1 /*===---- arm_acle.h - ARM Non-Neon intrinsics -----------------------------===
3 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 * See https://llvm.org/LICENSE.txt for license information.
5 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 *===-----------------------------------------------------------------------===
14 #error "ACLE intrinsics support not enabled."
19 #if defined(__cplusplus)
23 /* 8 SYNCHRONIZATION, BARRIER AND HINT INTRINSICS */
24 /* 8.3 Memory barriers */
25 #if !__has_builtin(__dmb)
26 #define __dmb(i) __builtin_arm_dmb(i)
28 #if !__has_builtin(__dsb)
29 #define __dsb(i) __builtin_arm_dsb(i)
31 #if !__has_builtin(__isb)
32 #define __isb(i) __builtin_arm_isb(i)
37 #if !__has_builtin(__wfi)
38 static __inline__
void __attribute__((__always_inline__
, __nodebug__
)) __wfi(void) {
43 #if !__has_builtin(__wfe)
44 static __inline__
void __attribute__((__always_inline__
, __nodebug__
)) __wfe(void) {
49 #if !__has_builtin(__sev)
50 static __inline__
void __attribute__((__always_inline__
, __nodebug__
)) __sev(void) {
55 #if !__has_builtin(__sevl)
56 static __inline__
void __attribute__((__always_inline__
, __nodebug__
)) __sevl(void) {
61 #if !__has_builtin(__yield)
62 static __inline__
void __attribute__((__always_inline__
, __nodebug__
)) __yield(void) {
63 __builtin_arm_yield();
67 #if defined(__ARM_32BIT_STATE) && __ARM_32BIT_STATE
68 #define __dbg(t) __builtin_arm_dbg(t)
72 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
))
73 __swp(uint32_t __x
, volatile uint32_t *__p
) {
76 v
= __builtin_arm_ldrex(__p
);
77 while (__builtin_arm_strex(__x
, __p
));
81 /* 8.6 Memory prefetch intrinsics */
82 /* 8.6.1 Data prefetch */
83 #define __pld(addr) __pldx(0, 0, 0, addr)
85 #if defined(__ARM_32BIT_STATE) && __ARM_32BIT_STATE
86 #define __pldx(access_kind, cache_level, retention_policy, addr) \
87 __builtin_arm_prefetch(addr, access_kind, 1)
89 #define __pldx(access_kind, cache_level, retention_policy, addr) \
90 __builtin_arm_prefetch(addr, access_kind, cache_level, retention_policy, 1)
93 /* 8.6.2 Instruction prefetch */
94 #define __pli(addr) __plix(0, 0, addr)
96 #if defined(__ARM_32BIT_STATE) && __ARM_32BIT_STATE
97 #define __plix(cache_level, retention_policy, addr) \
98 __builtin_arm_prefetch(addr, 0, 0)
100 #define __plix(cache_level, retention_policy, addr) \
101 __builtin_arm_prefetch(addr, 0, cache_level, retention_policy, 0)
105 #if !defined(_MSC_VER) || !defined(__aarch64__)
106 static __inline__
void __attribute__((__always_inline__
, __nodebug__
)) __nop(void) {
111 /* 9 DATA-PROCESSING INTRINSICS */
112 /* 9.2 Miscellaneous data-processing intrinsics */
114 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
))
115 __ror(uint32_t __x
, uint32_t __y
) {
119 return (__x
>> __y
) | (__x
<< (32 - __y
));
122 static __inline__
uint64_t __attribute__((__always_inline__
, __nodebug__
))
123 __rorll(uint64_t __x
, uint32_t __y
) {
127 return (__x
>> __y
) | (__x
<< (64 - __y
));
130 static __inline__
unsigned long __attribute__((__always_inline__
, __nodebug__
))
131 __rorl(unsigned long __x
, uint32_t __y
) {
132 #if __SIZEOF_LONG__ == 4
133 return __ror(__x
, __y
);
135 return __rorll(__x
, __y
);
141 static __inline__
unsigned int __attribute__((__always_inline__
, __nodebug__
))
142 __clz(uint32_t __t
) {
143 return __builtin_arm_clz(__t
);
146 static __inline__
unsigned int __attribute__((__always_inline__
, __nodebug__
))
147 __clzl(unsigned long __t
) {
148 #if __SIZEOF_LONG__ == 4
149 return __builtin_arm_clz(__t
);
151 return __builtin_arm_clz64(__t
);
155 static __inline__
unsigned int __attribute__((__always_inline__
, __nodebug__
))
156 __clzll(uint64_t __t
) {
157 return __builtin_arm_clz64(__t
);
161 static __inline__
unsigned int __attribute__((__always_inline__
, __nodebug__
))
162 __cls(uint32_t __t
) {
163 return __builtin_arm_cls(__t
);
166 static __inline__
unsigned int __attribute__((__always_inline__
, __nodebug__
))
167 __clsl(unsigned long __t
) {
168 #if __SIZEOF_LONG__ == 4
169 return __builtin_arm_cls(__t
);
171 return __builtin_arm_cls64(__t
);
175 static __inline__
unsigned int __attribute__((__always_inline__
, __nodebug__
))
176 __clsll(uint64_t __t
) {
177 return __builtin_arm_cls64(__t
);
181 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
))
182 __rev(uint32_t __t
) {
183 return __builtin_bswap32(__t
);
186 static __inline__
unsigned long __attribute__((__always_inline__
, __nodebug__
))
187 __revl(unsigned long __t
) {
188 #if __SIZEOF_LONG__ == 4
189 return __builtin_bswap32(__t
);
191 return __builtin_bswap64(__t
);
195 static __inline__
uint64_t __attribute__((__always_inline__
, __nodebug__
))
196 __revll(uint64_t __t
) {
197 return __builtin_bswap64(__t
);
201 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
))
202 __rev16(uint32_t __t
) {
203 return __ror(__rev(__t
), 16);
206 static __inline__
uint64_t __attribute__((__always_inline__
, __nodebug__
))
207 __rev16ll(uint64_t __t
) {
208 return (((uint64_t)__rev16(__t
>> 32)) << 32) | (uint64_t)__rev16((uint32_t)__t
);
211 static __inline__
unsigned long __attribute__((__always_inline__
, __nodebug__
))
212 __rev16l(unsigned long __t
) {
213 #if __SIZEOF_LONG__ == 4
216 return __rev16ll(__t
);
221 static __inline__
int16_t __attribute__((__always_inline__
, __nodebug__
))
222 __revsh(int16_t __t
) {
223 return (int16_t)__builtin_bswap16((uint16_t)__t
);
227 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
))
228 __rbit(uint32_t __t
) {
229 return __builtin_arm_rbit(__t
);
232 static __inline__
uint64_t __attribute__((__always_inline__
, __nodebug__
))
233 __rbitll(uint64_t __t
) {
234 #if defined(__ARM_32BIT_STATE) && __ARM_32BIT_STATE
235 return (((uint64_t)__builtin_arm_rbit(__t
)) << 32) |
236 __builtin_arm_rbit(__t
>> 32);
238 return __builtin_arm_rbit64(__t
);
242 static __inline__
unsigned long __attribute__((__always_inline__
, __nodebug__
))
243 __rbitl(unsigned long __t
) {
244 #if __SIZEOF_LONG__ == 4
247 return __rbitll(__t
);
252 * 9.3 16-bit multiplications
254 #if defined(__ARM_FEATURE_DSP) && __ARM_FEATURE_DSP
255 static __inline__
int32_t __attribute__((__always_inline__
,__nodebug__
))
256 __smulbb(int32_t __a
, int32_t __b
) {
257 return __builtin_arm_smulbb(__a
, __b
);
259 static __inline__
int32_t __attribute__((__always_inline__
,__nodebug__
))
260 __smulbt(int32_t __a
, int32_t __b
) {
261 return __builtin_arm_smulbt(__a
, __b
);
263 static __inline__
int32_t __attribute__((__always_inline__
,__nodebug__
))
264 __smultb(int32_t __a
, int32_t __b
) {
265 return __builtin_arm_smultb(__a
, __b
);
267 static __inline__
int32_t __attribute__((__always_inline__
,__nodebug__
))
268 __smultt(int32_t __a
, int32_t __b
) {
269 return __builtin_arm_smultt(__a
, __b
);
271 static __inline__
int32_t __attribute__((__always_inline__
,__nodebug__
))
272 __smulwb(int32_t __a
, int32_t __b
) {
273 return __builtin_arm_smulwb(__a
, __b
);
275 static __inline__
int32_t __attribute__((__always_inline__
,__nodebug__
))
276 __smulwt(int32_t __a
, int32_t __b
) {
277 return __builtin_arm_smulwt(__a
, __b
);
282 * 9.4 Saturating intrinsics
284 * FIXME: Change guard to their corresponding __ARM_FEATURE flag when Q flag
285 * intrinsics are implemented and the flag is enabled.
287 /* 9.4.1 Width-specified saturation intrinsics */
288 #if defined(__ARM_FEATURE_SAT) && __ARM_FEATURE_SAT
289 #define __ssat(x, y) __builtin_arm_ssat(x, y)
290 #define __usat(x, y) __builtin_arm_usat(x, y)
293 /* 9.4.2 Saturating addition and subtraction intrinsics */
294 #if defined(__ARM_FEATURE_DSP) && __ARM_FEATURE_DSP
295 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
296 __qadd(int32_t __t
, int32_t __v
) {
297 return __builtin_arm_qadd(__t
, __v
);
300 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
301 __qsub(int32_t __t
, int32_t __v
) {
302 return __builtin_arm_qsub(__t
, __v
);
305 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
306 __qdbl(int32_t __t
) {
307 return __builtin_arm_qadd(__t
, __t
);
311 /* 9.4.3 Accumultating multiplications */
312 #if defined(__ARM_FEATURE_DSP) && __ARM_FEATURE_DSP
313 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
314 __smlabb(int32_t __a
, int32_t __b
, int32_t __c
) {
315 return __builtin_arm_smlabb(__a
, __b
, __c
);
317 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
318 __smlabt(int32_t __a
, int32_t __b
, int32_t __c
) {
319 return __builtin_arm_smlabt(__a
, __b
, __c
);
321 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
322 __smlatb(int32_t __a
, int32_t __b
, int32_t __c
) {
323 return __builtin_arm_smlatb(__a
, __b
, __c
);
325 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
326 __smlatt(int32_t __a
, int32_t __b
, int32_t __c
) {
327 return __builtin_arm_smlatt(__a
, __b
, __c
);
329 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
330 __smlawb(int32_t __a
, int32_t __b
, int32_t __c
) {
331 return __builtin_arm_smlawb(__a
, __b
, __c
);
333 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
334 __smlawt(int32_t __a
, int32_t __b
, int32_t __c
) {
335 return __builtin_arm_smlawt(__a
, __b
, __c
);
340 /* 9.5.4 Parallel 16-bit saturation */
341 #if defined(__ARM_FEATURE_SIMD32) && __ARM_FEATURE_SIMD32
342 #define __ssat16(x, y) __builtin_arm_ssat16(x, y)
343 #define __usat16(x, y) __builtin_arm_usat16(x, y)
346 /* 9.5.5 Packing and unpacking */
347 #if defined(__ARM_FEATURE_SIMD32) && __ARM_FEATURE_SIMD32
348 typedef int32_t int8x4_t
;
349 typedef int32_t int16x2_t
;
350 typedef uint32_t uint8x4_t
;
351 typedef uint32_t uint16x2_t
;
353 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
354 __sxtab16(int16x2_t __a
, int8x4_t __b
) {
355 return __builtin_arm_sxtab16(__a
, __b
);
357 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
358 __sxtb16(int8x4_t __a
) {
359 return __builtin_arm_sxtb16(__a
);
361 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
362 __uxtab16(int16x2_t __a
, int8x4_t __b
) {
363 return __builtin_arm_uxtab16(__a
, __b
);
365 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
366 __uxtb16(int8x4_t __a
) {
367 return __builtin_arm_uxtb16(__a
);
371 /* 9.5.6 Parallel selection */
372 #if defined(__ARM_FEATURE_SIMD32) && __ARM_FEATURE_SIMD32
373 static __inline__ uint8x4_t
__attribute__((__always_inline__
, __nodebug__
))
374 __sel(uint8x4_t __a
, uint8x4_t __b
) {
375 return __builtin_arm_sel(__a
, __b
);
379 /* 9.5.7 Parallel 8-bit addition and subtraction */
380 #if defined(__ARM_FEATURE_SIMD32) && __ARM_FEATURE_SIMD32
381 static __inline__ int8x4_t
__attribute__((__always_inline__
, __nodebug__
))
382 __qadd8(int8x4_t __a
, int8x4_t __b
) {
383 return __builtin_arm_qadd8(__a
, __b
);
385 static __inline__ int8x4_t
__attribute__((__always_inline__
, __nodebug__
))
386 __qsub8(int8x4_t __a
, int8x4_t __b
) {
387 return __builtin_arm_qsub8(__a
, __b
);
389 static __inline__ int8x4_t
__attribute__((__always_inline__
, __nodebug__
))
390 __sadd8(int8x4_t __a
, int8x4_t __b
) {
391 return __builtin_arm_sadd8(__a
, __b
);
393 static __inline__ int8x4_t
__attribute__((__always_inline__
, __nodebug__
))
394 __shadd8(int8x4_t __a
, int8x4_t __b
) {
395 return __builtin_arm_shadd8(__a
, __b
);
397 static __inline__ int8x4_t
__attribute__((__always_inline__
, __nodebug__
))
398 __shsub8(int8x4_t __a
, int8x4_t __b
) {
399 return __builtin_arm_shsub8(__a
, __b
);
401 static __inline__ int8x4_t
__attribute__((__always_inline__
, __nodebug__
))
402 __ssub8(int8x4_t __a
, int8x4_t __b
) {
403 return __builtin_arm_ssub8(__a
, __b
);
405 static __inline__ uint8x4_t
__attribute__((__always_inline__
, __nodebug__
))
406 __uadd8(uint8x4_t __a
, uint8x4_t __b
) {
407 return __builtin_arm_uadd8(__a
, __b
);
409 static __inline__ uint8x4_t
__attribute__((__always_inline__
, __nodebug__
))
410 __uhadd8(uint8x4_t __a
, uint8x4_t __b
) {
411 return __builtin_arm_uhadd8(__a
, __b
);
413 static __inline__ uint8x4_t
__attribute__((__always_inline__
, __nodebug__
))
414 __uhsub8(uint8x4_t __a
, uint8x4_t __b
) {
415 return __builtin_arm_uhsub8(__a
, __b
);
417 static __inline__ uint8x4_t
__attribute__((__always_inline__
, __nodebug__
))
418 __uqadd8(uint8x4_t __a
, uint8x4_t __b
) {
419 return __builtin_arm_uqadd8(__a
, __b
);
421 static __inline__ uint8x4_t
__attribute__((__always_inline__
, __nodebug__
))
422 __uqsub8(uint8x4_t __a
, uint8x4_t __b
) {
423 return __builtin_arm_uqsub8(__a
, __b
);
425 static __inline__ uint8x4_t
__attribute__((__always_inline__
, __nodebug__
))
426 __usub8(uint8x4_t __a
, uint8x4_t __b
) {
427 return __builtin_arm_usub8(__a
, __b
);
431 /* 9.5.8 Sum of 8-bit absolute differences */
432 #if defined(__ARM_FEATURE_SIMD32) && __ARM_FEATURE_SIMD32
433 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
))
434 __usad8(uint8x4_t __a
, uint8x4_t __b
) {
435 return __builtin_arm_usad8(__a
, __b
);
437 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
))
438 __usada8(uint8x4_t __a
, uint8x4_t __b
, uint32_t __c
) {
439 return __builtin_arm_usada8(__a
, __b
, __c
);
443 /* 9.5.9 Parallel 16-bit addition and subtraction */
444 #if defined(__ARM_FEATURE_SIMD32) && __ARM_FEATURE_SIMD32
445 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
446 __qadd16(int16x2_t __a
, int16x2_t __b
) {
447 return __builtin_arm_qadd16(__a
, __b
);
449 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
450 __qasx(int16x2_t __a
, int16x2_t __b
) {
451 return __builtin_arm_qasx(__a
, __b
);
453 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
454 __qsax(int16x2_t __a
, int16x2_t __b
) {
455 return __builtin_arm_qsax(__a
, __b
);
457 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
458 __qsub16(int16x2_t __a
, int16x2_t __b
) {
459 return __builtin_arm_qsub16(__a
, __b
);
461 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
462 __sadd16(int16x2_t __a
, int16x2_t __b
) {
463 return __builtin_arm_sadd16(__a
, __b
);
465 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
466 __sasx(int16x2_t __a
, int16x2_t __b
) {
467 return __builtin_arm_sasx(__a
, __b
);
469 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
470 __shadd16(int16x2_t __a
, int16x2_t __b
) {
471 return __builtin_arm_shadd16(__a
, __b
);
473 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
474 __shasx(int16x2_t __a
, int16x2_t __b
) {
475 return __builtin_arm_shasx(__a
, __b
);
477 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
478 __shsax(int16x2_t __a
, int16x2_t __b
) {
479 return __builtin_arm_shsax(__a
, __b
);
481 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
482 __shsub16(int16x2_t __a
, int16x2_t __b
) {
483 return __builtin_arm_shsub16(__a
, __b
);
485 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
486 __ssax(int16x2_t __a
, int16x2_t __b
) {
487 return __builtin_arm_ssax(__a
, __b
);
489 static __inline__ int16x2_t
__attribute__((__always_inline__
, __nodebug__
))
490 __ssub16(int16x2_t __a
, int16x2_t __b
) {
491 return __builtin_arm_ssub16(__a
, __b
);
493 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
494 __uadd16(uint16x2_t __a
, uint16x2_t __b
) {
495 return __builtin_arm_uadd16(__a
, __b
);
497 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
498 __uasx(uint16x2_t __a
, uint16x2_t __b
) {
499 return __builtin_arm_uasx(__a
, __b
);
501 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
502 __uhadd16(uint16x2_t __a
, uint16x2_t __b
) {
503 return __builtin_arm_uhadd16(__a
, __b
);
505 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
506 __uhasx(uint16x2_t __a
, uint16x2_t __b
) {
507 return __builtin_arm_uhasx(__a
, __b
);
509 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
510 __uhsax(uint16x2_t __a
, uint16x2_t __b
) {
511 return __builtin_arm_uhsax(__a
, __b
);
513 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
514 __uhsub16(uint16x2_t __a
, uint16x2_t __b
) {
515 return __builtin_arm_uhsub16(__a
, __b
);
517 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
518 __uqadd16(uint16x2_t __a
, uint16x2_t __b
) {
519 return __builtin_arm_uqadd16(__a
, __b
);
521 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
522 __uqasx(uint16x2_t __a
, uint16x2_t __b
) {
523 return __builtin_arm_uqasx(__a
, __b
);
525 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
526 __uqsax(uint16x2_t __a
, uint16x2_t __b
) {
527 return __builtin_arm_uqsax(__a
, __b
);
529 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
530 __uqsub16(uint16x2_t __a
, uint16x2_t __b
) {
531 return __builtin_arm_uqsub16(__a
, __b
);
533 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
534 __usax(uint16x2_t __a
, uint16x2_t __b
) {
535 return __builtin_arm_usax(__a
, __b
);
537 static __inline__ uint16x2_t
__attribute__((__always_inline__
, __nodebug__
))
538 __usub16(uint16x2_t __a
, uint16x2_t __b
) {
539 return __builtin_arm_usub16(__a
, __b
);
543 /* 9.5.10 Parallel 16-bit multiplications */
544 #if defined(__ARM_FEATURE_SIMD32) && __ARM_FEATURE_SIMD32
545 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
546 __smlad(int16x2_t __a
, int16x2_t __b
, int32_t __c
) {
547 return __builtin_arm_smlad(__a
, __b
, __c
);
549 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
550 __smladx(int16x2_t __a
, int16x2_t __b
, int32_t __c
) {
551 return __builtin_arm_smladx(__a
, __b
, __c
);
553 static __inline__
int64_t __attribute__((__always_inline__
, __nodebug__
))
554 __smlald(int16x2_t __a
, int16x2_t __b
, int64_t __c
) {
555 return __builtin_arm_smlald(__a
, __b
, __c
);
557 static __inline__
int64_t __attribute__((__always_inline__
, __nodebug__
))
558 __smlaldx(int16x2_t __a
, int16x2_t __b
, int64_t __c
) {
559 return __builtin_arm_smlaldx(__a
, __b
, __c
);
561 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
562 __smlsd(int16x2_t __a
, int16x2_t __b
, int32_t __c
) {
563 return __builtin_arm_smlsd(__a
, __b
, __c
);
565 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
566 __smlsdx(int16x2_t __a
, int16x2_t __b
, int32_t __c
) {
567 return __builtin_arm_smlsdx(__a
, __b
, __c
);
569 static __inline__
int64_t __attribute__((__always_inline__
, __nodebug__
))
570 __smlsld(int16x2_t __a
, int16x2_t __b
, int64_t __c
) {
571 return __builtin_arm_smlsld(__a
, __b
, __c
);
573 static __inline__
int64_t __attribute__((__always_inline__
, __nodebug__
))
574 __smlsldx(int16x2_t __a
, int16x2_t __b
, int64_t __c
) {
575 return __builtin_arm_smlsldx(__a
, __b
, __c
);
577 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
578 __smuad(int16x2_t __a
, int16x2_t __b
) {
579 return __builtin_arm_smuad(__a
, __b
);
581 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
582 __smuadx(int16x2_t __a
, int16x2_t __b
) {
583 return __builtin_arm_smuadx(__a
, __b
);
585 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
586 __smusd(int16x2_t __a
, int16x2_t __b
) {
587 return __builtin_arm_smusd(__a
, __b
);
589 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
))
590 __smusdx(int16x2_t __a
, int16x2_t __b
) {
591 return __builtin_arm_smusdx(__a
, __b
);
595 /* 8.6 Floating-point data-processing intrinsics */
596 #if (defined(__ARM_FEATURE_DIRECTED_ROUNDING) && \
597 (__ARM_FEATURE_DIRECTED_ROUNDING)) && \
598 (defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE)
599 static __inline__
double __attribute__((__always_inline__
, __nodebug__
))
600 __rintn(double __a
) {
601 return __builtin_roundeven(__a
);
604 static __inline__
float __attribute__((__always_inline__
, __nodebug__
))
605 __rintnf(float __a
) {
606 return __builtin_roundevenf(__a
);
610 /* 9.7 CRC32 intrinsics */
611 #if (defined(__ARM_FEATURE_CRC32) && __ARM_FEATURE_CRC32) || \
612 (defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE)
613 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
, target("crc")))
614 __crc32b(uint32_t __a
, uint8_t __b
) {
615 return __builtin_arm_crc32b(__a
, __b
);
618 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
, target("crc")))
619 __crc32h(uint32_t __a
, uint16_t __b
) {
620 return __builtin_arm_crc32h(__a
, __b
);
623 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
, target("crc")))
624 __crc32w(uint32_t __a
, uint32_t __b
) {
625 return __builtin_arm_crc32w(__a
, __b
);
628 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
, target("crc")))
629 __crc32d(uint32_t __a
, uint64_t __b
) {
630 return __builtin_arm_crc32d(__a
, __b
);
633 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
, target("crc")))
634 __crc32cb(uint32_t __a
, uint8_t __b
) {
635 return __builtin_arm_crc32cb(__a
, __b
);
638 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
, target("crc")))
639 __crc32ch(uint32_t __a
, uint16_t __b
) {
640 return __builtin_arm_crc32ch(__a
, __b
);
643 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
, target("crc")))
644 __crc32cw(uint32_t __a
, uint32_t __b
) {
645 return __builtin_arm_crc32cw(__a
, __b
);
648 static __inline__
uint32_t __attribute__((__always_inline__
, __nodebug__
, target("crc")))
649 __crc32cd(uint32_t __a
, uint64_t __b
) {
650 return __builtin_arm_crc32cd(__a
, __b
);
654 /* Armv8.3-A Javascript conversion intrinsic */
655 #if defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE
656 static __inline__
int32_t __attribute__((__always_inline__
, __nodebug__
, target("v8.3a")))
658 return __builtin_arm_jcvt(__a
);
662 /* Armv8.5-A FP rounding intrinsics */
663 #if defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE
664 static __inline__
float __attribute__((__always_inline__
, __nodebug__
, target("v8.5a")))
665 __rint32zf(float __a
) {
666 return __builtin_arm_rint32zf(__a
);
669 static __inline__
double __attribute__((__always_inline__
, __nodebug__
, target("v8.5a")))
670 __rint32z(double __a
) {
671 return __builtin_arm_rint32z(__a
);
674 static __inline__
float __attribute__((__always_inline__
, __nodebug__
, target("v8.5a")))
675 __rint64zf(float __a
) {
676 return __builtin_arm_rint64zf(__a
);
679 static __inline__
double __attribute__((__always_inline__
, __nodebug__
, target("v8.5a")))
680 __rint64z(double __a
) {
681 return __builtin_arm_rint64z(__a
);
684 static __inline__
float __attribute__((__always_inline__
, __nodebug__
, target("v8.5a")))
685 __rint32xf(float __a
) {
686 return __builtin_arm_rint32xf(__a
);
689 static __inline__
double __attribute__((__always_inline__
, __nodebug__
, target("v8.5a")))
690 __rint32x(double __a
) {
691 return __builtin_arm_rint32x(__a
);
694 static __inline__
float __attribute__((__always_inline__
, __nodebug__
, target("v8.5a")))
695 __rint64xf(float __a
) {
696 return __builtin_arm_rint64xf(__a
);
699 static __inline__
double __attribute__((__always_inline__
, __nodebug__
, target("v8.5a")))
700 __rint64x(double __a
) {
701 return __builtin_arm_rint64x(__a
);
705 /* Armv8.7-A load/store 64-byte intrinsics */
706 #if defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE
711 static __inline__ data512_t
__attribute__((__always_inline__
, __nodebug__
, target("ls64")))
712 __arm_ld64b(const void *__addr
) {
714 __builtin_arm_ld64b(__addr
, __value
.val
);
717 static __inline__
void __attribute__((__always_inline__
, __nodebug__
, target("ls64")))
718 __arm_st64b(void *__addr
, data512_t __value
) {
719 __builtin_arm_st64b(__addr
, __value
.val
);
721 static __inline__
uint64_t __attribute__((__always_inline__
, __nodebug__
, target("ls64")))
722 __arm_st64bv(void *__addr
, data512_t __value
) {
723 return __builtin_arm_st64bv(__addr
, __value
.val
);
725 static __inline__
uint64_t __attribute__((__always_inline__
, __nodebug__
, target("ls64")))
726 __arm_st64bv0(void *__addr
, data512_t __value
) {
727 return __builtin_arm_st64bv0(__addr
, __value
.val
);
731 /* 10.1 Special register intrinsics */
732 #define __arm_rsr(sysreg) __builtin_arm_rsr(sysreg)
733 #define __arm_rsr64(sysreg) __builtin_arm_rsr64(sysreg)
734 #define __arm_rsr128(sysreg) __builtin_arm_rsr128(sysreg)
735 #define __arm_rsrp(sysreg) __builtin_arm_rsrp(sysreg)
736 #define __arm_rsrf(sysreg) __builtin_bit_cast(float, __arm_rsr(sysreg))
737 #define __arm_rsrf64(sysreg) __builtin_bit_cast(double, __arm_rsr64(sysreg))
738 #define __arm_wsr(sysreg, v) __builtin_arm_wsr(sysreg, v)
739 #define __arm_wsr64(sysreg, v) __builtin_arm_wsr64(sysreg, v)
740 #define __arm_wsr128(sysreg, v) __builtin_arm_wsr128(sysreg, v)
741 #define __arm_wsrp(sysreg, v) __builtin_arm_wsrp(sysreg, v)
742 #define __arm_wsrf(sysreg, v) __arm_wsr(sysreg, __builtin_bit_cast(uint32_t, v))
743 #define __arm_wsrf64(sysreg, v) __arm_wsr64(sysreg, __builtin_bit_cast(uint64_t, v))
745 /* Memory Tagging Extensions (MTE) Intrinsics */
746 #if defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE
747 #define __arm_mte_create_random_tag(__ptr, __mask) __builtin_arm_irg(__ptr, __mask)
748 #define __arm_mte_increment_tag(__ptr, __tag_offset) __builtin_arm_addg(__ptr, __tag_offset)
749 #define __arm_mte_exclude_tag(__ptr, __excluded) __builtin_arm_gmi(__ptr, __excluded)
750 #define __arm_mte_get_tag(__ptr) __builtin_arm_ldg(__ptr)
751 #define __arm_mte_set_tag(__ptr) __builtin_arm_stg(__ptr)
752 #define __arm_mte_ptrdiff(__ptra, __ptrb) __builtin_arm_subp(__ptra, __ptrb)
754 /* Memory Operations Intrinsics */
755 #define __arm_mops_memset_tag(__tagged_address, __value, __size) \
756 __builtin_arm_mops_memset_tag(__tagged_address, __value, __size)
759 /* Transactional Memory Extension (TME) Intrinsics */
760 #if defined(__ARM_FEATURE_TME) && __ARM_FEATURE_TME
762 #define _TMFAILURE_REASON 0x00007fffu
763 #define _TMFAILURE_RTRY 0x00008000u
764 #define _TMFAILURE_CNCL 0x00010000u
765 #define _TMFAILURE_MEM 0x00020000u
766 #define _TMFAILURE_IMP 0x00040000u
767 #define _TMFAILURE_ERR 0x00080000u
768 #define _TMFAILURE_SIZE 0x00100000u
769 #define _TMFAILURE_NEST 0x00200000u
770 #define _TMFAILURE_DBG 0x00400000u
771 #define _TMFAILURE_INT 0x00800000u
772 #define _TMFAILURE_TRIVIAL 0x01000000u
774 #define __tstart() __builtin_arm_tstart()
775 #define __tcommit() __builtin_arm_tcommit()
776 #define __tcancel(__arg) __builtin_arm_tcancel(__arg)
777 #define __ttest() __builtin_arm_ttest()
779 #endif /* __ARM_FEATURE_TME */
781 /* Armv8.5-A Random number generation intrinsics */
782 #if defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE
783 static __inline__
int __attribute__((__always_inline__
, __nodebug__
, target("rand")))
784 __rndr(uint64_t *__p
) {
785 return __builtin_arm_rndr(__p
);
787 static __inline__
int __attribute__((__always_inline__
, __nodebug__
, target("rand")))
788 __rndrrs(uint64_t *__p
) {
789 return __builtin_arm_rndrrs(__p
);
793 #if defined(__cplusplus)
797 #endif /* __ARM_ACLE_H */