves. */
987 poly_int64 initial_adjust;
989 /* The writeback value when pushing callee-save registers.
990 It is zero when no push is used. */
991 HOST_WIDE_INT callee_adjust;
993 /* The size of the stack adjustment before saving or after restoring
994 SVE registers. */
995 poly_int64 sve_callee_adjust;
997 /* The size of the stack adjustment after saving callee-saves. */
998 poly_int64 final_adjust;
1000 /* Store FP,LR and setup a frame pointer. */
1001 bool emit_frame_chain;
1003 /* In each frame, we can associate up to two register saves with the
1004 initial stack allocation. This happens in one of two ways:
1006 (1) Using an STR or STP with writeback to perform the initial
1007 stack allocation. When EMIT_FRAME_CHAIN, the registers will
1008 be those needed to create a frame chain.
1010 Indicated by CALLEE_ADJUST != 0.
1012 (2) Using a separate STP to set up the frame record, after the
1013 initial stack allocation but before setting up the frame pointer.
1014 This is used if the offset is too large to use writeback.
1016 Indicated by CALLEE_ADJUST == 0 && EMIT_FRAME_CHAIN.
1018 These fields indicate which registers we've decided to handle using
1019 (1) or (2), or INVALID_REGNUM if none.
1021 In some cases we don't always need to pop all registers in the push
1022 candidates, pop candidates record which registers need to be popped
1023 eventually. The initial value of a pop candidate is copied from its
1024 corresponding push candidate.
1026 Currently, different pop candidates are only used for shadow call
1027 stack. When "-fsanitize=shadow-call-stack" is specified, we replace
1028 x30 in the pop candidate with INVALID_REGNUM to ensure that x30 is
1029 not popped twice. */
1030 unsigned wb_push_candidate1;
1031 unsigned wb_push_candidate2;
1032 unsigned wb_pop_candidate1;
1033 unsigned wb_pop_candidate2;
1035 /* Big-endian SVE frames need a spare predicate register in order
1036 to save vector registers in the correct layout for unwinding.
1037 This is the register they should use. */
1038 unsigned spare_pred_reg;
1040 /* An SVE register that is saved below the hard frame pointer and that acts
1041 as a probe for later allocations, or INVALID_REGNUM if none. */
1042 unsigned sve_save_and_probe;
1044 /* A register that is saved at the hard frame pointer and that acts
1045 as a probe for later allocations, or INVALID_REGNUM if none. */
1046 unsigned hard_fp_save_and_probe;
1048 bool laid_out;
1050 /* True if shadow call stack should be enabled for the current function. */
1051 bool is_scs_enabled;
1054 /* Private to winnt.cc. */
1055 struct seh_frame_state;
1057 #ifdef hash_set_h
1058 typedef struct GTY (()) machine_function
1060 struct aarch64_frame frame;
1061 /* One entry for each hard register. */
1062 bool reg_is_wrapped_separately[LAST_SAVED_REGNUM + 1];
1063 /* One entry for each general purpose register. */
1064 rtx call_via[SP_REGNUM];
1066 /* A pseudo register that points to the function's TPIDR2 block, or null
1067 if the function doesn't have a TPIDR2 block. */
1068 rtx tpidr2_block;
1070 /* A pseudo register that points to the function's ZA save buffer,
1071 or null if none. */
1072 rtx za_save_buffer;
1074 /* A stack slot that stores the contents of the function's ZT0 state. */
1075 rtx zt0_save_buffer;
1077 bool label_is_assembled;
1079 /* True if we've expanded at least one call to a function that changes
1080 PSTATE.SM. This should only be used for saving compile time: false
1081 guarantees that no such mode switch exists. */
1082 bool call_switches_pstate_sm;
1084 /* Used to generated unique identifiers for each update to ZA by an
1085 asm statement. */
1086 unsigned int next_asm_update_za_id;
1088 /* A set of all decls that have been passed to a vld1 intrinsic in the
1089 current function. This is used to help guide the vector cost model. */
1090 hash_set<tree> *vector_load_decls;
1092 /* An instruction that was emitted at the start of the function to
1093 set an Advanced SIMD pseudo register to zero. If the instruction
1094 still exists and still fulfils its original purpose. the same register
1095 can be reused by other code. */
1096 rtx_insn *advsimd_zero_insn;
1098 /* During SEH output, this is non-null. */
1099 struct seh_frame_state * GTY ((skip (""))) seh;
1100 } machine_function;
1101 #endif
1102 #endif
1104 /* Which ABI to use. */
1105 enum aarch64_abi_type
1107 AARCH64_ABI_LP64 = 0,
1108 AARCH64_ABI_ILP32 = 1
1111 #ifndef AARCH64_ABI_DEFAULT
1112 #define AARCH64_ABI_DEFAULT AARCH64_ABI_LP64
1113 #endif
1115 #define TARGET_ILP32 (aarch64_abi & AARCH64_ABI_ILP32)
1117 enum arm_pcs
1119 ARM_PCS_AAPCS64, /* Base standard AAPCS for 64 bit. */
1120 ARM_PCS_SIMD, /* For aarch64_vector_pcs functions. */
1121 ARM_PCS_SVE, /* For functions that pass or return
1122 values in SVE registers. */
1123 ARM_PCS_TLSDESC, /* For targets of tlsdesc calls. */
1124 ARM_PCS_UNKNOWN
1130 /* We can't use machine_mode inside a generator file because it
1131 hasn't been created yet; we shouldn't be using any code that
1132 needs the real definition though, so this ought to be safe. */
1133 #ifdef GENERATOR_FILE
1134 #define MACHMODE int
1135 #else
1136 #include "insn-modes.h"
1137 #define MACHMODE machine_mode
1138 #endif
1140 #ifndef USED_FOR_TARGET
1141 /* AAPCS related state tracking. */
1142 typedef struct
1144 enum arm_pcs pcs_variant;
1145 aarch64_feature_flags isa_mode;
1146 int aapcs_arg_processed; /* No need to lay out this argument again. */
1147 int aapcs_ncrn; /* Next Core register number. */
1148 int aapcs_nextncrn; /* Next next core register number. */
1149 int aapcs_nvrn; /* Next Vector register number. */
1150 int aapcs_nextnvrn; /* Next Next Vector register number. */
1151 int aapcs_nprn; /* Next Predicate register number. */
1152 int aapcs_nextnprn; /* Next Next Predicate register number. */
1153 rtx aapcs_reg; /* Register assigned to this argument. This
1154 is NULL_RTX if this parameter goes on
1155 the stack. */
1156 MACHMODE aapcs_vfp_rmode;
1157 int aapcs_stack_words; /* If the argument is passed on the stack, this
1158 is the number of words needed, after rounding
1159 up. Only meaningful when
1160 aapcs_reg == NULL_RTX. */
1161 int aapcs_stack_size; /* The total size (in words, per 8 byte) of the
1162 stack arg area so far. */
1163 bool silent_p; /* True if we should act silently, rather than
1164 raise an error for invalid calls. */
1166 /* AARCH64_STATE_* flags that describe whether the function shares ZA
1167 and ZT0 with its callers. */
1168 unsigned int shared_za_flags;
1169 unsigned int shared_zt0_flags;
1171 /* A list of registers that need to be saved and restored around a
1172 change to PSTATE.SM. An auto_vec would be more convenient, but those
1173 can't be copied. */
1174 unsigned int num_sme_mode_switch_args;
1175 rtx sme_mode_switch_args[12];
1176 } CUMULATIVE_ARGS;
1177 #endif
1179 #define BLOCK_REG_PADDING(MODE, TYPE, FIRST) \
1180 (aarch64_pad_reg_upward (MODE, TYPE, FIRST) ? PAD_UPWARD : PAD_DOWNWARD)
1182 #define PAD_VARARGS_DOWN 0
1184 #define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \
1185 aarch64_init_cumulative_args (&(CUM), FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS)
1187 #define FUNCTION_ARG_REGNO_P(REGNO) \
1188 aarch64_function_arg_regno_p(REGNO)
1191 /* ISA Features. */
1193 /* Addressing modes, etc. */
1194 #define HAVE_POST_INCREMENT 1
1195 #define HAVE_PRE_INCREMENT 1
1196 #define HAVE_POST_DECREMENT 1
1197 #define HAVE_PRE_DECREMENT 1
1198 #define HAVE_POST_MODIFY_DISP 1
1199 #define HAVE_PRE_MODIFY_DISP 1
1201 #define MAX_REGS_PER_ADDRESS 2
1203 #define CONSTANT_ADDRESS_P(X) aarch64_constant_address_p(X)
1205 #define REGNO_OK_FOR_BASE_P(REGNO) \
1206 aarch64_regno_ok_for_base_p (REGNO, true)
1208 #define REGNO_OK_FOR_INDEX_P(REGNO) \
1209 aarch64_regno_ok_for_index_p (REGNO, true)
1211 #define LEGITIMATE_PIC_OPERAND_P(X) \
1212 aarch64_legitimate_pic_operand_p (X)
1214 #define CASE_VECTOR_MODE Pmode
1216 #define DEFAULT_SIGNED_CHAR 0
1218 /* An integer expression for the size in bits of the largest integer machine
1219 mode that should actually be used. We allow pairs of registers. */
1220 #define MAX_FIXED_MODE_SIZE GET_MODE_BITSIZE (TImode)
1222 /* Maximum bytes moved by a single instruction (load/store pair). */
1223 #define MOVE_MAX (UNITS_PER_WORD * 2)
1225 /* The base cost overhead of a memcpy call, for MOVE_RATIO and friends. */
1226 #define AARCH64_CALL_RATIO 8
1228 /* MOVE_RATIO dictates when we will use the move_by_pieces infrastructure.
1229 move_by_pieces will continually copy the largest safe chunks. So a
1230 7-byte copy is a 4-byte + 2-byte + byte copy. This proves inefficient
1231 for both size and speed of copy, so we will instead use the "cpymem"
1232 standard name to implement the copy. This logic does not apply when
1233 targeting -mstrict-align or TARGET_MOPS, so keep a sensible default in
1234 that case. */
1235 #define MOVE_RATIO(speed) \
1236 ((!STRICT_ALIGNMENT || TARGET_MOPS) ? 2 : (((speed) ? 15 : AARCH64_CALL_RATIO) / 2))
1238 /* Like MOVE_RATIO, without -mstrict-align, make decisions in "setmem" when
1239 we would use more than 3 scalar instructions.
1240 Otherwise follow a sensible default: when optimizing for size, give a better
1241 estimate of the length of a memset call, but use the default otherwise. */
1242 #define CLEAR_RATIO(speed) \
1243 (!STRICT_ALIGNMENT ? (TARGET_MOPS ? 0 : 4) : (speed) ? 15 : AARCH64_CALL_RATIO)
1245 /* SET_RATIO is similar to CLEAR_RATIO, but for a non-zero constant. Without
1246 -mstrict-align, make decisions in "setmem". Otherwise follow a sensible
1247 default: when optimizing for size adjust the ratio to account for the
1248 overhead of loading the constant. */
1249 #define SET_RATIO(speed) \
1250 ((!STRICT_ALIGNMENT || TARGET_MOPS) ? 0 : (speed) ? 15 : AARCH64_CALL_RATIO - 2)
1252 /* Disable auto-increment in move_by_pieces et al. Use of auto-increment is
1253 rarely a good idea in straight-line code since it adds an extra address
1254 dependency between each instruction. Better to use incrementing offsets. */
1255 #define USE_LOAD_POST_INCREMENT(MODE) 0
1256 #define USE_LOAD_POST_DECREMENT(MODE) 0
1257 #define USE_LOAD_PRE_INCREMENT(MODE) 0
1258 #define USE_LOAD_PRE_DECREMENT(MODE) 0
1259 #define USE_STORE_POST_INCREMENT(MODE) 0
1260 #define USE_STORE_POST_DECREMENT(MODE) 0
1261 #define USE_STORE_PRE_INCREMENT(MODE) 0
1262 #define USE_STORE_PRE_DECREMENT(MODE) 0
1264 /* WORD_REGISTER_OPERATIONS does not hold for AArch64.
1265 The assigned word_mode is DImode but operations narrower than SImode
1266 behave as 32-bit operations if using the W-form of the registers rather
1267 than as word_mode (64-bit) operations as WORD_REGISTER_OPERATIONS
1268 expects. */
1269 #define WORD_REGISTER_OPERATIONS 0
1271 /* Define if loading from memory in MODE, an integral mode narrower than
1272 BITS_PER_WORD will either zero-extend or sign-extend. The value of this
1273 macro should be the code that says which one of the two operations is
1274 implicitly done, or UNKNOWN if none. */
1275 #define LOAD_EXTEND_OP(MODE) ZERO_EXTEND
1277 /* Define this macro to be non-zero if instructions will fail to work
1278 if given data not on the nominal alignment. */
1279 #define STRICT_ALIGNMENT TARGET_STRICT_ALIGN
1281 /* Enable wide bitfield accesses for more efficient bitfield code. */
1282 #define SLOW_BYTE_ACCESS 1
1284 #define NO_FUNCTION_CSE 1
1286 /* Specify the machine mode that the hardware addresses have.
1287 After generation of rtl, the compiler makes no further distinction
1288 between pointers and any other objects of this machine mode. */
1289 #define Pmode DImode
1291 /* A C expression whose value is zero if pointers that need to be extended
1292 from being `POINTER_SIZE' bits wide to `Pmode' are sign-extended and
1293 greater then zero if they are zero-extended and less then zero if the
1294 ptr_extend instruction should be used. */
1295 #define POINTERS_EXTEND_UNSIGNED 1
1297 /* Mode of a function address in a call instruction (for indexing purposes). */
1298 #define FUNCTION_MODE Pmode
1300 #define SELECT_CC_MODE(OP, X, Y) aarch64_select_cc_mode (OP, X, Y)
1302 /* Having an integer comparison mode guarantees that we can use
1303 reverse_condition, but the usual restrictions apply to floating-point
1304 comparisons. */
1305 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPmode && (MODE) != CCFPEmode)
1307 #define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
1308 ((VALUE) = GET_MODE_UNIT_BITSIZE (MODE), 2)
1309 #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
1310 ((VALUE) = GET_MODE_UNIT_BITSIZE (MODE), 2)
1312 #define INCOMING_RETURN_ADDR_RTX gen_rtx_REG (Pmode, LR_REGNUM)
1314 #define RETURN_ADDR_RTX aarch64_return_addr
1316 /* BTI c + 3 insns
1317 + sls barrier of DSB + ISB.
1318 + 2 pointer-sized entries. */
1319 #define TRAMPOLINE_SIZE (24 + (TARGET_ILP32 ? 8 : 16))
1321 /* Trampolines contain dwords, so must be dword aligned. */
1322 #define TRAMPOLINE_ALIGNMENT 64
1324 /* Put trampolines in the text section so that mapping symbols work
1325 correctly. */
1326 #define TRAMPOLINE_SECTION text_section
1328 /* To start with. */
1329 #define BRANCH_COST(SPEED_P, PREDICTABLE_P) \
1330 (aarch64_branch_cost (SPEED_P, PREDICTABLE_P))
1333 /* Assembly output. */
1335 /* For now we'll make all jump tables pc-relative. */
1336 #define CASE_VECTOR_PC_RELATIVE 1
1338 #define CASE_VECTOR_SHORTEN_MODE(min, max, body) \
1339 ((min < -0x1fff0 || max > 0x1fff0) ? SImode \
1340 : (min < -0x1f0 || max > 0x1f0) ? HImode \
1341 : QImode)
1343 /* Jump table alignment is explicit in ASM_OUTPUT_CASE_LABEL. */
1344 #define ADDR_VEC_ALIGN(JUMPTABLE) 0
1346 #define MCOUNT_NAME "_mcount"
1348 #define NO_PROFILE_COUNTERS 1
1350 /* Emit rtl for profiling. Output assembler code to FILE
1351 to call "_mcount" for profiling a function entry. */
1352 #define PROFILE_HOOK(LABEL) \
1354 rtx fun, lr; \
1355 lr = aarch64_return_addr_rtx (); \
1356 fun = gen_rtx_SYMBOL_REF (Pmode, MCOUNT_NAME); \
1357 emit_library_call (fun, LCT_NORMAL, VOIDmode, lr, Pmode); \
1360 /* All the work done in PROFILE_HOOK, but still required. */
1361 #define FUNCTION_PROFILER(STREAM, LABELNO) do { } while (0)
1363 /* For some reason, the Linux headers think they know how to define
1364 these macros. They don't!!! */
1365 #undef ASM_APP_ON
1366 #undef ASM_APP_OFF
1367 #define ASM_APP_ON "\t" ASM_COMMENT_START " Start of user assembly\n"
1368 #define ASM_APP_OFF "\t" ASM_COMMENT_START " End of user assembly\n"
1370 #define CONSTANT_POOL_BEFORE_FUNCTION 0
1372 /* This definition should be relocated to aarch64-elf-raw.h. This macro
1373 should be undefined in aarch64-linux.h and a clear_cache pattern
1374 implmented to emit either the call to __aarch64_sync_cache_range()
1375 directly or preferably the appropriate sycall or cache clear
1376 instructions inline. */
1377 #define CLEAR_INSN_CACHE(beg, end) \
1378 extern void __aarch64_sync_cache_range (void *, void *); \
1379 __aarch64_sync_cache_range (beg, end)
1381 #define SHIFT_COUNT_TRUNCATED (!TARGET_SIMD)
1383 /* Choose appropriate mode for caller saves, so we do the minimum
1384 required size of load/store. */
1385 #define HARD_REGNO_CALLER_SAVE_MODE(REGNO, NREGS, MODE) \
1386 aarch64_hard_regno_caller_save_mode ((REGNO), (NREGS), (MODE))
1388 #undef SWITCHABLE_TARGET
1389 #define SWITCHABLE_TARGET 1
1391 /* Check TLS Descriptors mechanism is selected. */
1392 #define TARGET_TLS_DESC (aarch64_tls_dialect == TLS_DESCRIPTORS)
1394 extern enum aarch64_code_model aarch64_cmodel;
1396 /* When using the tiny addressing model conditional and unconditional branches
1397 can span the whole of the available address space (1MB). */
1398 #define HAS_LONG_COND_BRANCH \
1399 (aarch64_cmodel == AARCH64_CMODEL_TINY \
1400 || aarch64_cmodel == AARCH64_CMODEL_TINY_PIC)
1402 #define HAS_LONG_UNCOND_BRANCH \
1403 (aarch64_cmodel == AARCH64_CMODEL_TINY \
1404 || aarch64_cmodel == AARCH64_CMODEL_TINY_PIC)
1406 #define TARGET_HAS_FMV_TARGET_ATTRIBUTE 0
1408 #define TARGET_SUPPORTS_WIDE_INT 1
1410 /* Modes valid for AdvSIMD D registers, i.e. that fit in half a Q register. */
1411 #define AARCH64_VALID_SIMD_DREG_MODE(MODE) \
1412 ((MODE) == V2SImode || (MODE) == V4HImode || (MODE) == V8QImode \
1413 || (MODE) == V2SFmode || (MODE) == V4HFmode || (MODE) == DImode \
1414 || (MODE) == DFmode || (MODE) == V4BFmode)
1416 /* Modes valid for AdvSIMD Q registers. */
1417 #define AARCH64_VALID_SIMD_QREG_MODE(MODE) \
1418 ((MODE) == V4SImode || (MODE) == V8HImode || (MODE) == V16QImode \
1419 || (MODE) == V4SFmode || (MODE) == V8HFmode || (MODE) == V2DImode \
1420 || (MODE) == V2DFmode || (MODE) == V8BFmode)
1422 #define ENDIAN_LANE_N(NUNITS, N) \
1423 (BYTES_BIG_ENDIAN ? NUNITS - 1 - N : N)
1425 /* Extra specs when building a native AArch64-hosted compiler.
1426 Option rewriting rules based on host system. */
1427 #if defined(__aarch64__)
1428 extern const char *host_detect_local_cpu (int argc, const char **argv);
1429 #define HAVE_LOCAL_CPU_DETECT
1430 # define EXTRA_SPEC_FUNCTIONS \
1431 { "local_cpu_detect", host_detect_local_cpu }, \
1432 MCPU_TO_MARCH_SPEC_FUNCTIONS
1434 /* Rewrite -m{arch,cpu,tune}=native based on the host system information.
1435 When rewriting -march=native convert it into an -mcpu option if no other
1436 -mcpu or -mtune was given. */
1437 # define MCPU_MTUNE_NATIVE_SPECS \
1438 " %{march=native:%<march=native %:local_cpu_detect(%{mcpu=*|mtune=*:arch;:cpu})}" \
1439 " %{mcpu=native:%<mcpu=native %:local_cpu_detect(cpu)}" \
1440 " %{mtune=native:%<mtune=native %:local_cpu_detect(tune)}"
1441 /* This will be used in OPTION_DEFAULT_SPECS below.
1442 When GCC is configured with --with-tune we don't want to materialize an
1443 implicit -mtune would prevent the rewriting of -march=native into
1444 -mcpu=native as per the above rules. */
1445 #define CONFIG_TUNE_SPEC \
1446 { "tune", "%{!mcpu=*:%{!mtune=*:%{!march=native:-mtune=%(VALUE)}}}" },
1447 #else
1448 # define MCPU_MTUNE_NATIVE_SPECS ""
1449 # define EXTRA_SPEC_FUNCTIONS MCPU_TO_MARCH_SPEC_FUNCTIONS
1450 # define CONFIG_TUNE_SPEC \
1451 {"tune", "%{!mcpu=*:%{!mtune=*:-mtune=%(VALUE)}}"},
1452 #endif
1454 /* Support for configure-time --with-arch, --with-cpu and --with-tune.
1455 --with-arch and --with-cpu are ignored if either -mcpu or -march is used.
1456 --with-tune is ignored if either -mtune or -mcpu is used (but is not
1457 affected by -march, except in the -march=native case as per the
1458 CONFIG_TUNE_SPEC above). */
1459 #define OPTION_DEFAULT_SPECS \
1460 {"arch", "%{!march=*:%{!mcpu=*:-march=%(VALUE)}}" }, \
1461 {"cpu", "%{!march=*:%{!mcpu=*:-mcpu=%(VALUE)}}" }, \
1462 CONFIG_TUNE_SPEC
1464 #define MCPU_TO_MARCH_SPEC \
1465 " %{mcpu=*:-march=%:rewrite_mcpu(%{mcpu=*:%*})}"
1467 extern const char *aarch64_rewrite_mcpu (int argc, const char **argv);
1468 #define MCPU_TO_MARCH_SPEC_FUNCTIONS \
1469 { "rewrite_mcpu", aarch64_rewrite_mcpu },
1471 #define ASM_CPU_SPEC \
1472 MCPU_TO_MARCH_SPEC
1474 #define EXTRA_SPECS \
1475 { "asm_cpu_spec", ASM_CPU_SPEC }
1477 #define ASM_OUTPUT_POOL_EPILOGUE aarch64_asm_output_pool_epilogue
1479 /* This type is the user-visible __fp16, and a pointer to that type. We
1480 need it in many places in the backend. Defined in aarch64-builtins.cc. */
1481 extern GTY(()) tree aarch64_fp16_type_node;
1482 extern GTY(()) tree aarch64_fp16_ptr_type_node;
1484 /* Pointer to the user-visible __bf16 type. __bf16 itself is generic
1485 bfloat16_type_node. Defined in aarch64-builtins.cc. */
1486 extern GTY(()) tree aarch64_bf16_ptr_type_node;
1488 /* The generic unwind code in libgcc does not initialize the frame pointer.
1489 So in order to unwind a function using a frame pointer, the very first
1490 function that is unwound must save the frame pointer. That way the frame
1491 pointer is restored and its value is now valid - otherwise _Unwind_GetGR
1492 crashes. Libgcc can now be safely built with -fomit-frame-pointer. */
1493 #define LIBGCC2_UNWIND_ATTRIBUTE \
1494 __attribute__((optimize ("no-omit-frame-pointer")))
1496 #ifndef USED_FOR_TARGET
1497 extern poly_uint16 aarch64_sve_vg;
1499 /* The number of bits and bytes in an SVE vector. */
1500 #define BITS_PER_SVE_VECTOR (poly_uint16 (aarch64_sve_vg * 64))
1501 #define BYTES_PER_SVE_VECTOR (poly_uint16 (aarch64_sve_vg * 8))
1503 /* The number of bits and bytes in an SVE predicate. */
1504 #define BITS_PER_SVE_PRED BYTES_PER_SVE_VECTOR
1505 #define BYTES_PER_SVE_PRED aarch64_sve_vg
1507 /* The SVE mode for a vector of bytes. */
1508 #define SVE_BYTE_MODE VNx16QImode
1510 /* The maximum number of bytes in a fixed-size vector. This is 256 bytes
1511 (for -msve-vector-bits=2048) multiplied by the maximum number of
1512 vectors in a structure mode (4).
1514 This limit must not be used for variable-size vectors, since
1515 VL-agnostic code must work with arbitary vector lengths. */
1516 #define MAX_COMPILE_TIME_VEC_BYTES (256 * 4)
1517 #endif
1519 #define REGMODE_NATURAL_SIZE(MODE) aarch64_regmode_natural_size (MODE)
1521 /* Allocate a minimum of STACK_CLASH_MIN_BYTES_OUTGOING_ARGS bytes for the
1522 outgoing arguments if stack clash protection is enabled. This is essential
1523 as the extra arg space allows us to skip a check in alloca. */
1524 #undef STACK_DYNAMIC_OFFSET
1525 #define STACK_DYNAMIC_OFFSET(FUNDECL) \
1526 ((flag_stack_clash_protection \
1527 && cfun->calls_alloca \
1528 && known_lt (crtl->outgoing_args_size, \
1529 STACK_CLASH_MIN_BYTES_OUTGOING_ARGS)) \
1530 ? ROUND_UP (STACK_CLASH_MIN_BYTES_OUTGOING_ARGS, \
1531 STACK_BOUNDARY / BITS_PER_UNIT) \
1532 : (crtl->outgoing_args_size + STACK_POINTER_OFFSET))
1534 /* Filled in by aarch64_adjust_reg_alloc_order, which is called before
1535 the first relevant use. */
1536 #define REG_ALLOC_ORDER {}
1537 #define ADJUST_REG_ALLOC_ORDER aarch64_adjust_reg_alloc_order ()
1539 #define AARCH64_VALID_SHRN_OP(T,S) \
1540 ((T) == TRUNCATE \
1541 || ((T) == US_TRUNCATE && (S) == LSHIFTRT) \
1542 || ((T) == SS_TRUNCATE && (S) == ASHIFTRT))
1544 #ifndef USED_FOR_TARGET
1546 /* Enumerates the mode-switching "entities" for AArch64. */
1547 enum class aarch64_mode_entity : int
1549 /* An aarch64_tristate_mode that says whether we have created a local
1550 save buffer for the current function's ZA state. The only transition
1551 is from NO to YES. */
1552 HAVE_ZA_SAVE_BUFFER,
1554 /* An aarch64_local_sme_state that reflects the state of all data
1555 controlled by PSTATE.ZA. */
1556 LOCAL_SME_STATE
1559 /* Describes the state of all data controlled by PSTATE.ZA */
1560 enum class aarch64_local_sme_state : int
1562 /* ZA is in the off or dormant state. If it is dormant, the contents
1563 of ZA belong to a caller. */
1564 INACTIVE_CALLER,
1566 /* ZA is in the off state: PSTATE.ZA is 0 and TPIDR2_EL0 is null. */
1567 OFF,
1569 /* ZA is in the off or dormant state. If it is dormant, the contents
1570 of ZA belong to the current function. */
1571 INACTIVE_LOCAL,
1573 /* ZA is in the off state and the current function's ZA contents are
1574 stored in the lazy save buffer. This is the state on entry to
1575 exception handlers. */
1576 SAVED_LOCAL,
1578 /* ZA is in the active state: PSTATE.ZA is 1 and TPIDR2_EL0 is null.
1579 The contents of ZA are live. */
1580 ACTIVE_LIVE,
1582 /* ZA is in the active state: PSTATE.ZA is 1 and TPIDR2_EL0 is null.
1583 The contents of ZA are dead. */
1584 ACTIVE_DEAD,
1586 /* ZA could be in multiple states. */
1590 enum class aarch64_tristate_mode : int { NO, YES, MAYBE };
1592 #define OPTIMIZE_MODE_SWITCHING(ENTITY) \
1593 aarch64_optimize_mode_switching (aarch64_mode_entity (ENTITY))
1595 #define NUM_MODES_FOR_MODE_SWITCHING \
1596 { int (aarch64_tristate_mode::MAYBE), \
1597 int (aarch64_local_sme_state::ANY) }
1599 #endif
1601 #endif /* GCC_AARCH64_H */