aarch64-sve.md revision 1.1.1.3 1 ;; Machine description for AArch64 SVE.
2 ;; Copyright (C) 2009-2020 Free Software Foundation, Inc.
3 ;; Contributed by ARM Ltd.
4 ;;
5 ;; This file is part of GCC.
6 ;;
7 ;; GCC is free software; you can redistribute it and/or modify it
8 ;; under the terms of the GNU General Public License as published by
9 ;; the Free Software Foundation; either version 3, or (at your option)
10 ;; any later version.
11 ;;
12 ;; GCC is distributed in the hope that it will be useful, but
13 ;; WITHOUT ANY WARRANTY; without even the implied warranty of
14 ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 ;; General Public License for more details.
16 ;;
17 ;; You should have received a copy of the GNU General Public License
18 ;; along with GCC; see the file COPYING3. If not see
19 ;; <http://www.gnu.org/licenses/>.
20
21 ;; The file is organised into the following sections (search for the full
22 ;; line):
23 ;;
24 ;; == General notes
25 ;; ---- Note on the handling of big-endian SVE
26 ;; ---- Description of UNSPEC_PTEST
27 ;; ---- Description of UNSPEC_PRED_Z
28 ;; ---- Note on predicated integer arithemtic and UNSPEC_PRED_X
29 ;; ---- Note on predicated FP arithmetic patterns and GP "strictness"
30 ;; ---- Note on FFR handling
31 ;;
32 ;; == Moves
33 ;; ---- Moves of single vectors
34 ;; ---- Moves of multiple vectors
35 ;; ---- Moves of predicates
36 ;; ---- Moves relating to the FFR
37 ;;
38 ;; == Loads
39 ;; ---- Normal contiguous loads
40 ;; ---- Extending contiguous loads
41 ;; ---- First-faulting contiguous loads
42 ;; ---- First-faulting extending contiguous loads
43 ;; ---- Non-temporal contiguous loads
44 ;; ---- Normal gather loads
45 ;; ---- Extending gather loads
46 ;; ---- First-faulting gather loads
47 ;; ---- First-faulting extending gather loads
48 ;;
49 ;; == Prefetches
50 ;; ---- Contiguous prefetches
51 ;; ---- Gather prefetches
52 ;;
53 ;; == Stores
54 ;; ---- Normal contiguous stores
55 ;; ---- Truncating contiguous stores
56 ;; ---- Non-temporal contiguous stores
57 ;; ---- Normal scatter stores
58 ;; ---- Truncating scatter stores
59 ;;
60 ;; == Vector creation
61 ;; ---- [INT,FP] Duplicate element
62 ;; ---- [INT,FP] Initialize from individual elements
63 ;; ---- [INT] Linear series
64 ;; ---- [PRED] Duplicate element
65 ;;
66 ;; == Vector decomposition
67 ;; ---- [INT,FP] Extract index
68 ;; ---- [INT,FP] Extract active element
69 ;; ---- [PRED] Extract index
70 ;;
71 ;; == Unary arithmetic
72 ;; ---- [INT] General unary arithmetic corresponding to rtx codes
73 ;; ---- [INT] General unary arithmetic corresponding to unspecs
74 ;; ---- [INT] Sign and zero extension
75 ;; ---- [INT] Truncation
76 ;; ---- [INT] Logical inverse
77 ;; ---- [FP<-INT] General unary arithmetic that maps to unspecs
78 ;; ---- [FP] General unary arithmetic corresponding to unspecs
79 ;; ---- [FP] Square root
80 ;; ---- [FP] Reciprocal square root
81 ;; ---- [PRED] Inverse
82
83 ;; == Binary arithmetic
84 ;; ---- [INT] General binary arithmetic corresponding to rtx codes
85 ;; ---- [INT] Addition
86 ;; ---- [INT] Subtraction
87 ;; ---- [INT] Take address
88 ;; ---- [INT] Absolute difference
89 ;; ---- [INT] Saturating addition and subtraction
90 ;; ---- [INT] Highpart multiplication
91 ;; ---- [INT] Division
92 ;; ---- [INT] Binary logical operations
93 ;; ---- [INT] Binary logical operations (inverted second input)
94 ;; ---- [INT] Shifts (rounding towards -Inf)
95 ;; ---- [INT] Shifts (rounding towards 0)
96 ;; ---- [FP<-INT] General binary arithmetic corresponding to unspecs
97 ;; ---- [FP] General binary arithmetic corresponding to rtx codes
98 ;; ---- [FP] General binary arithmetic corresponding to unspecs
99 ;; ---- [FP] Addition
100 ;; ---- [FP] Complex addition
101 ;; ---- [FP] Subtraction
102 ;; ---- [FP] Absolute difference
103 ;; ---- [FP] Multiplication
104 ;; ---- [FP] Division
105 ;; ---- [FP] Binary logical operations
106 ;; ---- [FP] Sign copying
107 ;; ---- [FP] Maximum and minimum
108 ;; ---- [PRED] Binary logical operations
109 ;; ---- [PRED] Binary logical operations (inverted second input)
110 ;; ---- [PRED] Binary logical operations (inverted result)
111 ;;
112 ;; == Ternary arithmetic
113 ;; ---- [INT] MLA and MAD
114 ;; ---- [INT] MLS and MSB
115 ;; ---- [INT] Dot product
116 ;; ---- [INT] Sum of absolute differences
117 ;; ---- [INT] Matrix multiply-accumulate
118 ;; ---- [FP] General ternary arithmetic corresponding to unspecs
119 ;; ---- [FP] Complex multiply-add
120 ;; ---- [FP] Trigonometric multiply-add
121 ;; ---- [FP] Bfloat16 long ternary arithmetic (SF,BF,BF)
122 ;; ---- [FP] Matrix multiply-accumulate
123 ;;
124 ;; == Comparisons and selects
125 ;; ---- [INT,FP] Select based on predicates
126 ;; ---- [INT,FP] Compare and select
127 ;; ---- [INT] Comparisons
128 ;; ---- [INT] While tests
129 ;; ---- [FP] Direct comparisons
130 ;; ---- [FP] Absolute comparisons
131 ;; ---- [PRED] Select
132 ;; ---- [PRED] Test bits
133 ;;
134 ;; == Reductions
135 ;; ---- [INT,FP] Conditional reductions
136 ;; ---- [INT] Tree reductions
137 ;; ---- [FP] Tree reductions
138 ;; ---- [FP] Left-to-right reductions
139 ;;
140 ;; == Permutes
141 ;; ---- [INT,FP] General permutes
142 ;; ---- [INT,FP] Special-purpose unary permutes
143 ;; ---- [INT,FP] Special-purpose binary permutes
144 ;; ---- [PRED] Special-purpose unary permutes
145 ;; ---- [PRED] Special-purpose binary permutes
146 ;;
147 ;; == Conversions
148 ;; ---- [INT<-INT] Packs
149 ;; ---- [INT<-INT] Unpacks
150 ;; ---- [INT<-FP] Conversions
151 ;; ---- [INT<-FP] Packs
152 ;; ---- [INT<-FP] Unpacks
153 ;; ---- [FP<-INT] Conversions
154 ;; ---- [FP<-INT] Packs
155 ;; ---- [FP<-INT] Unpacks
156 ;; ---- [FP<-FP] Packs
157 ;; ---- [FP<-FP] Packs (bfloat16)
158 ;; ---- [FP<-FP] Unpacks
159 ;; ---- [PRED<-PRED] Packs
160 ;; ---- [PRED<-PRED] Unpacks
161 ;;
162 ;; == Vector partitioning
163 ;; ---- [PRED] Unary partitioning
164 ;; ---- [PRED] Binary partitioning
165 ;; ---- [PRED] Scalarization
166 ;;
167 ;; == Counting elements
168 ;; ---- [INT] Count elements in a pattern (scalar)
169 ;; ---- [INT] Increment by the number of elements in a pattern (scalar)
170 ;; ---- [INT] Increment by the number of elements in a pattern (vector)
171 ;; ---- [INT] Decrement by the number of elements in a pattern (scalar)
172 ;; ---- [INT] Decrement by the number of elements in a pattern (vector)
173 ;; ---- [INT] Count elements in a predicate (scalar)
174 ;; ---- [INT] Increment by the number of elements in a predicate (scalar)
175 ;; ---- [INT] Increment by the number of elements in a predicate (vector)
176 ;; ---- [INT] Decrement by the number of elements in a predicate (scalar)
177 ;; ---- [INT] Decrement by the number of elements in a predicate (vector)
178
179 ;; =========================================================================
180 ;; == General notes
181 ;; =========================================================================
182 ;;
183 ;; -------------------------------------------------------------------------
184 ;; ---- Note on the handling of big-endian SVE
185 ;; -------------------------------------------------------------------------
186 ;;
187 ;; On big-endian systems, Advanced SIMD mov<mode> patterns act in the
188 ;; same way as movdi or movti would: the first byte of memory goes
189 ;; into the most significant byte of the register and the last byte
190 ;; of memory goes into the least significant byte of the register.
191 ;; This is the most natural ordering for Advanced SIMD and matches
192 ;; the ABI layout for 64-bit and 128-bit vector types.
193 ;;
194 ;; As a result, the order of bytes within the register is what GCC
195 ;; expects for a big-endian target, and subreg offsets therefore work
196 ;; as expected, with the first element in memory having subreg offset 0
197 ;; and the last element in memory having the subreg offset associated
198 ;; with a big-endian lowpart. However, this ordering also means that
199 ;; GCC's lane numbering does not match the architecture's numbering:
200 ;; GCC always treats the element at the lowest address in memory
201 ;; (subreg offset 0) as element 0, while the architecture treats
202 ;; the least significant end of the register as element 0.
203 ;;
204 ;; The situation for SVE is different. We want the layout of the
205 ;; SVE register to be same for mov<mode> as it is for maskload<mode>:
206 ;; logically, a mov<mode> load must be indistinguishable from a
207 ;; maskload<mode> whose mask is all true. We therefore need the
208 ;; register layout to match LD1 rather than LDR. The ABI layout of
209 ;; SVE types also matches LD1 byte ordering rather than LDR byte ordering.
210 ;;
211 ;; As a result, the architecture lane numbering matches GCC's lane
212 ;; numbering, with element 0 always being the first in memory.
213 ;; However:
214 ;;
215 ;; - Applying a subreg offset to a register does not give the element
216 ;; that GCC expects: the first element in memory has the subreg offset
217 ;; associated with a big-endian lowpart while the last element in memory
218 ;; has subreg offset 0. We handle this via TARGET_CAN_CHANGE_MODE_CLASS.
219 ;;
220 ;; - We cannot use LDR and STR for spill slots that might be accessed
221 ;; via subregs, since although the elements have the order GCC expects,
222 ;; the order of the bytes within the elements is different. We instead
223 ;; access spill slots via LD1 and ST1, using secondary reloads to
224 ;; reserve a predicate register.
225 ;;
226 ;; -------------------------------------------------------------------------
227 ;; ---- Description of UNSPEC_PTEST
228 ;; -------------------------------------------------------------------------
229 ;;
230 ;; SVE provides a PTEST instruction for testing the active lanes of a
231 ;; predicate and setting the flags based on the result. The associated
232 ;; condition code tests are:
233 ;;
234 ;; - any (= ne): at least one active bit is set
235 ;; - none (= eq): all active bits are clear (*)
236 ;; - first (= mi): the first active bit is set
237 ;; - nfrst (= pl): the first active bit is clear (*)
238 ;; - last (= cc): the last active bit is set
239 ;; - nlast (= cs): the last active bit is clear (*)
240 ;;
241 ;; where the conditions marked (*) are also true when there are no active
242 ;; lanes (i.e. when the governing predicate is a PFALSE). The flags results
243 ;; of a PTEST use the condition code mode CC_NZC.
244 ;;
245 ;; PTEST is always a .B operation (i.e. it always operates on VNx16BI).
246 ;; This means that for other predicate modes, we need a governing predicate
247 ;; in which all bits are defined.
248 ;;
249 ;; For example, most predicated .H operations ignore the odd bits of the
250 ;; governing predicate, so that an active lane is represented by the
251 ;; bits "1x" and an inactive lane by the bits "0x", where "x" can be
252 ;; any value. To test a .H predicate, we instead need "10" and "00"
253 ;; respectively, so that the condition only tests the even bits of the
254 ;; predicate.
255 ;;
256 ;; Several instructions set the flags as a side-effect, in the same way
257 ;; that a separate PTEST would. It's important for code quality that we
258 ;; use these flags results as often as possible, particularly in the case
259 ;; of WHILE* and RDFFR.
260 ;;
261 ;; Also, some of the instructions that set the flags are unpredicated
262 ;; and instead implicitly test all .B, .H, .S or .D elements, as though
263 ;; they were predicated on a PTRUE of that size. For example, a .S
264 ;; WHILELO sets the flags in the same way as a PTEST with a .S PTRUE
265 ;; would.
266 ;;
267 ;; We therefore need to represent PTEST operations in a way that
268 ;; makes it easy to combine them with both predicated and unpredicated
269 ;; operations, while using a VNx16BI governing predicate for all
270 ;; predicate modes. We do this using:
271 ;;
272 ;; (unspec:CC_NZC [gp cast_gp ptrue_flag op] UNSPEC_PTEST)
273 ;;
274 ;; where:
275 ;;
276 ;; - GP is the real VNx16BI governing predicate
277 ;;
278 ;; - CAST_GP is GP cast to the mode of OP. All bits dropped by casting
279 ;; GP to CAST_GP are guaranteed to be clear in GP.
280 ;;
281 ;; - PTRUE_FLAG is a CONST_INT (conceptually of mode SI) that has the value
282 ;; SVE_KNOWN_PTRUE if we know that CAST_GP (rather than GP) is all-true and
283 ;; SVE_MAYBE_NOT_PTRUE otherwise.
284 ;;
285 ;; - OP is the predicate we want to test, of the same mode as CAST_GP.
286 ;;
287 ;; -------------------------------------------------------------------------
288 ;; ---- Description of UNSPEC_PRED_Z
289 ;; -------------------------------------------------------------------------
290 ;;
291 ;; SVE integer comparisons are predicated and return zero for inactive
292 ;; lanes. Sometimes we use them with predicates that are all-true and
293 ;; sometimes we use them with general predicates.
294 ;;
295 ;; The integer comparisons also set the flags and so build-in the effect
296 ;; of a PTEST. We therefore want to be able to combine integer comparison
297 ;; patterns with PTESTs of the result. One difficulty with doing this is
298 ;; that (as noted above) the PTEST is always a .B operation and so can place
299 ;; stronger requirements on the governing predicate than the comparison does.
300 ;;
301 ;; For example, when applying a separate PTEST to the result of a full-vector
302 ;; .H comparison, the PTEST must be predicated on a .H PTRUE instead of a
303 ;; .B PTRUE. In constrast, the comparison might be predicated on either
304 ;; a .H PTRUE or a .B PTRUE, since the values of odd-indexed predicate
305 ;; bits don't matter for .H operations.
306 ;;
307 ;; We therefore can't rely on a full-vector comparison using the same
308 ;; predicate register as a following PTEST. We instead need to remember
309 ;; whether a comparison is known to be a full-vector comparison and use
310 ;; this information in addition to a check for equal predicate registers.
311 ;; At the same time, it's useful to have a common representation for all
312 ;; integer comparisons, so that they can be handled by a single set of
313 ;; patterns.
314 ;;
315 ;; We therefore take a similar approach to UNSPEC_PTEST above and use:
316 ;;
317 ;; (unspec:<M:VPRED> [gp ptrue_flag (code:M op0 op1)] UNSPEC_PRED_Z)
318 ;;
319 ;; where:
320 ;;
321 ;; - GP is the governing predicate, of mode <M:VPRED>
322 ;;
323 ;; - PTRUE_FLAG is a CONST_INT (conceptually of mode SI) that has the value
324 ;; SVE_KNOWN_PTRUE if we know that GP is all-true and SVE_MAYBE_NOT_PTRUE
325 ;; otherwise
326 ;;
327 ;; - CODE is the comparison code
328 ;;
329 ;; - OP0 and OP1 are the values being compared, of mode M
330 ;;
331 ;; The "Z" in UNSPEC_PRED_Z indicates that inactive lanes are zero.
332 ;;
333 ;; -------------------------------------------------------------------------
334 ;; ---- Note on predicated integer arithemtic and UNSPEC_PRED_X
335 ;; -------------------------------------------------------------------------
336 ;;
337 ;; Many SVE integer operations are predicated. We can generate them
338 ;; from four sources:
339 ;;
340 ;; (1) Using normal unpredicated optabs. In this case we need to create
341 ;; an all-true predicate register to act as the governing predicate
342 ;; for the SVE instruction. There are no inactive lanes, and thus
343 ;; the values of inactive lanes don't matter.
344 ;;
345 ;; (2) Using _x ACLE functions. In this case the function provides a
346 ;; specific predicate and some lanes might be inactive. However,
347 ;; as for (1), the values of the inactive lanes don't matter.
348 ;; We can make extra lanes active without changing the behavior
349 ;; (although for code-quality reasons we should avoid doing so
350 ;; needlessly).
351 ;;
352 ;; (3) Using cond_* optabs that correspond to IFN_COND_* internal functions.
353 ;; These optabs have a predicate operand that specifies which lanes are
354 ;; active and another operand that provides the values of inactive lanes.
355 ;;
356 ;; (4) Using _m and _z ACLE functions. These functions map to the same
357 ;; patterns as (3), with the _z functions setting inactive lanes to zero
358 ;; and the _m functions setting the inactive lanes to one of the function
359 ;; arguments.
360 ;;
361 ;; For (1) and (2) we need a way of attaching the predicate to a normal
362 ;; unpredicated integer operation. We do this using:
363 ;;
364 ;; (unspec:M [pred (code:M (op0 op1 ...))] UNSPEC_PRED_X)
365 ;;
366 ;; where (code:M (op0 op1 ...)) is the normal integer operation and PRED
367 ;; is a predicate of mode <M:VPRED>. PRED might or might not be a PTRUE;
368 ;; it always is for (1), but might not be for (2).
369 ;;
370 ;; The unspec as a whole has the same value as (code:M ...) when PRED is
371 ;; all-true. It is always semantically valid to replace PRED with a PTRUE,
372 ;; but as noted above, we should only do so if there's a specific benefit.
373 ;;
374 ;; (The "_X" in the unspec is named after the ACLE functions in (2).)
375 ;;
376 ;; For (3) and (4) we can simply use the SVE port's normal representation
377 ;; of a predicate-based select:
378 ;;
379 ;; (unspec:M [pred (code:M (op0 op1 ...)) inactive] UNSPEC_SEL)
380 ;;
381 ;; where INACTIVE specifies the values of inactive lanes.
382 ;;
383 ;; We can also use the UNSPEC_PRED_X wrapper in the UNSPEC_SEL rather
384 ;; than inserting the integer operation directly. This is mostly useful
385 ;; if we want the combine pass to merge an integer operation with an explicit
386 ;; vcond_mask (in other words, with a following SEL instruction). However,
387 ;; it's generally better to merge such operations at the gimple level
388 ;; using (3).
389 ;;
390 ;; -------------------------------------------------------------------------
391 ;; ---- Note on predicated FP arithmetic patterns and GP "strictness"
392 ;; -------------------------------------------------------------------------
393 ;;
394 ;; Most SVE floating-point operations are predicated. We can generate
395 ;; them from four sources:
396 ;;
397 ;; (1) Using normal unpredicated optabs. In this case we need to create
398 ;; an all-true predicate register to act as the governing predicate
399 ;; for the SVE instruction. There are no inactive lanes, and thus
400 ;; the values of inactive lanes don't matter.
401 ;;
402 ;; (2) Using _x ACLE functions. In this case the function provides a
403 ;; specific predicate and some lanes might be inactive. However,
404 ;; as for (1), the values of the inactive lanes don't matter.
405 ;;
406 ;; The instruction must have the same exception behavior as the
407 ;; function call unless things like command-line flags specifically
408 ;; allow otherwise. For example, with -ffast-math, it is OK to
409 ;; raise exceptions for inactive lanes, but normally it isn't.
410 ;;
411 ;; (3) Using cond_* optabs that correspond to IFN_COND_* internal functions.
412 ;; These optabs have a predicate operand that specifies which lanes are
413 ;; active and another operand that provides the values of inactive lanes.
414 ;;
415 ;; (4) Using _m and _z ACLE functions. These functions map to the same
416 ;; patterns as (3), with the _z functions setting inactive lanes to zero
417 ;; and the _m functions setting the inactive lanes to one of the function
418 ;; arguments.
419 ;;
420 ;; So:
421 ;;
422 ;; - In (1), the predicate is known to be all true and the pattern can use
423 ;; unpredicated operations where available.
424 ;;
425 ;; - In (2), the predicate might or might not be all true. The pattern can
426 ;; use unpredicated instructions if the predicate is all-true or if things
427 ;; like command-line flags allow exceptions for inactive lanes.
428 ;;
429 ;; - (3) and (4) represent a native SVE predicated operation. Some lanes
430 ;; might be inactive and inactive lanes of the result must have specific
431 ;; values. There is no scope for using unpredicated instructions (and no
432 ;; reason to want to), so the question about command-line flags doesn't
433 ;; arise.
434 ;;
435 ;; It would be inaccurate to model (2) as an rtx code like (sqrt ...)
436 ;; in combination with a separate predicate operand, e.g.
437 ;;
438 ;; (unspec [(match_operand:<VPRED> 1 "register_operand" "Upl")
439 ;; (sqrt:SVE_FULL_F 2 "register_operand" "w")]
440 ;; ....)
441 ;;
442 ;; because (sqrt ...) can raise an exception for any lane, including
443 ;; inactive ones. We therefore need to use an unspec instead.
444 ;;
445 ;; Also, (2) requires some way of distinguishing the case in which the
446 ;; predicate might have inactive lanes and cannot be changed from the
447 ;; case in which the predicate has no inactive lanes or can be changed.
448 ;; This information is also useful when matching combined FP patterns
449 ;; in which the predicates might not be equal.
450 ;;
451 ;; We therefore model FP operations as an unspec of the form:
452 ;;
453 ;; (unspec [pred strictness op0 op1 ...] UNSPEC_COND_<MNEMONIC>)
454 ;;
455 ;; where:
456 ;;
457 ;; - PRED is the governing predicate.
458 ;;
459 ;; - STRICTNESS is a CONST_INT that conceptually has mode SI. It has the
460 ;; value SVE_STRICT_GP if PRED might have inactive lanes and if those
461 ;; lanes must remain inactive. It has the value SVE_RELAXED_GP otherwise.
462 ;;
463 ;; - OP0 OP1 ... are the normal input operands to the operation.
464 ;;
465 ;; - MNEMONIC is the mnemonic of the associated SVE instruction.
466 ;;
467 ;; For (3) and (4), we combine these operations with an UNSPEC_SEL
468 ;; that selects between the result of the FP operation and the "else"
469 ;; value. (This else value is a merge input for _m ACLE functions
470 ;; and zero for _z ACLE functions.) The outer pattern then has the form:
471 ;;
472 ;; (unspec [pred fp_operation else_value] UNSPEC_SEL)
473 ;;
474 ;; This means that the patterns for (3) and (4) have two predicates:
475 ;; one for the FP operation itself and one for the UNSPEC_SEL.
476 ;; This pattern is equivalent to the result of combining an instance
477 ;; of (1) or (2) with a separate vcond instruction, so these patterns
478 ;; are useful as combine targets too.
479 ;;
480 ;; However, in the combine case, the instructions that we want to
481 ;; combine might use different predicates. Then:
482 ;;
483 ;; - Some of the active lanes of the FP operation might be discarded
484 ;; by the UNSPEC_SEL. It's OK to drop the FP operation on those lanes,
485 ;; even for SVE_STRICT_GP, since the operations on those lanes are
486 ;; effectively dead code.
487 ;;
488 ;; - Some of the inactive lanes of the FP operation might be selected
489 ;; by the UNSPEC_SEL, giving unspecified values for those lanes.
490 ;; SVE_RELAXED_GP lets us extend the FP operation to cover these
491 ;; extra lanes, but SVE_STRICT_GP does not.
492 ;;
493 ;; Thus SVE_RELAXED_GP allows us to ignore the predicate on the FP operation
494 ;; and operate on exactly the lanes selected by the UNSPEC_SEL predicate.
495 ;; This typically leads to patterns like:
496 ;;
497 ;; (unspec [(match_operand 1 "register_operand" "Upl")
498 ;; (unspec [(match_operand N)
499 ;; (const_int SVE_RELAXED_GP)
500 ;; ...]
501 ;; UNSPEC_COND_<MNEMONIC>)
502 ;; ...])
503 ;;
504 ;; where operand N is allowed to be anything. These instructions then
505 ;; have rewrite rules to replace operand N with operand 1, which gives the
506 ;; instructions a canonical form and means that the original operand N is
507 ;; not kept live unnecessarily.
508 ;;
509 ;; In contrast, SVE_STRICT_GP only allows the UNSPEC_SEL predicate to be
510 ;; a subset of the FP operation predicate. This case isn't interesting
511 ;; for FP operations that have an all-true predicate, since such operations
512 ;; use SVE_RELAXED_GP instead. And it is not possible for instruction
513 ;; conditions to track the subset relationship for arbitrary registers.
514 ;; So in practice, the only useful case for SVE_STRICT_GP is the one
515 ;; in which the predicates match:
516 ;;
517 ;; (unspec [(match_operand 1 "register_operand" "Upl")
518 ;; (unspec [(match_dup 1)
519 ;; (const_int SVE_STRICT_GP)
520 ;; ...]
521 ;; UNSPEC_COND_<MNEMONIC>)
522 ;; ...])
523 ;;
524 ;; This pattern would also be correct for SVE_RELAXED_GP, but it would
525 ;; be redundant with the one above. However, if the combine pattern
526 ;; has multiple FP operations, using a match_operand allows combinations
527 ;; of SVE_STRICT_GP and SVE_RELAXED_GP in the same operation, provided
528 ;; that the predicates are the same:
529 ;;
530 ;; (unspec [(match_operand 1 "register_operand" "Upl")
531 ;; (...
532 ;; (unspec [(match_dup 1)
533 ;; (match_operand:SI N "aarch64_sve_gp_strictness")
534 ;; ...]
535 ;; UNSPEC_COND_<MNEMONIC1>)
536 ;; (unspec [(match_dup 1)
537 ;; (match_operand:SI M "aarch64_sve_gp_strictness")
538 ;; ...]
539 ;; UNSPEC_COND_<MNEMONIC2>) ...)
540 ;; ...])
541 ;;
542 ;; The fully-relaxed version of this pattern is:
543 ;;
544 ;; (unspec [(match_operand 1 "register_operand" "Upl")
545 ;; (...
546 ;; (unspec [(match_operand:SI N)
547 ;; (const_int SVE_RELAXED_GP)
548 ;; ...]
549 ;; UNSPEC_COND_<MNEMONIC1>)
550 ;; (unspec [(match_operand:SI M)
551 ;; (const_int SVE_RELAXED_GP)
552 ;; ...]
553 ;; UNSPEC_COND_<MNEMONIC2>) ...)
554 ;; ...])
555 ;;
556 ;; -------------------------------------------------------------------------
557 ;; ---- Note on FFR handling
558 ;; -------------------------------------------------------------------------
559 ;;
560 ;; Logically we want to divide FFR-related instructions into regions
561 ;; that contain exactly one of:
562 ;;
563 ;; - a single write to the FFR
564 ;; - any number of reads from the FFR (but only one read is likely)
565 ;; - any number of LDFF1 and LDNF1 instructions
566 ;;
567 ;; However, LDFF1 and LDNF1 instructions should otherwise behave like
568 ;; normal loads as far as possible. This means that they should be
569 ;; schedulable within a region in the same way that LD1 would be,
570 ;; and they should be deleted as dead if the result is unused. The loads
571 ;; should therefore not write to the FFR, since that would both serialize
572 ;; the loads with respect to each other and keep the loads live for any
573 ;; later RDFFR.
574 ;;
575 ;; We get around this by using a fake "FFR token" (FFRT) to help describe
576 ;; the dependencies. Writing to the FFRT starts a new "FFRT region",
577 ;; while using the FFRT keeps the instruction within its region.
578 ;; Specifically:
579 ;;
580 ;; - Writes start a new FFRT region as well as setting the FFR:
581 ;;
582 ;; W1: parallel (FFRT = <new value>, FFR = <actual FFR value>)
583 ;;
584 ;; - Loads use an LD1-like instruction that also uses the FFRT, so that the
585 ;; loads stay within the same FFRT region:
586 ;;
587 ;; L1: load data while using the FFRT
588 ;;
589 ;; In addition, any FFRT region that includes a load also has at least one
590 ;; instance of:
591 ;;
592 ;; L2: FFR = update(FFR, FFRT) [type == no_insn]
593 ;;
594 ;; to make it clear that the region both reads from and writes to the FFR.
595 ;;
596 ;; - Reads do the following:
597 ;;
598 ;; R1: FFRT = FFR [type == no_insn]
599 ;; R2: read from the FFRT
600 ;; R3: FFRT = update(FFRT) [type == no_insn]
601 ;;
602 ;; R1 and R3 both create new FFRT regions, so that previous LDFF1s and
603 ;; LDNF1s cannot move forwards across R1 and later LDFF1s and LDNF1s
604 ;; cannot move backwards across R3.
605 ;;
606 ;; This way, writes are only kept alive by later loads or reads,
607 ;; and write/read pairs fold normally. For two consecutive reads,
608 ;; the first R3 is made dead by the second R1, which in turn becomes
609 ;; redundant with the first R1. We then have:
610 ;;
611 ;; first R1: FFRT = FFR
612 ;; first read from the FFRT
613 ;; second read from the FFRT
614 ;; second R3: FFRT = update(FFRT)
615 ;;
616 ;; i.e. the two FFRT regions collapse into a single one with two
617 ;; independent reads.
618 ;;
619 ;; The model still prevents some valid optimizations though. For example,
620 ;; if all loads in an FFRT region are deleted as dead, nothing would remove
621 ;; the L2 instructions.
622
623 ;; =========================================================================
624 ;; == Moves
625 ;; =========================================================================
626
627 ;; -------------------------------------------------------------------------
628 ;; ---- Moves of single vectors
629 ;; -------------------------------------------------------------------------
630 ;; Includes:
631 ;; - MOV (including aliases)
632 ;; - LD1B (contiguous form)
633 ;; - LD1D ( " " )
634 ;; - LD1H ( " " )
635 ;; - LD1W ( " " )
636 ;; - LDR
637 ;; - ST1B (contiguous form)
638 ;; - ST1D ( " " )
639 ;; - ST1H ( " " )
640 ;; - ST1W ( " " )
641 ;; - STR
642 ;; -------------------------------------------------------------------------
643
644 (define_expand "mov<mode>"
645 [(set (match_operand:SVE_ALL 0 "nonimmediate_operand")
646 (match_operand:SVE_ALL 1 "general_operand"))]
647 "TARGET_SVE"
648 {
649 /* Use the predicated load and store patterns where possible.
650 This is required for big-endian targets (see the comment at the
651 head of the file) and increases the addressing choices for
652 little-endian. */
653 if ((MEM_P (operands[0]) || MEM_P (operands[1]))
654 && can_create_pseudo_p ())
655 {
656 aarch64_expand_sve_mem_move (operands[0], operands[1], <VPRED>mode);
657 DONE;
658 }
659
660 if (CONSTANT_P (operands[1]))
661 {
662 aarch64_expand_mov_immediate (operands[0], operands[1]);
663 DONE;
664 }
665
666 /* Optimize subregs on big-endian targets: we can use REV[BHW]
667 instead of going through memory. */
668 if (BYTES_BIG_ENDIAN
669 && aarch64_maybe_expand_sve_subreg_move (operands[0], operands[1]))
670 DONE;
671 }
672 )
673
674 (define_expand "movmisalign<mode>"
675 [(set (match_operand:SVE_ALL 0 "nonimmediate_operand")
676 (match_operand:SVE_ALL 1 "general_operand"))]
677 "TARGET_SVE"
678 {
679 /* Equivalent to a normal move for our purpooses. */
680 emit_move_insn (operands[0], operands[1]);
681 DONE;
682 }
683 )
684
685 ;; Unpredicated moves that can use LDR and STR, i.e. full vectors for which
686 ;; little-endian ordering is acceptable. Only allow memory operations during
687 ;; and after RA; before RA we want the predicated load and store patterns to
688 ;; be used instead.
689 (define_insn "*aarch64_sve_mov<mode>_ldr_str"
690 [(set (match_operand:SVE_FULL 0 "aarch64_sve_nonimmediate_operand" "=w, Utr, w, w")
691 (match_operand:SVE_FULL 1 "aarch64_sve_general_operand" "Utr, w, w, Dn"))]
692 "TARGET_SVE
693 && (<MODE>mode == VNx16QImode || !BYTES_BIG_ENDIAN)
694 && ((lra_in_progress || reload_completed)
695 || (register_operand (operands[0], <MODE>mode)
696 && nonmemory_operand (operands[1], <MODE>mode)))"
697 "@
698 ldr\t%0, %1
699 str\t%1, %0
700 mov\t%0.d, %1.d
701 * return aarch64_output_sve_mov_immediate (operands[1]);"
702 )
703
704 ;; Unpredicated moves that cannot use LDR and STR, i.e. partial vectors
705 ;; or vectors for which little-endian ordering isn't acceptable. Memory
706 ;; accesses require secondary reloads.
707 (define_insn "*aarch64_sve_mov<mode>_no_ldr_str"
708 [(set (match_operand:SVE_ALL 0 "register_operand" "=w, w")
709 (match_operand:SVE_ALL 1 "aarch64_nonmemory_operand" "w, Dn"))]
710 "TARGET_SVE
711 && <MODE>mode != VNx16QImode
712 && (BYTES_BIG_ENDIAN
713 || maybe_ne (BYTES_PER_SVE_VECTOR, GET_MODE_SIZE (<MODE>mode)))"
714 "@
715 mov\t%0.d, %1.d
716 * return aarch64_output_sve_mov_immediate (operands[1]);"
717 )
718
719 ;; Handle memory reloads for modes that can't use LDR and STR. We use
720 ;; byte PTRUE for all modes to try to encourage reuse. This pattern
721 ;; needs constraints because it is returned by TARGET_SECONDARY_RELOAD.
722 (define_expand "aarch64_sve_reload_mem"
723 [(parallel
724 [(set (match_operand 0)
725 (match_operand 1))
726 (clobber (match_operand:VNx16BI 2 "register_operand" "=Upl"))])]
727 "TARGET_SVE"
728 {
729 /* Create a PTRUE. */
730 emit_move_insn (operands[2], CONSTM1_RTX (VNx16BImode));
731
732 /* Refer to the PTRUE in the appropriate mode for this move. */
733 machine_mode mode = GET_MODE (operands[0]);
734 rtx pred = gen_lowpart (aarch64_sve_pred_mode (mode), operands[2]);
735
736 /* Emit a predicated load or store. */
737 aarch64_emit_sve_pred_move (operands[0], pred, operands[1]);
738 DONE;
739 }
740 )
741
742 ;; A predicated move in which the predicate is known to be all-true.
743 ;; Note that this pattern is generated directly by aarch64_emit_sve_pred_move,
744 ;; so changes to this pattern will need changes there as well.
745 (define_insn_and_split "@aarch64_pred_mov<mode>"
746 [(set (match_operand:SVE_ALL 0 "nonimmediate_operand" "=w, w, m")
747 (unspec:SVE_ALL
748 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
749 (match_operand:SVE_ALL 2 "nonimmediate_operand" "w, m, w")]
750 UNSPEC_PRED_X))]
751 "TARGET_SVE
752 && (register_operand (operands[0], <MODE>mode)
753 || register_operand (operands[2], <MODE>mode))"
754 "@
755 #
756 ld1<Vesize>\t%0.<Vctype>, %1/z, %2
757 st1<Vesize>\t%2.<Vctype>, %1, %0"
758 "&& register_operand (operands[0], <MODE>mode)
759 && register_operand (operands[2], <MODE>mode)"
760 [(set (match_dup 0) (match_dup 2))]
761 )
762
763 ;; A pattern for optimizing SUBREGs that have a reinterpreting effect
764 ;; on big-endian targets; see aarch64_maybe_expand_sve_subreg_move
765 ;; for details. We use a special predicate for operand 2 to reduce
766 ;; the number of patterns.
767 (define_insn_and_split "*aarch64_sve_mov<mode>_subreg_be"
768 [(set (match_operand:SVE_ALL 0 "aarch64_sve_nonimmediate_operand" "=w")
769 (unspec:SVE_ALL
770 [(match_operand:VNx16BI 1 "register_operand" "Upl")
771 (match_operand 2 "aarch64_any_register_operand" "w")]
772 UNSPEC_REV_SUBREG))]
773 "TARGET_SVE && BYTES_BIG_ENDIAN"
774 "#"
775 "&& reload_completed"
776 [(const_int 0)]
777 {
778 aarch64_split_sve_subreg_move (operands[0], operands[1], operands[2]);
779 DONE;
780 }
781 )
782
783 ;; Reinterpret operand 1 in operand 0's mode, without changing its contents.
784 ;; This is equivalent to a subreg on little-endian targets but not for
785 ;; big-endian; see the comment at the head of the file for details.
786 (define_expand "@aarch64_sve_reinterpret<mode>"
787 [(set (match_operand:SVE_ALL 0 "register_operand")
788 (unspec:SVE_ALL
789 [(match_operand 1 "aarch64_any_register_operand")]
790 UNSPEC_REINTERPRET))]
791 "TARGET_SVE"
792 {
793 machine_mode src_mode = GET_MODE (operands[1]);
794 if (targetm.can_change_mode_class (<MODE>mode, src_mode, FP_REGS))
795 {
796 emit_move_insn (operands[0], gen_lowpart (<MODE>mode, operands[1]));
797 DONE;
798 }
799 }
800 )
801
802 ;; A pattern for handling type punning on big-endian targets. We use a
803 ;; special predicate for operand 1 to reduce the number of patterns.
804 (define_insn_and_split "*aarch64_sve_reinterpret<mode>"
805 [(set (match_operand:SVE_ALL 0 "register_operand" "=w")
806 (unspec:SVE_ALL
807 [(match_operand 1 "aarch64_any_register_operand" "w")]
808 UNSPEC_REINTERPRET))]
809 "TARGET_SVE"
810 "#"
811 "&& reload_completed"
812 [(set (match_dup 0) (match_dup 1))]
813 {
814 operands[1] = aarch64_replace_reg_mode (operands[1], <MODE>mode);
815 }
816 )
817
818 ;; -------------------------------------------------------------------------
819 ;; ---- Moves of multiple vectors
820 ;; -------------------------------------------------------------------------
821 ;; All patterns in this section are synthetic and split to real
822 ;; instructions after reload.
823 ;; -------------------------------------------------------------------------
824
825 (define_expand "mov<mode>"
826 [(set (match_operand:SVE_STRUCT 0 "nonimmediate_operand")
827 (match_operand:SVE_STRUCT 1 "general_operand"))]
828 "TARGET_SVE"
829 {
830 /* Big-endian loads and stores need to be done via LD1 and ST1;
831 see the comment at the head of the file for details. */
832 if ((MEM_P (operands[0]) || MEM_P (operands[1]))
833 && BYTES_BIG_ENDIAN)
834 {
835 gcc_assert (can_create_pseudo_p ());
836 aarch64_expand_sve_mem_move (operands[0], operands[1], <VPRED>mode);
837 DONE;
838 }
839
840 if (CONSTANT_P (operands[1]))
841 {
842 aarch64_expand_mov_immediate (operands[0], operands[1]);
843 DONE;
844 }
845 }
846 )
847
848 ;; Unpredicated structure moves (little-endian).
849 (define_insn "*aarch64_sve_mov<mode>_le"
850 [(set (match_operand:SVE_STRUCT 0 "aarch64_sve_nonimmediate_operand" "=w, Utr, w, w")
851 (match_operand:SVE_STRUCT 1 "aarch64_sve_general_operand" "Utr, w, w, Dn"))]
852 "TARGET_SVE && !BYTES_BIG_ENDIAN"
853 "#"
854 [(set_attr "length" "<insn_length>")]
855 )
856
857 ;; Unpredicated structure moves (big-endian). Memory accesses require
858 ;; secondary reloads.
859 (define_insn "*aarch64_sve_mov<mode>_be"
860 [(set (match_operand:SVE_STRUCT 0 "register_operand" "=w, w")
861 (match_operand:SVE_STRUCT 1 "aarch64_nonmemory_operand" "w, Dn"))]
862 "TARGET_SVE && BYTES_BIG_ENDIAN"
863 "#"
864 [(set_attr "length" "<insn_length>")]
865 )
866
867 ;; Split unpredicated structure moves into pieces. This is the same
868 ;; for both big-endian and little-endian code, although it only needs
869 ;; to handle memory operands for little-endian code.
870 (define_split
871 [(set (match_operand:SVE_STRUCT 0 "aarch64_sve_nonimmediate_operand")
872 (match_operand:SVE_STRUCT 1 "aarch64_sve_general_operand"))]
873 "TARGET_SVE && reload_completed"
874 [(const_int 0)]
875 {
876 rtx dest = operands[0];
877 rtx src = operands[1];
878 if (REG_P (dest) && REG_P (src))
879 aarch64_simd_emit_reg_reg_move (operands, <VSINGLE>mode, <vector_count>);
880 else
881 for (unsigned int i = 0; i < <vector_count>; ++i)
882 {
883 rtx subdest = simplify_gen_subreg (<VSINGLE>mode, dest, <MODE>mode,
884 i * BYTES_PER_SVE_VECTOR);
885 rtx subsrc = simplify_gen_subreg (<VSINGLE>mode, src, <MODE>mode,
886 i * BYTES_PER_SVE_VECTOR);
887 emit_insn (gen_rtx_SET (subdest, subsrc));
888 }
889 DONE;
890 }
891 )
892
893 ;; Predicated structure moves. This works for both endiannesses but in
894 ;; practice is only useful for big-endian.
895 (define_insn_and_split "@aarch64_pred_mov<mode>"
896 [(set (match_operand:SVE_STRUCT 0 "aarch64_sve_struct_nonimmediate_operand" "=w, w, Utx")
897 (unspec:SVE_STRUCT
898 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
899 (match_operand:SVE_STRUCT 2 "aarch64_sve_struct_nonimmediate_operand" "w, Utx, w")]
900 UNSPEC_PRED_X))]
901 "TARGET_SVE
902 && (register_operand (operands[0], <MODE>mode)
903 || register_operand (operands[2], <MODE>mode))"
904 "#"
905 "&& reload_completed"
906 [(const_int 0)]
907 {
908 for (unsigned int i = 0; i < <vector_count>; ++i)
909 {
910 rtx subdest = simplify_gen_subreg (<VSINGLE>mode, operands[0],
911 <MODE>mode,
912 i * BYTES_PER_SVE_VECTOR);
913 rtx subsrc = simplify_gen_subreg (<VSINGLE>mode, operands[2],
914 <MODE>mode,
915 i * BYTES_PER_SVE_VECTOR);
916 aarch64_emit_sve_pred_move (subdest, operands[1], subsrc);
917 }
918 DONE;
919 }
920 [(set_attr "length" "<insn_length>")]
921 )
922
923 ;; -------------------------------------------------------------------------
924 ;; ---- Moves of predicates
925 ;; -------------------------------------------------------------------------
926 ;; Includes:
927 ;; - MOV
928 ;; - LDR
929 ;; - PFALSE
930 ;; - PTRUE
931 ;; - PTRUES
932 ;; - STR
933 ;; -------------------------------------------------------------------------
934
935 (define_expand "mov<mode>"
936 [(set (match_operand:PRED_ALL 0 "nonimmediate_operand")
937 (match_operand:PRED_ALL 1 "general_operand"))]
938 "TARGET_SVE"
939 {
940 if (GET_CODE (operands[0]) == MEM)
941 operands[1] = force_reg (<MODE>mode, operands[1]);
942
943 if (CONSTANT_P (operands[1]))
944 {
945 aarch64_expand_mov_immediate (operands[0], operands[1]);
946 DONE;
947 }
948 }
949 )
950
951 (define_insn "*aarch64_sve_mov<mode>"
952 [(set (match_operand:PRED_ALL 0 "nonimmediate_operand" "=Upa, m, Upa, Upa")
953 (match_operand:PRED_ALL 1 "aarch64_mov_operand" "Upa, Upa, m, Dn"))]
954 "TARGET_SVE
955 && (register_operand (operands[0], <MODE>mode)
956 || register_operand (operands[1], <MODE>mode))"
957 "@
958 mov\t%0.b, %1.b
959 str\t%1, %0
960 ldr\t%0, %1
961 * return aarch64_output_sve_mov_immediate (operands[1]);"
962 )
963
964 ;; Match PTRUES Pn.B when both the predicate and flags are useful.
965 (define_insn_and_rewrite "*aarch64_sve_ptruevnx16bi_cc"
966 [(set (reg:CC_NZC CC_REGNUM)
967 (unspec:CC_NZC
968 [(match_operand 2)
969 (match_operand 3)
970 (const_int SVE_KNOWN_PTRUE)
971 (match_operator:VNx16BI 1 "aarch64_sve_ptrue_svpattern_immediate"
972 [(unspec:VNx16BI
973 [(match_operand:SI 4 "const_int_operand")
974 (match_operand:VNx16BI 5 "aarch64_simd_imm_zero")]
975 UNSPEC_PTRUE)])]
976 UNSPEC_PTEST))
977 (set (match_operand:VNx16BI 0 "register_operand" "=Upa")
978 (match_dup 1))]
979 "TARGET_SVE"
980 {
981 return aarch64_output_sve_ptrues (operands[1]);
982 }
983 "&& (!CONSTANT_P (operands[2]) || !CONSTANT_P (operands[3]))"
984 {
985 operands[2] = operands[3] = CONSTM1_RTX (VNx16BImode);
986 }
987 )
988
989 ;; Match PTRUES Pn.[HSD] when both the predicate and flags are useful.
990 (define_insn_and_rewrite "*aarch64_sve_ptrue<mode>_cc"
991 [(set (reg:CC_NZC CC_REGNUM)
992 (unspec:CC_NZC
993 [(match_operand 2)
994 (match_operand 3)
995 (const_int SVE_KNOWN_PTRUE)
996 (subreg:PRED_HSD
997 (match_operator:VNx16BI 1 "aarch64_sve_ptrue_svpattern_immediate"
998 [(unspec:VNx16BI
999 [(match_operand:SI 4 "const_int_operand")
1000 (match_operand:PRED_HSD 5 "aarch64_simd_imm_zero")]
1001 UNSPEC_PTRUE)]) 0)]
1002 UNSPEC_PTEST))
1003 (set (match_operand:VNx16BI 0 "register_operand" "=Upa")
1004 (match_dup 1))]
1005 "TARGET_SVE"
1006 {
1007 return aarch64_output_sve_ptrues (operands[1]);
1008 }
1009 "&& (!CONSTANT_P (operands[2]) || !CONSTANT_P (operands[3]))"
1010 {
1011 operands[2] = CONSTM1_RTX (VNx16BImode);
1012 operands[3] = CONSTM1_RTX (<MODE>mode);
1013 }
1014 )
1015
1016 ;; Match PTRUES Pn.B when only the flags result is useful (which is
1017 ;; a way of testing VL).
1018 (define_insn_and_rewrite "*aarch64_sve_ptruevnx16bi_ptest"
1019 [(set (reg:CC_NZC CC_REGNUM)
1020 (unspec:CC_NZC
1021 [(match_operand 2)
1022 (match_operand 3)
1023 (const_int SVE_KNOWN_PTRUE)
1024 (match_operator:VNx16BI 1 "aarch64_sve_ptrue_svpattern_immediate"
1025 [(unspec:VNx16BI
1026 [(match_operand:SI 4 "const_int_operand")
1027 (match_operand:VNx16BI 5 "aarch64_simd_imm_zero")]
1028 UNSPEC_PTRUE)])]
1029 UNSPEC_PTEST))
1030 (clobber (match_scratch:VNx16BI 0 "=Upa"))]
1031 "TARGET_SVE"
1032 {
1033 return aarch64_output_sve_ptrues (operands[1]);
1034 }
1035 "&& (!CONSTANT_P (operands[2]) || !CONSTANT_P (operands[3]))"
1036 {
1037 operands[2] = operands[3] = CONSTM1_RTX (VNx16BImode);
1038 }
1039 )
1040
1041 ;; Match PTRUES Pn.[HWD] when only the flags result is useful (which is
1042 ;; a way of testing VL).
1043 (define_insn_and_rewrite "*aarch64_sve_ptrue<mode>_ptest"
1044 [(set (reg:CC_NZC CC_REGNUM)
1045 (unspec:CC_NZC
1046 [(match_operand 2)
1047 (match_operand 3)
1048 (const_int SVE_KNOWN_PTRUE)
1049 (subreg:PRED_HSD
1050 (match_operator:VNx16BI 1 "aarch64_sve_ptrue_svpattern_immediate"
1051 [(unspec:VNx16BI
1052 [(match_operand:SI 4 "const_int_operand")
1053 (match_operand:PRED_HSD 5 "aarch64_simd_imm_zero")]
1054 UNSPEC_PTRUE)]) 0)]
1055 UNSPEC_PTEST))
1056 (clobber (match_scratch:VNx16BI 0 "=Upa"))]
1057 "TARGET_SVE"
1058 {
1059 return aarch64_output_sve_ptrues (operands[1]);
1060 }
1061 "&& (!CONSTANT_P (operands[2]) || !CONSTANT_P (operands[3]))"
1062 {
1063 operands[2] = CONSTM1_RTX (VNx16BImode);
1064 operands[3] = CONSTM1_RTX (<MODE>mode);
1065 }
1066 )
1067
1068 ;; -------------------------------------------------------------------------
1069 ;; ---- Moves relating to the FFR
1070 ;; -------------------------------------------------------------------------
1071 ;; RDFFR
1072 ;; RDFFRS
1073 ;; SETFFR
1074 ;; WRFFR
1075 ;; -------------------------------------------------------------------------
1076
1077 ;; [W1 in the block comment above about FFR handling]
1078 ;;
1079 ;; Write to the FFR and start a new FFRT scheduling region.
1080 (define_insn "aarch64_wrffr"
1081 [(set (reg:VNx16BI FFR_REGNUM)
1082 (match_operand:VNx16BI 0 "aarch64_simd_reg_or_minus_one" "Dm, Upa"))
1083 (set (reg:VNx16BI FFRT_REGNUM)
1084 (unspec:VNx16BI [(match_dup 0)] UNSPEC_WRFFR))]
1085 "TARGET_SVE"
1086 "@
1087 setffr
1088 wrffr\t%0.b"
1089 )
1090
1091 ;; [L2 in the block comment above about FFR handling]
1092 ;;
1093 ;; Introduce a read from and write to the FFR in the current FFRT region,
1094 ;; so that the FFR value is live on entry to the region and so that the FFR
1095 ;; value visibly changes within the region. This is used (possibly multiple
1096 ;; times) in an FFRT region that includes LDFF1 or LDNF1 instructions.
1097 (define_insn "aarch64_update_ffr_for_load"
1098 [(set (reg:VNx16BI FFR_REGNUM)
1099 (unspec:VNx16BI [(reg:VNx16BI FFRT_REGNUM)
1100 (reg:VNx16BI FFR_REGNUM)] UNSPEC_UPDATE_FFR))]
1101 "TARGET_SVE"
1102 ""
1103 [(set_attr "type" "no_insn")]
1104 )
1105
1106 ;; [R1 in the block comment above about FFR handling]
1107 ;;
1108 ;; Notionally copy the FFR to the FFRT, so that the current FFR value
1109 ;; can be read from there by the RDFFR instructions below. This acts
1110 ;; as a scheduling barrier for earlier LDFF1 and LDNF1 instructions and
1111 ;; creates a natural dependency with earlier writes.
1112 (define_insn "aarch64_copy_ffr_to_ffrt"
1113 [(set (reg:VNx16BI FFRT_REGNUM)
1114 (reg:VNx16BI FFR_REGNUM))]
1115 "TARGET_SVE"
1116 ""
1117 [(set_attr "type" "no_insn")]
1118 )
1119
1120 ;; [R2 in the block comment above about FFR handling]
1121 ;;
1122 ;; Read the FFR via the FFRT.
1123 (define_insn "aarch64_rdffr"
1124 [(set (match_operand:VNx16BI 0 "register_operand" "=Upa")
1125 (reg:VNx16BI FFRT_REGNUM))]
1126 "TARGET_SVE"
1127 "rdffr\t%0.b"
1128 )
1129
1130 ;; Likewise with zero predication.
1131 (define_insn "aarch64_rdffr_z"
1132 [(set (match_operand:VNx16BI 0 "register_operand" "=Upa")
1133 (and:VNx16BI
1134 (reg:VNx16BI FFRT_REGNUM)
1135 (match_operand:VNx16BI 1 "register_operand" "Upa")))]
1136 "TARGET_SVE"
1137 "rdffr\t%0.b, %1/z"
1138 )
1139
1140 ;; Read the FFR to test for a fault, without using the predicate result.
1141 (define_insn "*aarch64_rdffr_z_ptest"
1142 [(set (reg:CC_NZC CC_REGNUM)
1143 (unspec:CC_NZC
1144 [(match_operand:VNx16BI 1 "register_operand" "Upa")
1145 (match_dup 1)
1146 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
1147 (and:VNx16BI
1148 (reg:VNx16BI FFRT_REGNUM)
1149 (match_dup 1))]
1150 UNSPEC_PTEST))
1151 (clobber (match_scratch:VNx16BI 0 "=Upa"))]
1152 "TARGET_SVE"
1153 "rdffrs\t%0.b, %1/z"
1154 )
1155
1156 ;; Same for unpredicated RDFFR when tested with a known PTRUE.
1157 (define_insn "*aarch64_rdffr_ptest"
1158 [(set (reg:CC_NZC CC_REGNUM)
1159 (unspec:CC_NZC
1160 [(match_operand:VNx16BI 1 "register_operand" "Upa")
1161 (match_dup 1)
1162 (const_int SVE_KNOWN_PTRUE)
1163 (reg:VNx16BI FFRT_REGNUM)]
1164 UNSPEC_PTEST))
1165 (clobber (match_scratch:VNx16BI 0 "=Upa"))]
1166 "TARGET_SVE"
1167 "rdffrs\t%0.b, %1/z"
1168 )
1169
1170 ;; Read the FFR with zero predication and test the result.
1171 (define_insn "*aarch64_rdffr_z_cc"
1172 [(set (reg:CC_NZC CC_REGNUM)
1173 (unspec:CC_NZC
1174 [(match_operand:VNx16BI 1 "register_operand" "Upa")
1175 (match_dup 1)
1176 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
1177 (and:VNx16BI
1178 (reg:VNx16BI FFRT_REGNUM)
1179 (match_dup 1))]
1180 UNSPEC_PTEST))
1181 (set (match_operand:VNx16BI 0 "register_operand" "=Upa")
1182 (and:VNx16BI
1183 (reg:VNx16BI FFRT_REGNUM)
1184 (match_dup 1)))]
1185 "TARGET_SVE"
1186 "rdffrs\t%0.b, %1/z"
1187 )
1188
1189 ;; Same for unpredicated RDFFR when tested with a known PTRUE.
1190 (define_insn "*aarch64_rdffr_cc"
1191 [(set (reg:CC_NZC CC_REGNUM)
1192 (unspec:CC_NZC
1193 [(match_operand:VNx16BI 1 "register_operand" "Upa")
1194 (match_dup 1)
1195 (const_int SVE_KNOWN_PTRUE)
1196 (reg:VNx16BI FFRT_REGNUM)]
1197 UNSPEC_PTEST))
1198 (set (match_operand:VNx16BI 0 "register_operand" "=Upa")
1199 (reg:VNx16BI FFRT_REGNUM))]
1200 "TARGET_SVE"
1201 "rdffrs\t%0.b, %1/z"
1202 )
1203
1204 ;; [R3 in the block comment above about FFR handling]
1205 ;;
1206 ;; Arbitrarily update the FFRT after a read from the FFR. This acts as
1207 ;; a scheduling barrier for later LDFF1 and LDNF1 instructions.
1208 (define_insn "aarch64_update_ffrt"
1209 [(set (reg:VNx16BI FFRT_REGNUM)
1210 (unspec:VNx16BI [(reg:VNx16BI FFRT_REGNUM)] UNSPEC_UPDATE_FFRT))]
1211 "TARGET_SVE"
1212 ""
1213 [(set_attr "type" "no_insn")]
1214 )
1215
1216 ;; =========================================================================
1217 ;; == Loads
1218 ;; =========================================================================
1219
1220 ;; -------------------------------------------------------------------------
1221 ;; ---- Normal contiguous loads
1222 ;; -------------------------------------------------------------------------
1223 ;; Includes contiguous forms of:
1224 ;; - LD1B
1225 ;; - LD1D
1226 ;; - LD1H
1227 ;; - LD1W
1228 ;; - LD2B
1229 ;; - LD2D
1230 ;; - LD2H
1231 ;; - LD2W
1232 ;; - LD3B
1233 ;; - LD3D
1234 ;; - LD3H
1235 ;; - LD3W
1236 ;; - LD4B
1237 ;; - LD4D
1238 ;; - LD4H
1239 ;; - LD4W
1240 ;; -------------------------------------------------------------------------
1241
1242 ;; Predicated LD1.
1243 (define_insn "maskload<mode><vpred>"
1244 [(set (match_operand:SVE_ALL 0 "register_operand" "=w")
1245 (unspec:SVE_ALL
1246 [(match_operand:<VPRED> 2 "register_operand" "Upl")
1247 (match_operand:SVE_ALL 1 "memory_operand" "m")]
1248 UNSPEC_LD1_SVE))]
1249 "TARGET_SVE"
1250 "ld1<Vesize>\t%0.<Vctype>, %2/z, %1"
1251 )
1252
1253 ;; Unpredicated LD[234].
1254 (define_expand "vec_load_lanes<mode><vsingle>"
1255 [(set (match_operand:SVE_STRUCT 0 "register_operand")
1256 (unspec:SVE_STRUCT
1257 [(match_dup 2)
1258 (match_operand:SVE_STRUCT 1 "memory_operand")]
1259 UNSPEC_LDN))]
1260 "TARGET_SVE"
1261 {
1262 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
1263 }
1264 )
1265
1266 ;; Predicated LD[234].
1267 (define_insn "vec_mask_load_lanes<mode><vsingle>"
1268 [(set (match_operand:SVE_STRUCT 0 "register_operand" "=w")
1269 (unspec:SVE_STRUCT
1270 [(match_operand:<VPRED> 2 "register_operand" "Upl")
1271 (match_operand:SVE_STRUCT 1 "memory_operand" "m")]
1272 UNSPEC_LDN))]
1273 "TARGET_SVE"
1274 "ld<vector_count><Vesize>\t%0, %2/z, %1"
1275 )
1276
1277 ;; -------------------------------------------------------------------------
1278 ;; ---- Extending contiguous loads
1279 ;; -------------------------------------------------------------------------
1280 ;; Includes contiguous forms of:
1281 ;; LD1B
1282 ;; LD1H
1283 ;; LD1SB
1284 ;; LD1SH
1285 ;; LD1SW
1286 ;; LD1W
1287 ;; -------------------------------------------------------------------------
1288
1289 ;; Predicated load and extend, with 8 elements per 128-bit block.
1290 (define_insn_and_rewrite "@aarch64_load_<ANY_EXTEND:optab><SVE_HSDI:mode><SVE_PARTIAL_I:mode>"
1291 [(set (match_operand:SVE_HSDI 0 "register_operand" "=w")
1292 (unspec:SVE_HSDI
1293 [(match_operand:<SVE_HSDI:VPRED> 3 "general_operand" "UplDnm")
1294 (ANY_EXTEND:SVE_HSDI
1295 (unspec:SVE_PARTIAL_I
1296 [(match_operand:<SVE_PARTIAL_I:VPRED> 2 "register_operand" "Upl")
1297 (match_operand:SVE_PARTIAL_I 1 "memory_operand" "m")]
1298 UNSPEC_LD1_SVE))]
1299 UNSPEC_PRED_X))]
1300 "TARGET_SVE && (~<SVE_HSDI:narrower_mask> & <SVE_PARTIAL_I:self_mask>) == 0"
1301 "ld1<ANY_EXTEND:s><SVE_PARTIAL_I:Vesize>\t%0.<SVE_HSDI:Vctype>, %2/z, %1"
1302 "&& !CONSTANT_P (operands[3])"
1303 {
1304 operands[3] = CONSTM1_RTX (<SVE_HSDI:VPRED>mode);
1305 }
1306 )
1307
1308 ;; -------------------------------------------------------------------------
1309 ;; ---- First-faulting contiguous loads
1310 ;; -------------------------------------------------------------------------
1311 ;; Includes contiguous forms of:
1312 ;; - LDFF1B
1313 ;; - LDFF1D
1314 ;; - LDFF1H
1315 ;; - LDFF1W
1316 ;; - LDNF1B
1317 ;; - LDNF1D
1318 ;; - LDNF1H
1319 ;; - LDNF1W
1320 ;; -------------------------------------------------------------------------
1321
1322 ;; Contiguous non-extending first-faulting or non-faulting loads.
1323 (define_insn "@aarch64_ld<fn>f1<mode>"
1324 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
1325 (unspec:SVE_FULL
1326 [(match_operand:<VPRED> 2 "register_operand" "Upl")
1327 (match_operand:SVE_FULL 1 "aarch64_sve_ld<fn>f1_operand" "Ut<fn>")
1328 (reg:VNx16BI FFRT_REGNUM)]
1329 SVE_LDFF1_LDNF1))]
1330 "TARGET_SVE"
1331 "ld<fn>f1<Vesize>\t%0.<Vetype>, %2/z, %1"
1332 )
1333
1334 ;; -------------------------------------------------------------------------
1335 ;; ---- First-faulting extending contiguous loads
1336 ;; -------------------------------------------------------------------------
1337 ;; Includes contiguous forms of:
1338 ;; - LDFF1B
1339 ;; - LDFF1H
1340 ;; - LDFF1SB
1341 ;; - LDFF1SH
1342 ;; - LDFF1SW
1343 ;; - LDFF1W
1344 ;; - LDNF1B
1345 ;; - LDNF1H
1346 ;; - LDNF1SB
1347 ;; - LDNF1SH
1348 ;; - LDNF1SW
1349 ;; - LDNF1W
1350 ;; -------------------------------------------------------------------------
1351
1352 ;; Predicated first-faulting or non-faulting load and extend.
1353 (define_insn_and_rewrite "@aarch64_ld<fn>f1_<ANY_EXTEND:optab><SVE_HSDI:mode><SVE_PARTIAL_I:mode>"
1354 [(set (match_operand:SVE_HSDI 0 "register_operand" "=w")
1355 (unspec:SVE_HSDI
1356 [(match_operand:<SVE_HSDI:VPRED> 3 "general_operand" "UplDnm")
1357 (ANY_EXTEND:SVE_HSDI
1358 (unspec:SVE_PARTIAL_I
1359 [(match_operand:<SVE_PARTIAL_I:VPRED> 2 "register_operand" "Upl")
1360 (match_operand:SVE_PARTIAL_I 1 "aarch64_sve_ld<fn>f1_operand" "Ut<fn>")
1361 (reg:VNx16BI FFRT_REGNUM)]
1362 SVE_LDFF1_LDNF1))]
1363 UNSPEC_PRED_X))]
1364 "TARGET_SVE && (~<SVE_HSDI:narrower_mask> & <SVE_PARTIAL_I:self_mask>) == 0"
1365 "ld<fn>f1<ANY_EXTEND:s><SVE_PARTIAL_I:Vesize>\t%0.<SVE_HSDI:Vctype>, %2/z, %1"
1366 "&& !CONSTANT_P (operands[3])"
1367 {
1368 operands[3] = CONSTM1_RTX (<SVE_HSDI:VPRED>mode);
1369 }
1370 )
1371
1372 ;; -------------------------------------------------------------------------
1373 ;; ---- Non-temporal contiguous loads
1374 ;; -------------------------------------------------------------------------
1375 ;; Includes:
1376 ;; - LDNT1B
1377 ;; - LDNT1D
1378 ;; - LDNT1H
1379 ;; - LDNT1W
1380 ;; -------------------------------------------------------------------------
1381
1382 ;; Predicated contiguous non-temporal load.
1383 (define_insn "@aarch64_ldnt1<mode>"
1384 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
1385 (unspec:SVE_FULL
1386 [(match_operand:<VPRED> 2 "register_operand" "Upl")
1387 (match_operand:SVE_FULL 1 "memory_operand" "m")]
1388 UNSPEC_LDNT1_SVE))]
1389 "TARGET_SVE"
1390 "ldnt1<Vesize>\t%0.<Vetype>, %2/z, %1"
1391 )
1392
1393 ;; -------------------------------------------------------------------------
1394 ;; ---- Normal gather loads
1395 ;; -------------------------------------------------------------------------
1396 ;; Includes gather forms of:
1397 ;; - LD1D
1398 ;; - LD1W
1399 ;; -------------------------------------------------------------------------
1400
1401 ;; Unpredicated gather loads.
1402 (define_expand "gather_load<mode><v_int_container>"
1403 [(set (match_operand:SVE_24 0 "register_operand")
1404 (unspec:SVE_24
1405 [(match_dup 5)
1406 (match_operand:DI 1 "aarch64_sve_gather_offset_<Vesize>")
1407 (match_operand:<V_INT_CONTAINER> 2 "register_operand")
1408 (match_operand:DI 3 "const_int_operand")
1409 (match_operand:DI 4 "aarch64_gather_scale_operand_<Vesize>")
1410 (mem:BLK (scratch))]
1411 UNSPEC_LD1_GATHER))]
1412 "TARGET_SVE"
1413 {
1414 operands[5] = aarch64_ptrue_reg (<VPRED>mode);
1415 }
1416 )
1417
1418 ;; Predicated gather loads for 32-bit elements. Operand 3 is true for
1419 ;; unsigned extension and false for signed extension.
1420 (define_insn "mask_gather_load<mode><v_int_container>"
1421 [(set (match_operand:SVE_4 0 "register_operand" "=w, w, w, w, w, w")
1422 (unspec:SVE_4
1423 [(match_operand:VNx4BI 5 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
1424 (match_operand:DI 1 "aarch64_sve_gather_offset_<Vesize>" "Z, vgw, rk, rk, rk, rk")
1425 (match_operand:VNx4SI 2 "register_operand" "w, w, w, w, w, w")
1426 (match_operand:DI 3 "const_int_operand" "Ui1, Ui1, Z, Ui1, Z, Ui1")
1427 (match_operand:DI 4 "aarch64_gather_scale_operand_<Vesize>" "Ui1, Ui1, Ui1, Ui1, i, i")
1428 (mem:BLK (scratch))]
1429 UNSPEC_LD1_GATHER))]
1430 "TARGET_SVE"
1431 "@
1432 ld1<Vesize>\t%0.s, %5/z, [%2.s]
1433 ld1<Vesize>\t%0.s, %5/z, [%2.s, #%1]
1434 ld1<Vesize>\t%0.s, %5/z, [%1, %2.s, sxtw]
1435 ld1<Vesize>\t%0.s, %5/z, [%1, %2.s, uxtw]
1436 ld1<Vesize>\t%0.s, %5/z, [%1, %2.s, sxtw %p4]
1437 ld1<Vesize>\t%0.s, %5/z, [%1, %2.s, uxtw %p4]"
1438 )
1439
1440 ;; Predicated gather loads for 64-bit elements. The value of operand 3
1441 ;; doesn't matter in this case.
1442 (define_insn "mask_gather_load<mode><v_int_container>"
1443 [(set (match_operand:SVE_2 0 "register_operand" "=w, w, w, w")
1444 (unspec:SVE_2
1445 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl, Upl, Upl")
1446 (match_operand:DI 1 "aarch64_sve_gather_offset_<Vesize>" "Z, vgd, rk, rk")
1447 (match_operand:VNx2DI 2 "register_operand" "w, w, w, w")
1448 (match_operand:DI 3 "const_int_operand")
1449 (match_operand:DI 4 "aarch64_gather_scale_operand_<Vesize>" "Ui1, Ui1, Ui1, i")
1450 (mem:BLK (scratch))]
1451 UNSPEC_LD1_GATHER))]
1452 "TARGET_SVE"
1453 "@
1454 ld1<Vesize>\t%0.d, %5/z, [%2.d]
1455 ld1<Vesize>\t%0.d, %5/z, [%2.d, #%1]
1456 ld1<Vesize>\t%0.d, %5/z, [%1, %2.d]
1457 ld1<Vesize>\t%0.d, %5/z, [%1, %2.d, lsl %p4]"
1458 )
1459
1460 ;; Likewise, but with the offset being extended from 32 bits.
1461 (define_insn_and_rewrite "*mask_gather_load<mode><v_int_container>_<su>xtw_unpacked"
1462 [(set (match_operand:SVE_2 0 "register_operand" "=w, w")
1463 (unspec:SVE_2
1464 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1465 (match_operand:DI 1 "register_operand" "rk, rk")
1466 (unspec:VNx2DI
1467 [(match_operand 6)
1468 (ANY_EXTEND:VNx2DI
1469 (match_operand:VNx2SI 2 "register_operand" "w, w"))]
1470 UNSPEC_PRED_X)
1471 (match_operand:DI 3 "const_int_operand")
1472 (match_operand:DI 4 "aarch64_gather_scale_operand_<Vesize>" "Ui1, i")
1473 (mem:BLK (scratch))]
1474 UNSPEC_LD1_GATHER))]
1475 "TARGET_SVE"
1476 "@
1477 ld1<Vesize>\t%0.d, %5/z, [%1, %2.d, <su>xtw]
1478 ld1<Vesize>\t%0.d, %5/z, [%1, %2.d, <su>xtw %p4]"
1479 "&& !CONSTANT_P (operands[6])"
1480 {
1481 operands[6] = CONSTM1_RTX (VNx2BImode);
1482 }
1483 )
1484
1485 ;; Likewise, but with the offset being truncated to 32 bits and then
1486 ;; sign-extended.
1487 (define_insn_and_rewrite "*mask_gather_load<mode><v_int_container>_sxtw"
1488 [(set (match_operand:SVE_2 0 "register_operand" "=w, w")
1489 (unspec:SVE_2
1490 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1491 (match_operand:DI 1 "register_operand" "rk, rk")
1492 (unspec:VNx2DI
1493 [(match_operand 6)
1494 (sign_extend:VNx2DI
1495 (truncate:VNx2SI
1496 (match_operand:VNx2DI 2 "register_operand" "w, w")))]
1497 UNSPEC_PRED_X)
1498 (match_operand:DI 3 "const_int_operand")
1499 (match_operand:DI 4 "aarch64_gather_scale_operand_<Vesize>" "Ui1, i")
1500 (mem:BLK (scratch))]
1501 UNSPEC_LD1_GATHER))]
1502 "TARGET_SVE"
1503 "@
1504 ld1<Vesize>\t%0.d, %5/z, [%1, %2.d, sxtw]
1505 ld1<Vesize>\t%0.d, %5/z, [%1, %2.d, sxtw %p4]"
1506 "&& !CONSTANT_P (operands[6])"
1507 {
1508 operands[6] = CONSTM1_RTX (VNx2BImode);
1509 }
1510 )
1511
1512 ;; Likewise, but with the offset being truncated to 32 bits and then
1513 ;; zero-extended.
1514 (define_insn "*mask_gather_load<mode><v_int_container>_uxtw"
1515 [(set (match_operand:SVE_2 0 "register_operand" "=w, w")
1516 (unspec:SVE_2
1517 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1518 (match_operand:DI 1 "register_operand" "rk, rk")
1519 (and:VNx2DI
1520 (match_operand:VNx2DI 2 "register_operand" "w, w")
1521 (match_operand:VNx2DI 6 "aarch64_sve_uxtw_immediate"))
1522 (match_operand:DI 3 "const_int_operand")
1523 (match_operand:DI 4 "aarch64_gather_scale_operand_<Vesize>" "Ui1, i")
1524 (mem:BLK (scratch))]
1525 UNSPEC_LD1_GATHER))]
1526 "TARGET_SVE"
1527 "@
1528 ld1<Vesize>\t%0.d, %5/z, [%1, %2.d, uxtw]
1529 ld1<Vesize>\t%0.d, %5/z, [%1, %2.d, uxtw %p4]"
1530 )
1531
1532 ;; -------------------------------------------------------------------------
1533 ;; ---- Extending gather loads
1534 ;; -------------------------------------------------------------------------
1535 ;; Includes gather forms of:
1536 ;; - LD1B
1537 ;; - LD1H
1538 ;; - LD1SB
1539 ;; - LD1SH
1540 ;; - LD1SW
1541 ;; - LD1W
1542 ;; -------------------------------------------------------------------------
1543
1544 ;; Predicated extending gather loads for 32-bit elements. Operand 3 is
1545 ;; true for unsigned extension and false for signed extension.
1546 (define_insn_and_rewrite "@aarch64_gather_load_<ANY_EXTEND:optab><SVE_4HSI:mode><SVE_4BHI:mode>"
1547 [(set (match_operand:SVE_4HSI 0 "register_operand" "=w, w, w, w, w, w")
1548 (unspec:SVE_4HSI
1549 [(match_operand:VNx4BI 6 "general_operand" "UplDnm, UplDnm, UplDnm, UplDnm, UplDnm, UplDnm")
1550 (ANY_EXTEND:SVE_4HSI
1551 (unspec:SVE_4BHI
1552 [(match_operand:VNx4BI 5 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
1553 (match_operand:DI 1 "aarch64_sve_gather_offset_<SVE_4BHI:Vesize>" "Z, vg<SVE_4BHI:Vesize>, rk, rk, rk, rk")
1554 (match_operand:VNx4SI 2 "register_operand" "w, w, w, w, w, w")
1555 (match_operand:DI 3 "const_int_operand" "Ui1, Ui1, Z, Ui1, Z, Ui1")
1556 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_4BHI:Vesize>" "Ui1, Ui1, Ui1, Ui1, i, i")
1557 (mem:BLK (scratch))]
1558 UNSPEC_LD1_GATHER))]
1559 UNSPEC_PRED_X))]
1560 "TARGET_SVE && (~<SVE_4HSI:narrower_mask> & <SVE_4BHI:self_mask>) == 0"
1561 "@
1562 ld1<ANY_EXTEND:s><SVE_4BHI:Vesize>\t%0.s, %5/z, [%2.s]
1563 ld1<ANY_EXTEND:s><SVE_4BHI:Vesize>\t%0.s, %5/z, [%2.s, #%1]
1564 ld1<ANY_EXTEND:s><SVE_4BHI:Vesize>\t%0.s, %5/z, [%1, %2.s, sxtw]
1565 ld1<ANY_EXTEND:s><SVE_4BHI:Vesize>\t%0.s, %5/z, [%1, %2.s, uxtw]
1566 ld1<ANY_EXTEND:s><SVE_4BHI:Vesize>\t%0.s, %5/z, [%1, %2.s, sxtw %p4]
1567 ld1<ANY_EXTEND:s><SVE_4BHI:Vesize>\t%0.s, %5/z, [%1, %2.s, uxtw %p4]"
1568 "&& !CONSTANT_P (operands[6])"
1569 {
1570 operands[6] = CONSTM1_RTX (VNx4BImode);
1571 }
1572 )
1573
1574 ;; Predicated extending gather loads for 64-bit elements. The value of
1575 ;; operand 3 doesn't matter in this case.
1576 (define_insn_and_rewrite "@aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode><SVE_2BHSI:mode>"
1577 [(set (match_operand:SVE_2HSDI 0 "register_operand" "=w, w, w, w")
1578 (unspec:SVE_2HSDI
1579 [(match_operand:VNx2BI 6 "general_operand" "UplDnm, UplDnm, UplDnm, UplDnm")
1580 (ANY_EXTEND:SVE_2HSDI
1581 (unspec:SVE_2BHSI
1582 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl, Upl, Upl")
1583 (match_operand:DI 1 "aarch64_sve_gather_offset_<SVE_2BHSI:Vesize>" "Z, vg<SVE_2BHSI:Vesize>, rk, rk")
1584 (match_operand:VNx2DI 2 "register_operand" "w, w, w, w")
1585 (match_operand:DI 3 "const_int_operand")
1586 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_2BHSI:Vesize>" "Ui1, Ui1, Ui1, i")
1587 (mem:BLK (scratch))]
1588 UNSPEC_LD1_GATHER))]
1589 UNSPEC_PRED_X))]
1590 "TARGET_SVE && (~<SVE_2HSDI:narrower_mask> & <SVE_2BHSI:self_mask>) == 0"
1591 "@
1592 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%2.d]
1593 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%2.d, #%1]
1594 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%1, %2.d]
1595 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%1, %2.d, lsl %p4]"
1596 "&& !CONSTANT_P (operands[6])"
1597 {
1598 operands[6] = CONSTM1_RTX (VNx2BImode);
1599 }
1600 )
1601
1602 ;; Likewise, but with the offset being extended from 32 bits.
1603 (define_insn_and_rewrite "*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode><SVE_2BHSI:mode>_<ANY_EXTEND2:su>xtw_unpacked"
1604 [(set (match_operand:SVE_2HSDI 0 "register_operand" "=w, w")
1605 (unspec:SVE_2HSDI
1606 [(match_operand 6)
1607 (ANY_EXTEND:SVE_2HSDI
1608 (unspec:SVE_2BHSI
1609 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1610 (match_operand:DI 1 "aarch64_reg_or_zero" "rk, rk")
1611 (unspec:VNx2DI
1612 [(match_operand 7)
1613 (ANY_EXTEND2:VNx2DI
1614 (match_operand:VNx2SI 2 "register_operand" "w, w"))]
1615 UNSPEC_PRED_X)
1616 (match_operand:DI 3 "const_int_operand")
1617 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_2BHSI:Vesize>" "Ui1, i")
1618 (mem:BLK (scratch))]
1619 UNSPEC_LD1_GATHER))]
1620 UNSPEC_PRED_X))]
1621 "TARGET_SVE && (~<SVE_2HSDI:narrower_mask> & <SVE_2BHSI:self_mask>) == 0"
1622 "@
1623 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%1, %2.d, <ANY_EXTEND2:su>xtw]
1624 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%1, %2.d, <ANY_EXTEND2:su>xtw %p4]"
1625 "&& (!CONSTANT_P (operands[6]) || !CONSTANT_P (operands[7]))"
1626 {
1627 operands[6] = CONSTM1_RTX (VNx2BImode);
1628 operands[7] = CONSTM1_RTX (VNx2BImode);
1629 }
1630 )
1631
1632 ;; Likewise, but with the offset being truncated to 32 bits and then
1633 ;; sign-extended.
1634 (define_insn_and_rewrite "*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode><SVE_2BHSI:mode>_sxtw"
1635 [(set (match_operand:SVE_2HSDI 0 "register_operand" "=w, w")
1636 (unspec:SVE_2HSDI
1637 [(match_operand 6)
1638 (ANY_EXTEND:SVE_2HSDI
1639 (unspec:SVE_2BHSI
1640 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1641 (match_operand:DI 1 "aarch64_reg_or_zero" "rk, rk")
1642 (unspec:VNx2DI
1643 [(match_operand 7)
1644 (sign_extend:VNx2DI
1645 (truncate:VNx2SI
1646 (match_operand:VNx2DI 2 "register_operand" "w, w")))]
1647 UNSPEC_PRED_X)
1648 (match_operand:DI 3 "const_int_operand")
1649 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_2BHSI:Vesize>" "Ui1, i")
1650 (mem:BLK (scratch))]
1651 UNSPEC_LD1_GATHER))]
1652 UNSPEC_PRED_X))]
1653 "TARGET_SVE && (~<SVE_2HSDI:narrower_mask> & <SVE_2BHSI:self_mask>) == 0"
1654 "@
1655 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%1, %2.d, sxtw]
1656 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%1, %2.d, sxtw %p4]"
1657 "&& (!CONSTANT_P (operands[6]) || !CONSTANT_P (operands[7]))"
1658 {
1659 operands[6] = CONSTM1_RTX (VNx2BImode);
1660 operands[7] = CONSTM1_RTX (VNx2BImode);
1661 }
1662 )
1663
1664 ;; Likewise, but with the offset being truncated to 32 bits and then
1665 ;; zero-extended.
1666 (define_insn_and_rewrite "*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode><SVE_2BHSI:mode>_uxtw"
1667 [(set (match_operand:SVE_2HSDI 0 "register_operand" "=w, w")
1668 (unspec:SVE_2HSDI
1669 [(match_operand 7)
1670 (ANY_EXTEND:SVE_2HSDI
1671 (unspec:SVE_2BHSI
1672 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1673 (match_operand:DI 1 "aarch64_reg_or_zero" "rk, rk")
1674 (and:VNx2DI
1675 (match_operand:VNx2DI 2 "register_operand" "w, w")
1676 (match_operand:VNx2DI 6 "aarch64_sve_uxtw_immediate"))
1677 (match_operand:DI 3 "const_int_operand")
1678 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_2BHSI:Vesize>" "Ui1, i")
1679 (mem:BLK (scratch))]
1680 UNSPEC_LD1_GATHER))]
1681 UNSPEC_PRED_X))]
1682 "TARGET_SVE && (~<SVE_2HSDI:narrower_mask> & <SVE_2BHSI:self_mask>) == 0"
1683 "@
1684 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%1, %2.d, uxtw]
1685 ld1<ANY_EXTEND:s><SVE_2BHSI:Vesize>\t%0.d, %5/z, [%1, %2.d, uxtw %p4]"
1686 "&& !CONSTANT_P (operands[7])"
1687 {
1688 operands[7] = CONSTM1_RTX (VNx2BImode);
1689 }
1690 )
1691
1692 ;; -------------------------------------------------------------------------
1693 ;; ---- First-faulting gather loads
1694 ;; -------------------------------------------------------------------------
1695 ;; Includes gather forms of:
1696 ;; - LDFF1D
1697 ;; - LDFF1W
1698 ;; -------------------------------------------------------------------------
1699
1700 ;; Predicated first-faulting gather loads for 32-bit elements. Operand
1701 ;; 3 is true for unsigned extension and false for signed extension.
1702 (define_insn "@aarch64_ldff1_gather<mode>"
1703 [(set (match_operand:SVE_FULL_S 0 "register_operand" "=w, w, w, w, w, w")
1704 (unspec:SVE_FULL_S
1705 [(match_operand:VNx4BI 5 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
1706 (match_operand:DI 1 "aarch64_sve_gather_offset_w" "Z, vgw, rk, rk, rk, rk")
1707 (match_operand:VNx4SI 2 "register_operand" "w, w, w, w, w, w")
1708 (match_operand:DI 3 "const_int_operand" "i, i, Z, Ui1, Z, Ui1")
1709 (match_operand:DI 4 "aarch64_gather_scale_operand_w" "Ui1, Ui1, Ui1, Ui1, i, i")
1710 (mem:BLK (scratch))
1711 (reg:VNx16BI FFRT_REGNUM)]
1712 UNSPEC_LDFF1_GATHER))]
1713 "TARGET_SVE"
1714 "@
1715 ldff1w\t%0.s, %5/z, [%2.s]
1716 ldff1w\t%0.s, %5/z, [%2.s, #%1]
1717 ldff1w\t%0.s, %5/z, [%1, %2.s, sxtw]
1718 ldff1w\t%0.s, %5/z, [%1, %2.s, uxtw]
1719 ldff1w\t%0.s, %5/z, [%1, %2.s, sxtw %p4]
1720 ldff1w\t%0.s, %5/z, [%1, %2.s, uxtw %p4]"
1721 )
1722
1723 ;; Predicated first-faulting gather loads for 64-bit elements. The value
1724 ;; of operand 3 doesn't matter in this case.
1725 (define_insn "@aarch64_ldff1_gather<mode>"
1726 [(set (match_operand:SVE_FULL_D 0 "register_operand" "=w, w, w, w")
1727 (unspec:SVE_FULL_D
1728 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl, Upl, Upl")
1729 (match_operand:DI 1 "aarch64_sve_gather_offset_d" "Z, vgd, rk, rk")
1730 (match_operand:VNx2DI 2 "register_operand" "w, w, w, w")
1731 (match_operand:DI 3 "const_int_operand")
1732 (match_operand:DI 4 "aarch64_gather_scale_operand_d" "Ui1, Ui1, Ui1, i")
1733 (mem:BLK (scratch))
1734 (reg:VNx16BI FFRT_REGNUM)]
1735 UNSPEC_LDFF1_GATHER))]
1736 "TARGET_SVE"
1737 "@
1738 ldff1d\t%0.d, %5/z, [%2.d]
1739 ldff1d\t%0.d, %5/z, [%2.d, #%1]
1740 ldff1d\t%0.d, %5/z, [%1, %2.d]
1741 ldff1d\t%0.d, %5/z, [%1, %2.d, lsl %p4]"
1742 )
1743
1744 ;; Likewise, but with the offset being sign-extended from 32 bits.
1745 (define_insn_and_rewrite "*aarch64_ldff1_gather<mode>_sxtw"
1746 [(set (match_operand:SVE_FULL_D 0 "register_operand" "=w, w")
1747 (unspec:SVE_FULL_D
1748 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1749 (match_operand:DI 1 "register_operand" "rk, rk")
1750 (unspec:VNx2DI
1751 [(match_operand 6)
1752 (sign_extend:VNx2DI
1753 (truncate:VNx2SI
1754 (match_operand:VNx2DI 2 "register_operand" "w, w")))]
1755 UNSPEC_PRED_X)
1756 (match_operand:DI 3 "const_int_operand")
1757 (match_operand:DI 4 "aarch64_gather_scale_operand_d" "Ui1, i")
1758 (mem:BLK (scratch))
1759 (reg:VNx16BI FFRT_REGNUM)]
1760 UNSPEC_LDFF1_GATHER))]
1761 "TARGET_SVE"
1762 "@
1763 ldff1d\t%0.d, %5/z, [%1, %2.d, sxtw]
1764 ldff1d\t%0.d, %5/z, [%1, %2.d, sxtw %p4]"
1765 "&& !CONSTANT_P (operands[6])"
1766 {
1767 operands[6] = CONSTM1_RTX (VNx2BImode);
1768 }
1769 )
1770
1771 ;; Likewise, but with the offset being zero-extended from 32 bits.
1772 (define_insn "*aarch64_ldff1_gather<mode>_uxtw"
1773 [(set (match_operand:SVE_FULL_D 0 "register_operand" "=w, w")
1774 (unspec:SVE_FULL_D
1775 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1776 (match_operand:DI 1 "register_operand" "rk, rk")
1777 (and:VNx2DI
1778 (match_operand:VNx2DI 2 "register_operand" "w, w")
1779 (match_operand:VNx2DI 6 "aarch64_sve_uxtw_immediate"))
1780 (match_operand:DI 3 "const_int_operand")
1781 (match_operand:DI 4 "aarch64_gather_scale_operand_d" "Ui1, i")
1782 (mem:BLK (scratch))
1783 (reg:VNx16BI FFRT_REGNUM)]
1784 UNSPEC_LDFF1_GATHER))]
1785 "TARGET_SVE"
1786 "@
1787 ldff1d\t%0.d, %5/z, [%1, %2.d, uxtw]
1788 ldff1d\t%0.d, %5/z, [%1, %2.d, uxtw %p4]"
1789 )
1790
1791 ;; -------------------------------------------------------------------------
1792 ;; ---- First-faulting extending gather loads
1793 ;; -------------------------------------------------------------------------
1794 ;; Includes gather forms of:
1795 ;; - LDFF1B
1796 ;; - LDFF1H
1797 ;; - LDFF1SB
1798 ;; - LDFF1SH
1799 ;; - LDFF1SW
1800 ;; - LDFF1W
1801 ;; -------------------------------------------------------------------------
1802
1803 ;; Predicated extending first-faulting gather loads for 32-bit elements.
1804 ;; Operand 3 is true for unsigned extension and false for signed extension.
1805 (define_insn_and_rewrite "@aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx4_WIDE:mode><VNx4_NARROW:mode>"
1806 [(set (match_operand:VNx4_WIDE 0 "register_operand" "=w, w, w, w, w, w")
1807 (unspec:VNx4_WIDE
1808 [(match_operand:VNx4BI 6 "general_operand" "UplDnm, UplDnm, UplDnm, UplDnm, UplDnm, UplDnm")
1809 (ANY_EXTEND:VNx4_WIDE
1810 (unspec:VNx4_NARROW
1811 [(match_operand:VNx4BI 5 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
1812 (match_operand:DI 1 "aarch64_sve_gather_offset_<VNx4_NARROW:Vesize>" "Z, vg<VNx4_NARROW:Vesize>, rk, rk, rk, rk")
1813 (match_operand:VNx4_WIDE 2 "register_operand" "w, w, w, w, w, w")
1814 (match_operand:DI 3 "const_int_operand" "i, i, Z, Ui1, Z, Ui1")
1815 (match_operand:DI 4 "aarch64_gather_scale_operand_<VNx4_NARROW:Vesize>" "Ui1, Ui1, Ui1, Ui1, i, i")
1816 (mem:BLK (scratch))
1817 (reg:VNx16BI FFRT_REGNUM)]
1818 UNSPEC_LDFF1_GATHER))]
1819 UNSPEC_PRED_X))]
1820 "TARGET_SVE"
1821 "@
1822 ldff1<ANY_EXTEND:s><VNx4_NARROW:Vesize>\t%0.s, %5/z, [%2.s]
1823 ldff1<ANY_EXTEND:s><VNx4_NARROW:Vesize>\t%0.s, %5/z, [%2.s, #%1]
1824 ldff1<ANY_EXTEND:s><VNx4_NARROW:Vesize>\t%0.s, %5/z, [%1, %2.s, sxtw]
1825 ldff1<ANY_EXTEND:s><VNx4_NARROW:Vesize>\t%0.s, %5/z, [%1, %2.s, uxtw]
1826 ldff1<ANY_EXTEND:s><VNx4_NARROW:Vesize>\t%0.s, %5/z, [%1, %2.s, sxtw %p4]
1827 ldff1<ANY_EXTEND:s><VNx4_NARROW:Vesize>\t%0.s, %5/z, [%1, %2.s, uxtw %p4]"
1828 "&& !CONSTANT_P (operands[6])"
1829 {
1830 operands[6] = CONSTM1_RTX (VNx4BImode);
1831 }
1832 )
1833
1834 ;; Predicated extending first-faulting gather loads for 64-bit elements.
1835 ;; The value of operand 3 doesn't matter in this case.
1836 (define_insn_and_rewrite "@aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode><VNx2_NARROW:mode>"
1837 [(set (match_operand:VNx2_WIDE 0 "register_operand" "=w, w, w, w")
1838 (unspec:VNx2_WIDE
1839 [(match_operand:VNx2BI 6 "general_operand" "UplDnm, UplDnm, UplDnm, UplDnm")
1840 (ANY_EXTEND:VNx2_WIDE
1841 (unspec:VNx2_NARROW
1842 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl, Upl, Upl")
1843 (match_operand:DI 1 "aarch64_sve_gather_offset_<VNx2_NARROW:Vesize>" "Z, vg<VNx2_NARROW:Vesize>, rk, rk")
1844 (match_operand:VNx2_WIDE 2 "register_operand" "w, w, w, w")
1845 (match_operand:DI 3 "const_int_operand")
1846 (match_operand:DI 4 "aarch64_gather_scale_operand_<VNx2_NARROW:Vesize>" "Ui1, Ui1, Ui1, i")
1847 (mem:BLK (scratch))
1848 (reg:VNx16BI FFRT_REGNUM)]
1849 UNSPEC_LDFF1_GATHER))]
1850 UNSPEC_PRED_X))]
1851 "TARGET_SVE"
1852 "@
1853 ldff1<ANY_EXTEND:s><VNx2_NARROW:Vesize>\t%0.d, %5/z, [%2.d]
1854 ldff1<ANY_EXTEND:s><VNx2_NARROW:Vesize>\t%0.d, %5/z, [%2.d, #%1]
1855 ldff1<ANY_EXTEND:s><VNx2_NARROW:Vesize>\t%0.d, %5/z, [%1, %2.d]
1856 ldff1<ANY_EXTEND:s><VNx2_NARROW:Vesize>\t%0.d, %5/z, [%1, %2.d, lsl %p4]"
1857 "&& !CONSTANT_P (operands[6])"
1858 {
1859 operands[6] = CONSTM1_RTX (VNx2BImode);
1860 }
1861 )
1862
1863 ;; Likewise, but with the offset being sign-extended from 32 bits.
1864 (define_insn_and_rewrite "*aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode><VNx2_NARROW:mode>_sxtw"
1865 [(set (match_operand:VNx2_WIDE 0 "register_operand" "=w, w")
1866 (unspec:VNx2_WIDE
1867 [(match_operand 6)
1868 (ANY_EXTEND:VNx2_WIDE
1869 (unspec:VNx2_NARROW
1870 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1871 (match_operand:DI 1 "aarch64_reg_or_zero" "rk, rk")
1872 (unspec:VNx2DI
1873 [(match_operand 7)
1874 (sign_extend:VNx2DI
1875 (truncate:VNx2SI
1876 (match_operand:VNx2DI 2 "register_operand" "w, w")))]
1877 UNSPEC_PRED_X)
1878 (match_operand:DI 3 "const_int_operand")
1879 (match_operand:DI 4 "aarch64_gather_scale_operand_<VNx2_NARROW:Vesize>" "Ui1, i")
1880 (mem:BLK (scratch))
1881 (reg:VNx16BI FFRT_REGNUM)]
1882 UNSPEC_LDFF1_GATHER))]
1883 UNSPEC_PRED_X))]
1884 "TARGET_SVE"
1885 "@
1886 ldff1<ANY_EXTEND:s><VNx2_NARROW:Vesize>\t%0.d, %5/z, [%1, %2.d, sxtw]
1887 ldff1<ANY_EXTEND:s><VNx2_NARROW:Vesize>\t%0.d, %5/z, [%1, %2.d, sxtw %p4]"
1888 "&& (!CONSTANT_P (operands[6]) || !CONSTANT_P (operands[7]))"
1889 {
1890 operands[6] = CONSTM1_RTX (VNx2BImode);
1891 operands[7] = CONSTM1_RTX (VNx2BImode);
1892 }
1893 )
1894
1895 ;; Likewise, but with the offset being zero-extended from 32 bits.
1896 (define_insn_and_rewrite "*aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode><VNx2_NARROW:mode>_uxtw"
1897 [(set (match_operand:VNx2_WIDE 0 "register_operand" "=w, w")
1898 (unspec:VNx2_WIDE
1899 [(match_operand 7)
1900 (ANY_EXTEND:VNx2_WIDE
1901 (unspec:VNx2_NARROW
1902 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
1903 (match_operand:DI 1 "aarch64_reg_or_zero" "rk, rk")
1904 (and:VNx2DI
1905 (match_operand:VNx2DI 2 "register_operand" "w, w")
1906 (match_operand:VNx2DI 6 "aarch64_sve_uxtw_immediate"))
1907 (match_operand:DI 3 "const_int_operand")
1908 (match_operand:DI 4 "aarch64_gather_scale_operand_<VNx2_NARROW:Vesize>" "Ui1, i")
1909 (mem:BLK (scratch))
1910 (reg:VNx16BI FFRT_REGNUM)]
1911 UNSPEC_LDFF1_GATHER))]
1912 UNSPEC_PRED_X))]
1913 "TARGET_SVE"
1914 "@
1915 ldff1<ANY_EXTEND:s><VNx2_NARROW:Vesize>\t%0.d, %5/z, [%1, %2.d, uxtw]
1916 ldff1<ANY_EXTEND:s><VNx2_NARROW:Vesize>\t%0.d, %5/z, [%1, %2.d, uxtw %p4]"
1917 "&& !CONSTANT_P (operands[7])"
1918 {
1919 operands[7] = CONSTM1_RTX (VNx2BImode);
1920 }
1921 )
1922
1923 ;; =========================================================================
1924 ;; == Prefetches
1925 ;; =========================================================================
1926
1927 ;; -------------------------------------------------------------------------
1928 ;; ---- Contiguous prefetches
1929 ;; -------------------------------------------------------------------------
1930 ;; Includes contiguous forms of:
1931 ;; - PRFB
1932 ;; - PRFD
1933 ;; - PRFH
1934 ;; - PRFW
1935 ;; -------------------------------------------------------------------------
1936
1937 ;; Contiguous predicated prefetches. Operand 2 gives the real prefetch
1938 ;; operation (as an svprfop), with operands 3 and 4 providing distilled
1939 ;; information.
1940 (define_insn "@aarch64_sve_prefetch<mode>"
1941 [(prefetch (unspec:DI
1942 [(match_operand:<VPRED> 0 "register_operand" "Upl")
1943 (match_operand:SVE_FULL_I 1 "aarch64_sve_prefetch_operand" "UP<Vesize>")
1944 (match_operand:DI 2 "const_int_operand")]
1945 UNSPEC_SVE_PREFETCH)
1946 (match_operand:DI 3 "const_int_operand")
1947 (match_operand:DI 4 "const_int_operand"))]
1948 "TARGET_SVE"
1949 {
1950 operands[1] = gen_rtx_MEM (<MODE>mode, operands[1]);
1951 return aarch64_output_sve_prefetch ("prf<Vesize>", operands[2], "%0, %1");
1952 }
1953 )
1954
1955 ;; -------------------------------------------------------------------------
1956 ;; ---- Gather prefetches
1957 ;; -------------------------------------------------------------------------
1958 ;; Includes gather forms of:
1959 ;; - PRFB
1960 ;; - PRFD
1961 ;; - PRFH
1962 ;; - PRFW
1963 ;; -------------------------------------------------------------------------
1964
1965 ;; Predicated gather prefetches for 32-bit bases and offsets. The operands
1966 ;; are:
1967 ;; 0: the governing predicate
1968 ;; 1: the scalar component of the address
1969 ;; 2: the vector component of the address
1970 ;; 3: 1 for zero extension, 0 for sign extension
1971 ;; 4: the scale multiplier
1972 ;; 5: a vector zero that identifies the mode of data being accessed
1973 ;; 6: the prefetch operator (an svprfop)
1974 ;; 7: the normal RTL prefetch rw flag
1975 ;; 8: the normal RTL prefetch locality value
1976 (define_insn "@aarch64_sve_gather_prefetch<SVE_FULL_I:mode><VNx4SI_ONLY:mode>"
1977 [(prefetch (unspec:DI
1978 [(match_operand:VNx4BI 0 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
1979 (match_operand:DI 1 "aarch64_sve_gather_offset_<SVE_FULL_I:Vesize>" "Z, vg<SVE_FULL_I:Vesize>, rk, rk, rk, rk")
1980 (match_operand:VNx4SI_ONLY 2 "register_operand" "w, w, w, w, w, w")
1981 (match_operand:DI 3 "const_int_operand" "i, i, Z, Ui1, Z, Ui1")
1982 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_FULL_I:Vesize>" "Ui1, Ui1, Ui1, Ui1, i, i")
1983 (match_operand:SVE_FULL_I 5 "aarch64_simd_imm_zero")
1984 (match_operand:DI 6 "const_int_operand")]
1985 UNSPEC_SVE_PREFETCH_GATHER)
1986 (match_operand:DI 7 "const_int_operand")
1987 (match_operand:DI 8 "const_int_operand"))]
1988 "TARGET_SVE"
1989 {
1990 static const char *const insns[][2] = {
1991 "prf<SVE_FULL_I:Vesize>", "%0, [%2.s]",
1992 "prf<SVE_FULL_I:Vesize>", "%0, [%2.s, #%1]",
1993 "prfb", "%0, [%1, %2.s, sxtw]",
1994 "prfb", "%0, [%1, %2.s, uxtw]",
1995 "prf<SVE_FULL_I:Vesize>", "%0, [%1, %2.s, sxtw %p4]",
1996 "prf<SVE_FULL_I:Vesize>", "%0, [%1, %2.s, uxtw %p4]"
1997 };
1998 const char *const *parts = insns[which_alternative];
1999 return aarch64_output_sve_prefetch (parts[0], operands[6], parts[1]);
2000 }
2001 )
2002
2003 ;; Predicated gather prefetches for 64-bit elements. The value of operand 3
2004 ;; doesn't matter in this case.
2005 (define_insn "@aarch64_sve_gather_prefetch<SVE_FULL_I:mode><VNx2DI_ONLY:mode>"
2006 [(prefetch (unspec:DI
2007 [(match_operand:VNx2BI 0 "register_operand" "Upl, Upl, Upl, Upl")
2008 (match_operand:DI 1 "aarch64_sve_gather_offset_<SVE_FULL_I:Vesize>" "Z, vg<SVE_FULL_I:Vesize>, rk, rk")
2009 (match_operand:VNx2DI_ONLY 2 "register_operand" "w, w, w, w")
2010 (match_operand:DI 3 "const_int_operand")
2011 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_FULL_I:Vesize>" "Ui1, Ui1, Ui1, i")
2012 (match_operand:SVE_FULL_I 5 "aarch64_simd_imm_zero")
2013 (match_operand:DI 6 "const_int_operand")]
2014 UNSPEC_SVE_PREFETCH_GATHER)
2015 (match_operand:DI 7 "const_int_operand")
2016 (match_operand:DI 8 "const_int_operand"))]
2017 "TARGET_SVE"
2018 {
2019 static const char *const insns[][2] = {
2020 "prf<SVE_FULL_I:Vesize>", "%0, [%2.d]",
2021 "prf<SVE_FULL_I:Vesize>", "%0, [%2.d, #%1]",
2022 "prfb", "%0, [%1, %2.d]",
2023 "prf<SVE_FULL_I:Vesize>", "%0, [%1, %2.d, lsl %p4]"
2024 };
2025 const char *const *parts = insns[which_alternative];
2026 return aarch64_output_sve_prefetch (parts[0], operands[6], parts[1]);
2027 }
2028 )
2029
2030 ;; Likewise, but with the offset being sign-extended from 32 bits.
2031 (define_insn_and_rewrite "*aarch64_sve_gather_prefetch<SVE_FULL_I:mode><VNx2DI_ONLY:mode>_sxtw"
2032 [(prefetch (unspec:DI
2033 [(match_operand:VNx2BI 0 "register_operand" "Upl, Upl")
2034 (match_operand:DI 1 "register_operand" "rk, rk")
2035 (unspec:VNx2DI_ONLY
2036 [(match_operand 9)
2037 (sign_extend:VNx2DI
2038 (truncate:VNx2SI
2039 (match_operand:VNx2DI 2 "register_operand" "w, w")))]
2040 UNSPEC_PRED_X)
2041 (match_operand:DI 3 "const_int_operand")
2042 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_FULL_I:Vesize>" "Ui1, i")
2043 (match_operand:SVE_FULL_I 5 "aarch64_simd_imm_zero")
2044 (match_operand:DI 6 "const_int_operand")]
2045 UNSPEC_SVE_PREFETCH_GATHER)
2046 (match_operand:DI 7 "const_int_operand")
2047 (match_operand:DI 8 "const_int_operand"))]
2048 "TARGET_SVE"
2049 {
2050 static const char *const insns[][2] = {
2051 "prfb", "%0, [%1, %2.d, sxtw]",
2052 "prf<SVE_FULL_I:Vesize>", "%0, [%1, %2.d, sxtw %p4]"
2053 };
2054 const char *const *parts = insns[which_alternative];
2055 return aarch64_output_sve_prefetch (parts[0], operands[6], parts[1]);
2056 }
2057 "&& !rtx_equal_p (operands[0], operands[9])"
2058 {
2059 operands[9] = copy_rtx (operands[0]);
2060 }
2061 )
2062
2063 ;; Likewise, but with the offset being zero-extended from 32 bits.
2064 (define_insn "*aarch64_sve_gather_prefetch<SVE_FULL_I:mode><VNx2DI_ONLY:mode>_uxtw"
2065 [(prefetch (unspec:DI
2066 [(match_operand:VNx2BI 0 "register_operand" "Upl, Upl")
2067 (match_operand:DI 1 "register_operand" "rk, rk")
2068 (and:VNx2DI_ONLY
2069 (match_operand:VNx2DI 2 "register_operand" "w, w")
2070 (match_operand:VNx2DI 9 "aarch64_sve_uxtw_immediate"))
2071 (match_operand:DI 3 "const_int_operand")
2072 (match_operand:DI 4 "aarch64_gather_scale_operand_<SVE_FULL_I:Vesize>" "Ui1, i")
2073 (match_operand:SVE_FULL_I 5 "aarch64_simd_imm_zero")
2074 (match_operand:DI 6 "const_int_operand")]
2075 UNSPEC_SVE_PREFETCH_GATHER)
2076 (match_operand:DI 7 "const_int_operand")
2077 (match_operand:DI 8 "const_int_operand"))]
2078 "TARGET_SVE"
2079 {
2080 static const char *const insns[][2] = {
2081 "prfb", "%0, [%1, %2.d, uxtw]",
2082 "prf<SVE_FULL_I:Vesize>", "%0, [%1, %2.d, uxtw %p4]"
2083 };
2084 const char *const *parts = insns[which_alternative];
2085 return aarch64_output_sve_prefetch (parts[0], operands[6], parts[1]);
2086 }
2087 )
2088
2089 ;; =========================================================================
2090 ;; == Stores
2091 ;; =========================================================================
2092
2093 ;; -------------------------------------------------------------------------
2094 ;; ---- Normal contiguous stores
2095 ;; -------------------------------------------------------------------------
2096 ;; Includes contiguous forms of:
2097 ;; - ST1B
2098 ;; - ST1D
2099 ;; - ST1H
2100 ;; - ST1W
2101 ;; - ST2B
2102 ;; - ST2D
2103 ;; - ST2H
2104 ;; - ST2W
2105 ;; - ST3B
2106 ;; - ST3D
2107 ;; - ST3H
2108 ;; - ST3W
2109 ;; - ST4B
2110 ;; - ST4D
2111 ;; - ST4H
2112 ;; - ST4W
2113 ;; -------------------------------------------------------------------------
2114
2115 ;; Predicated ST1.
2116 (define_insn "maskstore<mode><vpred>"
2117 [(set (match_operand:SVE_ALL 0 "memory_operand" "+m")
2118 (unspec:SVE_ALL
2119 [(match_operand:<VPRED> 2 "register_operand" "Upl")
2120 (match_operand:SVE_ALL 1 "register_operand" "w")
2121 (match_dup 0)]
2122 UNSPEC_ST1_SVE))]
2123 "TARGET_SVE"
2124 "st1<Vesize>\t%1.<Vctype>, %2, %0"
2125 )
2126
2127 ;; Unpredicated ST[234]. This is always a full update, so the dependence
2128 ;; on the old value of the memory location (via (match_dup 0)) is redundant.
2129 ;; There doesn't seem to be any obvious benefit to treating the all-true
2130 ;; case differently though. In particular, it's very unlikely that we'll
2131 ;; only find out during RTL that a store_lanes is dead.
2132 (define_expand "vec_store_lanes<mode><vsingle>"
2133 [(set (match_operand:SVE_STRUCT 0 "memory_operand")
2134 (unspec:SVE_STRUCT
2135 [(match_dup 2)
2136 (match_operand:SVE_STRUCT 1 "register_operand")
2137 (match_dup 0)]
2138 UNSPEC_STN))]
2139 "TARGET_SVE"
2140 {
2141 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
2142 }
2143 )
2144
2145 ;; Predicated ST[234].
2146 (define_insn "vec_mask_store_lanes<mode><vsingle>"
2147 [(set (match_operand:SVE_STRUCT 0 "memory_operand" "+m")
2148 (unspec:SVE_STRUCT
2149 [(match_operand:<VPRED> 2 "register_operand" "Upl")
2150 (match_operand:SVE_STRUCT 1 "register_operand" "w")
2151 (match_dup 0)]
2152 UNSPEC_STN))]
2153 "TARGET_SVE"
2154 "st<vector_count><Vesize>\t%1, %2, %0"
2155 )
2156
2157 ;; -------------------------------------------------------------------------
2158 ;; ---- Truncating contiguous stores
2159 ;; -------------------------------------------------------------------------
2160 ;; Includes:
2161 ;; - ST1B
2162 ;; - ST1H
2163 ;; - ST1W
2164 ;; -------------------------------------------------------------------------
2165
2166 ;; Predicated truncate and store, with 8 elements per 128-bit block.
2167 (define_insn "@aarch64_store_trunc<VNx8_NARROW:mode><VNx8_WIDE:mode>"
2168 [(set (match_operand:VNx8_NARROW 0 "memory_operand" "+m")
2169 (unspec:VNx8_NARROW
2170 [(match_operand:VNx8BI 2 "register_operand" "Upl")
2171 (truncate:VNx8_NARROW
2172 (match_operand:VNx8_WIDE 1 "register_operand" "w"))
2173 (match_dup 0)]
2174 UNSPEC_ST1_SVE))]
2175 "TARGET_SVE"
2176 "st1<VNx8_NARROW:Vesize>\t%1.<VNx8_WIDE:Vetype>, %2, %0"
2177 )
2178
2179 ;; Predicated truncate and store, with 4 elements per 128-bit block.
2180 (define_insn "@aarch64_store_trunc<VNx4_NARROW:mode><VNx4_WIDE:mode>"
2181 [(set (match_operand:VNx4_NARROW 0 "memory_operand" "+m")
2182 (unspec:VNx4_NARROW
2183 [(match_operand:VNx4BI 2 "register_operand" "Upl")
2184 (truncate:VNx4_NARROW
2185 (match_operand:VNx4_WIDE 1 "register_operand" "w"))
2186 (match_dup 0)]
2187 UNSPEC_ST1_SVE))]
2188 "TARGET_SVE"
2189 "st1<VNx4_NARROW:Vesize>\t%1.<VNx4_WIDE:Vetype>, %2, %0"
2190 )
2191
2192 ;; Predicated truncate and store, with 2 elements per 128-bit block.
2193 (define_insn "@aarch64_store_trunc<VNx2_NARROW:mode><VNx2_WIDE:mode>"
2194 [(set (match_operand:VNx2_NARROW 0 "memory_operand" "+m")
2195 (unspec:VNx2_NARROW
2196 [(match_operand:VNx2BI 2 "register_operand" "Upl")
2197 (truncate:VNx2_NARROW
2198 (match_operand:VNx2_WIDE 1 "register_operand" "w"))
2199 (match_dup 0)]
2200 UNSPEC_ST1_SVE))]
2201 "TARGET_SVE"
2202 "st1<VNx2_NARROW:Vesize>\t%1.<VNx2_WIDE:Vetype>, %2, %0"
2203 )
2204
2205 ;; -------------------------------------------------------------------------
2206 ;; ---- Non-temporal contiguous stores
2207 ;; -------------------------------------------------------------------------
2208 ;; Includes:
2209 ;; - STNT1B
2210 ;; - STNT1D
2211 ;; - STNT1H
2212 ;; - STNT1W
2213 ;; -------------------------------------------------------------------------
2214
2215 (define_insn "@aarch64_stnt1<mode>"
2216 [(set (match_operand:SVE_FULL 0 "memory_operand" "+m")
2217 (unspec:SVE_FULL
2218 [(match_operand:<VPRED> 2 "register_operand" "Upl")
2219 (match_operand:SVE_FULL 1 "register_operand" "w")
2220 (match_dup 0)]
2221 UNSPEC_STNT1_SVE))]
2222 "TARGET_SVE"
2223 "stnt1<Vesize>\t%1.<Vetype>, %2, %0"
2224 )
2225
2226 ;; -------------------------------------------------------------------------
2227 ;; ---- Normal scatter stores
2228 ;; -------------------------------------------------------------------------
2229 ;; Includes scatter forms of:
2230 ;; - ST1D
2231 ;; - ST1W
2232 ;; -------------------------------------------------------------------------
2233
2234 ;; Unpredicated scatter stores.
2235 (define_expand "scatter_store<mode><v_int_container>"
2236 [(set (mem:BLK (scratch))
2237 (unspec:BLK
2238 [(match_dup 5)
2239 (match_operand:DI 0 "aarch64_sve_gather_offset_<Vesize>")
2240 (match_operand:<V_INT_CONTAINER> 1 "register_operand")
2241 (match_operand:DI 2 "const_int_operand")
2242 (match_operand:DI 3 "aarch64_gather_scale_operand_<Vesize>")
2243 (match_operand:SVE_24 4 "register_operand")]
2244 UNSPEC_ST1_SCATTER))]
2245 "TARGET_SVE"
2246 {
2247 operands[5] = aarch64_ptrue_reg (<VPRED>mode);
2248 }
2249 )
2250
2251 ;; Predicated scatter stores for 32-bit elements. Operand 2 is true for
2252 ;; unsigned extension and false for signed extension.
2253 (define_insn "mask_scatter_store<mode><v_int_container>"
2254 [(set (mem:BLK (scratch))
2255 (unspec:BLK
2256 [(match_operand:VNx4BI 5 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
2257 (match_operand:DI 0 "aarch64_sve_gather_offset_<Vesize>" "Z, vgw, rk, rk, rk, rk")
2258 (match_operand:VNx4SI 1 "register_operand" "w, w, w, w, w, w")
2259 (match_operand:DI 2 "const_int_operand" "Ui1, Ui1, Z, Ui1, Z, Ui1")
2260 (match_operand:DI 3 "aarch64_gather_scale_operand_<Vesize>" "Ui1, Ui1, Ui1, Ui1, i, i")
2261 (match_operand:SVE_4 4 "register_operand" "w, w, w, w, w, w")]
2262 UNSPEC_ST1_SCATTER))]
2263 "TARGET_SVE"
2264 "@
2265 st1<Vesize>\t%4.s, %5, [%1.s]
2266 st1<Vesize>\t%4.s, %5, [%1.s, #%0]
2267 st1<Vesize>\t%4.s, %5, [%0, %1.s, sxtw]
2268 st1<Vesize>\t%4.s, %5, [%0, %1.s, uxtw]
2269 st1<Vesize>\t%4.s, %5, [%0, %1.s, sxtw %p3]
2270 st1<Vesize>\t%4.s, %5, [%0, %1.s, uxtw %p3]"
2271 )
2272
2273 ;; Predicated scatter stores for 64-bit elements. The value of operand 2
2274 ;; doesn't matter in this case.
2275 (define_insn "mask_scatter_store<mode><v_int_container>"
2276 [(set (mem:BLK (scratch))
2277 (unspec:BLK
2278 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl, Upl, Upl")
2279 (match_operand:DI 0 "aarch64_sve_gather_offset_<Vesize>" "Z, vgd, rk, rk")
2280 (match_operand:VNx2DI 1 "register_operand" "w, w, w, w")
2281 (match_operand:DI 2 "const_int_operand")
2282 (match_operand:DI 3 "aarch64_gather_scale_operand_<Vesize>" "Ui1, Ui1, Ui1, i")
2283 (match_operand:SVE_2 4 "register_operand" "w, w, w, w")]
2284 UNSPEC_ST1_SCATTER))]
2285 "TARGET_SVE"
2286 "@
2287 st1<Vesize>\t%4.d, %5, [%1.d]
2288 st1<Vesize>\t%4.d, %5, [%1.d, #%0]
2289 st1<Vesize>\t%4.d, %5, [%0, %1.d]
2290 st1<Vesize>\t%4.d, %5, [%0, %1.d, lsl %p3]"
2291 )
2292
2293 ;; Likewise, but with the offset being extended from 32 bits.
2294 (define_insn_and_rewrite "*mask_scatter_store<mode><v_int_container>_<su>xtw_unpacked"
2295 [(set (mem:BLK (scratch))
2296 (unspec:BLK
2297 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
2298 (match_operand:DI 0 "register_operand" "rk, rk")
2299 (unspec:VNx2DI
2300 [(match_operand 6)
2301 (ANY_EXTEND:VNx2DI
2302 (match_operand:VNx2SI 1 "register_operand" "w, w"))]
2303 UNSPEC_PRED_X)
2304 (match_operand:DI 2 "const_int_operand")
2305 (match_operand:DI 3 "aarch64_gather_scale_operand_<Vesize>" "Ui1, i")
2306 (match_operand:SVE_2 4 "register_operand" "w, w")]
2307 UNSPEC_ST1_SCATTER))]
2308 "TARGET_SVE"
2309 "@
2310 st1<Vesize>\t%4.d, %5, [%0, %1.d, <su>xtw]
2311 st1<Vesize>\t%4.d, %5, [%0, %1.d, <su>xtw %p3]"
2312 "&& !CONSTANT_P (operands[6])"
2313 {
2314 operands[6] = CONSTM1_RTX (<VPRED>mode);
2315 }
2316 )
2317
2318 ;; Likewise, but with the offset being truncated to 32 bits and then
2319 ;; sign-extended.
2320 (define_insn_and_rewrite "*mask_scatter_store<mode><v_int_container>_sxtw"
2321 [(set (mem:BLK (scratch))
2322 (unspec:BLK
2323 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
2324 (match_operand:DI 0 "register_operand" "rk, rk")
2325 (unspec:VNx2DI
2326 [(match_operand 6)
2327 (sign_extend:VNx2DI
2328 (truncate:VNx2SI
2329 (match_operand:VNx2DI 1 "register_operand" "w, w")))]
2330 UNSPEC_PRED_X)
2331 (match_operand:DI 2 "const_int_operand")
2332 (match_operand:DI 3 "aarch64_gather_scale_operand_<Vesize>" "Ui1, i")
2333 (match_operand:SVE_2 4 "register_operand" "w, w")]
2334 UNSPEC_ST1_SCATTER))]
2335 "TARGET_SVE"
2336 "@
2337 st1<Vesize>\t%4.d, %5, [%0, %1.d, sxtw]
2338 st1<Vesize>\t%4.d, %5, [%0, %1.d, sxtw %p3]"
2339 "&& !CONSTANT_P (operands[6])"
2340 {
2341 operands[6] = CONSTM1_RTX (<VPRED>mode);
2342 }
2343 )
2344
2345 ;; Likewise, but with the offset being truncated to 32 bits and then
2346 ;; zero-extended.
2347 (define_insn "*mask_scatter_store<mode><v_int_container>_uxtw"
2348 [(set (mem:BLK (scratch))
2349 (unspec:BLK
2350 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
2351 (match_operand:DI 0 "aarch64_reg_or_zero" "rk, rk")
2352 (and:VNx2DI
2353 (match_operand:VNx2DI 1 "register_operand" "w, w")
2354 (match_operand:VNx2DI 6 "aarch64_sve_uxtw_immediate"))
2355 (match_operand:DI 2 "const_int_operand")
2356 (match_operand:DI 3 "aarch64_gather_scale_operand_<Vesize>" "Ui1, i")
2357 (match_operand:SVE_2 4 "register_operand" "w, w")]
2358 UNSPEC_ST1_SCATTER))]
2359 "TARGET_SVE"
2360 "@
2361 st1<Vesize>\t%4.d, %5, [%0, %1.d, uxtw]
2362 st1<Vesize>\t%4.d, %5, [%0, %1.d, uxtw %p3]"
2363 )
2364
2365 ;; -------------------------------------------------------------------------
2366 ;; ---- Truncating scatter stores
2367 ;; -------------------------------------------------------------------------
2368 ;; Includes scatter forms of:
2369 ;; - ST1B
2370 ;; - ST1H
2371 ;; - ST1W
2372 ;; -------------------------------------------------------------------------
2373
2374 ;; Predicated truncating scatter stores for 32-bit elements. Operand 2 is
2375 ;; true for unsigned extension and false for signed extension.
2376 (define_insn "@aarch64_scatter_store_trunc<VNx4_NARROW:mode><VNx4_WIDE:mode>"
2377 [(set (mem:BLK (scratch))
2378 (unspec:BLK
2379 [(match_operand:VNx4BI 5 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
2380 (match_operand:DI 0 "aarch64_sve_gather_offset_<VNx4_NARROW:Vesize>" "Z, vg<VNx4_NARROW:Vesize>, rk, rk, rk, rk")
2381 (match_operand:VNx4SI 1 "register_operand" "w, w, w, w, w, w")
2382 (match_operand:DI 2 "const_int_operand" "Ui1, Ui1, Z, Ui1, Z, Ui1")
2383 (match_operand:DI 3 "aarch64_gather_scale_operand_<VNx4_NARROW:Vesize>" "Ui1, Ui1, Ui1, Ui1, i, i")
2384 (truncate:VNx4_NARROW
2385 (match_operand:VNx4_WIDE 4 "register_operand" "w, w, w, w, w, w"))]
2386 UNSPEC_ST1_SCATTER))]
2387 "TARGET_SVE"
2388 "@
2389 st1<VNx4_NARROW:Vesize>\t%4.s, %5, [%1.s]
2390 st1<VNx4_NARROW:Vesize>\t%4.s, %5, [%1.s, #%0]
2391 st1<VNx4_NARROW:Vesize>\t%4.s, %5, [%0, %1.s, sxtw]
2392 st1<VNx4_NARROW:Vesize>\t%4.s, %5, [%0, %1.s, uxtw]
2393 st1<VNx4_NARROW:Vesize>\t%4.s, %5, [%0, %1.s, sxtw %p3]
2394 st1<VNx4_NARROW:Vesize>\t%4.s, %5, [%0, %1.s, uxtw %p3]"
2395 )
2396
2397 ;; Predicated truncating scatter stores for 64-bit elements. The value of
2398 ;; operand 2 doesn't matter in this case.
2399 (define_insn "@aarch64_scatter_store_trunc<VNx2_NARROW:mode><VNx2_WIDE:mode>"
2400 [(set (mem:BLK (scratch))
2401 (unspec:BLK
2402 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl, Upl, Upl")
2403 (match_operand:DI 0 "aarch64_sve_gather_offset_<VNx2_NARROW:Vesize>" "Z, vg<VNx2_NARROW:Vesize>, rk, rk")
2404 (match_operand:VNx2DI 1 "register_operand" "w, w, w, w")
2405 (match_operand:DI 2 "const_int_operand")
2406 (match_operand:DI 3 "aarch64_gather_scale_operand_<VNx2_NARROW:Vesize>" "Ui1, Ui1, Ui1, i")
2407 (truncate:VNx2_NARROW
2408 (match_operand:VNx2_WIDE 4 "register_operand" "w, w, w, w"))]
2409 UNSPEC_ST1_SCATTER))]
2410 "TARGET_SVE"
2411 "@
2412 st1<VNx2_NARROW:Vesize>\t%4.d, %5, [%1.d]
2413 st1<VNx2_NARROW:Vesize>\t%4.d, %5, [%1.d, #%0]
2414 st1<VNx2_NARROW:Vesize>\t%4.d, %5, [%0, %1.d]
2415 st1<VNx2_NARROW:Vesize>\t%4.d, %5, [%0, %1.d, lsl %p3]"
2416 )
2417
2418 ;; Likewise, but with the offset being sign-extended from 32 bits.
2419 (define_insn_and_rewrite "*aarch64_scatter_store_trunc<VNx2_NARROW:mode><VNx2_WIDE:mode>_sxtw"
2420 [(set (mem:BLK (scratch))
2421 (unspec:BLK
2422 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
2423 (match_operand:DI 0 "register_operand" "rk, rk")
2424 (unspec:VNx2DI
2425 [(match_operand 6)
2426 (sign_extend:VNx2DI
2427 (truncate:VNx2SI
2428 (match_operand:VNx2DI 1 "register_operand" "w, w")))]
2429 UNSPEC_PRED_X)
2430 (match_operand:DI 2 "const_int_operand")
2431 (match_operand:DI 3 "aarch64_gather_scale_operand_<VNx2_NARROW:Vesize>" "Ui1, i")
2432 (truncate:VNx2_NARROW
2433 (match_operand:VNx2_WIDE 4 "register_operand" "w, w"))]
2434 UNSPEC_ST1_SCATTER))]
2435 "TARGET_SVE"
2436 "@
2437 st1<VNx2_NARROW:Vesize>\t%4.d, %5, [%0, %1.d, sxtw]
2438 st1<VNx2_NARROW:Vesize>\t%4.d, %5, [%0, %1.d, sxtw %p3]"
2439 "&& !rtx_equal_p (operands[5], operands[6])"
2440 {
2441 operands[6] = copy_rtx (operands[5]);
2442 }
2443 )
2444
2445 ;; Likewise, but with the offset being zero-extended from 32 bits.
2446 (define_insn "*aarch64_scatter_store_trunc<VNx2_NARROW:mode><VNx2_WIDE:mode>_uxtw"
2447 [(set (mem:BLK (scratch))
2448 (unspec:BLK
2449 [(match_operand:VNx2BI 5 "register_operand" "Upl, Upl")
2450 (match_operand:DI 0 "aarch64_reg_or_zero" "rk, rk")
2451 (and:VNx2DI
2452 (match_operand:VNx2DI 1 "register_operand" "w, w")
2453 (match_operand:VNx2DI 6 "aarch64_sve_uxtw_immediate"))
2454 (match_operand:DI 2 "const_int_operand")
2455 (match_operand:DI 3 "aarch64_gather_scale_operand_<VNx2_NARROW:Vesize>" "Ui1, i")
2456 (truncate:VNx2_NARROW
2457 (match_operand:VNx2_WIDE 4 "register_operand" "w, w"))]
2458 UNSPEC_ST1_SCATTER))]
2459 "TARGET_SVE"
2460 "@
2461 st1<VNx2_NARROW:Vesize>\t%4.d, %5, [%0, %1.d, uxtw]
2462 st1<VNx2_NARROW:Vesize>\t%4.d, %5, [%0, %1.d, uxtw %p3]"
2463 )
2464
2465 ;; =========================================================================
2466 ;; == Vector creation
2467 ;; =========================================================================
2468
2469 ;; -------------------------------------------------------------------------
2470 ;; ---- [INT,FP] Duplicate element
2471 ;; -------------------------------------------------------------------------
2472 ;; Includes:
2473 ;; - DUP
2474 ;; - MOV
2475 ;; - LD1RB
2476 ;; - LD1RD
2477 ;; - LD1RH
2478 ;; - LD1RW
2479 ;; - LD1ROB (F64MM)
2480 ;; - LD1ROD (F64MM)
2481 ;; - LD1ROH (F64MM)
2482 ;; - LD1ROW (F64MM)
2483 ;; - LD1RQB
2484 ;; - LD1RQD
2485 ;; - LD1RQH
2486 ;; - LD1RQW
2487 ;; -------------------------------------------------------------------------
2488
2489 (define_expand "vec_duplicate<mode>"
2490 [(parallel
2491 [(set (match_operand:SVE_ALL 0 "register_operand")
2492 (vec_duplicate:SVE_ALL
2493 (match_operand:<VEL> 1 "aarch64_sve_dup_operand")))
2494 (clobber (scratch:VNx16BI))])]
2495 "TARGET_SVE"
2496 {
2497 if (MEM_P (operands[1]))
2498 {
2499 rtx ptrue = aarch64_ptrue_reg (<VPRED>mode);
2500 emit_insn (gen_sve_ld1r<mode> (operands[0], ptrue, operands[1],
2501 CONST0_RTX (<MODE>mode)));
2502 DONE;
2503 }
2504 }
2505 )
2506
2507 ;; Accept memory operands for the benefit of combine, and also in case
2508 ;; the scalar input gets spilled to memory during RA. We want to split
2509 ;; the load at the first opportunity in order to allow the PTRUE to be
2510 ;; optimized with surrounding code.
2511 (define_insn_and_split "*vec_duplicate<mode>_reg"
2512 [(set (match_operand:SVE_ALL 0 "register_operand" "=w, w, w")
2513 (vec_duplicate:SVE_ALL
2514 (match_operand:<VEL> 1 "aarch64_sve_dup_operand" "r, w, Uty")))
2515 (clobber (match_scratch:VNx16BI 2 "=X, X, Upl"))]
2516 "TARGET_SVE"
2517 "@
2518 mov\t%0.<Vetype>, %<vwcore>1
2519 mov\t%0.<Vetype>, %<Vetype>1
2520 #"
2521 "&& MEM_P (operands[1])"
2522 [(const_int 0)]
2523 {
2524 if (GET_CODE (operands[2]) == SCRATCH)
2525 operands[2] = gen_reg_rtx (VNx16BImode);
2526 emit_move_insn (operands[2], CONSTM1_RTX (VNx16BImode));
2527 rtx gp = gen_lowpart (<VPRED>mode, operands[2]);
2528 emit_insn (gen_sve_ld1r<mode> (operands[0], gp, operands[1],
2529 CONST0_RTX (<MODE>mode)));
2530 DONE;
2531 }
2532 [(set_attr "length" "4,4,8")]
2533 )
2534
2535 ;; Duplicate an Advanced SIMD vector to fill an SVE vector (LE version).
2536 (define_insn "@aarch64_vec_duplicate_vq<mode>_le"
2537 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
2538 (vec_duplicate:SVE_FULL
2539 (match_operand:<V128> 1 "register_operand" "w")))]
2540 "TARGET_SVE && !BYTES_BIG_ENDIAN"
2541 {
2542 operands[1] = gen_rtx_REG (<MODE>mode, REGNO (operands[1]));
2543 return "dup\t%0.q, %1.q[0]";
2544 }
2545 )
2546
2547 ;; Duplicate an Advanced SIMD vector to fill an SVE vector (BE version).
2548 ;; The SVE register layout puts memory lane N into (architectural)
2549 ;; register lane N, whereas the Advanced SIMD layout puts the memory
2550 ;; lsb into the register lsb. We therefore have to describe this in rtl
2551 ;; terms as a reverse of the V128 vector followed by a duplicate.
2552 (define_insn "@aarch64_vec_duplicate_vq<mode>_be"
2553 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
2554 (vec_duplicate:SVE_FULL
2555 (vec_select:<V128>
2556 (match_operand:<V128> 1 "register_operand" "w")
2557 (match_operand 2 "descending_int_parallel"))))]
2558 "TARGET_SVE
2559 && BYTES_BIG_ENDIAN
2560 && known_eq (INTVAL (XVECEXP (operands[2], 0, 0)),
2561 GET_MODE_NUNITS (<V128>mode) - 1)"
2562 {
2563 operands[1] = gen_rtx_REG (<MODE>mode, REGNO (operands[1]));
2564 return "dup\t%0.q, %1.q[0]";
2565 }
2566 )
2567
2568 ;; This is used for vec_duplicate<mode>s from memory, but can also
2569 ;; be used by combine to optimize selects of a vec_duplicate<mode>
2570 ;; with zero.
2571 (define_insn "sve_ld1r<mode>"
2572 [(set (match_operand:SVE_ALL 0 "register_operand" "=w")
2573 (unspec:SVE_ALL
2574 [(match_operand:<VPRED> 1 "register_operand" "Upl")
2575 (vec_duplicate:SVE_ALL
2576 (match_operand:<VEL> 2 "aarch64_sve_ld1r_operand" "Uty"))
2577 (match_operand:SVE_ALL 3 "aarch64_simd_imm_zero")]
2578 UNSPEC_SEL))]
2579 "TARGET_SVE"
2580 "ld1r<Vesize>\t%0.<Vetype>, %1/z, %2"
2581 )
2582
2583 ;; Load 128 bits from memory under predicate control and duplicate to
2584 ;; fill a vector.
2585 (define_insn "@aarch64_sve_ld1rq<mode>"
2586 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
2587 (unspec:SVE_FULL
2588 [(match_operand:<VPRED> 2 "register_operand" "Upl")
2589 (match_operand:<V128> 1 "aarch64_sve_ld1rq_operand" "UtQ")]
2590 UNSPEC_LD1RQ))]
2591 "TARGET_SVE"
2592 {
2593 operands[1] = gen_rtx_MEM (<VEL>mode, XEXP (operands[1], 0));
2594 return "ld1rq<Vesize>\t%0.<Vetype>, %2/z, %1";
2595 }
2596 )
2597
2598 (define_insn "@aarch64_sve_ld1ro<mode>"
2599 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
2600 (unspec:SVE_FULL
2601 [(match_operand:<VPRED> 2 "register_operand" "Upl")
2602 (match_operand:OI 1 "aarch64_sve_ld1ro_operand_<Vesize>"
2603 "UO<Vesize>")]
2604 UNSPEC_LD1RO))]
2605 "TARGET_SVE_F64MM"
2606 {
2607 operands[1] = gen_rtx_MEM (<VEL>mode, XEXP (operands[1], 0));
2608 return "ld1ro<Vesize>\t%0.<Vetype>, %2/z, %1";
2609 }
2610 )
2611
2612 ;; -------------------------------------------------------------------------
2613 ;; ---- [INT,FP] Initialize from individual elements
2614 ;; -------------------------------------------------------------------------
2615 ;; Includes:
2616 ;; - INSR
2617 ;; -------------------------------------------------------------------------
2618
2619 (define_expand "vec_init<mode><Vel>"
2620 [(match_operand:SVE_FULL 0 "register_operand")
2621 (match_operand 1 "")]
2622 "TARGET_SVE"
2623 {
2624 aarch64_sve_expand_vector_init (operands[0], operands[1]);
2625 DONE;
2626 }
2627 )
2628
2629 ;; Shift an SVE vector left and insert a scalar into element 0.
2630 (define_insn "vec_shl_insert_<mode>"
2631 [(set (match_operand:SVE_FULL 0 "register_operand" "=?w, w, ??&w, ?&w")
2632 (unspec:SVE_FULL
2633 [(match_operand:SVE_FULL 1 "register_operand" "0, 0, w, w")
2634 (match_operand:<VEL> 2 "aarch64_reg_or_zero" "rZ, w, rZ, w")]
2635 UNSPEC_INSR))]
2636 "TARGET_SVE"
2637 "@
2638 insr\t%0.<Vetype>, %<vwcore>2
2639 insr\t%0.<Vetype>, %<Vetype>2
2640 movprfx\t%0, %1\;insr\t%0.<Vetype>, %<vwcore>2
2641 movprfx\t%0, %1\;insr\t%0.<Vetype>, %<Vetype>2"
2642 [(set_attr "movprfx" "*,*,yes,yes")]
2643 )
2644
2645 ;; -------------------------------------------------------------------------
2646 ;; ---- [INT] Linear series
2647 ;; -------------------------------------------------------------------------
2648 ;; Includes:
2649 ;; - INDEX
2650 ;; -------------------------------------------------------------------------
2651
2652 (define_insn "vec_series<mode>"
2653 [(set (match_operand:SVE_I 0 "register_operand" "=w, w, w")
2654 (vec_series:SVE_I
2655 (match_operand:<VEL> 1 "aarch64_sve_index_operand" "Usi, r, r")
2656 (match_operand:<VEL> 2 "aarch64_sve_index_operand" "r, Usi, r")))]
2657 "TARGET_SVE"
2658 "@
2659 index\t%0.<Vctype>, #%1, %<vccore>2
2660 index\t%0.<Vctype>, %<vccore>1, #%2
2661 index\t%0.<Vctype>, %<vccore>1, %<vccore>2"
2662 )
2663
2664 ;; Optimize {x, x, x, x, ...} + {0, n, 2*n, 3*n, ...} if n is in range
2665 ;; of an INDEX instruction.
2666 (define_insn "*vec_series<mode>_plus"
2667 [(set (match_operand:SVE_I 0 "register_operand" "=w")
2668 (plus:SVE_I
2669 (vec_duplicate:SVE_I
2670 (match_operand:<VEL> 1 "register_operand" "r"))
2671 (match_operand:SVE_I 2 "immediate_operand")))]
2672 "TARGET_SVE && aarch64_check_zero_based_sve_index_immediate (operands[2])"
2673 {
2674 operands[2] = aarch64_check_zero_based_sve_index_immediate (operands[2]);
2675 return "index\t%0.<Vctype>, %<vccore>1, #%2";
2676 }
2677 )
2678
2679 ;; -------------------------------------------------------------------------
2680 ;; ---- [PRED] Duplicate element
2681 ;; -------------------------------------------------------------------------
2682 ;; The patterns in this section are synthetic.
2683 ;; -------------------------------------------------------------------------
2684
2685 ;; Implement a predicate broadcast by shifting the low bit of the scalar
2686 ;; input into the top bit and using a WHILELO. An alternative would be to
2687 ;; duplicate the input and do a compare with zero.
2688 (define_expand "vec_duplicate<mode>"
2689 [(set (match_operand:PRED_ALL 0 "register_operand")
2690 (vec_duplicate:PRED_ALL (match_operand:QI 1 "register_operand")))]
2691 "TARGET_SVE"
2692 {
2693 rtx tmp = gen_reg_rtx (DImode);
2694 rtx op1 = gen_lowpart (DImode, operands[1]);
2695 emit_insn (gen_ashldi3 (tmp, op1, gen_int_mode (63, DImode)));
2696 emit_insn (gen_while_ultdi<mode> (operands[0], const0_rtx, tmp));
2697 DONE;
2698 }
2699 )
2700
2701 ;; =========================================================================
2702 ;; == Vector decomposition
2703 ;; =========================================================================
2704
2705 ;; -------------------------------------------------------------------------
2706 ;; ---- [INT,FP] Extract index
2707 ;; -------------------------------------------------------------------------
2708 ;; Includes:
2709 ;; - DUP (Advanced SIMD)
2710 ;; - DUP (SVE)
2711 ;; - EXT (SVE)
2712 ;; - ST1 (Advanced SIMD)
2713 ;; - UMOV (Advanced SIMD)
2714 ;; -------------------------------------------------------------------------
2715
2716 (define_expand "vec_extract<mode><Vel>"
2717 [(set (match_operand:<VEL> 0 "register_operand")
2718 (vec_select:<VEL>
2719 (match_operand:SVE_FULL 1 "register_operand")
2720 (parallel [(match_operand:SI 2 "nonmemory_operand")])))]
2721 "TARGET_SVE"
2722 {
2723 poly_int64 val;
2724 if (poly_int_rtx_p (operands[2], &val)
2725 && known_eq (val, GET_MODE_NUNITS (<MODE>mode) - 1))
2726 {
2727 /* The last element can be extracted with a LASTB and a false
2728 predicate. */
2729 rtx sel = aarch64_pfalse_reg (<VPRED>mode);
2730 emit_insn (gen_extract_last_<mode> (operands[0], sel, operands[1]));
2731 DONE;
2732 }
2733 if (!CONST_INT_P (operands[2]))
2734 {
2735 /* Create an index with operand[2] as the base and -1 as the step.
2736 It will then be zero for the element we care about. */
2737 rtx index = gen_lowpart (<VEL_INT>mode, operands[2]);
2738 index = force_reg (<VEL_INT>mode, index);
2739 rtx series = gen_reg_rtx (<V_INT_EQUIV>mode);
2740 emit_insn (gen_vec_series<v_int_equiv> (series, index, constm1_rtx));
2741
2742 /* Get a predicate that is true for only that element. */
2743 rtx zero = CONST0_RTX (<V_INT_EQUIV>mode);
2744 rtx cmp = gen_rtx_EQ (<V_INT_EQUIV>mode, series, zero);
2745 rtx sel = gen_reg_rtx (<VPRED>mode);
2746 emit_insn (gen_vec_cmp<v_int_equiv><vpred> (sel, cmp, series, zero));
2747
2748 /* Select the element using LASTB. */
2749 emit_insn (gen_extract_last_<mode> (operands[0], sel, operands[1]));
2750 DONE;
2751 }
2752 }
2753 )
2754
2755 ;; Extract element zero. This is a special case because we want to force
2756 ;; the registers to be the same for the second alternative, and then
2757 ;; split the instruction into nothing after RA.
2758 (define_insn_and_split "*vec_extract<mode><Vel>_0"
2759 [(set (match_operand:<VEL> 0 "aarch64_simd_nonimmediate_operand" "=r, w, Utv")
2760 (vec_select:<VEL>
2761 (match_operand:SVE_FULL 1 "register_operand" "w, 0, w")
2762 (parallel [(const_int 0)])))]
2763 "TARGET_SVE"
2764 {
2765 operands[1] = gen_rtx_REG (<V128>mode, REGNO (operands[1]));
2766 switch (which_alternative)
2767 {
2768 case 0:
2769 return "umov\\t%<vwcore>0, %1.<Vetype>[0]";
2770 case 1:
2771 return "#";
2772 case 2:
2773 return "st1\\t{%1.<Vetype>}[0], %0";
2774 default:
2775 gcc_unreachable ();
2776 }
2777 }
2778 "&& reload_completed
2779 && REG_P (operands[0])
2780 && REGNO (operands[0]) == REGNO (operands[1])"
2781 [(const_int 0)]
2782 {
2783 emit_note (NOTE_INSN_DELETED);
2784 DONE;
2785 }
2786 [(set_attr "type" "neon_to_gp_q, untyped, neon_store1_one_lane_q")]
2787 )
2788
2789 ;; Extract an element from the Advanced SIMD portion of the register.
2790 ;; We don't just reuse the aarch64-simd.md pattern because we don't
2791 ;; want any change in lane number on big-endian targets.
2792 (define_insn "*vec_extract<mode><Vel>_v128"
2793 [(set (match_operand:<VEL> 0 "aarch64_simd_nonimmediate_operand" "=r, w, Utv")
2794 (vec_select:<VEL>
2795 (match_operand:SVE_FULL 1 "register_operand" "w, w, w")
2796 (parallel [(match_operand:SI 2 "const_int_operand")])))]
2797 "TARGET_SVE
2798 && IN_RANGE (INTVAL (operands[2]) * GET_MODE_SIZE (<VEL>mode), 1, 15)"
2799 {
2800 operands[1] = gen_rtx_REG (<V128>mode, REGNO (operands[1]));
2801 switch (which_alternative)
2802 {
2803 case 0:
2804 return "umov\\t%<vwcore>0, %1.<Vetype>[%2]";
2805 case 1:
2806 return "dup\\t%<Vetype>0, %1.<Vetype>[%2]";
2807 case 2:
2808 return "st1\\t{%1.<Vetype>}[%2], %0";
2809 default:
2810 gcc_unreachable ();
2811 }
2812 }
2813 [(set_attr "type" "neon_to_gp_q, neon_dup_q, neon_store1_one_lane_q")]
2814 )
2815
2816 ;; Extract an element in the range of DUP. This pattern allows the
2817 ;; source and destination to be different.
2818 (define_insn "*vec_extract<mode><Vel>_dup"
2819 [(set (match_operand:<VEL> 0 "register_operand" "=w")
2820 (vec_select:<VEL>
2821 (match_operand:SVE_FULL 1 "register_operand" "w")
2822 (parallel [(match_operand:SI 2 "const_int_operand")])))]
2823 "TARGET_SVE
2824 && IN_RANGE (INTVAL (operands[2]) * GET_MODE_SIZE (<VEL>mode), 16, 63)"
2825 {
2826 operands[0] = gen_rtx_REG (<MODE>mode, REGNO (operands[0]));
2827 return "dup\t%0.<Vetype>, %1.<Vetype>[%2]";
2828 }
2829 )
2830
2831 ;; Extract an element outside the range of DUP. This pattern requires the
2832 ;; source and destination to be the same.
2833 (define_insn "*vec_extract<mode><Vel>_ext"
2834 [(set (match_operand:<VEL> 0 "register_operand" "=w, ?&w")
2835 (vec_select:<VEL>
2836 (match_operand:SVE_FULL 1 "register_operand" "0, w")
2837 (parallel [(match_operand:SI 2 "const_int_operand")])))]
2838 "TARGET_SVE && INTVAL (operands[2]) * GET_MODE_SIZE (<VEL>mode) >= 64"
2839 {
2840 operands[0] = gen_rtx_REG (<MODE>mode, REGNO (operands[0]));
2841 operands[2] = GEN_INT (INTVAL (operands[2]) * GET_MODE_SIZE (<VEL>mode));
2842 return (which_alternative == 0
2843 ? "ext\t%0.b, %0.b, %0.b, #%2"
2844 : "movprfx\t%0, %1\;ext\t%0.b, %0.b, %1.b, #%2");
2845 }
2846 [(set_attr "movprfx" "*,yes")]
2847 )
2848
2849 ;; -------------------------------------------------------------------------
2850 ;; ---- [INT,FP] Extract active element
2851 ;; -------------------------------------------------------------------------
2852 ;; Includes:
2853 ;; - LASTA
2854 ;; - LASTB
2855 ;; -------------------------------------------------------------------------
2856
2857 ;; Extract the last active element of operand 1 into operand 0.
2858 ;; If no elements are active, extract the last inactive element instead.
2859 (define_insn "@extract_<last_op>_<mode>"
2860 [(set (match_operand:<VEL> 0 "register_operand" "=?r, w")
2861 (unspec:<VEL>
2862 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
2863 (match_operand:SVE_FULL 2 "register_operand" "w, w")]
2864 LAST))]
2865 "TARGET_SVE"
2866 "@
2867 last<ab>\t%<vwcore>0, %1, %2.<Vetype>
2868 last<ab>\t%<Vetype>0, %1, %2.<Vetype>"
2869 )
2870
2871 ;; -------------------------------------------------------------------------
2872 ;; ---- [PRED] Extract index
2873 ;; -------------------------------------------------------------------------
2874 ;; The patterns in this section are synthetic.
2875 ;; -------------------------------------------------------------------------
2876
2877 ;; Handle extractions from a predicate by converting to an integer vector
2878 ;; and extracting from there.
2879 (define_expand "vec_extract<vpred><Vel>"
2880 [(match_operand:<VEL> 0 "register_operand")
2881 (match_operand:<VPRED> 1 "register_operand")
2882 (match_operand:SI 2 "nonmemory_operand")
2883 ;; Dummy operand to which we can attach the iterator.
2884 (reg:SVE_FULL_I V0_REGNUM)]
2885 "TARGET_SVE"
2886 {
2887 rtx tmp = gen_reg_rtx (<MODE>mode);
2888 emit_insn (gen_vcond_mask_<mode><vpred> (tmp, operands[1],
2889 CONST1_RTX (<MODE>mode),
2890 CONST0_RTX (<MODE>mode)));
2891 emit_insn (gen_vec_extract<mode><Vel> (operands[0], tmp, operands[2]));
2892 DONE;
2893 }
2894 )
2895
2896 ;; =========================================================================
2897 ;; == Unary arithmetic
2898 ;; =========================================================================
2899
2900 ;; -------------------------------------------------------------------------
2901 ;; ---- [INT] General unary arithmetic corresponding to rtx codes
2902 ;; -------------------------------------------------------------------------
2903 ;; Includes:
2904 ;; - ABS
2905 ;; - CLS (= clrsb)
2906 ;; - CLZ
2907 ;; - CNT (= popcount)
2908 ;; - NEG
2909 ;; - NOT
2910 ;; -------------------------------------------------------------------------
2911
2912 ;; Unpredicated integer unary arithmetic.
2913 (define_expand "<optab><mode>2"
2914 [(set (match_operand:SVE_FULL_I 0 "register_operand")
2915 (unspec:SVE_FULL_I
2916 [(match_dup 2)
2917 (SVE_INT_UNARY:SVE_FULL_I
2918 (match_operand:SVE_FULL_I 1 "register_operand"))]
2919 UNSPEC_PRED_X))]
2920 "TARGET_SVE"
2921 {
2922 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
2923 }
2924 )
2925
2926 ;; Integer unary arithmetic predicated with a PTRUE.
2927 (define_insn "@aarch64_pred_<optab><mode>"
2928 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
2929 (unspec:SVE_FULL_I
2930 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
2931 (SVE_INT_UNARY:SVE_FULL_I
2932 (match_operand:SVE_FULL_I 2 "register_operand" "0, w"))]
2933 UNSPEC_PRED_X))]
2934 "TARGET_SVE"
2935 "@
2936 <sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
2937 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
2938 [(set_attr "movprfx" "*,yes")]
2939 )
2940
2941 ;; Predicated integer unary arithmetic with merging.
2942 (define_expand "@cond_<optab><mode>"
2943 [(set (match_operand:SVE_FULL_I 0 "register_operand")
2944 (unspec:SVE_FULL_I
2945 [(match_operand:<VPRED> 1 "register_operand")
2946 (SVE_INT_UNARY:SVE_FULL_I
2947 (match_operand:SVE_FULL_I 2 "register_operand"))
2948 (match_operand:SVE_FULL_I 3 "aarch64_simd_reg_or_zero")]
2949 UNSPEC_SEL))]
2950 "TARGET_SVE"
2951 )
2952
2953 ;; Predicated integer unary arithmetic, merging with the first input.
2954 (define_insn "*cond_<optab><mode>_2"
2955 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
2956 (unspec:SVE_FULL_I
2957 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
2958 (SVE_INT_UNARY:SVE_FULL_I
2959 (match_operand:SVE_FULL_I 2 "register_operand" "0, w"))
2960 (match_dup 2)]
2961 UNSPEC_SEL))]
2962 "TARGET_SVE"
2963 "@
2964 <sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>
2965 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
2966 [(set_attr "movprfx" "*,yes")]
2967 )
2968
2969 ;; Predicated integer unary arithmetic, merging with an independent value.
2970 ;;
2971 ;; The earlyclobber isn't needed for the first alternative, but omitting
2972 ;; it would only help the case in which operands 2 and 3 are the same,
2973 ;; which is handled above rather than here. Marking all the alternatives
2974 ;; as earlyclobber helps to make the instruction more regular to the
2975 ;; register allocator.
2976 (define_insn "*cond_<optab><mode>_any"
2977 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, ?&w, ?&w")
2978 (unspec:SVE_FULL_I
2979 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
2980 (SVE_INT_UNARY:SVE_FULL_I
2981 (match_operand:SVE_FULL_I 2 "register_operand" "w, w, w"))
2982 (match_operand:SVE_FULL_I 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
2983 UNSPEC_SEL))]
2984 "TARGET_SVE && !rtx_equal_p (operands[2], operands[3])"
2985 "@
2986 <sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
2987 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
2988 movprfx\t%0, %3\;<sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
2989 [(set_attr "movprfx" "*,yes,yes")]
2990 )
2991
2992 ;; -------------------------------------------------------------------------
2993 ;; ---- [INT] General unary arithmetic corresponding to unspecs
2994 ;; -------------------------------------------------------------------------
2995 ;; Includes
2996 ;; - RBIT
2997 ;; - REVB
2998 ;; - REVH
2999 ;; - REVW
3000 ;; -------------------------------------------------------------------------
3001
3002 ;; Predicated integer unary operations.
3003 (define_insn "@aarch64_pred_<optab><mode>"
3004 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
3005 (unspec:SVE_FULL_I
3006 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3007 (unspec:SVE_FULL_I
3008 [(match_operand:SVE_FULL_I 2 "register_operand" "0, w")]
3009 SVE_INT_UNARY)]
3010 UNSPEC_PRED_X))]
3011 "TARGET_SVE && <elem_bits> >= <min_elem_bits>"
3012 "@
3013 <sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
3014 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
3015 [(set_attr "movprfx" "*,yes")]
3016 )
3017
3018 ;; Predicated integer unary operations with merging.
3019 (define_insn "@cond_<optab><mode>"
3020 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w, ?&w")
3021 (unspec:SVE_FULL_I
3022 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
3023 (unspec:SVE_FULL_I
3024 [(match_operand:SVE_FULL_I 2 "register_operand" "w, w, w")]
3025 SVE_INT_UNARY)
3026 (match_operand:SVE_FULL_I 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
3027 UNSPEC_SEL))]
3028 "TARGET_SVE && <elem_bits> >= <min_elem_bits>"
3029 "@
3030 <sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
3031 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
3032 movprfx\t%0, %3\;<sve_int_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
3033 [(set_attr "movprfx" "*,yes,yes")]
3034 )
3035
3036 ;; -------------------------------------------------------------------------
3037 ;; ---- [INT] Sign and zero extension
3038 ;; -------------------------------------------------------------------------
3039 ;; Includes:
3040 ;; - SXTB
3041 ;; - SXTH
3042 ;; - SXTW
3043 ;; - UXTB
3044 ;; - UXTH
3045 ;; - UXTW
3046 ;; -------------------------------------------------------------------------
3047
3048 ;; Unpredicated sign and zero extension from a narrower mode.
3049 (define_expand "<optab><SVE_PARTIAL_I:mode><SVE_HSDI:mode>2"
3050 [(set (match_operand:SVE_HSDI 0 "register_operand")
3051 (unspec:SVE_HSDI
3052 [(match_dup 2)
3053 (ANY_EXTEND:SVE_HSDI
3054 (match_operand:SVE_PARTIAL_I 1 "register_operand"))]
3055 UNSPEC_PRED_X))]
3056 "TARGET_SVE && (~<SVE_HSDI:narrower_mask> & <SVE_PARTIAL_I:self_mask>) == 0"
3057 {
3058 operands[2] = aarch64_ptrue_reg (<SVE_HSDI:VPRED>mode);
3059 }
3060 )
3061
3062 ;; Predicated sign and zero extension from a narrower mode.
3063 (define_insn "*<optab><SVE_PARTIAL_I:mode><SVE_HSDI:mode>2"
3064 [(set (match_operand:SVE_HSDI 0 "register_operand" "=w, ?&w")
3065 (unspec:SVE_HSDI
3066 [(match_operand:<SVE_HSDI:VPRED> 1 "register_operand" "Upl, Upl")
3067 (ANY_EXTEND:SVE_HSDI
3068 (match_operand:SVE_PARTIAL_I 2 "register_operand" "0, w"))]
3069 UNSPEC_PRED_X))]
3070 "TARGET_SVE && (~<SVE_HSDI:narrower_mask> & <SVE_PARTIAL_I:self_mask>) == 0"
3071 "@
3072 <su>xt<SVE_PARTIAL_I:Vesize>\t%0.<SVE_HSDI:Vetype>, %1/m, %2.<SVE_HSDI:Vetype>
3073 movprfx\t%0, %2\;<su>xt<SVE_PARTIAL_I:Vesize>\t%0.<SVE_HSDI:Vetype>, %1/m, %2.<SVE_HSDI:Vetype>"
3074 [(set_attr "movprfx" "*,yes")]
3075 )
3076
3077 ;; Predicated truncate-and-sign-extend operations.
3078 (define_insn "@aarch64_pred_sxt<SVE_FULL_HSDI:mode><SVE_PARTIAL_I:mode>"
3079 [(set (match_operand:SVE_FULL_HSDI 0 "register_operand" "=w, ?&w")
3080 (unspec:SVE_FULL_HSDI
3081 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand" "Upl, Upl")
3082 (sign_extend:SVE_FULL_HSDI
3083 (truncate:SVE_PARTIAL_I
3084 (match_operand:SVE_FULL_HSDI 2 "register_operand" "0, w")))]
3085 UNSPEC_PRED_X))]
3086 "TARGET_SVE
3087 && (~<SVE_FULL_HSDI:narrower_mask> & <SVE_PARTIAL_I:self_mask>) == 0"
3088 "@
3089 sxt<SVE_PARTIAL_I:Vesize>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>
3090 movprfx\t%0, %2\;sxt<SVE_PARTIAL_I:Vesize>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>"
3091 [(set_attr "movprfx" "*,yes")]
3092 )
3093
3094 ;; Predicated truncate-and-sign-extend operations with merging.
3095 (define_insn "@aarch64_cond_sxt<SVE_FULL_HSDI:mode><SVE_PARTIAL_I:mode>"
3096 [(set (match_operand:SVE_FULL_HSDI 0 "register_operand" "=w, ?&w, ?&w")
3097 (unspec:SVE_FULL_HSDI
3098 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand" "Upl, Upl, Upl")
3099 (sign_extend:SVE_FULL_HSDI
3100 (truncate:SVE_PARTIAL_I
3101 (match_operand:SVE_FULL_HSDI 2 "register_operand" "w, w, w")))
3102 (match_operand:SVE_FULL_HSDI 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
3103 UNSPEC_SEL))]
3104 "TARGET_SVE
3105 && (~<SVE_FULL_HSDI:narrower_mask> & <SVE_PARTIAL_I:self_mask>) == 0"
3106 "@
3107 sxt<SVE_PARTIAL_I:Vesize>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>
3108 movprfx\t%0.<SVE_FULL_HSDI:Vetype>, %1/z, %2.<SVE_FULL_HSDI:Vetype>\;sxt<SVE_PARTIAL_I:Vesize>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>
3109 movprfx\t%0, %3\;sxt<SVE_PARTIAL_I:Vesize>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>"
3110 [(set_attr "movprfx" "*,yes,yes")]
3111 )
3112
3113 ;; Predicated truncate-and-zero-extend operations, merging with the
3114 ;; first input.
3115 ;;
3116 ;; The canonical form of this operation is an AND of a constant rather
3117 ;; than (zero_extend (truncate ...)).
3118 (define_insn "*cond_uxt<mode>_2"
3119 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
3120 (unspec:SVE_FULL_I
3121 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3122 (and:SVE_FULL_I
3123 (match_operand:SVE_FULL_I 2 "register_operand" "0, w")
3124 (match_operand:SVE_FULL_I 3 "aarch64_sve_uxt_immediate"))
3125 (match_dup 2)]
3126 UNSPEC_SEL))]
3127 "TARGET_SVE"
3128 "@
3129 uxt%e3\t%0.<Vetype>, %1/m, %0.<Vetype>
3130 movprfx\t%0, %2\;uxt%e3\t%0.<Vetype>, %1/m, %2.<Vetype>"
3131 [(set_attr "movprfx" "*,yes")]
3132 )
3133
3134 ;; Predicated truncate-and-zero-extend operations, merging with an
3135 ;; independent value.
3136 ;;
3137 ;; The earlyclobber isn't needed for the first alternative, but omitting
3138 ;; it would only help the case in which operands 2 and 4 are the same,
3139 ;; which is handled above rather than here. Marking all the alternatives
3140 ;; as early-clobber helps to make the instruction more regular to the
3141 ;; register allocator.
3142 (define_insn "*cond_uxt<mode>_any"
3143 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, ?&w, ?&w")
3144 (unspec:SVE_FULL_I
3145 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
3146 (and:SVE_FULL_I
3147 (match_operand:SVE_FULL_I 2 "register_operand" "w, w, w")
3148 (match_operand:SVE_FULL_I 3 "aarch64_sve_uxt_immediate"))
3149 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero" "0, Dz, w")]
3150 UNSPEC_SEL))]
3151 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
3152 "@
3153 uxt%e3\t%0.<Vetype>, %1/m, %2.<Vetype>
3154 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;uxt%e3\t%0.<Vetype>, %1/m, %2.<Vetype>
3155 movprfx\t%0, %4\;uxt%e3\t%0.<Vetype>, %1/m, %2.<Vetype>"
3156 [(set_attr "movprfx" "*,yes,yes")]
3157 )
3158
3159 ;; -------------------------------------------------------------------------
3160 ;; ---- [INT] Truncation
3161 ;; -------------------------------------------------------------------------
3162 ;; The patterns in this section are synthetic.
3163 ;; -------------------------------------------------------------------------
3164
3165 ;; Truncate to a partial SVE vector from either a full vector or a
3166 ;; wider partial vector. This is a no-op, because we can just ignore
3167 ;; the unused upper bits of the source.
3168 (define_insn_and_split "trunc<SVE_HSDI:mode><SVE_PARTIAL_I:mode>2"
3169 [(set (match_operand:SVE_PARTIAL_I 0 "register_operand" "=w")
3170 (truncate:SVE_PARTIAL_I
3171 (match_operand:SVE_HSDI 1 "register_operand" "w")))]
3172 "TARGET_SVE && (~<SVE_HSDI:narrower_mask> & <SVE_PARTIAL_I:self_mask>) == 0"
3173 "#"
3174 "&& reload_completed"
3175 [(set (match_dup 0) (match_dup 1))]
3176 {
3177 operands[1] = aarch64_replace_reg_mode (operands[1],
3178 <SVE_PARTIAL_I:MODE>mode);
3179 }
3180 )
3181
3182 ;; -------------------------------------------------------------------------
3183 ;; ---- [INT] Logical inverse
3184 ;; -------------------------------------------------------------------------
3185 ;; Includes:
3186 ;; - CNOT
3187 ;; -------------------------------------------------------------------------
3188
3189 ;; Predicated logical inverse.
3190 (define_expand "@aarch64_pred_cnot<mode>"
3191 [(set (match_operand:SVE_FULL_I 0 "register_operand")
3192 (unspec:SVE_FULL_I
3193 [(unspec:<VPRED>
3194 [(match_operand:<VPRED> 1 "register_operand")
3195 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
3196 (eq:<VPRED>
3197 (match_operand:SVE_FULL_I 3 "register_operand")
3198 (match_dup 4))]
3199 UNSPEC_PRED_Z)
3200 (match_dup 5)
3201 (match_dup 4)]
3202 UNSPEC_SEL))]
3203 "TARGET_SVE"
3204 {
3205 operands[4] = CONST0_RTX (<MODE>mode);
3206 operands[5] = CONST1_RTX (<MODE>mode);
3207 }
3208 )
3209
3210 (define_insn "*cnot<mode>"
3211 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
3212 (unspec:SVE_FULL_I
3213 [(unspec:<VPRED>
3214 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3215 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
3216 (eq:<VPRED>
3217 (match_operand:SVE_FULL_I 2 "register_operand" "0, w")
3218 (match_operand:SVE_FULL_I 3 "aarch64_simd_imm_zero"))]
3219 UNSPEC_PRED_Z)
3220 (match_operand:SVE_FULL_I 4 "aarch64_simd_imm_one")
3221 (match_dup 3)]
3222 UNSPEC_SEL))]
3223 "TARGET_SVE"
3224 "@
3225 cnot\t%0.<Vetype>, %1/m, %2.<Vetype>
3226 movprfx\t%0, %2\;cnot\t%0.<Vetype>, %1/m, %2.<Vetype>"
3227 [(set_attr "movprfx" "*,yes")]
3228 )
3229
3230 ;; Predicated logical inverse with merging.
3231 (define_expand "@cond_cnot<mode>"
3232 [(set (match_operand:SVE_FULL_I 0 "register_operand")
3233 (unspec:SVE_FULL_I
3234 [(match_operand:<VPRED> 1 "register_operand")
3235 (unspec:SVE_FULL_I
3236 [(unspec:<VPRED>
3237 [(match_dup 4)
3238 (const_int SVE_KNOWN_PTRUE)
3239 (eq:<VPRED>
3240 (match_operand:SVE_FULL_I 2 "register_operand")
3241 (match_dup 5))]
3242 UNSPEC_PRED_Z)
3243 (match_dup 6)
3244 (match_dup 5)]
3245 UNSPEC_SEL)
3246 (match_operand:SVE_FULL_I 3 "aarch64_simd_reg_or_zero")]
3247 UNSPEC_SEL))]
3248 "TARGET_SVE"
3249 {
3250 operands[4] = CONSTM1_RTX (<VPRED>mode);
3251 operands[5] = CONST0_RTX (<MODE>mode);
3252 operands[6] = CONST1_RTX (<MODE>mode);
3253 }
3254 )
3255
3256 ;; Predicated logical inverse, merging with the first input.
3257 (define_insn_and_rewrite "*cond_cnot<mode>_2"
3258 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
3259 (unspec:SVE_FULL_I
3260 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3261 ;; Logical inverse of operand 2 (as above).
3262 (unspec:SVE_FULL_I
3263 [(unspec:<VPRED>
3264 [(match_operand 5)
3265 (const_int SVE_KNOWN_PTRUE)
3266 (eq:<VPRED>
3267 (match_operand:SVE_FULL_I 2 "register_operand" "0, w")
3268 (match_operand:SVE_FULL_I 3 "aarch64_simd_imm_zero"))]
3269 UNSPEC_PRED_Z)
3270 (match_operand:SVE_FULL_I 4 "aarch64_simd_imm_one")
3271 (match_dup 3)]
3272 UNSPEC_SEL)
3273 (match_dup 2)]
3274 UNSPEC_SEL))]
3275 "TARGET_SVE"
3276 "@
3277 cnot\t%0.<Vetype>, %1/m, %0.<Vetype>
3278 movprfx\t%0, %2\;cnot\t%0.<Vetype>, %1/m, %2.<Vetype>"
3279 "&& !CONSTANT_P (operands[5])"
3280 {
3281 operands[5] = CONSTM1_RTX (<VPRED>mode);
3282 }
3283 [(set_attr "movprfx" "*,yes")]
3284 )
3285
3286 ;; Predicated logical inverse, merging with an independent value.
3287 ;;
3288 ;; The earlyclobber isn't needed for the first alternative, but omitting
3289 ;; it would only help the case in which operands 2 and 6 are the same,
3290 ;; which is handled above rather than here. Marking all the alternatives
3291 ;; as earlyclobber helps to make the instruction more regular to the
3292 ;; register allocator.
3293 (define_insn_and_rewrite "*cond_cnot<mode>_any"
3294 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, ?&w, ?&w")
3295 (unspec:SVE_FULL_I
3296 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
3297 ;; Logical inverse of operand 2 (as above).
3298 (unspec:SVE_FULL_I
3299 [(unspec:<VPRED>
3300 [(match_operand 5)
3301 (const_int SVE_KNOWN_PTRUE)
3302 (eq:<VPRED>
3303 (match_operand:SVE_FULL_I 2 "register_operand" "w, w, w")
3304 (match_operand:SVE_FULL_I 3 "aarch64_simd_imm_zero"))]
3305 UNSPEC_PRED_Z)
3306 (match_operand:SVE_FULL_I 4 "aarch64_simd_imm_one")
3307 (match_dup 3)]
3308 UNSPEC_SEL)
3309 (match_operand:SVE_FULL_I 6 "aarch64_simd_reg_or_zero" "0, Dz, w")]
3310 UNSPEC_SEL))]
3311 "TARGET_SVE && !rtx_equal_p (operands[2], operands[6])"
3312 "@
3313 cnot\t%0.<Vetype>, %1/m, %2.<Vetype>
3314 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;cnot\t%0.<Vetype>, %1/m, %2.<Vetype>
3315 movprfx\t%0, %6\;cnot\t%0.<Vetype>, %1/m, %2.<Vetype>"
3316 "&& !CONSTANT_P (operands[5])"
3317 {
3318 operands[5] = CONSTM1_RTX (<VPRED>mode);
3319 }
3320 [(set_attr "movprfx" "*,yes,yes")]
3321 )
3322
3323 ;; -------------------------------------------------------------------------
3324 ;; ---- [FP<-INT] General unary arithmetic that maps to unspecs
3325 ;; -------------------------------------------------------------------------
3326 ;; Includes:
3327 ;; - FEXPA
3328 ;; -------------------------------------------------------------------------
3329
3330 ;; Unpredicated unary operations that take an integer and return a float.
3331 (define_insn "@aarch64_sve_<optab><mode>"
3332 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w")
3333 (unspec:SVE_FULL_F
3334 [(match_operand:<V_INT_EQUIV> 1 "register_operand" "w")]
3335 SVE_FP_UNARY_INT))]
3336 "TARGET_SVE"
3337 "<sve_fp_op>\t%0.<Vetype>, %1.<Vetype>"
3338 )
3339
3340 ;; -------------------------------------------------------------------------
3341 ;; ---- [FP] General unary arithmetic corresponding to unspecs
3342 ;; -------------------------------------------------------------------------
3343 ;; Includes:
3344 ;; - FABS
3345 ;; - FNEG
3346 ;; - FRECPE
3347 ;; - FRECPX
3348 ;; - FRINTA
3349 ;; - FRINTI
3350 ;; - FRINTM
3351 ;; - FRINTN
3352 ;; - FRINTP
3353 ;; - FRINTX
3354 ;; - FRINTZ
3355 ;; - FRSQRTE
3356 ;; - FSQRT
3357 ;; -------------------------------------------------------------------------
3358
3359 ;; Unpredicated floating-point unary operations.
3360 (define_insn "@aarch64_sve_<optab><mode>"
3361 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w")
3362 (unspec:SVE_FULL_F
3363 [(match_operand:SVE_FULL_F 1 "register_operand" "w")]
3364 SVE_FP_UNARY))]
3365 "TARGET_SVE"
3366 "<sve_fp_op>\t%0.<Vetype>, %1.<Vetype>"
3367 )
3368
3369 ;; Unpredicated floating-point unary operations.
3370 (define_expand "<optab><mode>2"
3371 [(set (match_operand:SVE_FULL_F 0 "register_operand")
3372 (unspec:SVE_FULL_F
3373 [(match_dup 2)
3374 (const_int SVE_RELAXED_GP)
3375 (match_operand:SVE_FULL_F 1 "register_operand")]
3376 SVE_COND_FP_UNARY_OPTAB))]
3377 "TARGET_SVE"
3378 {
3379 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
3380 }
3381 )
3382
3383 ;; Predicated floating-point unary operations.
3384 (define_insn "@aarch64_pred_<optab><mode>"
3385 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
3386 (unspec:SVE_FULL_F
3387 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3388 (match_operand:SI 3 "aarch64_sve_gp_strictness")
3389 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")]
3390 SVE_COND_FP_UNARY))]
3391 "TARGET_SVE"
3392 "@
3393 <sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
3394 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
3395 [(set_attr "movprfx" "*,yes")]
3396 )
3397
3398 ;; Predicated floating-point unary arithmetic with merging.
3399 (define_expand "@cond_<optab><mode>"
3400 [(set (match_operand:SVE_FULL_F 0 "register_operand")
3401 (unspec:SVE_FULL_F
3402 [(match_operand:<VPRED> 1 "register_operand")
3403 (unspec:SVE_FULL_F
3404 [(match_dup 1)
3405 (const_int SVE_STRICT_GP)
3406 (match_operand:SVE_FULL_F 2 "register_operand")]
3407 SVE_COND_FP_UNARY)
3408 (match_operand:SVE_FULL_F 3 "aarch64_simd_reg_or_zero")]
3409 UNSPEC_SEL))]
3410 "TARGET_SVE"
3411 )
3412
3413 ;; Predicated floating-point unary arithmetic, merging with the first input.
3414 (define_insn_and_rewrite "*cond_<optab><mode>_2_relaxed"
3415 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
3416 (unspec:SVE_FULL_F
3417 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3418 (unspec:SVE_FULL_F
3419 [(match_operand 3)
3420 (const_int SVE_RELAXED_GP)
3421 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")]
3422 SVE_COND_FP_UNARY)
3423 (match_dup 2)]
3424 UNSPEC_SEL))]
3425 "TARGET_SVE"
3426 "@
3427 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>
3428 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
3429 "&& !rtx_equal_p (operands[1], operands[3])"
3430 {
3431 operands[3] = copy_rtx (operands[1]);
3432 }
3433 [(set_attr "movprfx" "*,yes")]
3434 )
3435
3436 (define_insn "*cond_<optab><mode>_2_strict"
3437 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
3438 (unspec:SVE_FULL_F
3439 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3440 (unspec:SVE_FULL_F
3441 [(match_dup 1)
3442 (const_int SVE_STRICT_GP)
3443 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")]
3444 SVE_COND_FP_UNARY)
3445 (match_dup 2)]
3446 UNSPEC_SEL))]
3447 "TARGET_SVE"
3448 "@
3449 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>
3450 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
3451 [(set_attr "movprfx" "*,yes")]
3452 )
3453
3454 ;; Predicated floating-point unary arithmetic, merging with an independent
3455 ;; value.
3456 ;;
3457 ;; The earlyclobber isn't needed for the first alternative, but omitting
3458 ;; it would only help the case in which operands 2 and 3 are the same,
3459 ;; which is handled above rather than here. Marking all the alternatives
3460 ;; as earlyclobber helps to make the instruction more regular to the
3461 ;; register allocator.
3462 (define_insn_and_rewrite "*cond_<optab><mode>_any_relaxed"
3463 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, ?&w, ?&w")
3464 (unspec:SVE_FULL_F
3465 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
3466 (unspec:SVE_FULL_F
3467 [(match_operand 4)
3468 (const_int SVE_RELAXED_GP)
3469 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w")]
3470 SVE_COND_FP_UNARY)
3471 (match_operand:SVE_FULL_F 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
3472 UNSPEC_SEL))]
3473 "TARGET_SVE && !rtx_equal_p (operands[2], operands[3])"
3474 "@
3475 <sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
3476 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
3477 movprfx\t%0, %3\;<sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
3478 "&& !rtx_equal_p (operands[1], operands[4])"
3479 {
3480 operands[4] = copy_rtx (operands[1]);
3481 }
3482 [(set_attr "movprfx" "*,yes,yes")]
3483 )
3484
3485 (define_insn "*cond_<optab><mode>_any_strict"
3486 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, ?&w, ?&w")
3487 (unspec:SVE_FULL_F
3488 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
3489 (unspec:SVE_FULL_F
3490 [(match_dup 1)
3491 (const_int SVE_STRICT_GP)
3492 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w")]
3493 SVE_COND_FP_UNARY)
3494 (match_operand:SVE_FULL_F 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
3495 UNSPEC_SEL))]
3496 "TARGET_SVE && !rtx_equal_p (operands[2], operands[3])"
3497 "@
3498 <sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
3499 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>
3500 movprfx\t%0, %3\;<sve_fp_op>\t%0.<Vetype>, %1/m, %2.<Vetype>"
3501 [(set_attr "movprfx" "*,yes,yes")]
3502 )
3503
3504 ;; -------------------------------------------------------------------------
3505 ;; ---- [FP] Square root
3506 ;; -------------------------------------------------------------------------
3507
3508 (define_expand "sqrt<mode>2"
3509 [(set (match_operand:SVE_FULL_F 0 "register_operand")
3510 (unspec:SVE_FULL_F
3511 [(match_dup 2)
3512 (const_int SVE_RELAXED_GP)
3513 (match_operand:SVE_FULL_F 1 "register_operand")]
3514 UNSPEC_COND_FSQRT))]
3515 "TARGET_SVE"
3516 {
3517 if (aarch64_emit_approx_sqrt (operands[0], operands[1], false))
3518 DONE;
3519 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
3520 })
3521
3522 ;; -------------------------------------------------------------------------
3523 ;; ---- [FP] Reciprocal square root
3524 ;; -------------------------------------------------------------------------
3525
3526 (define_expand "rsqrt<mode>2"
3527 [(set (match_operand:SVE_FULL_SDF 0 "register_operand")
3528 (unspec:SVE_FULL_SDF
3529 [(match_operand:SVE_FULL_SDF 1 "register_operand")]
3530 UNSPEC_RSQRT))]
3531 "TARGET_SVE"
3532 {
3533 aarch64_emit_approx_sqrt (operands[0], operands[1], true);
3534 DONE;
3535 })
3536
3537 (define_expand "@aarch64_rsqrte<mode>"
3538 [(set (match_operand:SVE_FULL_SDF 0 "register_operand")
3539 (unspec:SVE_FULL_SDF
3540 [(match_operand:SVE_FULL_SDF 1 "register_operand")]
3541 UNSPEC_RSQRTE))]
3542 "TARGET_SVE"
3543 )
3544
3545 (define_expand "@aarch64_rsqrts<mode>"
3546 [(set (match_operand:SVE_FULL_SDF 0 "register_operand")
3547 (unspec:SVE_FULL_SDF
3548 [(match_operand:SVE_FULL_SDF 1 "register_operand")
3549 (match_operand:SVE_FULL_SDF 2 "register_operand")]
3550 UNSPEC_RSQRTS))]
3551 "TARGET_SVE"
3552 )
3553
3554 ;; -------------------------------------------------------------------------
3555 ;; ---- [PRED] Inverse
3556 ;; -------------------------------------------------------------------------
3557 ;; Includes:
3558 ;; - NOT
3559 ;; -------------------------------------------------------------------------
3560
3561 ;; Unpredicated predicate inverse.
3562 (define_expand "one_cmpl<mode>2"
3563 [(set (match_operand:PRED_ALL 0 "register_operand")
3564 (and:PRED_ALL
3565 (not:PRED_ALL (match_operand:PRED_ALL 1 "register_operand"))
3566 (match_dup 2)))]
3567 "TARGET_SVE"
3568 {
3569 operands[2] = aarch64_ptrue_reg (<MODE>mode);
3570 }
3571 )
3572
3573 ;; Predicated predicate inverse.
3574 (define_insn "*one_cmpl<mode>3"
3575 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
3576 (and:PRED_ALL
3577 (not:PRED_ALL (match_operand:PRED_ALL 2 "register_operand" "Upa"))
3578 (match_operand:PRED_ALL 1 "register_operand" "Upa")))]
3579 "TARGET_SVE"
3580 "not\t%0.b, %1/z, %2.b"
3581 )
3582
3583 ;; =========================================================================
3584 ;; == Binary arithmetic
3585 ;; =========================================================================
3586
3587 ;; -------------------------------------------------------------------------
3588 ;; ---- [INT] General binary arithmetic corresponding to rtx codes
3589 ;; -------------------------------------------------------------------------
3590 ;; Includes:
3591 ;; - ADD (merging form only)
3592 ;; - AND (merging form only)
3593 ;; - ASR (merging form only)
3594 ;; - EOR (merging form only)
3595 ;; - LSL (merging form only)
3596 ;; - LSR (merging form only)
3597 ;; - MUL
3598 ;; - ORR (merging form only)
3599 ;; - SMAX
3600 ;; - SMIN
3601 ;; - SQADD (SVE2 merging form only)
3602 ;; - SQSUB (SVE2 merging form only)
3603 ;; - SUB (merging form only)
3604 ;; - UMAX
3605 ;; - UMIN
3606 ;; - UQADD (SVE2 merging form only)
3607 ;; - UQSUB (SVE2 merging form only)
3608 ;; -------------------------------------------------------------------------
3609
3610 ;; Unpredicated integer binary operations that have an immediate form.
3611 (define_expand "<optab><mode>3"
3612 [(set (match_operand:SVE_FULL_I 0 "register_operand")
3613 (unspec:SVE_FULL_I
3614 [(match_dup 3)
3615 (SVE_INT_BINARY_IMM:SVE_FULL_I
3616 (match_operand:SVE_FULL_I 1 "register_operand")
3617 (match_operand:SVE_FULL_I 2 "aarch64_sve_<sve_imm_con>_operand"))]
3618 UNSPEC_PRED_X))]
3619 "TARGET_SVE"
3620 {
3621 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
3622 }
3623 )
3624
3625 ;; Integer binary operations that have an immediate form, predicated
3626 ;; with a PTRUE. We don't actually need the predicate for the first
3627 ;; and third alternatives, but using Upa or X isn't likely to gain much
3628 ;; and would make the instruction seem less uniform to the register
3629 ;; allocator.
3630 (define_insn_and_split "@aarch64_pred_<optab><mode>"
3631 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, w, ?&w, ?&w")
3632 (unspec:SVE_FULL_I
3633 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
3634 (SVE_INT_BINARY_IMM:SVE_FULL_I
3635 (match_operand:SVE_FULL_I 2 "register_operand" "%0, 0, w, w")
3636 (match_operand:SVE_FULL_I 3 "aarch64_sve_<sve_imm_con>_operand" "<sve_imm_con>, w, <sve_imm_con>, w"))]
3637 UNSPEC_PRED_X))]
3638 "TARGET_SVE"
3639 "@
3640 #
3641 <sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
3642 #
3643 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
3644 ; Split the unpredicated form after reload, so that we don't have
3645 ; the unnecessary PTRUE.
3646 "&& reload_completed
3647 && !register_operand (operands[3], <MODE>mode)"
3648 [(set (match_dup 0)
3649 (SVE_INT_BINARY_IMM:SVE_FULL_I (match_dup 2) (match_dup 3)))]
3650 ""
3651 [(set_attr "movprfx" "*,*,yes,yes")]
3652 )
3653
3654 ;; Unpredicated binary operations with a constant (post-RA only).
3655 ;; These are generated by splitting a predicated instruction whose
3656 ;; predicate is unused.
3657 (define_insn "*post_ra_<optab><mode>3"
3658 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
3659 (SVE_INT_BINARY_IMM:SVE_FULL_I
3660 (match_operand:SVE_FULL_I 1 "register_operand" "0, w")
3661 (match_operand:SVE_FULL_I 2 "aarch64_sve_<sve_imm_con>_immediate")))]
3662 "TARGET_SVE && reload_completed"
3663 "@
3664 <sve_int_op>\t%0.<Vetype>, %0.<Vetype>, #%<sve_imm_prefix>2
3665 movprfx\t%0, %1\;<sve_int_op>\t%0.<Vetype>, %0.<Vetype>, #%<sve_imm_prefix>2"
3666 [(set_attr "movprfx" "*,yes")]
3667 )
3668
3669 ;; Predicated integer operations with merging.
3670 (define_expand "@cond_<optab><mode>"
3671 [(set (match_operand:SVE_FULL_I 0 "register_operand")
3672 (unspec:SVE_FULL_I
3673 [(match_operand:<VPRED> 1 "register_operand")
3674 (SVE_INT_BINARY:SVE_FULL_I
3675 (match_operand:SVE_FULL_I 2 "register_operand")
3676 (match_operand:SVE_FULL_I 3 "<sve_pred_int_rhs2_operand>"))
3677 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero")]
3678 UNSPEC_SEL))]
3679 "TARGET_SVE"
3680 )
3681
3682 ;; Predicated integer operations, merging with the first input.
3683 (define_insn "*cond_<optab><mode>_2"
3684 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
3685 (unspec:SVE_FULL_I
3686 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3687 (SVE_INT_BINARY:SVE_FULL_I
3688 (match_operand:SVE_FULL_I 2 "register_operand" "0, w")
3689 (match_operand:SVE_FULL_I 3 "register_operand" "w, w"))
3690 (match_dup 2)]
3691 UNSPEC_SEL))]
3692 "TARGET_SVE"
3693 "@
3694 <sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
3695 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
3696 [(set_attr "movprfx" "*,yes")]
3697 )
3698
3699 ;; Predicated integer operations, merging with the second input.
3700 (define_insn "*cond_<optab><mode>_3"
3701 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
3702 (unspec:SVE_FULL_I
3703 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3704 (SVE_INT_BINARY:SVE_FULL_I
3705 (match_operand:SVE_FULL_I 2 "register_operand" "w, w")
3706 (match_operand:SVE_FULL_I 3 "register_operand" "0, w"))
3707 (match_dup 3)]
3708 UNSPEC_SEL))]
3709 "TARGET_SVE"
3710 "@
3711 <sve_int_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
3712 movprfx\t%0, %3\;<sve_int_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>"
3713 [(set_attr "movprfx" "*,yes")]
3714 )
3715
3716 ;; Predicated integer operations, merging with an independent value.
3717 (define_insn_and_rewrite "*cond_<optab><mode>_any"
3718 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, &w, &w, &w, ?&w")
3719 (unspec:SVE_FULL_I
3720 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl")
3721 (SVE_INT_BINARY:SVE_FULL_I
3722 (match_operand:SVE_FULL_I 2 "register_operand" "0, w, w, w, w")
3723 (match_operand:SVE_FULL_I 3 "register_operand" "w, 0, w, w, w"))
3724 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, 0, w")]
3725 UNSPEC_SEL))]
3726 "TARGET_SVE
3727 && !rtx_equal_p (operands[2], operands[4])
3728 && !rtx_equal_p (operands[3], operands[4])"
3729 "@
3730 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
3731 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_int_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
3732 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
3733 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
3734 #"
3735 "&& reload_completed
3736 && register_operand (operands[4], <MODE>mode)
3737 && !rtx_equal_p (operands[0], operands[4])"
3738 {
3739 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
3740 operands[4], operands[1]));
3741 operands[4] = operands[2] = operands[0];
3742 }
3743 [(set_attr "movprfx" "yes")]
3744 )
3745
3746 ;; -------------------------------------------------------------------------
3747 ;; ---- [INT] Addition
3748 ;; -------------------------------------------------------------------------
3749 ;; Includes:
3750 ;; - ADD
3751 ;; - DECB
3752 ;; - DECD
3753 ;; - DECH
3754 ;; - DECW
3755 ;; - INCB
3756 ;; - INCD
3757 ;; - INCH
3758 ;; - INCW
3759 ;; - SUB
3760 ;; -------------------------------------------------------------------------
3761
3762 (define_insn "add<mode>3"
3763 [(set (match_operand:SVE_I 0 "register_operand" "=w, w, w, ?w, ?w, w")
3764 (plus:SVE_I
3765 (match_operand:SVE_I 1 "register_operand" "%0, 0, 0, w, w, w")
3766 (match_operand:SVE_I 2 "aarch64_sve_add_operand" "vsa, vsn, vsi, vsa, vsn, w")))]
3767 "TARGET_SVE"
3768 "@
3769 add\t%0.<Vetype>, %0.<Vetype>, #%D2
3770 sub\t%0.<Vetype>, %0.<Vetype>, #%N2
3771 * return aarch64_output_sve_vector_inc_dec (\"%0.<Vetype>\", operands[2]);
3772 movprfx\t%0, %1\;add\t%0.<Vetype>, %0.<Vetype>, #%D2
3773 movprfx\t%0, %1\;sub\t%0.<Vetype>, %0.<Vetype>, #%N2
3774 add\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
3775 [(set_attr "movprfx" "*,*,*,yes,yes,*")]
3776 )
3777
3778 ;; Merging forms are handled through SVE_INT_BINARY.
3779
3780 ;; -------------------------------------------------------------------------
3781 ;; ---- [INT] Subtraction
3782 ;; -------------------------------------------------------------------------
3783 ;; Includes:
3784 ;; - SUB
3785 ;; - SUBR
3786 ;; -------------------------------------------------------------------------
3787
3788 (define_insn "sub<mode>3"
3789 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, w, ?&w")
3790 (minus:SVE_FULL_I
3791 (match_operand:SVE_FULL_I 1 "aarch64_sve_arith_operand" "w, vsa, vsa")
3792 (match_operand:SVE_FULL_I 2 "register_operand" "w, 0, w")))]
3793 "TARGET_SVE"
3794 "@
3795 sub\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>
3796 subr\t%0.<Vetype>, %0.<Vetype>, #%D1
3797 movprfx\t%0, %2\;subr\t%0.<Vetype>, %0.<Vetype>, #%D1"
3798 [(set_attr "movprfx" "*,*,yes")]
3799 )
3800
3801 ;; Merging forms are handled through SVE_INT_BINARY.
3802
3803 ;; -------------------------------------------------------------------------
3804 ;; ---- [INT] Take address
3805 ;; -------------------------------------------------------------------------
3806 ;; Includes:
3807 ;; - ADR
3808 ;; -------------------------------------------------------------------------
3809
3810 ;; An unshifted and unscaled ADR. This is functionally equivalent to an ADD,
3811 ;; but the svadrb intrinsics should preserve the user's choice.
3812 (define_insn "@aarch64_adr<mode>"
3813 [(set (match_operand:SVE_FULL_SDI 0 "register_operand" "=w")
3814 (unspec:SVE_FULL_SDI
3815 [(match_operand:SVE_FULL_SDI 1 "register_operand" "w")
3816 (match_operand:SVE_FULL_SDI 2 "register_operand" "w")]
3817 UNSPEC_ADR))]
3818 "TARGET_SVE"
3819 "adr\t%0.<Vetype>, [%1.<Vetype>, %2.<Vetype>]"
3820 )
3821
3822 ;; Same, but with the offset being sign-extended from the low 32 bits.
3823 (define_insn_and_rewrite "*aarch64_adr_sxtw"
3824 [(set (match_operand:VNx2DI 0 "register_operand" "=w")
3825 (unspec:VNx2DI
3826 [(match_operand:VNx2DI 1 "register_operand" "w")
3827 (unspec:VNx2DI
3828 [(match_operand 3)
3829 (sign_extend:VNx2DI
3830 (truncate:VNx2SI
3831 (match_operand:VNx2DI 2 "register_operand" "w")))]
3832 UNSPEC_PRED_X)]
3833 UNSPEC_ADR))]
3834 "TARGET_SVE"
3835 "adr\t%0.d, [%1.d, %2.d, sxtw]"
3836 "&& !CONSTANT_P (operands[3])"
3837 {
3838 operands[3] = CONSTM1_RTX (VNx2BImode);
3839 }
3840 )
3841
3842 ;; Same, but with the offset being zero-extended from the low 32 bits.
3843 (define_insn "*aarch64_adr_uxtw_unspec"
3844 [(set (match_operand:VNx2DI 0 "register_operand" "=w")
3845 (unspec:VNx2DI
3846 [(match_operand:VNx2DI 1 "register_operand" "w")
3847 (and:VNx2DI
3848 (match_operand:VNx2DI 2 "register_operand" "w")
3849 (match_operand:VNx2DI 3 "aarch64_sve_uxtw_immediate"))]
3850 UNSPEC_ADR))]
3851 "TARGET_SVE"
3852 "adr\t%0.d, [%1.d, %2.d, uxtw]"
3853 )
3854
3855 ;; Same, matching as a PLUS rather than unspec.
3856 (define_insn "*aarch64_adr_uxtw_and"
3857 [(set (match_operand:VNx2DI 0 "register_operand" "=w")
3858 (plus:VNx2DI
3859 (and:VNx2DI
3860 (match_operand:VNx2DI 2 "register_operand" "w")
3861 (match_operand:VNx2DI 3 "aarch64_sve_uxtw_immediate"))
3862 (match_operand:VNx2DI 1 "register_operand" "w")))]
3863 "TARGET_SVE"
3864 "adr\t%0.d, [%1.d, %2.d, uxtw]"
3865 )
3866
3867 ;; ADR with a nonzero shift.
3868 (define_expand "@aarch64_adr<mode>_shift"
3869 [(set (match_operand:SVE_FULL_SDI 0 "register_operand")
3870 (plus:SVE_FULL_SDI
3871 (unspec:SVE_FULL_SDI
3872 [(match_dup 4)
3873 (ashift:SVE_FULL_SDI
3874 (match_operand:SVE_FULL_SDI 2 "register_operand")
3875 (match_operand:SVE_FULL_SDI 3 "const_1_to_3_operand"))]
3876 UNSPEC_PRED_X)
3877 (match_operand:SVE_FULL_SDI 1 "register_operand")))]
3878 "TARGET_SVE"
3879 {
3880 operands[4] = CONSTM1_RTX (<VPRED>mode);
3881 }
3882 )
3883
3884 (define_insn_and_rewrite "*aarch64_adr<mode>_shift"
3885 [(set (match_operand:SVE_FULL_SDI 0 "register_operand" "=w")
3886 (plus:SVE_FULL_SDI
3887 (unspec:SVE_FULL_SDI
3888 [(match_operand 4)
3889 (ashift:SVE_FULL_SDI
3890 (match_operand:SVE_FULL_SDI 2 "register_operand" "w")
3891 (match_operand:SVE_FULL_SDI 3 "const_1_to_3_operand"))]
3892 UNSPEC_PRED_X)
3893 (match_operand:SVE_FULL_SDI 1 "register_operand" "w")))]
3894 "TARGET_SVE"
3895 "adr\t%0.<Vetype>, [%1.<Vetype>, %2.<Vetype>, lsl %3]"
3896 "&& !CONSTANT_P (operands[4])"
3897 {
3898 operands[4] = CONSTM1_RTX (<VPRED>mode);
3899 }
3900 )
3901
3902 ;; Same, but with the index being sign-extended from the low 32 bits.
3903 (define_insn_and_rewrite "*aarch64_adr_shift_sxtw"
3904 [(set (match_operand:VNx2DI 0 "register_operand" "=w")
3905 (plus:VNx2DI
3906 (unspec:VNx2DI
3907 [(match_operand 4)
3908 (ashift:VNx2DI
3909 (unspec:VNx2DI
3910 [(match_operand 5)
3911 (sign_extend:VNx2DI
3912 (truncate:VNx2SI
3913 (match_operand:VNx2DI 2 "register_operand" "w")))]
3914 UNSPEC_PRED_X)
3915 (match_operand:VNx2DI 3 "const_1_to_3_operand"))]
3916 UNSPEC_PRED_X)
3917 (match_operand:VNx2DI 1 "register_operand" "w")))]
3918 "TARGET_SVE"
3919 "adr\t%0.d, [%1.d, %2.d, sxtw %3]"
3920 "&& (!CONSTANT_P (operands[4]) || !CONSTANT_P (operands[5]))"
3921 {
3922 operands[5] = operands[4] = CONSTM1_RTX (VNx2BImode);
3923 }
3924 )
3925
3926 ;; Same, but with the index being zero-extended from the low 32 bits.
3927 (define_insn_and_rewrite "*aarch64_adr_shift_uxtw"
3928 [(set (match_operand:VNx2DI 0 "register_operand" "=w")
3929 (plus:VNx2DI
3930 (unspec:VNx2DI
3931 [(match_operand 5)
3932 (ashift:VNx2DI
3933 (and:VNx2DI
3934 (match_operand:VNx2DI 2 "register_operand" "w")
3935 (match_operand:VNx2DI 4 "aarch64_sve_uxtw_immediate"))
3936 (match_operand:VNx2DI 3 "const_1_to_3_operand"))]
3937 UNSPEC_PRED_X)
3938 (match_operand:VNx2DI 1 "register_operand" "w")))]
3939 "TARGET_SVE"
3940 "adr\t%0.d, [%1.d, %2.d, uxtw %3]"
3941 "&& !CONSTANT_P (operands[5])"
3942 {
3943 operands[5] = CONSTM1_RTX (VNx2BImode);
3944 }
3945 )
3946
3947 ;; -------------------------------------------------------------------------
3948 ;; ---- [INT] Absolute difference
3949 ;; -------------------------------------------------------------------------
3950 ;; Includes:
3951 ;; - SABD
3952 ;; - UABD
3953 ;; -------------------------------------------------------------------------
3954
3955 ;; Unpredicated integer absolute difference.
3956 (define_expand "<su>abd<mode>_3"
3957 [(use (match_operand:SVE_FULL_I 0 "register_operand"))
3958 (USMAX:SVE_FULL_I
3959 (match_operand:SVE_FULL_I 1 "register_operand")
3960 (match_operand:SVE_FULL_I 2 "register_operand"))]
3961 "TARGET_SVE"
3962 {
3963 rtx pred = aarch64_ptrue_reg (<VPRED>mode);
3964 emit_insn (gen_aarch64_pred_<su>abd<mode> (operands[0], pred, operands[1],
3965 operands[2]));
3966 DONE;
3967 }
3968 )
3969
3970 ;; Predicated integer absolute difference.
3971 (define_insn "@aarch64_pred_<su>abd<mode>"
3972 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
3973 (unspec:SVE_FULL_I
3974 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
3975 (minus:SVE_FULL_I
3976 (USMAX:SVE_FULL_I
3977 (match_operand:SVE_FULL_I 2 "register_operand" "%0, w")
3978 (match_operand:SVE_FULL_I 3 "register_operand" "w, w"))
3979 (<max_opp>:SVE_FULL_I
3980 (match_dup 2)
3981 (match_dup 3)))]
3982 UNSPEC_PRED_X))]
3983 "TARGET_SVE"
3984 "@
3985 <su>abd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
3986 movprfx\t%0, %2\;<su>abd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
3987 [(set_attr "movprfx" "*,yes")]
3988 )
3989
3990 (define_expand "@aarch64_cond_<su>abd<mode>"
3991 [(set (match_operand:SVE_FULL_I 0 "register_operand")
3992 (unspec:SVE_FULL_I
3993 [(match_operand:<VPRED> 1 "register_operand")
3994 (minus:SVE_FULL_I
3995 (unspec:SVE_FULL_I
3996 [(match_dup 1)
3997 (USMAX:SVE_FULL_I
3998 (match_operand:SVE_FULL_I 2 "register_operand")
3999 (match_operand:SVE_FULL_I 3 "register_operand"))]
4000 UNSPEC_PRED_X)
4001 (unspec:SVE_FULL_I
4002 [(match_dup 1)
4003 (<max_opp>:SVE_FULL_I
4004 (match_dup 2)
4005 (match_dup 3))]
4006 UNSPEC_PRED_X))
4007 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero")]
4008 UNSPEC_SEL))]
4009 "TARGET_SVE"
4010 {
4011 if (rtx_equal_p (operands[3], operands[4]))
4012 std::swap (operands[2], operands[3]);
4013 })
4014
4015 ;; Predicated integer absolute difference, merging with the first input.
4016 (define_insn_and_rewrite "*aarch64_cond_<su>abd<mode>_2"
4017 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
4018 (unspec:SVE_FULL_I
4019 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4020 (minus:SVE_FULL_I
4021 (unspec:SVE_FULL_I
4022 [(match_operand 4)
4023 (USMAX:SVE_FULL_I
4024 (match_operand:SVE_FULL_I 2 "register_operand" "0, w")
4025 (match_operand:SVE_FULL_I 3 "register_operand" "w, w"))]
4026 UNSPEC_PRED_X)
4027 (unspec:SVE_FULL_I
4028 [(match_operand 5)
4029 (<max_opp>:SVE_FULL_I
4030 (match_dup 2)
4031 (match_dup 3))]
4032 UNSPEC_PRED_X))
4033 (match_dup 2)]
4034 UNSPEC_SEL))]
4035 "TARGET_SVE"
4036 "@
4037 <su>abd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4038 movprfx\t%0, %2\;<su>abd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4039 "&& (!CONSTANT_P (operands[4]) || !CONSTANT_P (operands[5]))"
4040 {
4041 operands[4] = operands[5] = CONSTM1_RTX (<VPRED>mode);
4042 }
4043 [(set_attr "movprfx" "*,yes")]
4044 )
4045
4046 ;; Predicated integer absolute difference, merging with an independent value.
4047 (define_insn_and_rewrite "*aarch64_cond_<su>abd<mode>_any"
4048 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, &w, &w, &w, ?&w")
4049 (unspec:SVE_FULL_I
4050 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl")
4051 (minus:SVE_FULL_I
4052 (unspec:SVE_FULL_I
4053 [(match_operand 5)
4054 (USMAX:SVE_FULL_I
4055 (match_operand:SVE_FULL_I 2 "register_operand" "0, w, w, w, w")
4056 (match_operand:SVE_FULL_I 3 "register_operand" "w, 0, w, w, w"))]
4057 UNSPEC_PRED_X)
4058 (unspec:SVE_FULL_I
4059 [(match_operand 6)
4060 (<max_opp>:SVE_FULL_I
4061 (match_dup 2)
4062 (match_dup 3))]
4063 UNSPEC_PRED_X))
4064 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, 0, w")]
4065 UNSPEC_SEL))]
4066 "TARGET_SVE
4067 && !rtx_equal_p (operands[2], operands[4])
4068 && !rtx_equal_p (operands[3], operands[4])"
4069 "@
4070 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<su>abd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4071 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<su>abd\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
4072 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<su>abd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4073 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<su>abd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4074 #"
4075 "&& 1"
4076 {
4077 if (!CONSTANT_P (operands[5]) || !CONSTANT_P (operands[6]))
4078 operands[5] = operands[6] = CONSTM1_RTX (<VPRED>mode);
4079 else if (reload_completed
4080 && register_operand (operands[4], <MODE>mode)
4081 && !rtx_equal_p (operands[0], operands[4]))
4082 {
4083 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
4084 operands[4], operands[1]));
4085 operands[4] = operands[2] = operands[0];
4086 }
4087 else
4088 FAIL;
4089 }
4090 [(set_attr "movprfx" "yes")]
4091 )
4092
4093 ;; -------------------------------------------------------------------------
4094 ;; ---- [INT] Saturating addition and subtraction
4095 ;; -------------------------------------------------------------------------
4096 ;; - SQADD
4097 ;; - SQSUB
4098 ;; - UQADD
4099 ;; - UQSUB
4100 ;; -------------------------------------------------------------------------
4101
4102 ;; Unpredicated saturating signed addition and subtraction.
4103 (define_insn "@aarch64_sve_<optab><mode>"
4104 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, w, ?&w, ?&w, w")
4105 (SBINQOPS:SVE_FULL_I
4106 (match_operand:SVE_FULL_I 1 "register_operand" "0, 0, w, w, w")
4107 (match_operand:SVE_FULL_I 2 "aarch64_sve_sqadd_operand" "vsQ, vsS, vsQ, vsS, w")))]
4108 "TARGET_SVE"
4109 "@
4110 <binqops_op>\t%0.<Vetype>, %0.<Vetype>, #%D2
4111 <binqops_op_rev>\t%0.<Vetype>, %0.<Vetype>, #%N2
4112 movprfx\t%0, %1\;<binqops_op>\t%0.<Vetype>, %0.<Vetype>, #%D2
4113 movprfx\t%0, %1\;<binqops_op_rev>\t%0.<Vetype>, %0.<Vetype>, #%N2
4114 <binqops_op>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
4115 [(set_attr "movprfx" "*,*,yes,yes,*")]
4116 )
4117
4118 ;; Unpredicated saturating unsigned addition and subtraction.
4119 (define_insn "@aarch64_sve_<optab><mode>"
4120 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w, w")
4121 (UBINQOPS:SVE_FULL_I
4122 (match_operand:SVE_FULL_I 1 "register_operand" "0, w, w")
4123 (match_operand:SVE_FULL_I 2 "aarch64_sve_arith_operand" "vsa, vsa, w")))]
4124 "TARGET_SVE"
4125 "@
4126 <binqops_op>\t%0.<Vetype>, %0.<Vetype>, #%D2
4127 movprfx\t%0, %1\;<binqops_op>\t%0.<Vetype>, %0.<Vetype>, #%D2
4128 <binqops_op>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
4129 [(set_attr "movprfx" "*,yes,*")]
4130 )
4131
4132 ;; -------------------------------------------------------------------------
4133 ;; ---- [INT] Highpart multiplication
4134 ;; -------------------------------------------------------------------------
4135 ;; Includes:
4136 ;; - SMULH
4137 ;; - UMULH
4138 ;; -------------------------------------------------------------------------
4139
4140 ;; Unpredicated highpart multiplication.
4141 (define_expand "<su>mul<mode>3_highpart"
4142 [(set (match_operand:SVE_FULL_I 0 "register_operand")
4143 (unspec:SVE_FULL_I
4144 [(match_dup 3)
4145 (unspec:SVE_FULL_I
4146 [(match_operand:SVE_FULL_I 1 "register_operand")
4147 (match_operand:SVE_FULL_I 2 "register_operand")]
4148 MUL_HIGHPART)]
4149 UNSPEC_PRED_X))]
4150 "TARGET_SVE"
4151 {
4152 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
4153 }
4154 )
4155
4156 ;; Predicated highpart multiplication.
4157 (define_insn "@aarch64_pred_<optab><mode>"
4158 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
4159 (unspec:SVE_FULL_I
4160 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4161 (unspec:SVE_FULL_I
4162 [(match_operand:SVE_FULL_I 2 "register_operand" "%0, w")
4163 (match_operand:SVE_FULL_I 3 "register_operand" "w, w")]
4164 MUL_HIGHPART)]
4165 UNSPEC_PRED_X))]
4166 "TARGET_SVE"
4167 "@
4168 <su>mulh\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4169 movprfx\t%0, %2\;<su>mulh\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4170 [(set_attr "movprfx" "*,yes")]
4171 )
4172
4173 ;; Predicated highpart multiplications with merging.
4174 (define_expand "@cond_<optab><mode>"
4175 [(set (match_operand:SVE_FULL_I 0 "register_operand")
4176 (unspec:SVE_FULL_I
4177 [(match_operand:<VPRED> 1 "register_operand")
4178 (unspec:SVE_FULL_I
4179 [(match_operand:SVE_FULL_I 2 "register_operand")
4180 (match_operand:SVE_FULL_I 3 "register_operand")]
4181 MUL_HIGHPART)
4182 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero")]
4183 UNSPEC_SEL))]
4184 "TARGET_SVE"
4185 {
4186 /* Only target code is aware of these operations, so we don't need
4187 to handle the fully-general case. */
4188 gcc_assert (rtx_equal_p (operands[2], operands[4])
4189 || CONSTANT_P (operands[4]));
4190 })
4191
4192 ;; Predicated highpart multiplications, merging with the first input.
4193 (define_insn "*cond_<optab><mode>_2"
4194 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
4195 (unspec:SVE_FULL_I
4196 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4197 (unspec:SVE_FULL_I
4198 [(match_operand:SVE_FULL_I 2 "register_operand" "0, w")
4199 (match_operand:SVE_FULL_I 3 "register_operand" "w, w")]
4200 MUL_HIGHPART)
4201 (match_dup 2)]
4202 UNSPEC_SEL))]
4203 "TARGET_SVE"
4204 "@
4205 <sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4206 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4207 [(set_attr "movprfx" "*,yes")])
4208
4209 ;; Predicated highpart multiplications, merging with zero.
4210 (define_insn "*cond_<optab><mode>_z"
4211 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, &w")
4212 (unspec:SVE_FULL_I
4213 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4214 (unspec:SVE_FULL_I
4215 [(match_operand:SVE_FULL_I 2 "register_operand" "%0, w")
4216 (match_operand:SVE_FULL_I 3 "register_operand" "w, w")]
4217 MUL_HIGHPART)
4218 (match_operand:SVE_FULL_I 4 "aarch64_simd_imm_zero")]
4219 UNSPEC_SEL))]
4220 "TARGET_SVE"
4221 "@
4222 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4223 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4224 [(set_attr "movprfx" "yes")])
4225
4226 ;; -------------------------------------------------------------------------
4227 ;; ---- [INT] Division
4228 ;; -------------------------------------------------------------------------
4229 ;; Includes:
4230 ;; - SDIV
4231 ;; - SDIVR
4232 ;; - UDIV
4233 ;; - UDIVR
4234 ;; -------------------------------------------------------------------------
4235
4236 ;; Unpredicated integer division.
4237 (define_expand "<optab><mode>3"
4238 [(set (match_operand:SVE_FULL_SDI 0 "register_operand")
4239 (unspec:SVE_FULL_SDI
4240 [(match_dup 3)
4241 (SVE_INT_BINARY_SD:SVE_FULL_SDI
4242 (match_operand:SVE_FULL_SDI 1 "register_operand")
4243 (match_operand:SVE_FULL_SDI 2 "register_operand"))]
4244 UNSPEC_PRED_X))]
4245 "TARGET_SVE"
4246 {
4247 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
4248 }
4249 )
4250
4251 ;; Integer division predicated with a PTRUE.
4252 (define_insn "@aarch64_pred_<optab><mode>"
4253 [(set (match_operand:SVE_FULL_SDI 0 "register_operand" "=w, w, ?&w")
4254 (unspec:SVE_FULL_SDI
4255 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
4256 (SVE_INT_BINARY_SD:SVE_FULL_SDI
4257 (match_operand:SVE_FULL_SDI 2 "register_operand" "0, w, w")
4258 (match_operand:SVE_FULL_SDI 3 "register_operand" "w, 0, w"))]
4259 UNSPEC_PRED_X))]
4260 "TARGET_SVE"
4261 "@
4262 <sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4263 <sve_int_op>r\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
4264 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4265 [(set_attr "movprfx" "*,*,yes")]
4266 )
4267
4268 ;; Predicated integer division with merging.
4269 (define_expand "@cond_<optab><mode>"
4270 [(set (match_operand:SVE_FULL_SDI 0 "register_operand")
4271 (unspec:SVE_FULL_SDI
4272 [(match_operand:<VPRED> 1 "register_operand")
4273 (SVE_INT_BINARY_SD:SVE_FULL_SDI
4274 (match_operand:SVE_FULL_SDI 2 "register_operand")
4275 (match_operand:SVE_FULL_SDI 3 "register_operand"))
4276 (match_operand:SVE_FULL_SDI 4 "aarch64_simd_reg_or_zero")]
4277 UNSPEC_SEL))]
4278 "TARGET_SVE"
4279 )
4280
4281 ;; Predicated integer division, merging with the first input.
4282 (define_insn "*cond_<optab><mode>_2"
4283 [(set (match_operand:SVE_FULL_SDI 0 "register_operand" "=w, ?&w")
4284 (unspec:SVE_FULL_SDI
4285 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4286 (SVE_INT_BINARY_SD:SVE_FULL_SDI
4287 (match_operand:SVE_FULL_SDI 2 "register_operand" "0, w")
4288 (match_operand:SVE_FULL_SDI 3 "register_operand" "w, w"))
4289 (match_dup 2)]
4290 UNSPEC_SEL))]
4291 "TARGET_SVE"
4292 "@
4293 <sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4294 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4295 [(set_attr "movprfx" "*,yes")]
4296 )
4297
4298 ;; Predicated integer division, merging with the second input.
4299 (define_insn "*cond_<optab><mode>_3"
4300 [(set (match_operand:SVE_FULL_SDI 0 "register_operand" "=w, ?&w")
4301 (unspec:SVE_FULL_SDI
4302 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4303 (SVE_INT_BINARY_SD:SVE_FULL_SDI
4304 (match_operand:SVE_FULL_SDI 2 "register_operand" "w, w")
4305 (match_operand:SVE_FULL_SDI 3 "register_operand" "0, w"))
4306 (match_dup 3)]
4307 UNSPEC_SEL))]
4308 "TARGET_SVE"
4309 "@
4310 <sve_int_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
4311 movprfx\t%0, %3\;<sve_int_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>"
4312 [(set_attr "movprfx" "*,yes")]
4313 )
4314
4315 ;; Predicated integer division, merging with an independent value.
4316 (define_insn_and_rewrite "*cond_<optab><mode>_any"
4317 [(set (match_operand:SVE_FULL_SDI 0 "register_operand" "=&w, &w, &w, &w, ?&w")
4318 (unspec:SVE_FULL_SDI
4319 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl")
4320 (SVE_INT_BINARY_SD:SVE_FULL_SDI
4321 (match_operand:SVE_FULL_SDI 2 "register_operand" "0, w, w, w, w")
4322 (match_operand:SVE_FULL_SDI 3 "register_operand" "w, 0, w, w, w"))
4323 (match_operand:SVE_FULL_SDI 4 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, 0, w")]
4324 UNSPEC_SEL))]
4325 "TARGET_SVE
4326 && !rtx_equal_p (operands[2], operands[4])
4327 && !rtx_equal_p (operands[3], operands[4])"
4328 "@
4329 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4330 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_int_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
4331 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4332 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4333 #"
4334 "&& reload_completed
4335 && register_operand (operands[4], <MODE>mode)
4336 && !rtx_equal_p (operands[0], operands[4])"
4337 {
4338 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
4339 operands[4], operands[1]));
4340 operands[4] = operands[2] = operands[0];
4341 }
4342 [(set_attr "movprfx" "yes")]
4343 )
4344
4345 ;; -------------------------------------------------------------------------
4346 ;; ---- [INT] Binary logical operations
4347 ;; -------------------------------------------------------------------------
4348 ;; Includes:
4349 ;; - AND
4350 ;; - EOR
4351 ;; - ORR
4352 ;; -------------------------------------------------------------------------
4353
4354 ;; Unpredicated integer binary logical operations.
4355 (define_insn "<optab><mode>3"
4356 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?w, w")
4357 (LOGICAL:SVE_FULL_I
4358 (match_operand:SVE_FULL_I 1 "register_operand" "%0, w, w")
4359 (match_operand:SVE_FULL_I 2 "aarch64_sve_logical_operand" "vsl, vsl, w")))]
4360 "TARGET_SVE"
4361 "@
4362 <logical>\t%0.<Vetype>, %0.<Vetype>, #%C2
4363 movprfx\t%0, %1\;<logical>\t%0.<Vetype>, %0.<Vetype>, #%C2
4364 <logical>\t%0.d, %1.d, %2.d"
4365 [(set_attr "movprfx" "*,yes,*")]
4366 )
4367
4368 ;; Merging forms are handled through SVE_INT_BINARY.
4369
4370 ;; -------------------------------------------------------------------------
4371 ;; ---- [INT] Binary logical operations (inverted second input)
4372 ;; -------------------------------------------------------------------------
4373 ;; Includes:
4374 ;; - BIC
4375 ;; -------------------------------------------------------------------------
4376
4377 ;; Unpredicated BIC.
4378 (define_expand "@aarch64_bic<mode>"
4379 [(set (match_operand:SVE_FULL_I 0 "register_operand")
4380 (and:SVE_FULL_I
4381 (unspec:SVE_FULL_I
4382 [(match_dup 3)
4383 (not:SVE_FULL_I (match_operand:SVE_FULL_I 2 "register_operand"))]
4384 UNSPEC_PRED_X)
4385 (match_operand:SVE_FULL_I 1 "register_operand")))]
4386 "TARGET_SVE"
4387 {
4388 operands[3] = CONSTM1_RTX (<VPRED>mode);
4389 }
4390 )
4391
4392 ;; Predicated BIC.
4393 (define_insn_and_rewrite "*bic<mode>3"
4394 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w")
4395 (and:SVE_FULL_I
4396 (unspec:SVE_FULL_I
4397 [(match_operand 3)
4398 (not:SVE_FULL_I
4399 (match_operand:SVE_FULL_I 2 "register_operand" "w"))]
4400 UNSPEC_PRED_X)
4401 (match_operand:SVE_FULL_I 1 "register_operand" "w")))]
4402 "TARGET_SVE"
4403 "bic\t%0.d, %1.d, %2.d"
4404 "&& !CONSTANT_P (operands[3])"
4405 {
4406 operands[3] = CONSTM1_RTX (<VPRED>mode);
4407 }
4408 )
4409
4410 ;; Predicated BIC with merging.
4411 (define_expand "@cond_bic<mode>"
4412 [(set (match_operand:SVE_FULL_I 0 "register_operand")
4413 (unspec:SVE_FULL_I
4414 [(match_operand:<VPRED> 1 "register_operand")
4415 (and:SVE_FULL_I
4416 (not:SVE_FULL_I (match_operand:SVE_FULL_I 3 "register_operand"))
4417 (match_operand:SVE_FULL_I 2 "register_operand"))
4418 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero")]
4419 UNSPEC_SEL))]
4420 "TARGET_SVE"
4421 )
4422
4423 ;; Predicated integer BIC, merging with the first input.
4424 (define_insn "*cond_bic<mode>_2"
4425 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
4426 (unspec:SVE_FULL_I
4427 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4428 (and:SVE_FULL_I
4429 (not:SVE_FULL_I
4430 (match_operand:SVE_FULL_I 3 "register_operand" "w, w"))
4431 (match_operand:SVE_FULL_I 2 "register_operand" "0, w"))
4432 (match_dup 2)]
4433 UNSPEC_SEL))]
4434 "TARGET_SVE"
4435 "@
4436 bic\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4437 movprfx\t%0, %2\;bic\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4438 [(set_attr "movprfx" "*,yes")]
4439 )
4440
4441 ;; Predicated integer BIC, merging with an independent value.
4442 (define_insn_and_rewrite "*cond_bic<mode>_any"
4443 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, &w, &w, ?&w")
4444 (unspec:SVE_FULL_I
4445 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
4446 (and:SVE_FULL_I
4447 (not:SVE_FULL_I
4448 (match_operand:SVE_FULL_I 3 "register_operand" "w, w, w, w"))
4449 (match_operand:SVE_FULL_I 2 "register_operand" "0, w, w, w"))
4450 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, w")]
4451 UNSPEC_SEL))]
4452 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
4453 "@
4454 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;bic\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4455 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;bic\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4456 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;bic\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4457 #"
4458 "&& reload_completed
4459 && register_operand (operands[4], <MODE>mode)
4460 && !rtx_equal_p (operands[0], operands[4])"
4461 {
4462 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
4463 operands[4], operands[1]));
4464 operands[4] = operands[2] = operands[0];
4465 }
4466 [(set_attr "movprfx" "yes")]
4467 )
4468
4469 ;; -------------------------------------------------------------------------
4470 ;; ---- [INT] Shifts (rounding towards -Inf)
4471 ;; -------------------------------------------------------------------------
4472 ;; Includes:
4473 ;; - ASR
4474 ;; - ASRR
4475 ;; - LSL
4476 ;; - LSLR
4477 ;; - LSR
4478 ;; - LSRR
4479 ;; -------------------------------------------------------------------------
4480
4481 ;; Unpredicated shift by a scalar, which expands into one of the vector
4482 ;; shifts below.
4483 (define_expand "<ASHIFT:optab><mode>3"
4484 [(set (match_operand:SVE_FULL_I 0 "register_operand")
4485 (ASHIFT:SVE_FULL_I
4486 (match_operand:SVE_FULL_I 1 "register_operand")
4487 (match_operand:<VEL> 2 "general_operand")))]
4488 "TARGET_SVE"
4489 {
4490 rtx amount;
4491 if (CONST_INT_P (operands[2]))
4492 {
4493 amount = gen_const_vec_duplicate (<MODE>mode, operands[2]);
4494 if (!aarch64_sve_<lr>shift_operand (operands[2], <MODE>mode))
4495 amount = force_reg (<MODE>mode, amount);
4496 }
4497 else
4498 {
4499 amount = gen_reg_rtx (<MODE>mode);
4500 emit_insn (gen_vec_duplicate<mode> (amount,
4501 convert_to_mode (<VEL>mode,
4502 operands[2], 0)));
4503 }
4504 emit_insn (gen_v<optab><mode>3 (operands[0], operands[1], amount));
4505 DONE;
4506 }
4507 )
4508
4509 ;; Unpredicated shift by a vector.
4510 (define_expand "v<optab><mode>3"
4511 [(set (match_operand:SVE_FULL_I 0 "register_operand")
4512 (unspec:SVE_FULL_I
4513 [(match_dup 3)
4514 (ASHIFT:SVE_FULL_I
4515 (match_operand:SVE_FULL_I 1 "register_operand")
4516 (match_operand:SVE_FULL_I 2 "aarch64_sve_<lr>shift_operand"))]
4517 UNSPEC_PRED_X))]
4518 "TARGET_SVE"
4519 {
4520 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
4521 }
4522 )
4523
4524 ;; Shift by a vector, predicated with a PTRUE. We don't actually need
4525 ;; the predicate for the first alternative, but using Upa or X isn't
4526 ;; likely to gain much and would make the instruction seem less uniform
4527 ;; to the register allocator.
4528 (define_insn_and_split "@aarch64_pred_<optab><mode>"
4529 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, w, w, ?&w")
4530 (unspec:SVE_FULL_I
4531 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
4532 (ASHIFT:SVE_FULL_I
4533 (match_operand:SVE_FULL_I 2 "register_operand" "w, 0, w, w")
4534 (match_operand:SVE_FULL_I 3 "aarch64_sve_<lr>shift_operand" "D<lr>, w, 0, w"))]
4535 UNSPEC_PRED_X))]
4536 "TARGET_SVE"
4537 "@
4538 #
4539 <shift>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4540 <shift>r\t%0.<Vetype>, %1/m, %3.<Vetype>, %2.<Vetype>
4541 movprfx\t%0, %2\;<shift>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4542 "&& reload_completed
4543 && !register_operand (operands[3], <MODE>mode)"
4544 [(set (match_dup 0) (ASHIFT:SVE_FULL_I (match_dup 2) (match_dup 3)))]
4545 ""
4546 [(set_attr "movprfx" "*,*,*,yes")]
4547 )
4548
4549 ;; Unpredicated shift operations by a constant (post-RA only).
4550 ;; These are generated by splitting a predicated instruction whose
4551 ;; predicate is unused.
4552 (define_insn "*post_ra_v<optab><mode>3"
4553 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w")
4554 (ASHIFT:SVE_FULL_I
4555 (match_operand:SVE_FULL_I 1 "register_operand" "w")
4556 (match_operand:SVE_FULL_I 2 "aarch64_simd_<lr>shift_imm")))]
4557 "TARGET_SVE && reload_completed"
4558 "<shift>\t%0.<Vetype>, %1.<Vetype>, #%2"
4559 )
4560
4561 ;; Predicated integer shift, merging with the first input.
4562 (define_insn "*cond_<optab><mode>_2_const"
4563 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
4564 (unspec:SVE_FULL_I
4565 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4566 (ASHIFT:SVE_FULL_I
4567 (match_operand:SVE_FULL_I 2 "register_operand" "0, w")
4568 (match_operand:SVE_FULL_I 3 "aarch64_simd_<lr>shift_imm"))
4569 (match_dup 2)]
4570 UNSPEC_SEL))]
4571 "TARGET_SVE"
4572 "@
4573 <shift>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
4574 movprfx\t%0, %2\;<shift>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3"
4575 [(set_attr "movprfx" "*,yes")]
4576 )
4577
4578 ;; Predicated integer shift, merging with an independent value.
4579 (define_insn_and_rewrite "*cond_<optab><mode>_any_const"
4580 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, &w, ?&w")
4581 (unspec:SVE_FULL_I
4582 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
4583 (ASHIFT:SVE_FULL_I
4584 (match_operand:SVE_FULL_I 2 "register_operand" "w, w, w")
4585 (match_operand:SVE_FULL_I 3 "aarch64_simd_<lr>shift_imm"))
4586 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero" "Dz, 0, w")]
4587 UNSPEC_SEL))]
4588 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
4589 "@
4590 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<shift>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
4591 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<shift>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
4592 #"
4593 "&& reload_completed
4594 && register_operand (operands[4], <MODE>mode)
4595 && !rtx_equal_p (operands[0], operands[4])"
4596 {
4597 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
4598 operands[4], operands[1]));
4599 operands[4] = operands[2] = operands[0];
4600 }
4601 [(set_attr "movprfx" "yes")]
4602 )
4603
4604 ;; Unpredicated shifts of narrow elements by 64-bit amounts.
4605 (define_insn "@aarch64_sve_<sve_int_op><mode>"
4606 [(set (match_operand:SVE_FULL_BHSI 0 "register_operand" "=w")
4607 (unspec:SVE_FULL_BHSI
4608 [(match_operand:SVE_FULL_BHSI 1 "register_operand" "w")
4609 (match_operand:VNx2DI 2 "register_operand" "w")]
4610 SVE_SHIFT_WIDE))]
4611 "TARGET_SVE"
4612 "<sve_int_op>\t%0.<Vetype>, %1.<Vetype>, %2.d"
4613 )
4614
4615 ;; Merging predicated shifts of narrow elements by 64-bit amounts.
4616 (define_expand "@cond_<sve_int_op><mode>"
4617 [(set (match_operand:SVE_FULL_BHSI 0 "register_operand")
4618 (unspec:SVE_FULL_BHSI
4619 [(match_operand:<VPRED> 1 "register_operand")
4620 (unspec:SVE_FULL_BHSI
4621 [(match_operand:SVE_FULL_BHSI 2 "register_operand")
4622 (match_operand:VNx2DI 3 "register_operand")]
4623 SVE_SHIFT_WIDE)
4624 (match_operand:SVE_FULL_BHSI 4 "aarch64_simd_reg_or_zero")]
4625 UNSPEC_SEL))]
4626 "TARGET_SVE"
4627 )
4628
4629 ;; Predicated shifts of narrow elements by 64-bit amounts, merging with
4630 ;; the first input.
4631 (define_insn "*cond_<sve_int_op><mode>_m"
4632 [(set (match_operand:SVE_FULL_BHSI 0 "register_operand" "=w, ?&w")
4633 (unspec:SVE_FULL_BHSI
4634 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4635 (unspec:SVE_FULL_BHSI
4636 [(match_operand:SVE_FULL_BHSI 2 "register_operand" "0, w")
4637 (match_operand:VNx2DI 3 "register_operand" "w, w")]
4638 SVE_SHIFT_WIDE)
4639 (match_dup 2)]
4640 UNSPEC_SEL))]
4641 "TARGET_SVE"
4642 "@
4643 <sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.d
4644 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.d"
4645 [(set_attr "movprfx" "*, yes")])
4646
4647 ;; Predicated shifts of narrow elements by 64-bit amounts, merging with zero.
4648 (define_insn "*cond_<sve_int_op><mode>_z"
4649 [(set (match_operand:SVE_FULL_BHSI 0 "register_operand" "=&w, &w")
4650 (unspec:SVE_FULL_BHSI
4651 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4652 (unspec:SVE_FULL_BHSI
4653 [(match_operand:SVE_FULL_BHSI 2 "register_operand" "0, w")
4654 (match_operand:VNx2DI 3 "register_operand" "w, w")]
4655 SVE_SHIFT_WIDE)
4656 (match_operand:SVE_FULL_BHSI 4 "aarch64_simd_imm_zero")]
4657 UNSPEC_SEL))]
4658 "TARGET_SVE"
4659 "@
4660 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.d
4661 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.d"
4662 [(set_attr "movprfx" "yes")])
4663
4664 ;; -------------------------------------------------------------------------
4665 ;; ---- [INT] Shifts (rounding towards 0)
4666 ;; -------------------------------------------------------------------------
4667 ;; Includes:
4668 ;; - ASRD
4669 ;; - SQSHLU (SVE2)
4670 ;; - SRSHR (SVE2)
4671 ;; - URSHR (SVE2)
4672 ;; -------------------------------------------------------------------------
4673
4674 ;; Unpredicated <SVE_INT_OP>.
4675 (define_expand "sdiv_pow2<mode>3"
4676 [(set (match_operand:SVE_FULL_I 0 "register_operand")
4677 (unspec:SVE_FULL_I
4678 [(match_dup 3)
4679 (unspec:SVE_FULL_I
4680 [(match_operand:SVE_FULL_I 1 "register_operand")
4681 (match_operand 2 "aarch64_simd_rshift_imm")]
4682 UNSPEC_ASRD)
4683 (match_dup 1)]
4684 UNSPEC_SEL))]
4685 "TARGET_SVE"
4686 {
4687 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
4688 }
4689 )
4690
4691 ;; Predicated right shift with merging.
4692 (define_expand "@cond_<sve_int_op><mode>"
4693 [(set (match_operand:SVE_FULL_I 0 "register_operand")
4694 (unspec:SVE_FULL_I
4695 [(match_operand:<VPRED> 1 "register_operand")
4696 (unspec:SVE_FULL_I
4697 [(match_operand:SVE_FULL_I 2 "register_operand")
4698 (match_operand:SVE_FULL_I 3 "aarch64_simd_<lr>shift_imm")]
4699 SVE_INT_SHIFT_IMM)
4700 (match_operand:SVE_FULL_I 4 "aarch64_simd_reg_or_zero")]
4701 UNSPEC_SEL))]
4702 "TARGET_SVE"
4703 )
4704
4705 ;; Predicated right shift, merging with the first input.
4706 (define_insn "*cond_<sve_int_op><mode>_2"
4707 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
4708 (unspec:SVE_FULL_I
4709 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4710 (unspec:SVE_FULL_I
4711 [(match_operand:SVE_FULL_I 2 "register_operand" "0, w")
4712 (match_operand:SVE_FULL_I 3 "aarch64_simd_<lr>shift_imm")]
4713 SVE_INT_SHIFT_IMM)
4714 (match_dup 2)]
4715 UNSPEC_SEL))]
4716 "TARGET_SVE"
4717 "@
4718 <sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
4719 movprfx\t%0, %2\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3"
4720 [(set_attr "movprfx" "*,yes")])
4721
4722 ;; Predicated right shift, merging with zero.
4723 (define_insn "*cond_<sve_int_op><mode>_z"
4724 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w")
4725 (unspec:SVE_FULL_I
4726 [(match_operand:<VPRED> 1 "register_operand" "Upl")
4727 (unspec:SVE_FULL_I
4728 [(match_operand:SVE_FULL_I 2 "register_operand" "w")
4729 (match_operand:SVE_FULL_I 3 "aarch64_simd_<lr>shift_imm")]
4730 SVE_INT_SHIFT_IMM)
4731 (match_operand:SVE_FULL_I 4 "aarch64_simd_imm_zero")]
4732 UNSPEC_SEL))]
4733 "TARGET_SVE"
4734 "movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_int_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3"
4735 [(set_attr "movprfx" "yes")])
4736
4737 ;; -------------------------------------------------------------------------
4738 ;; ---- [FP<-INT] General binary arithmetic corresponding to unspecs
4739 ;; -------------------------------------------------------------------------
4740 ;; Includes:
4741 ;; - FSCALE
4742 ;; - FTSMUL
4743 ;; - FTSSEL
4744 ;; -------------------------------------------------------------------------
4745
4746 ;; Unpredicated floating-point binary operations that take an integer as
4747 ;; their second operand.
4748 (define_insn "@aarch64_sve_<optab><mode>"
4749 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w")
4750 (unspec:SVE_FULL_F
4751 [(match_operand:SVE_FULL_F 1 "register_operand" "w")
4752 (match_operand:<V_INT_EQUIV> 2 "register_operand" "w")]
4753 SVE_FP_BINARY_INT))]
4754 "TARGET_SVE"
4755 "<sve_fp_op>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
4756 )
4757
4758 ;; Predicated floating-point binary operations that take an integer
4759 ;; as their second operand.
4760 (define_insn "@aarch64_pred_<optab><mode>"
4761 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
4762 (unspec:SVE_FULL_F
4763 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4764 (match_operand:SI 4 "aarch64_sve_gp_strictness")
4765 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
4766 (match_operand:<V_INT_EQUIV> 3 "register_operand" "w, w")]
4767 SVE_COND_FP_BINARY_INT))]
4768 "TARGET_SVE"
4769 "@
4770 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4771 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4772 [(set_attr "movprfx" "*,yes")]
4773 )
4774
4775 ;; Predicated floating-point binary operations with merging, taking an
4776 ;; integer as their second operand.
4777 (define_expand "@cond_<optab><mode>"
4778 [(set (match_operand:SVE_FULL_F 0 "register_operand")
4779 (unspec:SVE_FULL_F
4780 [(match_operand:<VPRED> 1 "register_operand")
4781 (unspec:SVE_FULL_F
4782 [(match_dup 1)
4783 (const_int SVE_STRICT_GP)
4784 (match_operand:SVE_FULL_F 2 "register_operand")
4785 (match_operand:<V_INT_EQUIV> 3 "register_operand")]
4786 SVE_COND_FP_BINARY_INT)
4787 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero")]
4788 UNSPEC_SEL))]
4789 "TARGET_SVE"
4790 )
4791
4792 ;; Predicated floating-point binary operations that take an integer as their
4793 ;; second operand, with inactive lanes coming from the first operand.
4794 (define_insn_and_rewrite "*cond_<optab><mode>_2_relaxed"
4795 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
4796 (unspec:SVE_FULL_F
4797 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4798 (unspec:SVE_FULL_F
4799 [(match_operand 4)
4800 (const_int SVE_RELAXED_GP)
4801 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
4802 (match_operand:<V_INT_EQUIV> 3 "register_operand" "w, w")]
4803 SVE_COND_FP_BINARY_INT)
4804 (match_dup 2)]
4805 UNSPEC_SEL))]
4806 "TARGET_SVE"
4807 "@
4808 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4809 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4810 "&& !rtx_equal_p (operands[1], operands[4])"
4811 {
4812 operands[4] = copy_rtx (operands[1]);
4813 }
4814 [(set_attr "movprfx" "*,yes")]
4815 )
4816
4817 (define_insn "*cond_<optab><mode>_2_strict"
4818 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
4819 (unspec:SVE_FULL_F
4820 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
4821 (unspec:SVE_FULL_F
4822 [(match_dup 1)
4823 (const_int SVE_STRICT_GP)
4824 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
4825 (match_operand:<V_INT_EQUIV> 3 "register_operand" "w, w")]
4826 SVE_COND_FP_BINARY_INT)
4827 (match_dup 2)]
4828 UNSPEC_SEL))]
4829 "TARGET_SVE"
4830 "@
4831 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4832 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4833 [(set_attr "movprfx" "*,yes")]
4834 )
4835
4836 ;; Predicated floating-point binary operations that take an integer as
4837 ;; their second operand, with the values of inactive lanes being distinct
4838 ;; from the other inputs.
4839 (define_insn_and_rewrite "*cond_<optab><mode>_any_relaxed"
4840 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, ?&w")
4841 (unspec:SVE_FULL_F
4842 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
4843 (unspec:SVE_FULL_F
4844 [(match_operand 5)
4845 (const_int SVE_RELAXED_GP)
4846 (match_operand:SVE_FULL_F 2 "register_operand" "0, w, w, w")
4847 (match_operand:<V_INT_EQUIV> 3 "register_operand" "w, w, w, w")]
4848 SVE_COND_FP_BINARY_INT)
4849 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, w")]
4850 UNSPEC_SEL))]
4851 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
4852 "@
4853 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4854 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4855 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4856 #"
4857 "&& 1"
4858 {
4859 if (reload_completed
4860 && register_operand (operands[4], <MODE>mode)
4861 && !rtx_equal_p (operands[0], operands[4]))
4862 {
4863 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
4864 operands[4], operands[1]));
4865 operands[4] = operands[2] = operands[0];
4866 }
4867 else if (!rtx_equal_p (operands[1], operands[5]))
4868 operands[5] = copy_rtx (operands[1]);
4869 else
4870 FAIL;
4871 }
4872 [(set_attr "movprfx" "yes")]
4873 )
4874
4875 (define_insn_and_rewrite "*cond_<optab><mode>_any_strict"
4876 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, ?&w")
4877 (unspec:SVE_FULL_F
4878 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
4879 (unspec:SVE_FULL_F
4880 [(match_dup 1)
4881 (const_int SVE_STRICT_GP)
4882 (match_operand:SVE_FULL_F 2 "register_operand" "0, w, w, w")
4883 (match_operand:<V_INT_EQUIV> 3 "register_operand" "w, w, w, w")]
4884 SVE_COND_FP_BINARY_INT)
4885 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, w")]
4886 UNSPEC_SEL))]
4887 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
4888 "@
4889 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4890 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4891 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4892 #"
4893 "&& reload_completed
4894 && register_operand (operands[4], <MODE>mode)
4895 && !rtx_equal_p (operands[0], operands[4])"
4896 {
4897 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
4898 operands[4], operands[1]));
4899 operands[4] = operands[2] = operands[0];
4900 }
4901 [(set_attr "movprfx" "yes")]
4902 )
4903
4904 ;; -------------------------------------------------------------------------
4905 ;; ---- [FP] General binary arithmetic corresponding to rtx codes
4906 ;; -------------------------------------------------------------------------
4907 ;; Includes post-RA forms of:
4908 ;; - FADD
4909 ;; - FMUL
4910 ;; - FSUB
4911 ;; -------------------------------------------------------------------------
4912
4913 ;; Unpredicated floating-point binary operations (post-RA only).
4914 ;; These are generated by splitting a predicated instruction whose
4915 ;; predicate is unused.
4916 (define_insn "*post_ra_<sve_fp_op><mode>3"
4917 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w")
4918 (SVE_UNPRED_FP_BINARY:SVE_FULL_F
4919 (match_operand:SVE_FULL_F 1 "register_operand" "w")
4920 (match_operand:SVE_FULL_F 2 "register_operand" "w")))]
4921 "TARGET_SVE && reload_completed"
4922 "<sve_fp_op>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>")
4923
4924 ;; -------------------------------------------------------------------------
4925 ;; ---- [FP] General binary arithmetic corresponding to unspecs
4926 ;; -------------------------------------------------------------------------
4927 ;; Includes merging forms of:
4928 ;; - FADD (constant forms handled in the "Addition" section)
4929 ;; - FDIV
4930 ;; - FDIVR
4931 ;; - FMAX
4932 ;; - FMAXNM (including #0.0 and #1.0)
4933 ;; - FMIN
4934 ;; - FMINNM (including #0.0 and #1.0)
4935 ;; - FMUL (including #0.5 and #2.0)
4936 ;; - FMULX
4937 ;; - FRECPS
4938 ;; - FRSQRTS
4939 ;; - FSUB (constant forms handled in the "Addition" section)
4940 ;; - FSUBR (constant forms handled in the "Subtraction" section)
4941 ;; -------------------------------------------------------------------------
4942
4943 ;; Unpredicated floating-point binary operations.
4944 (define_insn "@aarch64_sve_<optab><mode>"
4945 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w")
4946 (unspec:SVE_FULL_F
4947 [(match_operand:SVE_FULL_F 1 "register_operand" "w")
4948 (match_operand:SVE_FULL_F 2 "register_operand" "w")]
4949 SVE_FP_BINARY))]
4950 "TARGET_SVE"
4951 "<sve_fp_op>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
4952 )
4953
4954 ;; Unpredicated floating-point binary operations that need to be predicated
4955 ;; for SVE.
4956 (define_expand "<optab><mode>3"
4957 [(set (match_operand:SVE_FULL_F 0 "register_operand")
4958 (unspec:SVE_FULL_F
4959 [(match_dup 3)
4960 (const_int SVE_RELAXED_GP)
4961 (match_operand:SVE_FULL_F 1 "<sve_pred_fp_rhs1_operand>")
4962 (match_operand:SVE_FULL_F 2 "<sve_pred_fp_rhs2_operand>")]
4963 SVE_COND_FP_BINARY_OPTAB))]
4964 "TARGET_SVE"
4965 {
4966 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
4967 }
4968 )
4969
4970 ;; Predicated floating-point binary operations that have no immediate forms.
4971 (define_insn "@aarch64_pred_<optab><mode>"
4972 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?&w")
4973 (unspec:SVE_FULL_F
4974 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
4975 (match_operand:SI 4 "aarch64_sve_gp_strictness")
4976 (match_operand:SVE_FULL_F 2 "register_operand" "0, w, w")
4977 (match_operand:SVE_FULL_F 3 "register_operand" "w, 0, w")]
4978 SVE_COND_FP_BINARY_REG))]
4979 "TARGET_SVE"
4980 "@
4981 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
4982 <sve_fp_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
4983 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
4984 [(set_attr "movprfx" "*,*,yes")]
4985 )
4986
4987 ;; Predicated floating-point operations with merging.
4988 (define_expand "@cond_<optab><mode>"
4989 [(set (match_operand:SVE_FULL_F 0 "register_operand")
4990 (unspec:SVE_FULL_F
4991 [(match_operand:<VPRED> 1 "register_operand")
4992 (unspec:SVE_FULL_F
4993 [(match_dup 1)
4994 (const_int SVE_STRICT_GP)
4995 (match_operand:SVE_FULL_F 2 "<sve_pred_fp_rhs1_operand>")
4996 (match_operand:SVE_FULL_F 3 "<sve_pred_fp_rhs2_operand>")]
4997 SVE_COND_FP_BINARY)
4998 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero")]
4999 UNSPEC_SEL))]
5000 "TARGET_SVE"
5001 )
5002
5003 ;; Predicated floating-point operations, merging with the first input.
5004 (define_insn_and_rewrite "*cond_<optab><mode>_2_relaxed"
5005 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5006 (unspec:SVE_FULL_F
5007 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5008 (unspec:SVE_FULL_F
5009 [(match_operand 4)
5010 (const_int SVE_RELAXED_GP)
5011 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5012 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5013 SVE_COND_FP_BINARY)
5014 (match_dup 2)]
5015 UNSPEC_SEL))]
5016 "TARGET_SVE"
5017 "@
5018 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5019 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
5020 "&& !rtx_equal_p (operands[1], operands[4])"
5021 {
5022 operands[4] = copy_rtx (operands[1]);
5023 }
5024 [(set_attr "movprfx" "*,yes")]
5025 )
5026
5027 (define_insn "*cond_<optab><mode>_2_strict"
5028 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5029 (unspec:SVE_FULL_F
5030 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5031 (unspec:SVE_FULL_F
5032 [(match_dup 1)
5033 (const_int SVE_STRICT_GP)
5034 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5035 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5036 SVE_COND_FP_BINARY)
5037 (match_dup 2)]
5038 UNSPEC_SEL))]
5039 "TARGET_SVE"
5040 "@
5041 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5042 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
5043 [(set_attr "movprfx" "*,yes")]
5044 )
5045
5046 ;; Same for operations that take a 1-bit constant.
5047 (define_insn_and_rewrite "*cond_<optab><mode>_2_const_relaxed"
5048 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?w")
5049 (unspec:SVE_FULL_F
5050 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5051 (unspec:SVE_FULL_F
5052 [(match_operand 4)
5053 (const_int SVE_RELAXED_GP)
5054 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5055 (match_operand:SVE_FULL_F 3 "<sve_pred_fp_rhs2_immediate>")]
5056 SVE_COND_FP_BINARY_I1)
5057 (match_dup 2)]
5058 UNSPEC_SEL))]
5059 "TARGET_SVE"
5060 "@
5061 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5062 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3"
5063 "&& !rtx_equal_p (operands[1], operands[4])"
5064 {
5065 operands[4] = copy_rtx (operands[1]);
5066 }
5067 [(set_attr "movprfx" "*,yes")]
5068 )
5069
5070 (define_insn "*cond_<optab><mode>_2_const_strict"
5071 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?w")
5072 (unspec:SVE_FULL_F
5073 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5074 (unspec:SVE_FULL_F
5075 [(match_dup 1)
5076 (const_int SVE_STRICT_GP)
5077 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5078 (match_operand:SVE_FULL_F 3 "<sve_pred_fp_rhs2_immediate>")]
5079 SVE_COND_FP_BINARY_I1)
5080 (match_dup 2)]
5081 UNSPEC_SEL))]
5082 "TARGET_SVE"
5083 "@
5084 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5085 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3"
5086 [(set_attr "movprfx" "*,yes")]
5087 )
5088
5089 ;; Predicated floating-point operations, merging with the second input.
5090 (define_insn_and_rewrite "*cond_<optab><mode>_3_relaxed"
5091 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5092 (unspec:SVE_FULL_F
5093 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5094 (unspec:SVE_FULL_F
5095 [(match_operand 4)
5096 (const_int SVE_RELAXED_GP)
5097 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
5098 (match_operand:SVE_FULL_F 3 "register_operand" "0, w")]
5099 SVE_COND_FP_BINARY)
5100 (match_dup 3)]
5101 UNSPEC_SEL))]
5102 "TARGET_SVE"
5103 "@
5104 <sve_fp_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5105 movprfx\t%0, %3\;<sve_fp_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>"
5106 "&& !rtx_equal_p (operands[1], operands[4])"
5107 {
5108 operands[4] = copy_rtx (operands[1]);
5109 }
5110 [(set_attr "movprfx" "*,yes")]
5111 )
5112
5113 (define_insn "*cond_<optab><mode>_3_strict"
5114 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5115 (unspec:SVE_FULL_F
5116 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5117 (unspec:SVE_FULL_F
5118 [(match_dup 1)
5119 (const_int SVE_STRICT_GP)
5120 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
5121 (match_operand:SVE_FULL_F 3 "register_operand" "0, w")]
5122 SVE_COND_FP_BINARY)
5123 (match_dup 3)]
5124 UNSPEC_SEL))]
5125 "TARGET_SVE"
5126 "@
5127 <sve_fp_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5128 movprfx\t%0, %3\;<sve_fp_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>"
5129 [(set_attr "movprfx" "*,yes")]
5130 )
5131
5132 ;; Predicated floating-point operations, merging with an independent value.
5133 (define_insn_and_rewrite "*cond_<optab><mode>_any_relaxed"
5134 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, &w, ?&w")
5135 (unspec:SVE_FULL_F
5136 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl")
5137 (unspec:SVE_FULL_F
5138 [(match_operand 5)
5139 (const_int SVE_RELAXED_GP)
5140 (match_operand:SVE_FULL_F 2 "register_operand" "0, w, w, w, w")
5141 (match_operand:SVE_FULL_F 3 "register_operand" "w, 0, w, w, w")]
5142 SVE_COND_FP_BINARY)
5143 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, 0, w")]
5144 UNSPEC_SEL))]
5145 "TARGET_SVE
5146 && !rtx_equal_p (operands[2], operands[4])
5147 && !rtx_equal_p (operands[3], operands[4])"
5148 "@
5149 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5150 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fp_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5151 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5152 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5153 #"
5154 "&& 1"
5155 {
5156 if (reload_completed
5157 && register_operand (operands[4], <MODE>mode)
5158 && !rtx_equal_p (operands[0], operands[4]))
5159 {
5160 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
5161 operands[4], operands[1]));
5162 operands[4] = operands[2] = operands[0];
5163 }
5164 else if (!rtx_equal_p (operands[1], operands[5]))
5165 operands[5] = copy_rtx (operands[1]);
5166 else
5167 FAIL;
5168 }
5169 [(set_attr "movprfx" "yes")]
5170 )
5171
5172 (define_insn_and_rewrite "*cond_<optab><mode>_any_strict"
5173 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, &w, ?&w")
5174 (unspec:SVE_FULL_F
5175 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl")
5176 (unspec:SVE_FULL_F
5177 [(match_dup 1)
5178 (const_int SVE_STRICT_GP)
5179 (match_operand:SVE_FULL_F 2 "register_operand" "0, w, w, w, w")
5180 (match_operand:SVE_FULL_F 3 "register_operand" "w, 0, w, w, w")]
5181 SVE_COND_FP_BINARY)
5182 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, 0, w")]
5183 UNSPEC_SEL))]
5184 "TARGET_SVE
5185 && !rtx_equal_p (operands[2], operands[4])
5186 && !rtx_equal_p (operands[3], operands[4])"
5187 "@
5188 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5189 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fp_op_rev>\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5190 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5191 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5192 #"
5193 "&& reload_completed
5194 && register_operand (operands[4], <MODE>mode)
5195 && !rtx_equal_p (operands[0], operands[4])"
5196 {
5197 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
5198 operands[4], operands[1]));
5199 operands[4] = operands[2] = operands[0];
5200 }
5201 [(set_attr "movprfx" "yes")]
5202 )
5203
5204 ;; Same for operations that take a 1-bit constant.
5205 (define_insn_and_rewrite "*cond_<optab><mode>_any_const_relaxed"
5206 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?w")
5207 (unspec:SVE_FULL_F
5208 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
5209 (unspec:SVE_FULL_F
5210 [(match_operand 5)
5211 (const_int SVE_RELAXED_GP)
5212 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w")
5213 (match_operand:SVE_FULL_F 3 "<sve_pred_fp_rhs2_immediate>")]
5214 SVE_COND_FP_BINARY_I1)
5215 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, 0, w")]
5216 UNSPEC_SEL))]
5217 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
5218 "@
5219 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5220 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5221 #"
5222 "&& 1"
5223 {
5224 if (reload_completed
5225 && register_operand (operands[4], <MODE>mode)
5226 && !rtx_equal_p (operands[0], operands[4]))
5227 {
5228 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
5229 operands[4], operands[1]));
5230 operands[4] = operands[2] = operands[0];
5231 }
5232 else if (!rtx_equal_p (operands[1], operands[5]))
5233 operands[5] = copy_rtx (operands[1]);
5234 else
5235 FAIL;
5236 }
5237 [(set_attr "movprfx" "yes")]
5238 )
5239
5240 (define_insn_and_rewrite "*cond_<optab><mode>_any_const_strict"
5241 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?w")
5242 (unspec:SVE_FULL_F
5243 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
5244 (unspec:SVE_FULL_F
5245 [(match_dup 1)
5246 (const_int SVE_STRICT_GP)
5247 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w")
5248 (match_operand:SVE_FULL_F 3 "<sve_pred_fp_rhs2_immediate>")]
5249 SVE_COND_FP_BINARY_I1)
5250 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, 0, w")]
5251 UNSPEC_SEL))]
5252 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
5253 "@
5254 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5255 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5256 #"
5257 "&& reload_completed
5258 && register_operand (operands[4], <MODE>mode)
5259 && !rtx_equal_p (operands[0], operands[4])"
5260 {
5261 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
5262 operands[4], operands[1]));
5263 operands[4] = operands[2] = operands[0];
5264 }
5265 [(set_attr "movprfx" "yes")]
5266 )
5267
5268 ;; -------------------------------------------------------------------------
5269 ;; ---- [FP] Addition
5270 ;; -------------------------------------------------------------------------
5271 ;; Includes:
5272 ;; - FADD
5273 ;; - FSUB
5274 ;; -------------------------------------------------------------------------
5275
5276 ;; Predicated floating-point addition.
5277 (define_insn_and_split "@aarch64_pred_<optab><mode>"
5278 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, w, w, ?&w, ?&w, ?&w")
5279 (unspec:SVE_FULL_F
5280 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl, Upl")
5281 (match_operand:SI 4 "aarch64_sve_gp_strictness" "i, i, Z, Ui1, i, i, Ui1")
5282 (match_operand:SVE_FULL_F 2 "register_operand" "%0, 0, w, 0, w, w, w")
5283 (match_operand:SVE_FULL_F 3 "aarch64_sve_float_arith_with_sub_operand" "vsA, vsN, w, w, vsA, vsN, w")]
5284 SVE_COND_FP_ADD))]
5285 "TARGET_SVE"
5286 "@
5287 fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5288 fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3
5289 #
5290 fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5291 movprfx\t%0, %2\;fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5292 movprfx\t%0, %2\;fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3
5293 movprfx\t%0, %2\;fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
5294 ; Split the unpredicated form after reload, so that we don't have
5295 ; the unnecessary PTRUE.
5296 "&& reload_completed
5297 && register_operand (operands[3], <MODE>mode)
5298 && INTVAL (operands[4]) == SVE_RELAXED_GP"
5299 [(set (match_dup 0) (plus:SVE_FULL_F (match_dup 2) (match_dup 3)))]
5300 ""
5301 [(set_attr "movprfx" "*,*,*,*,yes,yes,yes")]
5302 )
5303
5304 ;; Predicated floating-point addition of a constant, merging with the
5305 ;; first input.
5306 (define_insn_and_rewrite "*cond_add<mode>_2_const_relaxed"
5307 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?w, ?w")
5308 (unspec:SVE_FULL_F
5309 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
5310 (unspec:SVE_FULL_F
5311 [(match_operand 4)
5312 (const_int SVE_RELAXED_GP)
5313 (match_operand:SVE_FULL_F 2 "register_operand" "0, 0, w, w")
5314 (match_operand:SVE_FULL_F 3 "aarch64_sve_float_arith_with_sub_immediate" "vsA, vsN, vsA, vsN")]
5315 UNSPEC_COND_FADD)
5316 (match_dup 2)]
5317 UNSPEC_SEL))]
5318 "TARGET_SVE"
5319 "@
5320 fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5321 fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3
5322 movprfx\t%0, %2\;fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5323 movprfx\t%0, %2\;fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3"
5324 "&& !rtx_equal_p (operands[1], operands[4])"
5325 {
5326 operands[4] = copy_rtx (operands[1]);
5327 }
5328 [(set_attr "movprfx" "*,*,yes,yes")]
5329 )
5330
5331 (define_insn "*cond_add<mode>_2_const_strict"
5332 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?w, ?w")
5333 (unspec:SVE_FULL_F
5334 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
5335 (unspec:SVE_FULL_F
5336 [(match_dup 1)
5337 (const_int SVE_STRICT_GP)
5338 (match_operand:SVE_FULL_F 2 "register_operand" "0, 0, w, w")
5339 (match_operand:SVE_FULL_F 3 "aarch64_sve_float_arith_with_sub_immediate" "vsA, vsN, vsA, vsN")]
5340 UNSPEC_COND_FADD)
5341 (match_dup 2)]
5342 UNSPEC_SEL))]
5343 "TARGET_SVE"
5344 "@
5345 fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5346 fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3
5347 movprfx\t%0, %2\;fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5348 movprfx\t%0, %2\;fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3"
5349 [(set_attr "movprfx" "*,*,yes,yes")]
5350 )
5351
5352 ;; Predicated floating-point addition of a constant, merging with an
5353 ;; independent value.
5354 (define_insn_and_rewrite "*cond_add<mode>_any_const_relaxed"
5355 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, w, w, ?w, ?w")
5356 (unspec:SVE_FULL_F
5357 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
5358 (unspec:SVE_FULL_F
5359 [(match_operand 5)
5360 (const_int SVE_RELAXED_GP)
5361 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w, w, w, w")
5362 (match_operand:SVE_FULL_F 3 "aarch64_sve_float_arith_with_sub_immediate" "vsA, vsN, vsA, vsN, vsA, vsN")]
5363 UNSPEC_COND_FADD)
5364 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, 0, w, w")]
5365 UNSPEC_SEL))]
5366 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
5367 "@
5368 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5369 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3
5370 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5371 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3
5372 #
5373 #"
5374 "&& 1"
5375 {
5376 if (reload_completed
5377 && register_operand (operands[4], <MODE>mode)
5378 && !rtx_equal_p (operands[0], operands[4]))
5379 {
5380 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
5381 operands[4], operands[1]));
5382 operands[4] = operands[2] = operands[0];
5383 }
5384 else if (!rtx_equal_p (operands[1], operands[5]))
5385 operands[5] = copy_rtx (operands[1]);
5386 else
5387 FAIL;
5388 }
5389 [(set_attr "movprfx" "yes")]
5390 )
5391
5392 (define_insn_and_rewrite "*cond_add<mode>_any_const_strict"
5393 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, w, w, ?w, ?w")
5394 (unspec:SVE_FULL_F
5395 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
5396 (unspec:SVE_FULL_F
5397 [(match_dup 1)
5398 (const_int SVE_STRICT_GP)
5399 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w, w, w, w")
5400 (match_operand:SVE_FULL_F 3 "aarch64_sve_float_arith_with_sub_immediate" "vsA, vsN, vsA, vsN, vsA, vsN")]
5401 UNSPEC_COND_FADD)
5402 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, 0, w, w")]
5403 UNSPEC_SEL))]
5404 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
5405 "@
5406 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5407 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3
5408 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;fadd\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
5409 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, #%N3
5410 #
5411 #"
5412 "&& reload_completed
5413 && register_operand (operands[4], <MODE>mode)
5414 && !rtx_equal_p (operands[0], operands[4])"
5415 {
5416 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
5417 operands[4], operands[1]));
5418 operands[4] = operands[2] = operands[0];
5419 }
5420 [(set_attr "movprfx" "yes")]
5421 )
5422
5423 ;; Register merging forms are handled through SVE_COND_FP_BINARY.
5424
5425 ;; -------------------------------------------------------------------------
5426 ;; ---- [FP] Complex addition
5427 ;; -------------------------------------------------------------------------
5428 ;; Includes:
5429 ;; - FCADD
5430 ;; -------------------------------------------------------------------------
5431
5432 ;; Predicated FCADD.
5433 (define_insn "@aarch64_pred_<optab><mode>"
5434 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5435 (unspec:SVE_FULL_F
5436 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5437 (match_operand:SI 4 "aarch64_sve_gp_strictness")
5438 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5439 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5440 SVE_COND_FCADD))]
5441 "TARGET_SVE"
5442 "@
5443 fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5444 movprfx\t%0, %2\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>"
5445 [(set_attr "movprfx" "*,yes")]
5446 )
5447
5448 ;; Predicated FCADD with merging.
5449 (define_expand "@cond_<optab><mode>"
5450 [(set (match_operand:SVE_FULL_F 0 "register_operand")
5451 (unspec:SVE_FULL_F
5452 [(match_operand:<VPRED> 1 "register_operand")
5453 (unspec:SVE_FULL_F
5454 [(match_dup 1)
5455 (const_int SVE_STRICT_GP)
5456 (match_operand:SVE_FULL_F 2 "register_operand")
5457 (match_operand:SVE_FULL_F 3 "register_operand")]
5458 SVE_COND_FCADD)
5459 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero")]
5460 UNSPEC_SEL))]
5461 "TARGET_SVE"
5462 )
5463
5464 ;; Predicated FCADD, merging with the first input.
5465 (define_insn_and_rewrite "*cond_<optab><mode>_2_relaxed"
5466 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5467 (unspec:SVE_FULL_F
5468 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5469 (unspec:SVE_FULL_F
5470 [(match_operand 4)
5471 (const_int SVE_RELAXED_GP)
5472 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5473 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5474 SVE_COND_FCADD)
5475 (match_dup 2)]
5476 UNSPEC_SEL))]
5477 "TARGET_SVE"
5478 "@
5479 fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5480 movprfx\t%0, %2\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>"
5481 "&& !rtx_equal_p (operands[1], operands[4])"
5482 {
5483 operands[4] = copy_rtx (operands[1]);
5484 }
5485 [(set_attr "movprfx" "*,yes")]
5486 )
5487
5488 (define_insn "*cond_<optab><mode>_2_strict"
5489 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5490 (unspec:SVE_FULL_F
5491 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5492 (unspec:SVE_FULL_F
5493 [(match_dup 1)
5494 (const_int SVE_STRICT_GP)
5495 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5496 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5497 SVE_COND_FCADD)
5498 (match_dup 2)]
5499 UNSPEC_SEL))]
5500 "TARGET_SVE"
5501 "@
5502 fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5503 movprfx\t%0, %2\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>"
5504 [(set_attr "movprfx" "*,yes")]
5505 )
5506
5507 ;; Predicated FCADD, merging with an independent value.
5508 (define_insn_and_rewrite "*cond_<optab><mode>_any_relaxed"
5509 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, ?&w")
5510 (unspec:SVE_FULL_F
5511 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
5512 (unspec:SVE_FULL_F
5513 [(match_operand 5)
5514 (const_int SVE_RELAXED_GP)
5515 (match_operand:SVE_FULL_F 2 "register_operand" "w, 0, w, w")
5516 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w, w")]
5517 SVE_COND_FCADD)
5518 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, w")]
5519 UNSPEC_SEL))]
5520 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
5521 "@
5522 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5523 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5524 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5525 #"
5526 "&& 1"
5527 {
5528 if (reload_completed
5529 && register_operand (operands[4], <MODE>mode)
5530 && !rtx_equal_p (operands[0], operands[4]))
5531 {
5532 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
5533 operands[4], operands[1]));
5534 operands[4] = operands[2] = operands[0];
5535 }
5536 else if (!rtx_equal_p (operands[1], operands[5]))
5537 operands[5] = copy_rtx (operands[1]);
5538 else
5539 FAIL;
5540 }
5541 [(set_attr "movprfx" "yes")]
5542 )
5543
5544 (define_insn_and_rewrite "*cond_<optab><mode>_any_strict"
5545 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, ?&w")
5546 (unspec:SVE_FULL_F
5547 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
5548 (unspec:SVE_FULL_F
5549 [(match_dup 1)
5550 (const_int SVE_STRICT_GP)
5551 (match_operand:SVE_FULL_F 2 "register_operand" "w, 0, w, w")
5552 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w, w")]
5553 SVE_COND_FCADD)
5554 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, w")]
5555 UNSPEC_SEL))]
5556 "TARGET_SVE && !rtx_equal_p (operands[2], operands[4])"
5557 "@
5558 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5559 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5560 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;fcadd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>, #<rot>
5561 #"
5562 "&& reload_completed
5563 && register_operand (operands[4], <MODE>mode)
5564 && !rtx_equal_p (operands[0], operands[4])"
5565 {
5566 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[2],
5567 operands[4], operands[1]));
5568 operands[4] = operands[2] = operands[0];
5569 }
5570 [(set_attr "movprfx" "yes")]
5571 )
5572
5573 ;; -------------------------------------------------------------------------
5574 ;; ---- [FP] Subtraction
5575 ;; -------------------------------------------------------------------------
5576 ;; Includes:
5577 ;; - FSUB
5578 ;; - FSUBR
5579 ;; -------------------------------------------------------------------------
5580
5581 ;; Predicated floating-point subtraction.
5582 (define_insn_and_split "@aarch64_pred_<optab><mode>"
5583 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, w, w, ?&w, ?&w")
5584 (unspec:SVE_FULL_F
5585 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
5586 (match_operand:SI 4 "aarch64_sve_gp_strictness" "i, Z, Ui1, Ui1, i, Ui1")
5587 (match_operand:SVE_FULL_F 2 "aarch64_sve_float_arith_operand" "vsA, w, 0, w, vsA, w")
5588 (match_operand:SVE_FULL_F 3 "register_operand" "0, w, w, 0, w, w")]
5589 SVE_COND_FP_SUB))]
5590 "TARGET_SVE"
5591 "@
5592 fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2
5593 #
5594 fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5595 fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5596 movprfx\t%0, %3\;fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2
5597 movprfx\t%0, %2\;fsub\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
5598 ; Split the unpredicated form after reload, so that we don't have
5599 ; the unnecessary PTRUE.
5600 "&& reload_completed
5601 && register_operand (operands[2], <MODE>mode)
5602 && INTVAL (operands[4]) == SVE_RELAXED_GP"
5603 [(set (match_dup 0) (minus:SVE_FULL_F (match_dup 2) (match_dup 3)))]
5604 ""
5605 [(set_attr "movprfx" "*,*,*,*,yes,yes")]
5606 )
5607
5608 ;; Predicated floating-point subtraction from a constant, merging with the
5609 ;; second input.
5610 (define_insn_and_rewrite "*cond_sub<mode>_3_const_relaxed"
5611 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?w")
5612 (unspec:SVE_FULL_F
5613 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5614 (unspec:SVE_FULL_F
5615 [(match_operand 4)
5616 (const_int SVE_RELAXED_GP)
5617 (match_operand:SVE_FULL_F 2 "aarch64_sve_float_arith_immediate")
5618 (match_operand:SVE_FULL_F 3 "register_operand" "0, w")]
5619 UNSPEC_COND_FSUB)
5620 (match_dup 3)]
5621 UNSPEC_SEL))]
5622 "TARGET_SVE"
5623 "@
5624 fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2
5625 movprfx\t%0, %3\;fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2"
5626 "&& !rtx_equal_p (operands[1], operands[4])"
5627 {
5628 operands[4] = copy_rtx (operands[1]);
5629 }
5630 [(set_attr "movprfx" "*,yes")]
5631 )
5632
5633 (define_insn "*cond_sub<mode>_3_const_strict"
5634 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?w")
5635 (unspec:SVE_FULL_F
5636 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5637 (unspec:SVE_FULL_F
5638 [(match_dup 1)
5639 (const_int SVE_STRICT_GP)
5640 (match_operand:SVE_FULL_F 2 "aarch64_sve_float_arith_immediate")
5641 (match_operand:SVE_FULL_F 3 "register_operand" "0, w")]
5642 UNSPEC_COND_FSUB)
5643 (match_dup 3)]
5644 UNSPEC_SEL))]
5645 "TARGET_SVE"
5646 "@
5647 fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2
5648 movprfx\t%0, %3\;fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2"
5649 [(set_attr "movprfx" "*,yes")]
5650 )
5651
5652 ;; Predicated floating-point subtraction from a constant, merging with an
5653 ;; independent value.
5654 (define_insn_and_rewrite "*cond_sub<mode>_const_relaxed"
5655 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?w")
5656 (unspec:SVE_FULL_F
5657 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
5658 (unspec:SVE_FULL_F
5659 [(match_operand 5)
5660 (const_int SVE_RELAXED_GP)
5661 (match_operand:SVE_FULL_F 2 "aarch64_sve_float_arith_immediate")
5662 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w")]
5663 UNSPEC_COND_FSUB)
5664 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, 0, w")]
5665 UNSPEC_SEL))]
5666 "TARGET_SVE && !rtx_equal_p (operands[3], operands[4])"
5667 "@
5668 movprfx\t%0.<Vetype>, %1/z, %3.<Vetype>\;fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2
5669 movprfx\t%0.<Vetype>, %1/m, %3.<Vetype>\;fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2
5670 #"
5671 "&& 1"
5672 {
5673 if (reload_completed
5674 && register_operand (operands[4], <MODE>mode)
5675 && !rtx_equal_p (operands[0], operands[4]))
5676 {
5677 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[3],
5678 operands[4], operands[1]));
5679 operands[4] = operands[3] = operands[0];
5680 }
5681 else if (!rtx_equal_p (operands[1], operands[5]))
5682 operands[5] = copy_rtx (operands[1]);
5683 else
5684 FAIL;
5685 }
5686 [(set_attr "movprfx" "yes")]
5687 )
5688
5689 (define_insn_and_rewrite "*cond_sub<mode>_const_strict"
5690 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?w")
5691 (unspec:SVE_FULL_F
5692 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
5693 (unspec:SVE_FULL_F
5694 [(match_dup 1)
5695 (const_int SVE_STRICT_GP)
5696 (match_operand:SVE_FULL_F 2 "aarch64_sve_float_arith_immediate")
5697 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w")]
5698 UNSPEC_COND_FSUB)
5699 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, 0, w")]
5700 UNSPEC_SEL))]
5701 "TARGET_SVE && !rtx_equal_p (operands[3], operands[4])"
5702 "@
5703 movprfx\t%0.<Vetype>, %1/z, %3.<Vetype>\;fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2
5704 movprfx\t%0.<Vetype>, %1/m, %3.<Vetype>\;fsubr\t%0.<Vetype>, %1/m, %0.<Vetype>, #%2
5705 #"
5706 "&& reload_completed
5707 && register_operand (operands[4], <MODE>mode)
5708 && !rtx_equal_p (operands[0], operands[4])"
5709 {
5710 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[3],
5711 operands[4], operands[1]));
5712 operands[4] = operands[3] = operands[0];
5713 }
5714 [(set_attr "movprfx" "yes")]
5715 )
5716 ;; Register merging forms are handled through SVE_COND_FP_BINARY.
5717
5718 ;; -------------------------------------------------------------------------
5719 ;; ---- [FP] Absolute difference
5720 ;; -------------------------------------------------------------------------
5721 ;; Includes:
5722 ;; - FABD
5723 ;; -------------------------------------------------------------------------
5724
5725 ;; Predicated floating-point absolute difference.
5726 (define_expand "@aarch64_pred_abd<mode>"
5727 [(set (match_operand:SVE_FULL_F 0 "register_operand")
5728 (unspec:SVE_FULL_F
5729 [(match_operand:<VPRED> 1 "register_operand")
5730 (match_operand:SI 4 "aarch64_sve_gp_strictness")
5731 (unspec:SVE_FULL_F
5732 [(match_dup 1)
5733 (match_dup 4)
5734 (match_operand:SVE_FULL_F 2 "register_operand")
5735 (match_operand:SVE_FULL_F 3 "register_operand")]
5736 UNSPEC_COND_FSUB)]
5737 UNSPEC_COND_FABS))]
5738 "TARGET_SVE"
5739 )
5740
5741 ;; Predicated floating-point absolute difference.
5742 (define_insn_and_rewrite "*aarch64_pred_abd<mode>_relaxed"
5743 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5744 (unspec:SVE_FULL_F
5745 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5746 (match_operand:SI 4 "aarch64_sve_gp_strictness")
5747 (unspec:SVE_FULL_F
5748 [(match_operand 5)
5749 (const_int SVE_RELAXED_GP)
5750 (match_operand:SVE_FULL_F 2 "register_operand" "%0, w")
5751 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5752 UNSPEC_COND_FSUB)]
5753 UNSPEC_COND_FABS))]
5754 "TARGET_SVE"
5755 "@
5756 fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5757 movprfx\t%0, %2\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
5758 "&& !rtx_equal_p (operands[1], operands[5])"
5759 {
5760 operands[5] = copy_rtx (operands[1]);
5761 }
5762 [(set_attr "movprfx" "*,yes")]
5763 )
5764
5765 (define_insn "*aarch64_pred_abd<mode>_strict"
5766 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5767 (unspec:SVE_FULL_F
5768 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5769 (match_operand:SI 4 "aarch64_sve_gp_strictness")
5770 (unspec:SVE_FULL_F
5771 [(match_dup 1)
5772 (const_int SVE_STRICT_GP)
5773 (match_operand:SVE_FULL_F 2 "register_operand" "%0, w")
5774 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5775 UNSPEC_COND_FSUB)]
5776 UNSPEC_COND_FABS))]
5777 "TARGET_SVE"
5778 "@
5779 fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5780 movprfx\t%0, %2\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
5781 [(set_attr "movprfx" "*,yes")]
5782 )
5783
5784 (define_expand "@aarch64_cond_abd<mode>"
5785 [(set (match_operand:SVE_FULL_F 0 "register_operand")
5786 (unspec:SVE_FULL_F
5787 [(match_operand:<VPRED> 1 "register_operand")
5788 (unspec:SVE_FULL_F
5789 [(match_dup 1)
5790 (const_int SVE_STRICT_GP)
5791 (unspec:SVE_FULL_F
5792 [(match_dup 1)
5793 (const_int SVE_STRICT_GP)
5794 (match_operand:SVE_FULL_F 2 "register_operand")
5795 (match_operand:SVE_FULL_F 3 "register_operand")]
5796 UNSPEC_COND_FSUB)]
5797 UNSPEC_COND_FABS)
5798 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero")]
5799 UNSPEC_SEL))]
5800 "TARGET_SVE"
5801 {
5802 if (rtx_equal_p (operands[3], operands[4]))
5803 std::swap (operands[2], operands[3]);
5804 })
5805
5806 ;; Predicated floating-point absolute difference, merging with the first
5807 ;; input.
5808 (define_insn_and_rewrite "*aarch64_cond_abd<mode>_2_relaxed"
5809 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5810 (unspec:SVE_FULL_F
5811 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5812 (unspec:SVE_FULL_F
5813 [(match_operand 4)
5814 (const_int SVE_RELAXED_GP)
5815 (unspec:SVE_FULL_F
5816 [(match_operand 5)
5817 (const_int SVE_RELAXED_GP)
5818 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5819 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5820 UNSPEC_COND_FSUB)]
5821 UNSPEC_COND_FABS)
5822 (match_dup 2)]
5823 UNSPEC_SEL))]
5824 "TARGET_SVE"
5825 "@
5826 fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5827 movprfx\t%0, %2\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
5828 "&& (!rtx_equal_p (operands[1], operands[4])
5829 || !rtx_equal_p (operands[1], operands[5]))"
5830 {
5831 operands[4] = copy_rtx (operands[1]);
5832 operands[5] = copy_rtx (operands[1]);
5833 }
5834 [(set_attr "movprfx" "*,yes")]
5835 )
5836
5837 (define_insn "*aarch64_cond_abd<mode>_2_strict"
5838 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5839 (unspec:SVE_FULL_F
5840 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5841 (unspec:SVE_FULL_F
5842 [(match_dup 1)
5843 (match_operand:SI 4 "aarch64_sve_gp_strictness")
5844 (unspec:SVE_FULL_F
5845 [(match_dup 1)
5846 (match_operand:SI 5 "aarch64_sve_gp_strictness")
5847 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
5848 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")]
5849 UNSPEC_COND_FSUB)]
5850 UNSPEC_COND_FABS)
5851 (match_dup 2)]
5852 UNSPEC_SEL))]
5853 "TARGET_SVE"
5854 "@
5855 fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5856 movprfx\t%0, %2\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
5857 [(set_attr "movprfx" "*,yes")]
5858 )
5859
5860 ;; Predicated floating-point absolute difference, merging with the second
5861 ;; input.
5862 (define_insn_and_rewrite "*aarch64_cond_abd<mode>_3_relaxed"
5863 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5864 (unspec:SVE_FULL_F
5865 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5866 (unspec:SVE_FULL_F
5867 [(match_operand 4)
5868 (const_int SVE_RELAXED_GP)
5869 (unspec:SVE_FULL_F
5870 [(match_operand 5)
5871 (const_int SVE_RELAXED_GP)
5872 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
5873 (match_operand:SVE_FULL_F 3 "register_operand" "0, w")]
5874 UNSPEC_COND_FSUB)]
5875 UNSPEC_COND_FABS)
5876 (match_dup 3)]
5877 UNSPEC_SEL))]
5878 "TARGET_SVE"
5879 "@
5880 fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5881 movprfx\t%0, %3\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>"
5882 "&& (!rtx_equal_p (operands[1], operands[4])
5883 || !rtx_equal_p (operands[1], operands[5]))"
5884 {
5885 operands[4] = copy_rtx (operands[1]);
5886 operands[5] = copy_rtx (operands[1]);
5887 }
5888 [(set_attr "movprfx" "*,yes")]
5889 )
5890
5891 (define_insn "*aarch64_cond_abd<mode>_3_strict"
5892 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
5893 (unspec:SVE_FULL_F
5894 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
5895 (unspec:SVE_FULL_F
5896 [(match_dup 1)
5897 (match_operand:SI 4 "aarch64_sve_gp_strictness")
5898 (unspec:SVE_FULL_F
5899 [(match_dup 1)
5900 (match_operand:SI 5 "aarch64_sve_gp_strictness")
5901 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
5902 (match_operand:SVE_FULL_F 3 "register_operand" "0, w")]
5903 UNSPEC_COND_FSUB)]
5904 UNSPEC_COND_FABS)
5905 (match_dup 3)]
5906 UNSPEC_SEL))]
5907 "TARGET_SVE"
5908 "@
5909 fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5910 movprfx\t%0, %3\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>"
5911 [(set_attr "movprfx" "*,yes")]
5912 )
5913
5914 ;; Predicated floating-point absolute difference, merging with an
5915 ;; independent value.
5916 (define_insn_and_rewrite "*aarch64_cond_abd<mode>_any_relaxed"
5917 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, &w, ?&w")
5918 (unspec:SVE_FULL_F
5919 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl")
5920 (unspec:SVE_FULL_F
5921 [(match_operand 5)
5922 (const_int SVE_RELAXED_GP)
5923 (unspec:SVE_FULL_F
5924 [(match_operand 6)
5925 (const_int SVE_RELAXED_GP)
5926 (match_operand:SVE_FULL_F 2 "register_operand" "0, w, w, w, w")
5927 (match_operand:SVE_FULL_F 3 "register_operand" "w, 0, w, w, w")]
5928 UNSPEC_COND_FSUB)]
5929 UNSPEC_COND_FABS)
5930 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, 0, w")]
5931 UNSPEC_SEL))]
5932 "TARGET_SVE
5933 && !rtx_equal_p (operands[2], operands[4])
5934 && !rtx_equal_p (operands[3], operands[4])"
5935 "@
5936 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5937 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5938 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5939 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5940 #"
5941 "&& 1"
5942 {
5943 if (reload_completed
5944 && register_operand (operands[4], <MODE>mode)
5945 && !rtx_equal_p (operands[0], operands[4]))
5946 {
5947 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[3],
5948 operands[4], operands[1]));
5949 operands[4] = operands[3] = operands[0];
5950 }
5951 else if (!rtx_equal_p (operands[1], operands[5])
5952 || !rtx_equal_p (operands[1], operands[6]))
5953 {
5954 operands[5] = copy_rtx (operands[1]);
5955 operands[6] = copy_rtx (operands[1]);
5956 }
5957 else
5958 FAIL;
5959 }
5960 [(set_attr "movprfx" "yes")]
5961 )
5962
5963 (define_insn_and_rewrite "*aarch64_cond_abd<mode>_any_strict"
5964 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, &w, ?&w")
5965 (unspec:SVE_FULL_F
5966 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl")
5967 (unspec:SVE_FULL_F
5968 [(match_dup 1)
5969 (match_operand:SI 5 "aarch64_sve_gp_strictness")
5970 (unspec:SVE_FULL_F
5971 [(match_dup 1)
5972 (match_operand:SI 6 "aarch64_sve_gp_strictness")
5973 (match_operand:SVE_FULL_F 2 "register_operand" "0, w, w, w, w")
5974 (match_operand:SVE_FULL_F 3 "register_operand" "w, 0, w, w, w")]
5975 UNSPEC_COND_FSUB)]
5976 UNSPEC_COND_FABS)
5977 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, 0, w")]
5978 UNSPEC_SEL))]
5979 "TARGET_SVE
5980 && !rtx_equal_p (operands[2], operands[4])
5981 && !rtx_equal_p (operands[3], operands[4])"
5982 "@
5983 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5984 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %2.<Vetype>
5985 movprfx\t%0.<Vetype>, %1/z, %2.<Vetype>\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5986 movprfx\t%0.<Vetype>, %1/m, %2.<Vetype>\;fabd\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
5987 #"
5988 "&& reload_completed
5989 && register_operand (operands[4], <MODE>mode)
5990 && !rtx_equal_p (operands[0], operands[4])"
5991 {
5992 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[3],
5993 operands[4], operands[1]));
5994 operands[4] = operands[3] = operands[0];
5995 }
5996 [(set_attr "movprfx" "yes")]
5997 )
5998
5999 ;; -------------------------------------------------------------------------
6000 ;; ---- [FP] Multiplication
6001 ;; -------------------------------------------------------------------------
6002 ;; Includes:
6003 ;; - FMUL
6004 ;; -------------------------------------------------------------------------
6005
6006 ;; Predicated floating-point multiplication.
6007 (define_insn_and_split "@aarch64_pred_<optab><mode>"
6008 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, w, ?&w, ?&w")
6009 (unspec:SVE_FULL_F
6010 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl")
6011 (match_operand:SI 4 "aarch64_sve_gp_strictness" "i, Z, Ui1, i, Ui1")
6012 (match_operand:SVE_FULL_F 2 "register_operand" "%0, w, 0, w, w")
6013 (match_operand:SVE_FULL_F 3 "aarch64_sve_float_mul_operand" "vsM, w, w, vsM, w")]
6014 SVE_COND_FP_MUL))]
6015 "TARGET_SVE"
6016 "@
6017 fmul\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
6018 #
6019 fmul\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
6020 movprfx\t%0, %2\;fmul\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
6021 movprfx\t%0, %2\;fmul\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
6022 ; Split the unpredicated form after reload, so that we don't have
6023 ; the unnecessary PTRUE.
6024 "&& reload_completed
6025 && register_operand (operands[3], <MODE>mode)
6026 && INTVAL (operands[4]) == SVE_RELAXED_GP"
6027 [(set (match_dup 0) (mult:SVE_FULL_F (match_dup 2) (match_dup 3)))]
6028 ""
6029 [(set_attr "movprfx" "*,*,*,yes,yes")]
6030 )
6031
6032 ;; Merging forms are handled through SVE_COND_FP_BINARY and
6033 ;; SVE_COND_FP_BINARY_I1.
6034
6035 ;; Unpredicated multiplication by selected lanes.
6036 (define_insn "@aarch64_mul_lane_<mode>"
6037 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w")
6038 (mult:SVE_FULL_F
6039 (unspec:SVE_FULL_F
6040 [(match_operand:SVE_FULL_F 2 "register_operand" "<sve_lane_con>")
6041 (match_operand:SI 3 "const_int_operand")]
6042 UNSPEC_SVE_LANE_SELECT)
6043 (match_operand:SVE_FULL_F 1 "register_operand" "w")))]
6044 "TARGET_SVE"
6045 "fmul\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>[%3]"
6046 )
6047
6048 ;; -------------------------------------------------------------------------
6049 ;; ---- [FP] Division
6050 ;; -------------------------------------------------------------------------
6051 ;; The patterns in this section are synthetic.
6052 ;; -------------------------------------------------------------------------
6053
6054 (define_expand "div<mode>3"
6055 [(set (match_operand:SVE_FULL_F 0 "register_operand")
6056 (unspec:SVE_FULL_F
6057 [(match_dup 3)
6058 (const_int SVE_RELAXED_GP)
6059 (match_operand:SVE_FULL_F 1 "nonmemory_operand")
6060 (match_operand:SVE_FULL_F 2 "register_operand")]
6061 UNSPEC_COND_FDIV))]
6062 "TARGET_SVE"
6063 {
6064 if (aarch64_emit_approx_div (operands[0], operands[1], operands[2]))
6065 DONE;
6066
6067 operands[1] = force_reg (<MODE>mode, operands[1]);
6068 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
6069 }
6070 )
6071
6072 (define_expand "@aarch64_frecpe<mode>"
6073 [(set (match_operand:SVE_FULL_F 0 "register_operand")
6074 (unspec:SVE_FULL_F
6075 [(match_operand:SVE_FULL_F 1 "register_operand")]
6076 UNSPEC_FRECPE))]
6077 "TARGET_SVE"
6078 )
6079
6080 (define_expand "@aarch64_frecps<mode>"
6081 [(set (match_operand:SVE_FULL_F 0 "register_operand")
6082 (unspec:SVE_FULL_F
6083 [(match_operand:SVE_FULL_F 1 "register_operand")
6084 (match_operand:SVE_FULL_F 2 "register_operand")]
6085 UNSPEC_FRECPS))]
6086 "TARGET_SVE"
6087 )
6088
6089 ;; -------------------------------------------------------------------------
6090 ;; ---- [FP] Binary logical operations
6091 ;; -------------------------------------------------------------------------
6092 ;; Includes
6093 ;; - AND
6094 ;; - EOR
6095 ;; - ORR
6096 ;; -------------------------------------------------------------------------
6097
6098 ;; Binary logical operations on floating-point modes. We avoid subregs
6099 ;; by providing this, but we need to use UNSPECs since rtx logical ops
6100 ;; aren't defined for floating-point modes.
6101 (define_insn "*<optab><mode>3"
6102 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w")
6103 (unspec:SVE_FULL_F
6104 [(match_operand:SVE_FULL_F 1 "register_operand" "w")
6105 (match_operand:SVE_FULL_F 2 "register_operand" "w")]
6106 LOGICALF))]
6107 "TARGET_SVE"
6108 "<logicalf_op>\t%0.d, %1.d, %2.d"
6109 )
6110
6111 ;; -------------------------------------------------------------------------
6112 ;; ---- [FP] Sign copying
6113 ;; -------------------------------------------------------------------------
6114 ;; The patterns in this section are synthetic.
6115 ;; -------------------------------------------------------------------------
6116
6117 (define_expand "copysign<mode>3"
6118 [(match_operand:SVE_FULL_F 0 "register_operand")
6119 (match_operand:SVE_FULL_F 1 "register_operand")
6120 (match_operand:SVE_FULL_F 2 "register_operand")]
6121 "TARGET_SVE"
6122 {
6123 rtx sign = gen_reg_rtx (<V_INT_EQUIV>mode);
6124 rtx mant = gen_reg_rtx (<V_INT_EQUIV>mode);
6125 rtx int_res = gen_reg_rtx (<V_INT_EQUIV>mode);
6126 int bits = GET_MODE_UNIT_BITSIZE (<MODE>mode) - 1;
6127
6128 rtx arg1 = lowpart_subreg (<V_INT_EQUIV>mode, operands[1], <MODE>mode);
6129 rtx arg2 = lowpart_subreg (<V_INT_EQUIV>mode, operands[2], <MODE>mode);
6130
6131 emit_insn (gen_and<v_int_equiv>3
6132 (sign, arg2,
6133 aarch64_simd_gen_const_vector_dup (<V_INT_EQUIV>mode,
6134 HOST_WIDE_INT_M1U
6135 << bits)));
6136 emit_insn (gen_and<v_int_equiv>3
6137 (mant, arg1,
6138 aarch64_simd_gen_const_vector_dup (<V_INT_EQUIV>mode,
6139 ~(HOST_WIDE_INT_M1U
6140 << bits))));
6141 emit_insn (gen_ior<v_int_equiv>3 (int_res, sign, mant));
6142 emit_move_insn (operands[0], gen_lowpart (<MODE>mode, int_res));
6143 DONE;
6144 }
6145 )
6146
6147 (define_expand "xorsign<mode>3"
6148 [(match_operand:SVE_FULL_F 0 "register_operand")
6149 (match_operand:SVE_FULL_F 1 "register_operand")
6150 (match_operand:SVE_FULL_F 2 "register_operand")]
6151 "TARGET_SVE"
6152 {
6153 rtx sign = gen_reg_rtx (<V_INT_EQUIV>mode);
6154 rtx int_res = gen_reg_rtx (<V_INT_EQUIV>mode);
6155 int bits = GET_MODE_UNIT_BITSIZE (<MODE>mode) - 1;
6156
6157 rtx arg1 = lowpart_subreg (<V_INT_EQUIV>mode, operands[1], <MODE>mode);
6158 rtx arg2 = lowpart_subreg (<V_INT_EQUIV>mode, operands[2], <MODE>mode);
6159
6160 emit_insn (gen_and<v_int_equiv>3
6161 (sign, arg2,
6162 aarch64_simd_gen_const_vector_dup (<V_INT_EQUIV>mode,
6163 HOST_WIDE_INT_M1U
6164 << bits)));
6165 emit_insn (gen_xor<v_int_equiv>3 (int_res, arg1, sign));
6166 emit_move_insn (operands[0], gen_lowpart (<MODE>mode, int_res));
6167 DONE;
6168 }
6169 )
6170
6171 ;; -------------------------------------------------------------------------
6172 ;; ---- [FP] Maximum and minimum
6173 ;; -------------------------------------------------------------------------
6174 ;; Includes:
6175 ;; - FMAX
6176 ;; - FMAXNM
6177 ;; - FMIN
6178 ;; - FMINNM
6179 ;; -------------------------------------------------------------------------
6180
6181 ;; Unpredicated fmax/fmin (the libm functions). The optabs for the
6182 ;; smin/smax rtx codes are handled in the generic section above.
6183 (define_expand "<maxmin_uns><mode>3"
6184 [(set (match_operand:SVE_FULL_F 0 "register_operand")
6185 (unspec:SVE_FULL_F
6186 [(match_dup 3)
6187 (const_int SVE_RELAXED_GP)
6188 (match_operand:SVE_FULL_F 1 "register_operand")
6189 (match_operand:SVE_FULL_F 2 "aarch64_sve_float_maxmin_operand")]
6190 SVE_COND_FP_MAXMIN_PUBLIC))]
6191 "TARGET_SVE"
6192 {
6193 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
6194 }
6195 )
6196
6197 ;; Predicated floating-point maximum/minimum.
6198 (define_insn "@aarch64_pred_<optab><mode>"
6199 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?&w, ?&w")
6200 (unspec:SVE_FULL_F
6201 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
6202 (match_operand:SI 4 "aarch64_sve_gp_strictness")
6203 (match_operand:SVE_FULL_F 2 "register_operand" "%0, 0, w, w")
6204 (match_operand:SVE_FULL_F 3 "aarch64_sve_float_maxmin_operand" "vsB, w, vsB, w")]
6205 SVE_COND_FP_MAXMIN))]
6206 "TARGET_SVE"
6207 "@
6208 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
6209 <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
6210 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, #%3
6211 movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
6212 [(set_attr "movprfx" "*,*,yes,yes")]
6213 )
6214
6215 ;; Merging forms are handled through SVE_COND_FP_BINARY and
6216 ;; SVE_COND_FP_BINARY_I1.
6217
6218 ;; -------------------------------------------------------------------------
6219 ;; ---- [PRED] Binary logical operations
6220 ;; -------------------------------------------------------------------------
6221 ;; Includes:
6222 ;; - AND
6223 ;; - ANDS
6224 ;; - EOR
6225 ;; - EORS
6226 ;; - ORR
6227 ;; - ORRS
6228 ;; -------------------------------------------------------------------------
6229
6230 ;; Predicate AND. We can reuse one of the inputs as the GP.
6231 ;; Doubling the second operand is the preferred implementation
6232 ;; of the MOV alias, so we use that instead of %1/z, %1, %2.
6233 (define_insn "and<mode>3"
6234 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
6235 (and:PRED_ALL (match_operand:PRED_ALL 1 "register_operand" "Upa")
6236 (match_operand:PRED_ALL 2 "register_operand" "Upa")))]
6237 "TARGET_SVE"
6238 "and\t%0.b, %1/z, %2.b, %2.b"
6239 )
6240
6241 ;; Unpredicated predicate EOR and ORR.
6242 (define_expand "<optab><mode>3"
6243 [(set (match_operand:PRED_ALL 0 "register_operand")
6244 (and:PRED_ALL
6245 (LOGICAL_OR:PRED_ALL
6246 (match_operand:PRED_ALL 1 "register_operand")
6247 (match_operand:PRED_ALL 2 "register_operand"))
6248 (match_dup 3)))]
6249 "TARGET_SVE"
6250 {
6251 operands[3] = aarch64_ptrue_reg (<MODE>mode);
6252 }
6253 )
6254
6255 ;; Predicated predicate AND, EOR and ORR.
6256 (define_insn "@aarch64_pred_<optab><mode>_z"
6257 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
6258 (and:PRED_ALL
6259 (LOGICAL:PRED_ALL
6260 (match_operand:PRED_ALL 2 "register_operand" "Upa")
6261 (match_operand:PRED_ALL 3 "register_operand" "Upa"))
6262 (match_operand:PRED_ALL 1 "register_operand" "Upa")))]
6263 "TARGET_SVE"
6264 "<logical>\t%0.b, %1/z, %2.b, %3.b"
6265 )
6266
6267 ;; Perform a logical operation on operands 2 and 3, using operand 1 as
6268 ;; the GP. Store the result in operand 0 and set the flags in the same
6269 ;; way as for PTEST.
6270 (define_insn "*<optab><mode>3_cc"
6271 [(set (reg:CC_NZC CC_REGNUM)
6272 (unspec:CC_NZC
6273 [(match_operand:VNx16BI 1 "register_operand" "Upa")
6274 (match_operand 4)
6275 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
6276 (and:PRED_ALL
6277 (LOGICAL:PRED_ALL
6278 (match_operand:PRED_ALL 2 "register_operand" "Upa")
6279 (match_operand:PRED_ALL 3 "register_operand" "Upa"))
6280 (match_dup 4))]
6281 UNSPEC_PTEST))
6282 (set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
6283 (and:PRED_ALL (LOGICAL:PRED_ALL (match_dup 2) (match_dup 3))
6284 (match_dup 4)))]
6285 "TARGET_SVE"
6286 "<logical>s\t%0.b, %1/z, %2.b, %3.b"
6287 )
6288
6289 ;; Same with just the flags result.
6290 (define_insn "*<optab><mode>3_ptest"
6291 [(set (reg:CC_NZC CC_REGNUM)
6292 (unspec:CC_NZC
6293 [(match_operand:VNx16BI 1 "register_operand" "Upa")
6294 (match_operand 4)
6295 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
6296 (and:PRED_ALL
6297 (LOGICAL:PRED_ALL
6298 (match_operand:PRED_ALL 2 "register_operand" "Upa")
6299 (match_operand:PRED_ALL 3 "register_operand" "Upa"))
6300 (match_dup 4))]
6301 UNSPEC_PTEST))
6302 (clobber (match_scratch:VNx16BI 0 "=Upa"))]
6303 "TARGET_SVE"
6304 "<logical>s\t%0.b, %1/z, %2.b, %3.b"
6305 )
6306
6307 ;; -------------------------------------------------------------------------
6308 ;; ---- [PRED] Binary logical operations (inverted second input)
6309 ;; -------------------------------------------------------------------------
6310 ;; Includes:
6311 ;; - BIC
6312 ;; - ORN
6313 ;; -------------------------------------------------------------------------
6314
6315 ;; Predicated predicate BIC and ORN.
6316 (define_insn "aarch64_pred_<nlogical><mode>_z"
6317 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
6318 (and:PRED_ALL
6319 (NLOGICAL:PRED_ALL
6320 (not:PRED_ALL (match_operand:PRED_ALL 3 "register_operand" "Upa"))
6321 (match_operand:PRED_ALL 2 "register_operand" "Upa"))
6322 (match_operand:PRED_ALL 1 "register_operand" "Upa")))]
6323 "TARGET_SVE"
6324 "<nlogical>\t%0.b, %1/z, %2.b, %3.b"
6325 )
6326
6327 ;; Same, but set the flags as a side-effect.
6328 (define_insn "*<nlogical><mode>3_cc"
6329 [(set (reg:CC_NZC CC_REGNUM)
6330 (unspec:CC_NZC
6331 [(match_operand:VNx16BI 1 "register_operand" "Upa")
6332 (match_operand 4)
6333 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
6334 (and:PRED_ALL
6335 (NLOGICAL:PRED_ALL
6336 (not:PRED_ALL
6337 (match_operand:PRED_ALL 3 "register_operand" "Upa"))
6338 (match_operand:PRED_ALL 2 "register_operand" "Upa"))
6339 (match_dup 4))]
6340 UNSPEC_PTEST))
6341 (set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
6342 (and:PRED_ALL (NLOGICAL:PRED_ALL
6343 (not:PRED_ALL (match_dup 3))
6344 (match_dup 2))
6345 (match_dup 4)))]
6346 "TARGET_SVE"
6347 "<nlogical>s\t%0.b, %1/z, %2.b, %3.b"
6348 )
6349
6350 ;; Same with just the flags result.
6351 (define_insn "*<nlogical><mode>3_ptest"
6352 [(set (reg:CC_NZC CC_REGNUM)
6353 (unspec:CC_NZC
6354 [(match_operand:VNx16BI 1 "register_operand" "Upa")
6355 (match_operand 4)
6356 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
6357 (and:PRED_ALL
6358 (NLOGICAL:PRED_ALL
6359 (not:PRED_ALL
6360 (match_operand:PRED_ALL 3 "register_operand" "Upa"))
6361 (match_operand:PRED_ALL 2 "register_operand" "Upa"))
6362 (match_dup 4))]
6363 UNSPEC_PTEST))
6364 (clobber (match_scratch:VNx16BI 0 "=Upa"))]
6365 "TARGET_SVE"
6366 "<nlogical>s\t%0.b, %1/z, %2.b, %3.b"
6367 )
6368
6369 ;; -------------------------------------------------------------------------
6370 ;; ---- [PRED] Binary logical operations (inverted result)
6371 ;; -------------------------------------------------------------------------
6372 ;; Includes:
6373 ;; - NAND
6374 ;; - NOR
6375 ;; -------------------------------------------------------------------------
6376
6377 ;; Predicated predicate NAND and NOR.
6378 (define_insn "aarch64_pred_<logical_nn><mode>_z"
6379 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
6380 (and:PRED_ALL
6381 (NLOGICAL:PRED_ALL
6382 (not:PRED_ALL (match_operand:PRED_ALL 2 "register_operand" "Upa"))
6383 (not:PRED_ALL (match_operand:PRED_ALL 3 "register_operand" "Upa")))
6384 (match_operand:PRED_ALL 1 "register_operand" "Upa")))]
6385 "TARGET_SVE"
6386 "<logical_nn>\t%0.b, %1/z, %2.b, %3.b"
6387 )
6388
6389 ;; Same, but set the flags as a side-effect.
6390 (define_insn "*<logical_nn><mode>3_cc"
6391 [(set (reg:CC_NZC CC_REGNUM)
6392 (unspec:CC_NZC
6393 [(match_operand:VNx16BI 1 "register_operand" "Upa")
6394 (match_operand 4)
6395 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
6396 (and:PRED_ALL
6397 (NLOGICAL:PRED_ALL
6398 (not:PRED_ALL
6399 (match_operand:PRED_ALL 2 "register_operand" "Upa"))
6400 (not:PRED_ALL
6401 (match_operand:PRED_ALL 3 "register_operand" "Upa")))
6402 (match_dup 4))]
6403 UNSPEC_PTEST))
6404 (set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
6405 (and:PRED_ALL (NLOGICAL:PRED_ALL
6406 (not:PRED_ALL (match_dup 2))
6407 (not:PRED_ALL (match_dup 3)))
6408 (match_dup 4)))]
6409 "TARGET_SVE"
6410 "<logical_nn>s\t%0.b, %1/z, %2.b, %3.b"
6411 )
6412
6413 ;; Same with just the flags result.
6414 (define_insn "*<logical_nn><mode>3_ptest"
6415 [(set (reg:CC_NZC CC_REGNUM)
6416 (unspec:CC_NZC
6417 [(match_operand:VNx16BI 1 "register_operand" "Upa")
6418 (match_operand 4)
6419 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
6420 (and:PRED_ALL
6421 (NLOGICAL:PRED_ALL
6422 (not:PRED_ALL
6423 (match_operand:PRED_ALL 2 "register_operand" "Upa"))
6424 (not:PRED_ALL
6425 (match_operand:PRED_ALL 3 "register_operand" "Upa")))
6426 (match_dup 4))]
6427 UNSPEC_PTEST))
6428 (clobber (match_scratch:VNx16BI 0 "=Upa"))]
6429 "TARGET_SVE"
6430 "<logical_nn>s\t%0.b, %1/z, %2.b, %3.b"
6431 )
6432
6433 ;; =========================================================================
6434 ;; == Ternary arithmetic
6435 ;; =========================================================================
6436
6437 ;; -------------------------------------------------------------------------
6438 ;; ---- [INT] MLA and MAD
6439 ;; -------------------------------------------------------------------------
6440 ;; Includes:
6441 ;; - MAD
6442 ;; - MLA
6443 ;; -------------------------------------------------------------------------
6444
6445 ;; Unpredicated integer addition of product.
6446 (define_expand "fma<mode>4"
6447 [(set (match_operand:SVE_FULL_I 0 "register_operand")
6448 (plus:SVE_FULL_I
6449 (unspec:SVE_FULL_I
6450 [(match_dup 4)
6451 (mult:SVE_FULL_I
6452 (match_operand:SVE_FULL_I 1 "register_operand")
6453 (match_operand:SVE_FULL_I 2 "nonmemory_operand"))]
6454 UNSPEC_PRED_X)
6455 (match_operand:SVE_FULL_I 3 "register_operand")))]
6456 "TARGET_SVE"
6457 {
6458 if (aarch64_prepare_sve_int_fma (operands, PLUS))
6459 DONE;
6460 operands[4] = aarch64_ptrue_reg (<VPRED>mode);
6461 }
6462 )
6463
6464 ;; Predicated integer addition of product.
6465 (define_insn "@aarch64_pred_fma<mode>"
6466 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, w, ?&w")
6467 (plus:SVE_FULL_I
6468 (unspec:SVE_FULL_I
6469 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
6470 (mult:SVE_FULL_I
6471 (match_operand:SVE_FULL_I 2 "register_operand" "%0, w, w")
6472 (match_operand:SVE_FULL_I 3 "register_operand" "w, w, w"))]
6473 UNSPEC_PRED_X)
6474 (match_operand:SVE_FULL_I 4 "register_operand" "w, 0, w")))]
6475 "TARGET_SVE"
6476 "@
6477 mad\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6478 mla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6479 movprfx\t%0, %4\;mla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>"
6480 [(set_attr "movprfx" "*,*,yes")]
6481 )
6482
6483 ;; Predicated integer addition of product with merging.
6484 (define_expand "cond_fma<mode>"
6485 [(set (match_operand:SVE_FULL_I 0 "register_operand")
6486 (unspec:SVE_FULL_I
6487 [(match_operand:<VPRED> 1 "register_operand")
6488 (plus:SVE_FULL_I
6489 (mult:SVE_FULL_I
6490 (match_operand:SVE_FULL_I 2 "register_operand")
6491 (match_operand:SVE_FULL_I 3 "general_operand"))
6492 (match_operand:SVE_FULL_I 4 "register_operand"))
6493 (match_operand:SVE_FULL_I 5 "aarch64_simd_reg_or_zero")]
6494 UNSPEC_SEL))]
6495 "TARGET_SVE"
6496 {
6497 if (aarch64_prepare_sve_cond_int_fma (operands, PLUS))
6498 DONE;
6499 /* Swap the multiplication operands if the fallback value is the
6500 second of the two. */
6501 if (rtx_equal_p (operands[3], operands[5]))
6502 std::swap (operands[2], operands[3]);
6503 }
6504 )
6505
6506 ;; Predicated integer addition of product, merging with the first input.
6507 (define_insn "*cond_fma<mode>_2"
6508 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
6509 (unspec:SVE_FULL_I
6510 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
6511 (plus:SVE_FULL_I
6512 (mult:SVE_FULL_I
6513 (match_operand:SVE_FULL_I 2 "register_operand" "0, w")
6514 (match_operand:SVE_FULL_I 3 "register_operand" "w, w"))
6515 (match_operand:SVE_FULL_I 4 "register_operand" "w, w"))
6516 (match_dup 2)]
6517 UNSPEC_SEL))]
6518 "TARGET_SVE"
6519 "@
6520 mad\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6521 movprfx\t%0, %2\;mad\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>"
6522 [(set_attr "movprfx" "*,yes")]
6523 )
6524
6525 ;; Predicated integer addition of product, merging with the third input.
6526 (define_insn "*cond_fma<mode>_4"
6527 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
6528 (unspec:SVE_FULL_I
6529 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
6530 (plus:SVE_FULL_I
6531 (mult:SVE_FULL_I
6532 (match_operand:SVE_FULL_I 2 "register_operand" "w, w")
6533 (match_operand:SVE_FULL_I 3 "register_operand" "w, w"))
6534 (match_operand:SVE_FULL_I 4 "register_operand" "0, w"))
6535 (match_dup 4)]
6536 UNSPEC_SEL))]
6537 "TARGET_SVE"
6538 "@
6539 mla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6540 movprfx\t%0, %4\;mla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>"
6541 [(set_attr "movprfx" "*,yes")]
6542 )
6543
6544 ;; Predicated integer addition of product, merging with an independent value.
6545 (define_insn_and_rewrite "*cond_fma<mode>_any"
6546 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, &w, &w, &w, &w, ?&w")
6547 (unspec:SVE_FULL_I
6548 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
6549 (plus:SVE_FULL_I
6550 (mult:SVE_FULL_I
6551 (match_operand:SVE_FULL_I 2 "register_operand" "w, w, 0, w, w, w")
6552 (match_operand:SVE_FULL_I 3 "register_operand" "w, w, w, 0, w, w"))
6553 (match_operand:SVE_FULL_I 4 "register_operand" "w, 0, w, w, w, w"))
6554 (match_operand:SVE_FULL_I 5 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, Dz, 0, w")]
6555 UNSPEC_SEL))]
6556 "TARGET_SVE
6557 && !rtx_equal_p (operands[2], operands[5])
6558 && !rtx_equal_p (operands[3], operands[5])
6559 && !rtx_equal_p (operands[4], operands[5])"
6560 "@
6561 movprfx\t%0.<Vetype>, %1/z, %4.<Vetype>\;mla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6562 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;mla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6563 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;mad\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6564 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;mad\t%0.<Vetype>, %1/m, %2.<Vetype>, %4.<Vetype>
6565 movprfx\t%0.<Vetype>, %1/m, %4.<Vetype>\;mla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6566 #"
6567 "&& reload_completed
6568 && register_operand (operands[5], <MODE>mode)
6569 && !rtx_equal_p (operands[0], operands[5])"
6570 {
6571 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[4],
6572 operands[5], operands[1]));
6573 operands[5] = operands[4] = operands[0];
6574 }
6575 [(set_attr "movprfx" "yes")]
6576 )
6577
6578 ;; -------------------------------------------------------------------------
6579 ;; ---- [INT] MLS and MSB
6580 ;; -------------------------------------------------------------------------
6581 ;; Includes:
6582 ;; - MLS
6583 ;; - MSB
6584 ;; -------------------------------------------------------------------------
6585
6586 ;; Unpredicated integer subtraction of product.
6587 (define_expand "fnma<mode>4"
6588 [(set (match_operand:SVE_FULL_I 0 "register_operand")
6589 (minus:SVE_FULL_I
6590 (match_operand:SVE_FULL_I 3 "register_operand")
6591 (unspec:SVE_FULL_I
6592 [(match_dup 4)
6593 (mult:SVE_FULL_I
6594 (match_operand:SVE_FULL_I 1 "register_operand")
6595 (match_operand:SVE_FULL_I 2 "general_operand"))]
6596 UNSPEC_PRED_X)))]
6597 "TARGET_SVE"
6598 {
6599 if (aarch64_prepare_sve_int_fma (operands, MINUS))
6600 DONE;
6601 operands[4] = aarch64_ptrue_reg (<VPRED>mode);
6602 }
6603 )
6604
6605 ;; Predicated integer subtraction of product.
6606 (define_insn "@aarch64_pred_fnma<mode>"
6607 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, w, ?&w")
6608 (minus:SVE_FULL_I
6609 (match_operand:SVE_FULL_I 4 "register_operand" "w, 0, w")
6610 (unspec:SVE_FULL_I
6611 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
6612 (mult:SVE_FULL_I
6613 (match_operand:SVE_FULL_I 2 "register_operand" "%0, w, w")
6614 (match_operand:SVE_FULL_I 3 "register_operand" "w, w, w"))]
6615 UNSPEC_PRED_X)))]
6616 "TARGET_SVE"
6617 "@
6618 msb\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6619 mls\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6620 movprfx\t%0, %4\;mls\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>"
6621 [(set_attr "movprfx" "*,*,yes")]
6622 )
6623
6624 ;; Predicated integer subtraction of product with merging.
6625 (define_expand "cond_fnma<mode>"
6626 [(set (match_operand:SVE_FULL_I 0 "register_operand")
6627 (unspec:SVE_FULL_I
6628 [(match_operand:<VPRED> 1 "register_operand")
6629 (minus:SVE_FULL_I
6630 (match_operand:SVE_FULL_I 4 "register_operand")
6631 (mult:SVE_FULL_I
6632 (match_operand:SVE_FULL_I 2 "register_operand")
6633 (match_operand:SVE_FULL_I 3 "general_operand")))
6634 (match_operand:SVE_FULL_I 5 "aarch64_simd_reg_or_zero")]
6635 UNSPEC_SEL))]
6636 "TARGET_SVE"
6637 {
6638 if (aarch64_prepare_sve_cond_int_fma (operands, MINUS))
6639 DONE;
6640 /* Swap the multiplication operands if the fallback value is the
6641 second of the two. */
6642 if (rtx_equal_p (operands[3], operands[5]))
6643 std::swap (operands[2], operands[3]);
6644 }
6645 )
6646
6647 ;; Predicated integer subtraction of product, merging with the first input.
6648 (define_insn "*cond_fnma<mode>_2"
6649 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
6650 (unspec:SVE_FULL_I
6651 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
6652 (minus:SVE_FULL_I
6653 (match_operand:SVE_FULL_I 4 "register_operand" "w, w")
6654 (mult:SVE_FULL_I
6655 (match_operand:SVE_FULL_I 2 "register_operand" "0, w")
6656 (match_operand:SVE_FULL_I 3 "register_operand" "w, w")))
6657 (match_dup 2)]
6658 UNSPEC_SEL))]
6659 "TARGET_SVE"
6660 "@
6661 msb\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6662 movprfx\t%0, %2\;msb\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>"
6663 [(set_attr "movprfx" "*,yes")]
6664 )
6665
6666 ;; Predicated integer subtraction of product, merging with the third input.
6667 (define_insn "*cond_fnma<mode>_4"
6668 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=w, ?&w")
6669 (unspec:SVE_FULL_I
6670 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
6671 (minus:SVE_FULL_I
6672 (match_operand:SVE_FULL_I 4 "register_operand" "0, w")
6673 (mult:SVE_FULL_I
6674 (match_operand:SVE_FULL_I 2 "register_operand" "w, w")
6675 (match_operand:SVE_FULL_I 3 "register_operand" "w, w")))
6676 (match_dup 4)]
6677 UNSPEC_SEL))]
6678 "TARGET_SVE"
6679 "@
6680 mls\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6681 movprfx\t%0, %4\;mls\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>"
6682 [(set_attr "movprfx" "*,yes")]
6683 )
6684
6685 ;; Predicated integer subtraction of product, merging with an
6686 ;; independent value.
6687 (define_insn_and_rewrite "*cond_fnma<mode>_any"
6688 [(set (match_operand:SVE_FULL_I 0 "register_operand" "=&w, &w, &w, &w, &w, ?&w")
6689 (unspec:SVE_FULL_I
6690 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
6691 (minus:SVE_FULL_I
6692 (match_operand:SVE_FULL_I 4 "register_operand" "w, 0, w, w, w, w")
6693 (mult:SVE_FULL_I
6694 (match_operand:SVE_FULL_I 2 "register_operand" "w, w, 0, w, w, w")
6695 (match_operand:SVE_FULL_I 3 "register_operand" "w, w, w, 0, w, w")))
6696 (match_operand:SVE_FULL_I 5 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, Dz, 0, w")]
6697 UNSPEC_SEL))]
6698 "TARGET_SVE
6699 && !rtx_equal_p (operands[2], operands[5])
6700 && !rtx_equal_p (operands[3], operands[5])
6701 && !rtx_equal_p (operands[4], operands[5])"
6702 "@
6703 movprfx\t%0.<Vetype>, %1/z, %4.<Vetype>\;mls\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6704 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;mls\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6705 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;msb\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6706 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;msb\t%0.<Vetype>, %1/m, %2.<Vetype>, %4.<Vetype>
6707 movprfx\t%0.<Vetype>, %1/m, %4.<Vetype>\;mls\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6708 #"
6709 "&& reload_completed
6710 && register_operand (operands[5], <MODE>mode)
6711 && !rtx_equal_p (operands[0], operands[5])"
6712 {
6713 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[4],
6714 operands[5], operands[1]));
6715 operands[5] = operands[4] = operands[0];
6716 }
6717 [(set_attr "movprfx" "yes")]
6718 )
6719
6720 ;; -------------------------------------------------------------------------
6721 ;; ---- [INT] Dot product
6722 ;; -------------------------------------------------------------------------
6723 ;; Includes:
6724 ;; - SDOT
6725 ;; - SUDOT (I8MM)
6726 ;; - UDOT
6727 ;; - USDOT (I8MM)
6728 ;; -------------------------------------------------------------------------
6729
6730 ;; Four-element integer dot-product with accumulation.
6731 (define_insn "<sur>dot_prod<vsi2qi>"
6732 [(set (match_operand:SVE_FULL_SDI 0 "register_operand" "=w, ?&w")
6733 (plus:SVE_FULL_SDI
6734 (unspec:SVE_FULL_SDI
6735 [(match_operand:<VSI2QI> 1 "register_operand" "w, w")
6736 (match_operand:<VSI2QI> 2 "register_operand" "w, w")]
6737 DOTPROD)
6738 (match_operand:SVE_FULL_SDI 3 "register_operand" "0, w")))]
6739 "TARGET_SVE"
6740 "@
6741 <sur>dot\\t%0.<Vetype>, %1.<Vetype_fourth>, %2.<Vetype_fourth>
6742 movprfx\t%0, %3\;<sur>dot\\t%0.<Vetype>, %1.<Vetype_fourth>, %2.<Vetype_fourth>"
6743 [(set_attr "movprfx" "*,yes")]
6744 )
6745
6746 ;; Four-element integer dot-product by selected lanes with accumulation.
6747 (define_insn "@aarch64_<sur>dot_prod_lane<vsi2qi>"
6748 [(set (match_operand:SVE_FULL_SDI 0 "register_operand" "=w, ?&w")
6749 (plus:SVE_FULL_SDI
6750 (unspec:SVE_FULL_SDI
6751 [(match_operand:<VSI2QI> 1 "register_operand" "w, w")
6752 (unspec:<VSI2QI>
6753 [(match_operand:<VSI2QI> 2 "register_operand" "<sve_lane_con>, <sve_lane_con>")
6754 (match_operand:SI 3 "const_int_operand")]
6755 UNSPEC_SVE_LANE_SELECT)]
6756 DOTPROD)
6757 (match_operand:SVE_FULL_SDI 4 "register_operand" "0, w")))]
6758 "TARGET_SVE"
6759 "@
6760 <sur>dot\\t%0.<Vetype>, %1.<Vetype_fourth>, %2.<Vetype_fourth>[%3]
6761 movprfx\t%0, %4\;<sur>dot\\t%0.<Vetype>, %1.<Vetype_fourth>, %2.<Vetype_fourth>[%3]"
6762 [(set_attr "movprfx" "*,yes")]
6763 )
6764
6765 (define_insn "@aarch64_<sur>dot_prod<vsi2qi>"
6766 [(set (match_operand:VNx4SI_ONLY 0 "register_operand" "=w, ?&w")
6767 (plus:VNx4SI_ONLY
6768 (unspec:VNx4SI_ONLY
6769 [(match_operand:<VSI2QI> 1 "register_operand" "w, w")
6770 (match_operand:<VSI2QI> 2 "register_operand" "w, w")]
6771 DOTPROD_US_ONLY)
6772 (match_operand:VNx4SI_ONLY 3 "register_operand" "0, w")))]
6773 "TARGET_SVE_I8MM"
6774 "@
6775 <sur>dot\\t%0.s, %1.b, %2.b
6776 movprfx\t%0, %3\;<sur>dot\\t%0.s, %1.b, %2.b"
6777 [(set_attr "movprfx" "*,yes")]
6778 )
6779
6780 (define_insn "@aarch64_<sur>dot_prod_lane<vsi2qi>"
6781 [(set (match_operand:VNx4SI_ONLY 0 "register_operand" "=w, ?&w")
6782 (plus:VNx4SI_ONLY
6783 (unspec:VNx4SI_ONLY
6784 [(match_operand:<VSI2QI> 1 "register_operand" "w, w")
6785 (unspec:<VSI2QI>
6786 [(match_operand:<VSI2QI> 2 "register_operand" "y, y")
6787 (match_operand:SI 3 "const_int_operand")]
6788 UNSPEC_SVE_LANE_SELECT)]
6789 DOTPROD_I8MM)
6790 (match_operand:VNx4SI_ONLY 4 "register_operand" "0, w")))]
6791 "TARGET_SVE_I8MM"
6792 "@
6793 <sur>dot\\t%0.s, %1.b, %2.b[%3]
6794 movprfx\t%0, %4\;<sur>dot\\t%0.s, %1.b, %2.b[%3]"
6795 [(set_attr "movprfx" "*,yes")]
6796 )
6797
6798 ;; -------------------------------------------------------------------------
6799 ;; ---- [INT] Sum of absolute differences
6800 ;; -------------------------------------------------------------------------
6801 ;; The patterns in this section are synthetic.
6802 ;; -------------------------------------------------------------------------
6803
6804 ;; Emit a sequence to produce a sum-of-absolute-differences of the inputs in
6805 ;; operands 1 and 2. The sequence also has to perform a widening reduction of
6806 ;; the difference into a vector and accumulate that into operand 3 before
6807 ;; copying that into the result operand 0.
6808 ;; Perform that with a sequence of:
6809 ;; MOV ones.b, #1
6810 ;; [SU]ABD diff.b, p0/m, op1.b, op2.b
6811 ;; MOVPRFX op0, op3 // If necessary
6812 ;; UDOT op0.s, diff.b, ones.b
6813 (define_expand "<sur>sad<vsi2qi>"
6814 [(use (match_operand:SVE_FULL_SDI 0 "register_operand"))
6815 (unspec:<VSI2QI> [(use (match_operand:<VSI2QI> 1 "register_operand"))
6816 (use (match_operand:<VSI2QI> 2 "register_operand"))] ABAL)
6817 (use (match_operand:SVE_FULL_SDI 3 "register_operand"))]
6818 "TARGET_SVE"
6819 {
6820 rtx ones = force_reg (<VSI2QI>mode, CONST1_RTX (<VSI2QI>mode));
6821 rtx diff = gen_reg_rtx (<VSI2QI>mode);
6822 emit_insn (gen_<sur>abd<vsi2qi>_3 (diff, operands[1], operands[2]));
6823 emit_insn (gen_udot_prod<vsi2qi> (operands[0], diff, ones, operands[3]));
6824 DONE;
6825 }
6826 )
6827
6828 ;; -------------------------------------------------------------------------
6829 ;; ---- [INT] Matrix multiply-accumulate
6830 ;; -------------------------------------------------------------------------
6831 ;; Includes:
6832 ;; - SMMLA (I8MM)
6833 ;; - UMMLA (I8MM)
6834 ;; - USMMLA (I8MM)
6835 ;; -------------------------------------------------------------------------
6836
6837 (define_insn "@aarch64_sve_add_<optab><vsi2qi>"
6838 [(set (match_operand:VNx4SI_ONLY 0 "register_operand" "=w, ?&w")
6839 (plus:VNx4SI_ONLY
6840 (unspec:VNx4SI_ONLY
6841 [(match_operand:<VSI2QI> 2 "register_operand" "w, w")
6842 (match_operand:<VSI2QI> 3 "register_operand" "w, w")]
6843 MATMUL)
6844 (match_operand:VNx4SI_ONLY 1 "register_operand" "0, w")))]
6845 "TARGET_SVE_I8MM"
6846 "@
6847 <sur>mmla\\t%0.s, %2.b, %3.b
6848 movprfx\t%0, %1\;<sur>mmla\\t%0.s, %2.b, %3.b"
6849 [(set_attr "movprfx" "*,yes")]
6850 )
6851
6852 ;; -------------------------------------------------------------------------
6853 ;; ---- [FP] General ternary arithmetic corresponding to unspecs
6854 ;; -------------------------------------------------------------------------
6855 ;; Includes merging patterns for:
6856 ;; - FMAD
6857 ;; - FMLA
6858 ;; - FMLS
6859 ;; - FMSB
6860 ;; - FNMAD
6861 ;; - FNMLA
6862 ;; - FNMLS
6863 ;; - FNMSB
6864 ;; -------------------------------------------------------------------------
6865
6866 ;; Unpredicated floating-point ternary operations.
6867 (define_expand "<optab><mode>4"
6868 [(set (match_operand:SVE_FULL_F 0 "register_operand")
6869 (unspec:SVE_FULL_F
6870 [(match_dup 4)
6871 (const_int SVE_RELAXED_GP)
6872 (match_operand:SVE_FULL_F 1 "register_operand")
6873 (match_operand:SVE_FULL_F 2 "register_operand")
6874 (match_operand:SVE_FULL_F 3 "register_operand")]
6875 SVE_COND_FP_TERNARY))]
6876 "TARGET_SVE"
6877 {
6878 operands[4] = aarch64_ptrue_reg (<VPRED>mode);
6879 }
6880 )
6881
6882 ;; Predicated floating-point ternary operations.
6883 (define_insn "@aarch64_pred_<optab><mode>"
6884 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, w, ?&w")
6885 (unspec:SVE_FULL_F
6886 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl")
6887 (match_operand:SI 5 "aarch64_sve_gp_strictness")
6888 (match_operand:SVE_FULL_F 2 "register_operand" "%w, 0, w")
6889 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w")
6890 (match_operand:SVE_FULL_F 4 "register_operand" "0, w, w")]
6891 SVE_COND_FP_TERNARY))]
6892 "TARGET_SVE"
6893 "@
6894 <sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6895 <sve_fmad_op>\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6896 movprfx\t%0, %4\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>"
6897 [(set_attr "movprfx" "*,*,yes")]
6898 )
6899
6900 ;; Predicated floating-point ternary operations with merging.
6901 (define_expand "@cond_<optab><mode>"
6902 [(set (match_operand:SVE_FULL_F 0 "register_operand")
6903 (unspec:SVE_FULL_F
6904 [(match_operand:<VPRED> 1 "register_operand")
6905 (unspec:SVE_FULL_F
6906 [(match_dup 1)
6907 (const_int SVE_STRICT_GP)
6908 (match_operand:SVE_FULL_F 2 "register_operand")
6909 (match_operand:SVE_FULL_F 3 "register_operand")
6910 (match_operand:SVE_FULL_F 4 "register_operand")]
6911 SVE_COND_FP_TERNARY)
6912 (match_operand:SVE_FULL_F 5 "aarch64_simd_reg_or_zero")]
6913 UNSPEC_SEL))]
6914 "TARGET_SVE"
6915 {
6916 /* Swap the multiplication operands if the fallback value is the
6917 second of the two. */
6918 if (rtx_equal_p (operands[3], operands[5]))
6919 std::swap (operands[2], operands[3]);
6920 })
6921
6922 ;; Predicated floating-point ternary operations, merging with the
6923 ;; first input.
6924 (define_insn_and_rewrite "*cond_<optab><mode>_2_relaxed"
6925 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
6926 (unspec:SVE_FULL_F
6927 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
6928 (unspec:SVE_FULL_F
6929 [(match_operand 5)
6930 (const_int SVE_RELAXED_GP)
6931 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
6932 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")
6933 (match_operand:SVE_FULL_F 4 "register_operand" "w, w")]
6934 SVE_COND_FP_TERNARY)
6935 (match_dup 2)]
6936 UNSPEC_SEL))]
6937 "TARGET_SVE"
6938 "@
6939 <sve_fmad_op>\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6940 movprfx\t%0, %2\;<sve_fmad_op>\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>"
6941 "&& !rtx_equal_p (operands[1], operands[5])"
6942 {
6943 operands[5] = copy_rtx (operands[1]);
6944 }
6945 [(set_attr "movprfx" "*,yes")]
6946 )
6947
6948 (define_insn "*cond_<optab><mode>_2_strict"
6949 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
6950 (unspec:SVE_FULL_F
6951 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
6952 (unspec:SVE_FULL_F
6953 [(match_dup 1)
6954 (const_int SVE_STRICT_GP)
6955 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")
6956 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")
6957 (match_operand:SVE_FULL_F 4 "register_operand" "w, w")]
6958 SVE_COND_FP_TERNARY)
6959 (match_dup 2)]
6960 UNSPEC_SEL))]
6961 "TARGET_SVE"
6962 "@
6963 <sve_fmad_op>\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
6964 movprfx\t%0, %2\;<sve_fmad_op>\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>"
6965 [(set_attr "movprfx" "*,yes")]
6966 )
6967
6968 ;; Predicated floating-point ternary operations, merging with the
6969 ;; third input.
6970 (define_insn_and_rewrite "*cond_<optab><mode>_4_relaxed"
6971 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
6972 (unspec:SVE_FULL_F
6973 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
6974 (unspec:SVE_FULL_F
6975 [(match_operand 5)
6976 (const_int SVE_RELAXED_GP)
6977 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
6978 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")
6979 (match_operand:SVE_FULL_F 4 "register_operand" "0, w")]
6980 SVE_COND_FP_TERNARY)
6981 (match_dup 4)]
6982 UNSPEC_SEL))]
6983 "TARGET_SVE"
6984 "@
6985 <sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
6986 movprfx\t%0, %4\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>"
6987 "&& !rtx_equal_p (operands[1], operands[5])"
6988 {
6989 operands[5] = copy_rtx (operands[1]);
6990 }
6991 [(set_attr "movprfx" "*,yes")]
6992 )
6993
6994 (define_insn "*cond_<optab><mode>_4_strict"
6995 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
6996 (unspec:SVE_FULL_F
6997 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
6998 (unspec:SVE_FULL_F
6999 [(match_dup 1)
7000 (const_int SVE_STRICT_GP)
7001 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
7002 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")
7003 (match_operand:SVE_FULL_F 4 "register_operand" "0, w")]
7004 SVE_COND_FP_TERNARY)
7005 (match_dup 4)]
7006 UNSPEC_SEL))]
7007 "TARGET_SVE"
7008 "@
7009 <sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
7010 movprfx\t%0, %4\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>"
7011 [(set_attr "movprfx" "*,yes")]
7012 )
7013
7014 ;; Predicated floating-point ternary operations, merging with an
7015 ;; independent value.
7016 (define_insn_and_rewrite "*cond_<optab><mode>_any_relaxed"
7017 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, &w, &w, ?&w")
7018 (unspec:SVE_FULL_F
7019 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
7020 (unspec:SVE_FULL_F
7021 [(match_operand 6)
7022 (const_int SVE_RELAXED_GP)
7023 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, 0, w, w, w")
7024 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w, 0, w, w")
7025 (match_operand:SVE_FULL_F 4 "register_operand" "w, 0, w, w, w, w")]
7026 SVE_COND_FP_TERNARY)
7027 (match_operand:SVE_FULL_F 5 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, Dz, 0, w")]
7028 UNSPEC_SEL))]
7029 "TARGET_SVE
7030 && !rtx_equal_p (operands[2], operands[5])
7031 && !rtx_equal_p (operands[3], operands[5])
7032 && !rtx_equal_p (operands[4], operands[5])"
7033 "@
7034 movprfx\t%0.<Vetype>, %1/z, %4.<Vetype>\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
7035 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
7036 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fmad_op>\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
7037 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fmad_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %4.<Vetype>
7038 movprfx\t%0.<Vetype>, %1/m, %4.<Vetype>\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
7039 #"
7040 "&& 1"
7041 {
7042 if (reload_completed
7043 && register_operand (operands[5], <MODE>mode)
7044 && !rtx_equal_p (operands[0], operands[5]))
7045 {
7046 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[4],
7047 operands[5], operands[1]));
7048 operands[5] = operands[4] = operands[0];
7049 }
7050 else if (!rtx_equal_p (operands[1], operands[6]))
7051 operands[6] = copy_rtx (operands[1]);
7052 else
7053 FAIL;
7054 }
7055 [(set_attr "movprfx" "yes")]
7056 )
7057
7058 (define_insn_and_rewrite "*cond_<optab><mode>_any_strict"
7059 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, &w, &w, ?&w")
7060 (unspec:SVE_FULL_F
7061 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
7062 (unspec:SVE_FULL_F
7063 [(match_dup 1)
7064 (const_int SVE_STRICT_GP)
7065 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, 0, w, w, w")
7066 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w, 0, w, w")
7067 (match_operand:SVE_FULL_F 4 "register_operand" "w, 0, w, w, w, w")]
7068 SVE_COND_FP_TERNARY)
7069 (match_operand:SVE_FULL_F 5 "aarch64_simd_reg_or_zero" "Dz, Dz, Dz, Dz, 0, w")]
7070 UNSPEC_SEL))]
7071 "TARGET_SVE
7072 && !rtx_equal_p (operands[2], operands[5])
7073 && !rtx_equal_p (operands[3], operands[5])
7074 && !rtx_equal_p (operands[4], operands[5])"
7075 "@
7076 movprfx\t%0.<Vetype>, %1/z, %4.<Vetype>\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
7077 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
7078 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fmad_op>\t%0.<Vetype>, %1/m, %3.<Vetype>, %4.<Vetype>
7079 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;<sve_fmad_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %4.<Vetype>
7080 movprfx\t%0.<Vetype>, %1/m, %4.<Vetype>\;<sve_fmla_op>\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>
7081 #"
7082 "&& reload_completed
7083 && register_operand (operands[5], <MODE>mode)
7084 && !rtx_equal_p (operands[0], operands[5])"
7085 {
7086 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[4],
7087 operands[5], operands[1]));
7088 operands[5] = operands[4] = operands[0];
7089 }
7090 [(set_attr "movprfx" "yes")]
7091 )
7092
7093 ;; Unpredicated FMLA and FMLS by selected lanes. It doesn't seem worth using
7094 ;; (fma ...) since target-independent code won't understand the indexing.
7095 (define_insn "@aarch64_<optab>_lane_<mode>"
7096 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
7097 (unspec:SVE_FULL_F
7098 [(match_operand:SVE_FULL_F 1 "register_operand" "w, w")
7099 (unspec:SVE_FULL_F
7100 [(match_operand:SVE_FULL_F 2 "register_operand" "<sve_lane_con>, <sve_lane_con>")
7101 (match_operand:SI 3 "const_int_operand")]
7102 UNSPEC_SVE_LANE_SELECT)
7103 (match_operand:SVE_FULL_F 4 "register_operand" "0, w")]
7104 SVE_FP_TERNARY_LANE))]
7105 "TARGET_SVE"
7106 "@
7107 <sve_fp_op>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>[%3]
7108 movprfx\t%0, %4\;<sve_fp_op>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>[%3]"
7109 [(set_attr "movprfx" "*,yes")]
7110 )
7111
7112 ;; -------------------------------------------------------------------------
7113 ;; ---- [FP] Complex multiply-add
7114 ;; -------------------------------------------------------------------------
7115 ;; Includes merging patterns for:
7116 ;; - FCMLA
7117 ;; -------------------------------------------------------------------------
7118
7119 ;; Predicated FCMLA.
7120 (define_insn "@aarch64_pred_<optab><mode>"
7121 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
7122 (unspec:SVE_FULL_F
7123 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
7124 (match_operand:SI 5 "aarch64_sve_gp_strictness")
7125 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
7126 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")
7127 (match_operand:SVE_FULL_F 4 "register_operand" "0, w")]
7128 SVE_COND_FCMLA))]
7129 "TARGET_SVE"
7130 "@
7131 fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7132 movprfx\t%0, %4\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>"
7133 [(set_attr "movprfx" "*,yes")]
7134 )
7135
7136 ;; Predicated FCMLA with merging.
7137 (define_expand "@cond_<optab><mode>"
7138 [(set (match_operand:SVE_FULL_F 0 "register_operand")
7139 (unspec:SVE_FULL_F
7140 [(match_operand:<VPRED> 1 "register_operand")
7141 (unspec:SVE_FULL_F
7142 [(match_dup 1)
7143 (const_int SVE_STRICT_GP)
7144 (match_operand:SVE_FULL_F 2 "register_operand")
7145 (match_operand:SVE_FULL_F 3 "register_operand")
7146 (match_operand:SVE_FULL_F 4 "register_operand")]
7147 SVE_COND_FCMLA)
7148 (match_operand:SVE_FULL_F 5 "aarch64_simd_reg_or_zero")]
7149 UNSPEC_SEL))]
7150 "TARGET_SVE"
7151 )
7152
7153 ;; Predicated FCMLA, merging with the third input.
7154 (define_insn_and_rewrite "*cond_<optab><mode>_4_relaxed"
7155 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
7156 (unspec:SVE_FULL_F
7157 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
7158 (unspec:SVE_FULL_F
7159 [(match_operand 5)
7160 (const_int SVE_RELAXED_GP)
7161 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
7162 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")
7163 (match_operand:SVE_FULL_F 4 "register_operand" "0, w")]
7164 SVE_COND_FCMLA)
7165 (match_dup 4)]
7166 UNSPEC_SEL))]
7167 "TARGET_SVE"
7168 "@
7169 fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7170 movprfx\t%0, %4\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>"
7171 "&& !rtx_equal_p (operands[1], operands[5])"
7172 {
7173 operands[5] = copy_rtx (operands[1]);
7174 }
7175 [(set_attr "movprfx" "*,yes")]
7176 )
7177
7178 (define_insn "*cond_<optab><mode>_4_strict"
7179 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
7180 (unspec:SVE_FULL_F
7181 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
7182 (unspec:SVE_FULL_F
7183 [(match_dup 1)
7184 (const_int SVE_STRICT_GP)
7185 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
7186 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")
7187 (match_operand:SVE_FULL_F 4 "register_operand" "0, w")]
7188 SVE_COND_FCMLA)
7189 (match_dup 4)]
7190 UNSPEC_SEL))]
7191 "TARGET_SVE"
7192 "@
7193 fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7194 movprfx\t%0, %4\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>"
7195 [(set_attr "movprfx" "*,yes")]
7196 )
7197
7198 ;; Predicated FCMLA, merging with an independent value.
7199 (define_insn_and_rewrite "*cond_<optab><mode>_any_relaxed"
7200 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, ?&w")
7201 (unspec:SVE_FULL_F
7202 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
7203 (unspec:SVE_FULL_F
7204 [(match_operand 6)
7205 (const_int SVE_RELAXED_GP)
7206 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w, w")
7207 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w, w")
7208 (match_operand:SVE_FULL_F 4 "register_operand" "w, 0, w, w")]
7209 SVE_COND_FCMLA)
7210 (match_operand:SVE_FULL_F 5 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, w")]
7211 UNSPEC_SEL))]
7212 "TARGET_SVE && !rtx_equal_p (operands[4], operands[5])"
7213 "@
7214 movprfx\t%0.<Vetype>, %1/z, %4.<Vetype>\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7215 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7216 movprfx\t%0.<Vetype>, %1/m, %4.<Vetype>\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7217 #"
7218 "&& 1"
7219 {
7220 if (reload_completed
7221 && register_operand (operands[5], <MODE>mode)
7222 && !rtx_equal_p (operands[0], operands[5]))
7223 {
7224 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[4],
7225 operands[5], operands[1]));
7226 operands[5] = operands[4] = operands[0];
7227 }
7228 else if (!rtx_equal_p (operands[1], operands[6]))
7229 operands[6] = copy_rtx (operands[1]);
7230 else
7231 FAIL;
7232 }
7233 [(set_attr "movprfx" "yes")]
7234 )
7235
7236 (define_insn_and_rewrite "*cond_<optab><mode>_any_strict"
7237 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, &w, ?&w")
7238 (unspec:SVE_FULL_F
7239 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl, Upl, Upl")
7240 (unspec:SVE_FULL_F
7241 [(match_dup 1)
7242 (const_int SVE_STRICT_GP)
7243 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w, w")
7244 (match_operand:SVE_FULL_F 3 "register_operand" "w, w, w, w")
7245 (match_operand:SVE_FULL_F 4 "register_operand" "w, 0, w, w")]
7246 SVE_COND_FCMLA)
7247 (match_operand:SVE_FULL_F 5 "aarch64_simd_reg_or_zero" "Dz, Dz, 0, w")]
7248 UNSPEC_SEL))]
7249 "TARGET_SVE && !rtx_equal_p (operands[4], operands[5])"
7250 "@
7251 movprfx\t%0.<Vetype>, %1/z, %4.<Vetype>\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7252 movprfx\t%0.<Vetype>, %1/z, %0.<Vetype>\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7253 movprfx\t%0.<Vetype>, %1/m, %4.<Vetype>\;fcmla\t%0.<Vetype>, %1/m, %2.<Vetype>, %3.<Vetype>, #<rot>
7254 #"
7255 "&& reload_completed
7256 && register_operand (operands[5], <MODE>mode)
7257 && !rtx_equal_p (operands[0], operands[5])"
7258 {
7259 emit_insn (gen_vcond_mask_<mode><vpred> (operands[0], operands[4],
7260 operands[5], operands[1]));
7261 operands[5] = operands[4] = operands[0];
7262 }
7263 [(set_attr "movprfx" "yes")]
7264 )
7265
7266 ;; Unpredicated FCMLA with indexing.
7267 (define_insn "@aarch64_<optab>_lane_<mode>"
7268 [(set (match_operand:SVE_FULL_HSF 0 "register_operand" "=w, ?&w")
7269 (unspec:SVE_FULL_HSF
7270 [(match_operand:SVE_FULL_HSF 1 "register_operand" "w, w")
7271 (unspec:SVE_FULL_HSF
7272 [(match_operand:SVE_FULL_HSF 2 "register_operand" "<sve_lane_pair_con>, <sve_lane_pair_con>")
7273 (match_operand:SI 3 "const_int_operand")]
7274 UNSPEC_SVE_LANE_SELECT)
7275 (match_operand:SVE_FULL_HSF 4 "register_operand" "0, w")]
7276 FCMLA))]
7277 "TARGET_SVE"
7278 "@
7279 fcmla\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>[%3], #<rot>
7280 movprfx\t%0, %4\;fcmla\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>[%3], #<rot>"
7281 [(set_attr "movprfx" "*,yes")]
7282 )
7283
7284 ;; -------------------------------------------------------------------------
7285 ;; ---- [FP] Trigonometric multiply-add
7286 ;; -------------------------------------------------------------------------
7287 ;; Includes:
7288 ;; - FTMAD
7289 ;; -------------------------------------------------------------------------
7290
7291 (define_insn "@aarch64_sve_tmad<mode>"
7292 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
7293 (unspec:SVE_FULL_F
7294 [(match_operand:SVE_FULL_F 1 "register_operand" "0, w")
7295 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
7296 (match_operand:DI 3 "const_int_operand")]
7297 UNSPEC_FTMAD))]
7298 "TARGET_SVE"
7299 "@
7300 ftmad\t%0.<Vetype>, %0.<Vetype>, %2.<Vetype>, #%3
7301 movprfx\t%0, %1\;ftmad\t%0.<Vetype>, %0.<Vetype>, %2.<Vetype>, #%3"
7302 [(set_attr "movprfx" "*,yes")]
7303 )
7304
7305 ;; -------------------------------------------------------------------------
7306 ;; ---- [FP] Bfloat16 long ternary arithmetic (SF,BF,BF)
7307 ;; -------------------------------------------------------------------------
7308 ;; Includes:
7309 ;; - BFDOT (BF16)
7310 ;; - BFMLALB (BF16)
7311 ;; - BFMLALT (BF16)
7312 ;; - BFMMLA (BF16)
7313 ;; -------------------------------------------------------------------------
7314
7315 (define_insn "@aarch64_sve_<sve_fp_op>vnx4sf"
7316 [(set (match_operand:VNx4SF 0 "register_operand" "=w, ?&w")
7317 (unspec:VNx4SF
7318 [(match_operand:VNx4SF 1 "register_operand" "0, w")
7319 (match_operand:VNx8BF 2 "register_operand" "w, w")
7320 (match_operand:VNx8BF 3 "register_operand" "w, w")]
7321 SVE_BFLOAT_TERNARY_LONG))]
7322 "TARGET_SVE_BF16"
7323 "@
7324 <sve_fp_op>\t%0.s, %2.h, %3.h
7325 movprfx\t%0, %1\;<sve_fp_op>\t%0.s, %2.h, %3.h"
7326 [(set_attr "movprfx" "*,yes")]
7327 )
7328
7329 ;; The immediate range is enforced before generating the instruction.
7330 (define_insn "@aarch64_sve_<sve_fp_op>_lanevnx4sf"
7331 [(set (match_operand:VNx4SF 0 "register_operand" "=w, ?&w")
7332 (unspec:VNx4SF
7333 [(match_operand:VNx4SF 1 "register_operand" "0, w")
7334 (match_operand:VNx8BF 2 "register_operand" "w, w")
7335 (match_operand:VNx8BF 3 "register_operand" "y, y")
7336 (match_operand:SI 4 "const_int_operand")]
7337 SVE_BFLOAT_TERNARY_LONG_LANE))]
7338 "TARGET_SVE_BF16"
7339 "@
7340 <sve_fp_op>\t%0.s, %2.h, %3.h[%4]
7341 movprfx\t%0, %1\;<sve_fp_op>\t%0.s, %2.h, %3.h[%4]"
7342 [(set_attr "movprfx" "*,yes")]
7343 )
7344
7345 ;; -------------------------------------------------------------------------
7346 ;; ---- [FP] Matrix multiply-accumulate
7347 ;; -------------------------------------------------------------------------
7348 ;; Includes:
7349 ;; - FMMLA (F32MM,F64MM)
7350 ;; -------------------------------------------------------------------------
7351
7352 ;; The mode iterator enforces the target requirements.
7353 (define_insn "@aarch64_sve_<sve_fp_op><mode>"
7354 [(set (match_operand:SVE_MATMULF 0 "register_operand" "=w, ?&w")
7355 (unspec:SVE_MATMULF
7356 [(match_operand:SVE_MATMULF 2 "register_operand" "w, w")
7357 (match_operand:SVE_MATMULF 3 "register_operand" "w, w")
7358 (match_operand:SVE_MATMULF 1 "register_operand" "0, w")]
7359 FMMLA))]
7360 "TARGET_SVE"
7361 "@
7362 <sve_fp_op>\\t%0.<Vetype>, %2.<Vetype>, %3.<Vetype>
7363 movprfx\t%0, %1\;<sve_fp_op>\\t%0.<Vetype>, %2.<Vetype>, %3.<Vetype>"
7364 [(set_attr "movprfx" "*,yes")]
7365 )
7366
7367 ;; =========================================================================
7368 ;; == Comparisons and selects
7369 ;; =========================================================================
7370
7371 ;; -------------------------------------------------------------------------
7372 ;; ---- [INT,FP] Select based on predicates
7373 ;; -------------------------------------------------------------------------
7374 ;; Includes merging patterns for:
7375 ;; - FMOV
7376 ;; - MOV
7377 ;; - SEL
7378 ;; -------------------------------------------------------------------------
7379
7380 ;; vcond_mask operand order: true, false, mask
7381 ;; UNSPEC_SEL operand order: mask, true, false (as for VEC_COND_EXPR)
7382 ;; SEL operand order: mask, true, false
7383 (define_expand "@vcond_mask_<mode><vpred>"
7384 [(set (match_operand:SVE_FULL 0 "register_operand")
7385 (unspec:SVE_FULL
7386 [(match_operand:<VPRED> 3 "register_operand")
7387 (match_operand:SVE_FULL 1 "aarch64_sve_reg_or_dup_imm")
7388 (match_operand:SVE_FULL 2 "aarch64_simd_reg_or_zero")]
7389 UNSPEC_SEL))]
7390 "TARGET_SVE"
7391 {
7392 if (register_operand (operands[1], <MODE>mode))
7393 operands[2] = force_reg (<MODE>mode, operands[2]);
7394 }
7395 )
7396
7397 ;; Selects between:
7398 ;; - two registers
7399 ;; - a duplicated immediate and a register
7400 ;; - a duplicated immediate and zero
7401 (define_insn "*vcond_mask_<mode><vpred>"
7402 [(set (match_operand:SVE_FULL 0 "register_operand" "=w, w, w, w, ?w, ?&w, ?&w")
7403 (unspec:SVE_FULL
7404 [(match_operand:<VPRED> 3 "register_operand" "Upa, Upa, Upa, Upa, Upl, Upl, Upl")
7405 (match_operand:SVE_FULL 1 "aarch64_sve_reg_or_dup_imm" "w, vss, vss, Ufc, Ufc, vss, Ufc")
7406 (match_operand:SVE_FULL 2 "aarch64_simd_reg_or_zero" "w, 0, Dz, 0, Dz, w, w")]
7407 UNSPEC_SEL))]
7408 "TARGET_SVE
7409 && (!register_operand (operands[1], <MODE>mode)
7410 || register_operand (operands[2], <MODE>mode))"
7411 "@
7412 sel\t%0.<Vetype>, %3, %1.<Vetype>, %2.<Vetype>
7413 mov\t%0.<Vetype>, %3/m, #%I1
7414 mov\t%0.<Vetype>, %3/z, #%I1
7415 fmov\t%0.<Vetype>, %3/m, #%1
7416 movprfx\t%0.<Vetype>, %3/z, %0.<Vetype>\;fmov\t%0.<Vetype>, %3/m, #%1
7417 movprfx\t%0, %2\;mov\t%0.<Vetype>, %3/m, #%I1
7418 movprfx\t%0, %2\;fmov\t%0.<Vetype>, %3/m, #%1"
7419 [(set_attr "movprfx" "*,*,*,*,yes,yes,yes")]
7420 )
7421
7422 ;; Optimize selects between a duplicated scalar variable and another vector,
7423 ;; the latter of which can be a zero constant or a variable. Treat duplicates
7424 ;; of GPRs as being more expensive than duplicates of FPRs, since they
7425 ;; involve a cross-file move.
7426 (define_insn "@aarch64_sel_dup<mode>"
7427 [(set (match_operand:SVE_FULL 0 "register_operand" "=?w, w, ??w, ?&w, ??&w, ?&w")
7428 (unspec:SVE_FULL
7429 [(match_operand:<VPRED> 3 "register_operand" "Upl, Upl, Upl, Upl, Upl, Upl")
7430 (vec_duplicate:SVE_FULL
7431 (match_operand:<VEL> 1 "register_operand" "r, w, r, w, r, w"))
7432 (match_operand:SVE_FULL 2 "aarch64_simd_reg_or_zero" "0, 0, Dz, Dz, w, w")]
7433 UNSPEC_SEL))]
7434 "TARGET_SVE"
7435 "@
7436 mov\t%0.<Vetype>, %3/m, %<vwcore>1
7437 mov\t%0.<Vetype>, %3/m, %<Vetype>1
7438 movprfx\t%0.<Vetype>, %3/z, %0.<Vetype>\;mov\t%0.<Vetype>, %3/m, %<vwcore>1
7439 movprfx\t%0.<Vetype>, %3/z, %0.<Vetype>\;mov\t%0.<Vetype>, %3/m, %<Vetype>1
7440 movprfx\t%0, %2\;mov\t%0.<Vetype>, %3/m, %<vwcore>1
7441 movprfx\t%0, %2\;mov\t%0.<Vetype>, %3/m, %<Vetype>1"
7442 [(set_attr "movprfx" "*,*,yes,yes,yes,yes")]
7443 )
7444
7445 ;; -------------------------------------------------------------------------
7446 ;; ---- [INT,FP] Compare and select
7447 ;; -------------------------------------------------------------------------
7448 ;; The patterns in this section are synthetic.
7449 ;; -------------------------------------------------------------------------
7450
7451 ;; Integer (signed) vcond. Don't enforce an immediate range here, since it
7452 ;; depends on the comparison; leave it to aarch64_expand_sve_vcond instead.
7453 (define_expand "vcond<mode><v_int_equiv>"
7454 [(set (match_operand:SVE_FULL 0 "register_operand")
7455 (if_then_else:SVE_FULL
7456 (match_operator 3 "comparison_operator"
7457 [(match_operand:<V_INT_EQUIV> 4 "register_operand")
7458 (match_operand:<V_INT_EQUIV> 5 "nonmemory_operand")])
7459 (match_operand:SVE_FULL 1 "nonmemory_operand")
7460 (match_operand:SVE_FULL 2 "nonmemory_operand")))]
7461 "TARGET_SVE"
7462 {
7463 aarch64_expand_sve_vcond (<MODE>mode, <V_INT_EQUIV>mode, operands);
7464 DONE;
7465 }
7466 )
7467
7468 ;; Integer vcondu. Don't enforce an immediate range here, since it
7469 ;; depends on the comparison; leave it to aarch64_expand_sve_vcond instead.
7470 (define_expand "vcondu<mode><v_int_equiv>"
7471 [(set (match_operand:SVE_FULL 0 "register_operand")
7472 (if_then_else:SVE_FULL
7473 (match_operator 3 "comparison_operator"
7474 [(match_operand:<V_INT_EQUIV> 4 "register_operand")
7475 (match_operand:<V_INT_EQUIV> 5 "nonmemory_operand")])
7476 (match_operand:SVE_FULL 1 "nonmemory_operand")
7477 (match_operand:SVE_FULL 2 "nonmemory_operand")))]
7478 "TARGET_SVE"
7479 {
7480 aarch64_expand_sve_vcond (<MODE>mode, <V_INT_EQUIV>mode, operands);
7481 DONE;
7482 }
7483 )
7484
7485 ;; Floating-point vcond. All comparisons except FCMUO allow a zero operand;
7486 ;; aarch64_expand_sve_vcond handles the case of an FCMUO with zero.
7487 (define_expand "vcond<mode><v_fp_equiv>"
7488 [(set (match_operand:SVE_FULL_HSD 0 "register_operand")
7489 (if_then_else:SVE_FULL_HSD
7490 (match_operator 3 "comparison_operator"
7491 [(match_operand:<V_FP_EQUIV> 4 "register_operand")
7492 (match_operand:<V_FP_EQUIV> 5 "aarch64_simd_reg_or_zero")])
7493 (match_operand:SVE_FULL_HSD 1 "nonmemory_operand")
7494 (match_operand:SVE_FULL_HSD 2 "nonmemory_operand")))]
7495 "TARGET_SVE"
7496 {
7497 aarch64_expand_sve_vcond (<MODE>mode, <V_FP_EQUIV>mode, operands);
7498 DONE;
7499 }
7500 )
7501
7502 ;; -------------------------------------------------------------------------
7503 ;; ---- [INT] Comparisons
7504 ;; -------------------------------------------------------------------------
7505 ;; Includes:
7506 ;; - CMPEQ
7507 ;; - CMPGE
7508 ;; - CMPGT
7509 ;; - CMPHI
7510 ;; - CMPHS
7511 ;; - CMPLE
7512 ;; - CMPLO
7513 ;; - CMPLS
7514 ;; - CMPLT
7515 ;; - CMPNE
7516 ;; -------------------------------------------------------------------------
7517
7518 ;; Signed integer comparisons. Don't enforce an immediate range here, since
7519 ;; it depends on the comparison; leave it to aarch64_expand_sve_vec_cmp_int
7520 ;; instead.
7521 (define_expand "vec_cmp<mode><vpred>"
7522 [(parallel
7523 [(set (match_operand:<VPRED> 0 "register_operand")
7524 (match_operator:<VPRED> 1 "comparison_operator"
7525 [(match_operand:SVE_FULL_I 2 "register_operand")
7526 (match_operand:SVE_FULL_I 3 "nonmemory_operand")]))
7527 (clobber (reg:CC_NZC CC_REGNUM))])]
7528 "TARGET_SVE"
7529 {
7530 aarch64_expand_sve_vec_cmp_int (operands[0], GET_CODE (operands[1]),
7531 operands[2], operands[3]);
7532 DONE;
7533 }
7534 )
7535
7536 ;; Unsigned integer comparisons. Don't enforce an immediate range here, since
7537 ;; it depends on the comparison; leave it to aarch64_expand_sve_vec_cmp_int
7538 ;; instead.
7539 (define_expand "vec_cmpu<mode><vpred>"
7540 [(parallel
7541 [(set (match_operand:<VPRED> 0 "register_operand")
7542 (match_operator:<VPRED> 1 "comparison_operator"
7543 [(match_operand:SVE_FULL_I 2 "register_operand")
7544 (match_operand:SVE_FULL_I 3 "nonmemory_operand")]))
7545 (clobber (reg:CC_NZC CC_REGNUM))])]
7546 "TARGET_SVE"
7547 {
7548 aarch64_expand_sve_vec_cmp_int (operands[0], GET_CODE (operands[1]),
7549 operands[2], operands[3]);
7550 DONE;
7551 }
7552 )
7553
7554 ;; Predicated integer comparisons.
7555 (define_insn "@aarch64_pred_cmp<cmp_op><mode>"
7556 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
7557 (unspec:<VPRED>
7558 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
7559 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
7560 (SVE_INT_CMP:<VPRED>
7561 (match_operand:SVE_FULL_I 3 "register_operand" "w, w")
7562 (match_operand:SVE_FULL_I 4 "aarch64_sve_cmp_<sve_imm_con>_operand" "<sve_imm_con>, w"))]
7563 UNSPEC_PRED_Z))
7564 (clobber (reg:CC_NZC CC_REGNUM))]
7565 "TARGET_SVE"
7566 "@
7567 cmp<cmp_op>\t%0.<Vetype>, %1/z, %3.<Vetype>, #%4
7568 cmp<cmp_op>\t%0.<Vetype>, %1/z, %3.<Vetype>, %4.<Vetype>"
7569 )
7570
7571 ;; Predicated integer comparisons in which both the flag and predicate
7572 ;; results are interesting.
7573 (define_insn_and_rewrite "*cmp<cmp_op><mode>_cc"
7574 [(set (reg:CC_NZC CC_REGNUM)
7575 (unspec:CC_NZC
7576 [(match_operand:VNx16BI 1 "register_operand" "Upl, Upl")
7577 (match_operand 4)
7578 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
7579 (unspec:<VPRED>
7580 [(match_operand 6)
7581 (match_operand:SI 7 "aarch64_sve_ptrue_flag")
7582 (SVE_INT_CMP:<VPRED>
7583 (match_operand:SVE_FULL_I 2 "register_operand" "w, w")
7584 (match_operand:SVE_FULL_I 3 "aarch64_sve_cmp_<sve_imm_con>_operand" "<sve_imm_con>, w"))]
7585 UNSPEC_PRED_Z)]
7586 UNSPEC_PTEST))
7587 (set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
7588 (unspec:<VPRED>
7589 [(match_dup 6)
7590 (match_dup 7)
7591 (SVE_INT_CMP:<VPRED>
7592 (match_dup 2)
7593 (match_dup 3))]
7594 UNSPEC_PRED_Z))]
7595 "TARGET_SVE
7596 && aarch64_sve_same_pred_for_ptest_p (&operands[4], &operands[6])"
7597 "@
7598 cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, #%3
7599 cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
7600 "&& !rtx_equal_p (operands[4], operands[6])"
7601 {
7602 operands[6] = copy_rtx (operands[4]);
7603 operands[7] = operands[5];
7604 }
7605 )
7606
7607 ;; Predicated integer comparisons in which only the flags result is
7608 ;; interesting.
7609 (define_insn_and_rewrite "*cmp<cmp_op><mode>_ptest"
7610 [(set (reg:CC_NZC CC_REGNUM)
7611 (unspec:CC_NZC
7612 [(match_operand:VNx16BI 1 "register_operand" "Upl, Upl")
7613 (match_operand 4)
7614 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
7615 (unspec:<VPRED>
7616 [(match_operand 6)
7617 (match_operand:SI 7 "aarch64_sve_ptrue_flag")
7618 (SVE_INT_CMP:<VPRED>
7619 (match_operand:SVE_FULL_I 2 "register_operand" "w, w")
7620 (match_operand:SVE_FULL_I 3 "aarch64_sve_cmp_<sve_imm_con>_operand" "<sve_imm_con>, w"))]
7621 UNSPEC_PRED_Z)]
7622 UNSPEC_PTEST))
7623 (clobber (match_scratch:<VPRED> 0 "=Upa, Upa"))]
7624 "TARGET_SVE
7625 && aarch64_sve_same_pred_for_ptest_p (&operands[4], &operands[6])"
7626 "@
7627 cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, #%3
7628 cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
7629 "&& !rtx_equal_p (operands[4], operands[6])"
7630 {
7631 operands[6] = copy_rtx (operands[4]);
7632 operands[7] = operands[5];
7633 }
7634 )
7635
7636 ;; Predicated integer comparisons, formed by combining a PTRUE-predicated
7637 ;; comparison with an AND. Split the instruction into its preferred form
7638 ;; at the earliest opportunity, in order to get rid of the redundant
7639 ;; operand 4.
7640 (define_insn_and_split "*cmp<cmp_op><mode>_and"
7641 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
7642 (and:<VPRED>
7643 (unspec:<VPRED>
7644 [(match_operand 4)
7645 (const_int SVE_KNOWN_PTRUE)
7646 (SVE_INT_CMP:<VPRED>
7647 (match_operand:SVE_FULL_I 2 "register_operand" "w, w")
7648 (match_operand:SVE_FULL_I 3 "aarch64_sve_cmp_<sve_imm_con>_operand" "<sve_imm_con>, w"))]
7649 UNSPEC_PRED_Z)
7650 (match_operand:<VPRED> 1 "register_operand" "Upl, Upl")))
7651 (clobber (reg:CC_NZC CC_REGNUM))]
7652 "TARGET_SVE"
7653 "#"
7654 "&& 1"
7655 [(parallel
7656 [(set (match_dup 0)
7657 (unspec:<VPRED>
7658 [(match_dup 1)
7659 (const_int SVE_MAYBE_NOT_PTRUE)
7660 (SVE_INT_CMP:<VPRED>
7661 (match_dup 2)
7662 (match_dup 3))]
7663 UNSPEC_PRED_Z))
7664 (clobber (reg:CC_NZC CC_REGNUM))])]
7665 )
7666
7667 ;; Predicated integer wide comparisons.
7668 (define_insn "@aarch64_pred_cmp<cmp_op><mode>_wide"
7669 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa")
7670 (unspec:<VPRED>
7671 [(match_operand:VNx16BI 1 "register_operand" "Upl")
7672 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
7673 (unspec:<VPRED>
7674 [(match_operand:SVE_FULL_BHSI 3 "register_operand" "w")
7675 (match_operand:VNx2DI 4 "register_operand" "w")]
7676 SVE_COND_INT_CMP_WIDE)]
7677 UNSPEC_PRED_Z))
7678 (clobber (reg:CC_NZC CC_REGNUM))]
7679 "TARGET_SVE"
7680 "cmp<cmp_op>\t%0.<Vetype>, %1/z, %3.<Vetype>, %4.d"
7681 )
7682
7683 ;; Predicated integer wide comparisons in which both the flag and
7684 ;; predicate results are interesting.
7685 (define_insn "*aarch64_pred_cmp<cmp_op><mode>_wide_cc"
7686 [(set (reg:CC_NZC CC_REGNUM)
7687 (unspec:CC_NZC
7688 [(match_operand:VNx16BI 1 "register_operand" "Upl")
7689 (match_operand 4)
7690 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
7691 (unspec:<VPRED>
7692 [(match_operand:VNx16BI 6 "register_operand" "Upl")
7693 (match_operand:SI 7 "aarch64_sve_ptrue_flag")
7694 (unspec:<VPRED>
7695 [(match_operand:SVE_FULL_BHSI 2 "register_operand" "w")
7696 (match_operand:VNx2DI 3 "register_operand" "w")]
7697 SVE_COND_INT_CMP_WIDE)]
7698 UNSPEC_PRED_Z)]
7699 UNSPEC_PTEST))
7700 (set (match_operand:<VPRED> 0 "register_operand" "=Upa")
7701 (unspec:<VPRED>
7702 [(match_dup 6)
7703 (match_dup 7)
7704 (unspec:<VPRED>
7705 [(match_dup 2)
7706 (match_dup 3)]
7707 SVE_COND_INT_CMP_WIDE)]
7708 UNSPEC_PRED_Z))]
7709 "TARGET_SVE
7710 && aarch64_sve_same_pred_for_ptest_p (&operands[4], &operands[6])"
7711 "cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.d"
7712 )
7713
7714 ;; Predicated integer wide comparisons in which only the flags result
7715 ;; is interesting.
7716 (define_insn "*aarch64_pred_cmp<cmp_op><mode>_wide_ptest"
7717 [(set (reg:CC_NZC CC_REGNUM)
7718 (unspec:CC_NZC
7719 [(match_operand:VNx16BI 1 "register_operand" "Upl")
7720 (match_operand 4)
7721 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
7722 (unspec:<VPRED>
7723 [(match_operand:VNx16BI 6 "register_operand" "Upl")
7724 (match_operand:SI 7 "aarch64_sve_ptrue_flag")
7725 (unspec:<VPRED>
7726 [(match_operand:SVE_FULL_BHSI 2 "register_operand" "w")
7727 (match_operand:VNx2DI 3 "register_operand" "w")]
7728 SVE_COND_INT_CMP_WIDE)]
7729 UNSPEC_PRED_Z)]
7730 UNSPEC_PTEST))
7731 (clobber (match_scratch:<VPRED> 0 "=Upa"))]
7732 "TARGET_SVE
7733 && aarch64_sve_same_pred_for_ptest_p (&operands[4], &operands[6])"
7734 "cmp<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.d"
7735 )
7736
7737 ;; -------------------------------------------------------------------------
7738 ;; ---- [INT] While tests
7739 ;; -------------------------------------------------------------------------
7740 ;; Includes:
7741 ;; - WHILEGE (SVE2)
7742 ;; - WHILEGT (SVE2)
7743 ;; - WHILEHI (SVE2)
7744 ;; - WHILEHS (SVE2)
7745 ;; - WHILELE
7746 ;; - WHILELO
7747 ;; - WHILELS
7748 ;; - WHILELT
7749 ;; - WHILERW (SVE2)
7750 ;; - WHILEWR (SVE2)
7751 ;; -------------------------------------------------------------------------
7752
7753 ;; Set element I of the result if (cmp (plus operand1 J) operand2) is
7754 ;; true for all J in [0, I].
7755 (define_insn "@while_<while_optab_cmp><GPI:mode><PRED_ALL:mode>"
7756 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
7757 (unspec:PRED_ALL [(match_operand:GPI 1 "aarch64_reg_or_zero" "rZ")
7758 (match_operand:GPI 2 "aarch64_reg_or_zero" "rZ")]
7759 SVE_WHILE))
7760 (clobber (reg:CC_NZC CC_REGNUM))]
7761 "TARGET_SVE"
7762 "while<cmp_op>\t%0.<PRED_ALL:Vetype>, %<w>1, %<w>2"
7763 )
7764
7765 ;; The WHILE instructions set the flags in the same way as a PTEST with
7766 ;; a PTRUE GP. Handle the case in which both results are useful. The GP
7767 ;; operands to the PTEST aren't needed, so we allow them to be anything.
7768 (define_insn_and_rewrite "*while_<while_optab_cmp><GPI:mode><PRED_ALL:mode>_cc"
7769 [(set (reg:CC_NZC CC_REGNUM)
7770 (unspec:CC_NZC
7771 [(match_operand 3)
7772 (match_operand 4)
7773 (const_int SVE_KNOWN_PTRUE)
7774 (unspec:PRED_ALL
7775 [(match_operand:GPI 1 "aarch64_reg_or_zero" "rZ")
7776 (match_operand:GPI 2 "aarch64_reg_or_zero" "rZ")]
7777 SVE_WHILE)]
7778 UNSPEC_PTEST))
7779 (set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
7780 (unspec:PRED_ALL [(match_dup 1)
7781 (match_dup 2)]
7782 SVE_WHILE))]
7783 "TARGET_SVE"
7784 "while<cmp_op>\t%0.<PRED_ALL:Vetype>, %<w>1, %<w>2"
7785 ;; Force the compiler to drop the unused predicate operand, so that we
7786 ;; don't have an unnecessary PTRUE.
7787 "&& (!CONSTANT_P (operands[3]) || !CONSTANT_P (operands[4]))"
7788 {
7789 operands[3] = CONSTM1_RTX (VNx16BImode);
7790 operands[4] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
7791 }
7792 )
7793
7794 ;; Same, but handle the case in which only the flags result is useful.
7795 (define_insn_and_rewrite "@while_<while_optab_cmp><GPI:mode><PRED_ALL:mode>_ptest"
7796 [(set (reg:CC_NZC CC_REGNUM)
7797 (unspec:CC_NZC
7798 [(match_operand 3)
7799 (match_operand 4)
7800 (const_int SVE_KNOWN_PTRUE)
7801 (unspec:PRED_ALL
7802 [(match_operand:GPI 1 "aarch64_reg_or_zero" "rZ")
7803 (match_operand:GPI 2 "aarch64_reg_or_zero" "rZ")]
7804 SVE_WHILE)]
7805 UNSPEC_PTEST))
7806 (clobber (match_scratch:PRED_ALL 0 "=Upa"))]
7807 "TARGET_SVE"
7808 "while<cmp_op>\t%0.<PRED_ALL:Vetype>, %<w>1, %<w>2"
7809 ;; Force the compiler to drop the unused predicate operand, so that we
7810 ;; don't have an unnecessary PTRUE.
7811 "&& (!CONSTANT_P (operands[3]) || !CONSTANT_P (operands[4]))"
7812 {
7813 operands[3] = CONSTM1_RTX (VNx16BImode);
7814 operands[4] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
7815 }
7816 )
7817
7818 ;; -------------------------------------------------------------------------
7819 ;; ---- [FP] Direct comparisons
7820 ;; -------------------------------------------------------------------------
7821 ;; Includes:
7822 ;; - FCMEQ
7823 ;; - FCMGE
7824 ;; - FCMGT
7825 ;; - FCMLE
7826 ;; - FCMLT
7827 ;; - FCMNE
7828 ;; - FCMUO
7829 ;; -------------------------------------------------------------------------
7830
7831 ;; Floating-point comparisons. All comparisons except FCMUO allow a zero
7832 ;; operand; aarch64_expand_sve_vec_cmp_float handles the case of an FCMUO
7833 ;; with zero.
7834 (define_expand "vec_cmp<mode><vpred>"
7835 [(set (match_operand:<VPRED> 0 "register_operand")
7836 (match_operator:<VPRED> 1 "comparison_operator"
7837 [(match_operand:SVE_FULL_F 2 "register_operand")
7838 (match_operand:SVE_FULL_F 3 "aarch64_simd_reg_or_zero")]))]
7839 "TARGET_SVE"
7840 {
7841 aarch64_expand_sve_vec_cmp_float (operands[0], GET_CODE (operands[1]),
7842 operands[2], operands[3], false);
7843 DONE;
7844 }
7845 )
7846
7847 ;; Predicated floating-point comparisons.
7848 (define_insn "@aarch64_pred_fcm<cmp_op><mode>"
7849 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
7850 (unspec:<VPRED>
7851 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
7852 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
7853 (match_operand:SVE_FULL_F 3 "register_operand" "w, w")
7854 (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero" "Dz, w")]
7855 SVE_COND_FP_CMP_I0))]
7856 "TARGET_SVE"
7857 "@
7858 fcm<cmp_op>\t%0.<Vetype>, %1/z, %3.<Vetype>, #0.0
7859 fcm<cmp_op>\t%0.<Vetype>, %1/z, %3.<Vetype>, %4.<Vetype>"
7860 )
7861
7862 ;; Same for unordered comparisons.
7863 (define_insn "@aarch64_pred_fcmuo<mode>"
7864 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa")
7865 (unspec:<VPRED>
7866 [(match_operand:<VPRED> 1 "register_operand" "Upl")
7867 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
7868 (match_operand:SVE_FULL_F 3 "register_operand" "w")
7869 (match_operand:SVE_FULL_F 4 "register_operand" "w")]
7870 UNSPEC_COND_FCMUO))]
7871 "TARGET_SVE"
7872 "fcmuo\t%0.<Vetype>, %1/z, %3.<Vetype>, %4.<Vetype>"
7873 )
7874
7875 ;; Floating-point comparisons predicated on a PTRUE, with the results ANDed
7876 ;; with another predicate P. This does not have the same trapping behavior
7877 ;; as predicating the comparison itself on P, but it's a legitimate fold,
7878 ;; since we can drop any potentially-trapping operations whose results
7879 ;; are not needed.
7880 ;;
7881 ;; Split the instruction into its preferred form (below) at the earliest
7882 ;; opportunity, in order to get rid of the redundant operand 1.
7883 (define_insn_and_split "*fcm<cmp_op><mode>_and_combine"
7884 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa, Upa")
7885 (and:<VPRED>
7886 (unspec:<VPRED>
7887 [(match_operand:<VPRED> 1)
7888 (const_int SVE_KNOWN_PTRUE)
7889 (match_operand:SVE_FULL_F 2 "register_operand" "w, w")
7890 (match_operand:SVE_FULL_F 3 "aarch64_simd_reg_or_zero" "Dz, w")]
7891 SVE_COND_FP_CMP_I0)
7892 (match_operand:<VPRED> 4 "register_operand" "Upl, Upl")))]
7893 "TARGET_SVE"
7894 "#"
7895 "&& 1"
7896 [(set (match_dup 0)
7897 (unspec:<VPRED>
7898 [(match_dup 4)
7899 (const_int SVE_MAYBE_NOT_PTRUE)
7900 (match_dup 2)
7901 (match_dup 3)]
7902 SVE_COND_FP_CMP_I0))]
7903 )
7904
7905 ;; Same for unordered comparisons.
7906 (define_insn_and_split "*fcmuo<mode>_and_combine"
7907 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa")
7908 (and:<VPRED>
7909 (unspec:<VPRED>
7910 [(match_operand:<VPRED> 1)
7911 (const_int SVE_KNOWN_PTRUE)
7912 (match_operand:SVE_FULL_F 2 "register_operand" "w")
7913 (match_operand:SVE_FULL_F 3 "register_operand" "w")]
7914 UNSPEC_COND_FCMUO)
7915 (match_operand:<VPRED> 4 "register_operand" "Upl")))]
7916 "TARGET_SVE"
7917 "#"
7918 "&& 1"
7919 [(set (match_dup 0)
7920 (unspec:<VPRED>
7921 [(match_dup 4)
7922 (const_int SVE_MAYBE_NOT_PTRUE)
7923 (match_dup 2)
7924 (match_dup 3)]
7925 UNSPEC_COND_FCMUO))]
7926 )
7927
7928 ;; -------------------------------------------------------------------------
7929 ;; ---- [FP] Absolute comparisons
7930 ;; -------------------------------------------------------------------------
7931 ;; Includes:
7932 ;; - FACGE
7933 ;; - FACGT
7934 ;; - FACLE
7935 ;; - FACLT
7936 ;; -------------------------------------------------------------------------
7937
7938 ;; Predicated floating-point absolute comparisons.
7939 (define_expand "@aarch64_pred_fac<cmp_op><mode>"
7940 [(set (match_operand:<VPRED> 0 "register_operand")
7941 (unspec:<VPRED>
7942 [(match_operand:<VPRED> 1 "register_operand")
7943 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
7944 (unspec:SVE_FULL_F
7945 [(match_dup 1)
7946 (match_dup 2)
7947 (match_operand:SVE_FULL_F 3 "register_operand")]
7948 UNSPEC_COND_FABS)
7949 (unspec:SVE_FULL_F
7950 [(match_dup 1)
7951 (match_dup 2)
7952 (match_operand:SVE_FULL_F 4 "register_operand")]
7953 UNSPEC_COND_FABS)]
7954 SVE_COND_FP_ABS_CMP))]
7955 "TARGET_SVE"
7956 )
7957
7958 (define_insn_and_rewrite "*aarch64_pred_fac<cmp_op><mode>_relaxed"
7959 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa")
7960 (unspec:<VPRED>
7961 [(match_operand:<VPRED> 1 "register_operand" "Upl")
7962 (match_operand:SI 4 "aarch64_sve_ptrue_flag")
7963 (unspec:SVE_FULL_F
7964 [(match_operand 5)
7965 (const_int SVE_RELAXED_GP)
7966 (match_operand:SVE_FULL_F 2 "register_operand" "w")]
7967 UNSPEC_COND_FABS)
7968 (unspec:SVE_FULL_F
7969 [(match_operand 6)
7970 (const_int SVE_RELAXED_GP)
7971 (match_operand:SVE_FULL_F 3 "register_operand" "w")]
7972 UNSPEC_COND_FABS)]
7973 SVE_COND_FP_ABS_CMP))]
7974 "TARGET_SVE"
7975 "fac<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
7976 "&& (!rtx_equal_p (operands[1], operands[5])
7977 || !rtx_equal_p (operands[1], operands[6]))"
7978 {
7979 operands[5] = copy_rtx (operands[1]);
7980 operands[6] = copy_rtx (operands[1]);
7981 }
7982 )
7983
7984 (define_insn "*aarch64_pred_fac<cmp_op><mode>_strict"
7985 [(set (match_operand:<VPRED> 0 "register_operand" "=Upa")
7986 (unspec:<VPRED>
7987 [(match_operand:<VPRED> 1 "register_operand" "Upl")
7988 (match_operand:SI 4 "aarch64_sve_ptrue_flag")
7989 (unspec:SVE_FULL_F
7990 [(match_dup 1)
7991 (match_operand:SI 5 "aarch64_sve_gp_strictness")
7992 (match_operand:SVE_FULL_F 2 "register_operand" "w")]
7993 UNSPEC_COND_FABS)
7994 (unspec:SVE_FULL_F
7995 [(match_dup 1)
7996 (match_operand:SI 6 "aarch64_sve_gp_strictness")
7997 (match_operand:SVE_FULL_F 3 "register_operand" "w")]
7998 UNSPEC_COND_FABS)]
7999 SVE_COND_FP_ABS_CMP))]
8000 "TARGET_SVE"
8001 "fac<cmp_op>\t%0.<Vetype>, %1/z, %2.<Vetype>, %3.<Vetype>"
8002 )
8003
8004 ;; -------------------------------------------------------------------------
8005 ;; ---- [PRED] Select
8006 ;; -------------------------------------------------------------------------
8007 ;; Includes:
8008 ;; - SEL
8009 ;; -------------------------------------------------------------------------
8010
8011 (define_insn "@vcond_mask_<mode><mode>"
8012 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
8013 (ior:PRED_ALL
8014 (and:PRED_ALL
8015 (match_operand:PRED_ALL 3 "register_operand" "Upa")
8016 (match_operand:PRED_ALL 1 "register_operand" "Upa"))
8017 (and:PRED_ALL
8018 (not (match_dup 3))
8019 (match_operand:PRED_ALL 2 "register_operand" "Upa"))))]
8020 "TARGET_SVE"
8021 "sel\t%0.b, %3, %1.b, %2.b"
8022 )
8023
8024 ;; -------------------------------------------------------------------------
8025 ;; ---- [PRED] Test bits
8026 ;; -------------------------------------------------------------------------
8027 ;; Includes:
8028 ;; - PTEST
8029 ;; -------------------------------------------------------------------------
8030
8031 ;; Branch based on predicate equality or inequality.
8032 (define_expand "cbranch<mode>4"
8033 [(set (pc)
8034 (if_then_else
8035 (match_operator 0 "aarch64_equality_operator"
8036 [(match_operand:PRED_ALL 1 "register_operand")
8037 (match_operand:PRED_ALL 2 "aarch64_simd_reg_or_zero")])
8038 (label_ref (match_operand 3 ""))
8039 (pc)))]
8040 ""
8041 {
8042 rtx ptrue = force_reg (VNx16BImode, aarch64_ptrue_all (<data_bytes>));
8043 rtx cast_ptrue = gen_lowpart (<MODE>mode, ptrue);
8044 rtx ptrue_flag = gen_int_mode (SVE_KNOWN_PTRUE, SImode);
8045 rtx pred;
8046 if (operands[2] == CONST0_RTX (<MODE>mode))
8047 pred = operands[1];
8048 else
8049 {
8050 pred = gen_reg_rtx (<MODE>mode);
8051 emit_insn (gen_aarch64_pred_xor<mode>_z (pred, cast_ptrue, operands[1],
8052 operands[2]));
8053 }
8054 emit_insn (gen_aarch64_ptest<mode> (ptrue, cast_ptrue, ptrue_flag, pred));
8055 operands[1] = gen_rtx_REG (CC_NZCmode, CC_REGNUM);
8056 operands[2] = const0_rtx;
8057 }
8058 )
8059
8060 ;; See "Description of UNSPEC_PTEST" above for details.
8061 (define_insn "aarch64_ptest<mode>"
8062 [(set (reg:CC_NZC CC_REGNUM)
8063 (unspec:CC_NZC [(match_operand:VNx16BI 0 "register_operand" "Upa")
8064 (match_operand 1)
8065 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
8066 (match_operand:PRED_ALL 3 "register_operand" "Upa")]
8067 UNSPEC_PTEST))]
8068 "TARGET_SVE"
8069 "ptest\t%0, %3.b"
8070 )
8071
8072 ;; =========================================================================
8073 ;; == Reductions
8074 ;; =========================================================================
8075
8076 ;; -------------------------------------------------------------------------
8077 ;; ---- [INT,FP] Conditional reductions
8078 ;; -------------------------------------------------------------------------
8079 ;; Includes:
8080 ;; - CLASTA
8081 ;; - CLASTB
8082 ;; -------------------------------------------------------------------------
8083
8084 ;; Set operand 0 to the last active element in operand 3, or to tied
8085 ;; operand 1 if no elements are active.
8086 (define_insn "@fold_extract_<last_op>_<mode>"
8087 [(set (match_operand:<VEL> 0 "register_operand" "=?r, w")
8088 (unspec:<VEL>
8089 [(match_operand:<VEL> 1 "register_operand" "0, 0")
8090 (match_operand:<VPRED> 2 "register_operand" "Upl, Upl")
8091 (match_operand:SVE_FULL 3 "register_operand" "w, w")]
8092 CLAST))]
8093 "TARGET_SVE"
8094 "@
8095 clast<ab>\t%<vwcore>0, %2, %<vwcore>0, %3.<Vetype>
8096 clast<ab>\t%<Vetype>0, %2, %<Vetype>0, %3.<Vetype>"
8097 )
8098
8099 (define_insn "@aarch64_fold_extract_vector_<last_op>_<mode>"
8100 [(set (match_operand:SVE_FULL 0 "register_operand" "=w, ?&w")
8101 (unspec:SVE_FULL
8102 [(match_operand:SVE_FULL 1 "register_operand" "0, w")
8103 (match_operand:<VPRED> 2 "register_operand" "Upl, Upl")
8104 (match_operand:SVE_FULL 3 "register_operand" "w, w")]
8105 CLAST))]
8106 "TARGET_SVE"
8107 "@
8108 clast<ab>\t%0.<Vetype>, %2, %0.<Vetype>, %3.<Vetype>
8109 movprfx\t%0, %1\;clast<ab>\t%0.<Vetype>, %2, %0.<Vetype>, %3.<Vetype>"
8110 )
8111
8112 ;; -------------------------------------------------------------------------
8113 ;; ---- [INT] Tree reductions
8114 ;; -------------------------------------------------------------------------
8115 ;; Includes:
8116 ;; - ANDV
8117 ;; - EORV
8118 ;; - ORV
8119 ;; - SADDV
8120 ;; - SMAXV
8121 ;; - SMINV
8122 ;; - UADDV
8123 ;; - UMAXV
8124 ;; - UMINV
8125 ;; -------------------------------------------------------------------------
8126
8127 ;; Unpredicated integer add reduction.
8128 (define_expand "reduc_plus_scal_<mode>"
8129 [(match_operand:<VEL> 0 "register_operand")
8130 (match_operand:SVE_FULL_I 1 "register_operand")]
8131 "TARGET_SVE"
8132 {
8133 rtx pred = aarch64_ptrue_reg (<VPRED>mode);
8134 rtx tmp = <VEL>mode == DImode ? operands[0] : gen_reg_rtx (DImode);
8135 emit_insn (gen_aarch64_pred_reduc_uadd_<mode> (tmp, pred, operands[1]));
8136 if (tmp != operands[0])
8137 emit_move_insn (operands[0], gen_lowpart (<VEL>mode, tmp));
8138 DONE;
8139 }
8140 )
8141
8142 ;; Predicated integer add reduction. The result is always 64-bits.
8143 (define_insn "@aarch64_pred_reduc_<optab>_<mode>"
8144 [(set (match_operand:DI 0 "register_operand" "=w")
8145 (unspec:DI [(match_operand:<VPRED> 1 "register_operand" "Upl")
8146 (match_operand:SVE_FULL_I 2 "register_operand" "w")]
8147 SVE_INT_ADDV))]
8148 "TARGET_SVE && <max_elem_bits> >= <elem_bits>"
8149 "<su>addv\t%d0, %1, %2.<Vetype>"
8150 )
8151
8152 ;; Unpredicated integer reductions.
8153 (define_expand "reduc_<optab>_scal_<mode>"
8154 [(set (match_operand:<VEL> 0 "register_operand")
8155 (unspec:<VEL> [(match_dup 2)
8156 (match_operand:SVE_FULL_I 1 "register_operand")]
8157 SVE_INT_REDUCTION))]
8158 "TARGET_SVE"
8159 {
8160 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
8161 }
8162 )
8163
8164 ;; Predicated integer reductions.
8165 (define_insn "@aarch64_pred_reduc_<optab>_<mode>"
8166 [(set (match_operand:<VEL> 0 "register_operand" "=w")
8167 (unspec:<VEL> [(match_operand:<VPRED> 1 "register_operand" "Upl")
8168 (match_operand:SVE_FULL_I 2 "register_operand" "w")]
8169 SVE_INT_REDUCTION))]
8170 "TARGET_SVE"
8171 "<sve_int_op>\t%<Vetype>0, %1, %2.<Vetype>"
8172 )
8173
8174 ;; -------------------------------------------------------------------------
8175 ;; ---- [FP] Tree reductions
8176 ;; -------------------------------------------------------------------------
8177 ;; Includes:
8178 ;; - FADDV
8179 ;; - FMAXNMV
8180 ;; - FMAXV
8181 ;; - FMINNMV
8182 ;; - FMINV
8183 ;; -------------------------------------------------------------------------
8184
8185 ;; Unpredicated floating-point tree reductions.
8186 (define_expand "reduc_<optab>_scal_<mode>"
8187 [(set (match_operand:<VEL> 0 "register_operand")
8188 (unspec:<VEL> [(match_dup 2)
8189 (match_operand:SVE_FULL_F 1 "register_operand")]
8190 SVE_FP_REDUCTION))]
8191 "TARGET_SVE"
8192 {
8193 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
8194 }
8195 )
8196
8197 ;; Predicated floating-point tree reductions.
8198 (define_insn "@aarch64_pred_reduc_<optab>_<mode>"
8199 [(set (match_operand:<VEL> 0 "register_operand" "=w")
8200 (unspec:<VEL> [(match_operand:<VPRED> 1 "register_operand" "Upl")
8201 (match_operand:SVE_FULL_F 2 "register_operand" "w")]
8202 SVE_FP_REDUCTION))]
8203 "TARGET_SVE"
8204 "<sve_fp_op>\t%<Vetype>0, %1, %2.<Vetype>"
8205 )
8206
8207 ;; -------------------------------------------------------------------------
8208 ;; ---- [FP] Left-to-right reductions
8209 ;; -------------------------------------------------------------------------
8210 ;; Includes:
8211 ;; - FADDA
8212 ;; -------------------------------------------------------------------------
8213
8214 ;; Unpredicated in-order FP reductions.
8215 (define_expand "fold_left_plus_<mode>"
8216 [(set (match_operand:<VEL> 0 "register_operand")
8217 (unspec:<VEL> [(match_dup 3)
8218 (match_operand:<VEL> 1 "register_operand")
8219 (match_operand:SVE_FULL_F 2 "register_operand")]
8220 UNSPEC_FADDA))]
8221 "TARGET_SVE"
8222 {
8223 operands[3] = aarch64_ptrue_reg (<VPRED>mode);
8224 }
8225 )
8226
8227 ;; Predicated in-order FP reductions.
8228 (define_insn "mask_fold_left_plus_<mode>"
8229 [(set (match_operand:<VEL> 0 "register_operand" "=w")
8230 (unspec:<VEL> [(match_operand:<VPRED> 3 "register_operand" "Upl")
8231 (match_operand:<VEL> 1 "register_operand" "0")
8232 (match_operand:SVE_FULL_F 2 "register_operand" "w")]
8233 UNSPEC_FADDA))]
8234 "TARGET_SVE"
8235 "fadda\t%<Vetype>0, %3, %<Vetype>0, %2.<Vetype>"
8236 )
8237
8238 ;; =========================================================================
8239 ;; == Permutes
8240 ;; =========================================================================
8241
8242 ;; -------------------------------------------------------------------------
8243 ;; ---- [INT,FP] General permutes
8244 ;; -------------------------------------------------------------------------
8245 ;; Includes:
8246 ;; - TBL
8247 ;; -------------------------------------------------------------------------
8248
8249 (define_expand "vec_perm<mode>"
8250 [(match_operand:SVE_FULL 0 "register_operand")
8251 (match_operand:SVE_FULL 1 "register_operand")
8252 (match_operand:SVE_FULL 2 "register_operand")
8253 (match_operand:<V_INT_EQUIV> 3 "aarch64_sve_vec_perm_operand")]
8254 "TARGET_SVE && GET_MODE_NUNITS (<MODE>mode).is_constant ()"
8255 {
8256 aarch64_expand_sve_vec_perm (operands[0], operands[1],
8257 operands[2], operands[3]);
8258 DONE;
8259 }
8260 )
8261
8262 (define_insn "@aarch64_sve_tbl<mode>"
8263 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
8264 (unspec:SVE_FULL
8265 [(match_operand:SVE_FULL 1 "register_operand" "w")
8266 (match_operand:<V_INT_EQUIV> 2 "register_operand" "w")]
8267 UNSPEC_TBL))]
8268 "TARGET_SVE"
8269 "tbl\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
8270 )
8271
8272 ;; -------------------------------------------------------------------------
8273 ;; ---- [INT,FP] Special-purpose unary permutes
8274 ;; -------------------------------------------------------------------------
8275 ;; Includes:
8276 ;; - COMPACT
8277 ;; - DUP
8278 ;; - REV
8279 ;; -------------------------------------------------------------------------
8280
8281 ;; Compact active elements and pad with zeros.
8282 (define_insn "@aarch64_sve_compact<mode>"
8283 [(set (match_operand:SVE_FULL_SD 0 "register_operand" "=w")
8284 (unspec:SVE_FULL_SD
8285 [(match_operand:<VPRED> 1 "register_operand" "Upl")
8286 (match_operand:SVE_FULL_SD 2 "register_operand" "w")]
8287 UNSPEC_SVE_COMPACT))]
8288 "TARGET_SVE"
8289 "compact\t%0.<Vetype>, %1, %2.<Vetype>"
8290 )
8291
8292 ;; Duplicate one element of a vector.
8293 (define_insn "@aarch64_sve_dup_lane<mode>"
8294 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
8295 (vec_duplicate:SVE_FULL
8296 (vec_select:<VEL>
8297 (match_operand:SVE_FULL 1 "register_operand" "w")
8298 (parallel [(match_operand:SI 2 "const_int_operand")]))))]
8299 "TARGET_SVE
8300 && IN_RANGE (INTVAL (operands[2]) * GET_MODE_SIZE (<VEL>mode), 0, 63)"
8301 "dup\t%0.<Vetype>, %1.<Vetype>[%2]"
8302 )
8303
8304 ;; Use DUP.Q to duplicate a 128-bit segment of a register.
8305 ;;
8306 ;; The vec_select:<V128> sets memory lane number N of the V128 to lane
8307 ;; number op2 + N of op1. (We don't need to distinguish between memory
8308 ;; and architectural register lane numbering for op1 or op0, since the
8309 ;; two numbering schemes are the same for SVE.)
8310 ;;
8311 ;; The vec_duplicate:SVE_FULL then copies memory lane number N of the
8312 ;; V128 (and thus lane number op2 + N of op1) to lane numbers N + I * STEP
8313 ;; of op0. We therefore get the correct result for both endiannesses.
8314 ;;
8315 ;; The wrinkle is that for big-endian V128 registers, memory lane numbering
8316 ;; is in the opposite order to architectural register lane numbering.
8317 ;; Thus if we were to do this operation via a V128 temporary register,
8318 ;; the vec_select and vec_duplicate would both involve a reverse operation
8319 ;; for big-endian targets. In this fused pattern the two reverses cancel
8320 ;; each other out.
8321 (define_insn "@aarch64_sve_dupq_lane<mode>"
8322 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
8323 (vec_duplicate:SVE_FULL
8324 (vec_select:<V128>
8325 (match_operand:SVE_FULL 1 "register_operand" "w")
8326 (match_operand 2 "ascending_int_parallel"))))]
8327 "TARGET_SVE
8328 && (INTVAL (XVECEXP (operands[2], 0, 0))
8329 * GET_MODE_SIZE (<VEL>mode)) % 16 == 0
8330 && IN_RANGE (INTVAL (XVECEXP (operands[2], 0, 0))
8331 * GET_MODE_SIZE (<VEL>mode), 0, 63)"
8332 {
8333 unsigned int byte = (INTVAL (XVECEXP (operands[2], 0, 0))
8334 * GET_MODE_SIZE (<VEL>mode));
8335 operands[2] = gen_int_mode (byte / 16, DImode);
8336 return "dup\t%0.q, %1.q[%2]";
8337 }
8338 )
8339
8340 ;; Reverse the order of elements within a full vector.
8341 (define_insn "@aarch64_sve_rev<mode>"
8342 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
8343 (unspec:SVE_FULL
8344 [(match_operand:SVE_FULL 1 "register_operand" "w")]
8345 UNSPEC_REV))]
8346 "TARGET_SVE"
8347 "rev\t%0.<Vetype>, %1.<Vetype>")
8348
8349 ;; -------------------------------------------------------------------------
8350 ;; ---- [INT,FP] Special-purpose binary permutes
8351 ;; -------------------------------------------------------------------------
8352 ;; Includes:
8353 ;; - SPLICE
8354 ;; - TRN1
8355 ;; - TRN2
8356 ;; - UZP1
8357 ;; - UZP2
8358 ;; - ZIP1
8359 ;; - ZIP2
8360 ;; -------------------------------------------------------------------------
8361
8362 ;; Like EXT, but start at the first active element.
8363 (define_insn "@aarch64_sve_splice<mode>"
8364 [(set (match_operand:SVE_FULL 0 "register_operand" "=w, ?&w")
8365 (unspec:SVE_FULL
8366 [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
8367 (match_operand:SVE_FULL 2 "register_operand" "0, w")
8368 (match_operand:SVE_FULL 3 "register_operand" "w, w")]
8369 UNSPEC_SVE_SPLICE))]
8370 "TARGET_SVE"
8371 "@
8372 splice\t%0.<Vetype>, %1, %0.<Vetype>, %3.<Vetype>
8373 movprfx\t%0, %2\;splice\t%0.<Vetype>, %1, %0.<Vetype>, %3.<Vetype>"
8374 [(set_attr "movprfx" "*, yes")]
8375 )
8376
8377 ;; Permutes that take half the elements from one vector and half the
8378 ;; elements from the other.
8379 (define_insn "@aarch64_sve_<perm_insn><mode>"
8380 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
8381 (unspec:SVE_FULL
8382 [(match_operand:SVE_FULL 1 "register_operand" "w")
8383 (match_operand:SVE_FULL 2 "register_operand" "w")]
8384 PERMUTE))]
8385 "TARGET_SVE"
8386 "<perm_insn>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
8387 )
8388
8389 ;; Apply PERMUTE to 128-bit sequences. The behavior of these patterns
8390 ;; doesn't depend on the mode.
8391 (define_insn "@aarch64_sve_<optab><mode>"
8392 [(set (match_operand:SVE_FULL 0 "register_operand" "=w")
8393 (unspec:SVE_FULL
8394 [(match_operand:SVE_FULL 1 "register_operand" "w")
8395 (match_operand:SVE_FULL 2 "register_operand" "w")]
8396 PERMUTEQ))]
8397 "TARGET_SVE_F64MM"
8398 "<perm_insn>\t%0.q, %1.q, %2.q"
8399 )
8400
8401 ;; Concatenate two vectors and extract a subvector. Note that the
8402 ;; immediate (third) operand is the lane index not the byte index.
8403 (define_insn "@aarch64_sve_ext<mode>"
8404 [(set (match_operand:SVE_FULL 0 "register_operand" "=w, ?&w")
8405 (unspec:SVE_FULL
8406 [(match_operand:SVE_FULL 1 "register_operand" "0, w")
8407 (match_operand:SVE_FULL 2 "register_operand" "w, w")
8408 (match_operand:SI 3 "const_int_operand")]
8409 UNSPEC_EXT))]
8410 "TARGET_SVE
8411 && IN_RANGE (INTVAL (operands[3]) * GET_MODE_SIZE (<VEL>mode), 0, 255)"
8412 {
8413 operands[3] = GEN_INT (INTVAL (operands[3]) * GET_MODE_SIZE (<VEL>mode));
8414 return (which_alternative == 0
8415 ? "ext\\t%0.b, %0.b, %2.b, #%3"
8416 : "movprfx\t%0, %1\;ext\\t%0.b, %0.b, %2.b, #%3");
8417 }
8418 [(set_attr "movprfx" "*,yes")]
8419 )
8420
8421 ;; -------------------------------------------------------------------------
8422 ;; ---- [PRED] Special-purpose unary permutes
8423 ;; -------------------------------------------------------------------------
8424 ;; Includes:
8425 ;; - REV
8426 ;; -------------------------------------------------------------------------
8427
8428 (define_insn "@aarch64_sve_rev<mode>"
8429 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
8430 (unspec:PRED_ALL [(match_operand:PRED_ALL 1 "register_operand" "Upa")]
8431 UNSPEC_REV))]
8432 "TARGET_SVE"
8433 "rev\t%0.<Vetype>, %1.<Vetype>")
8434
8435 ;; -------------------------------------------------------------------------
8436 ;; ---- [PRED] Special-purpose binary permutes
8437 ;; -------------------------------------------------------------------------
8438 ;; Includes:
8439 ;; - TRN1
8440 ;; - TRN2
8441 ;; - UZP1
8442 ;; - UZP2
8443 ;; - ZIP1
8444 ;; - ZIP2
8445 ;; -------------------------------------------------------------------------
8446
8447 ;; Permutes that take half the elements from one vector and half the
8448 ;; elements from the other.
8449 (define_insn "@aarch64_sve_<perm_insn><mode>"
8450 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
8451 (unspec:PRED_ALL [(match_operand:PRED_ALL 1 "register_operand" "Upa")
8452 (match_operand:PRED_ALL 2 "register_operand" "Upa")]
8453 PERMUTE))]
8454 "TARGET_SVE"
8455 "<perm_insn>\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
8456 )
8457
8458 ;; Special purpose permute used by the predicate generation instructions.
8459 ;; Unlike the normal permute patterns, these instructions operate on VNx16BI
8460 ;; regardless of the element size, so that all input and output bits are
8461 ;; well-defined. Operand 3 then indicates the size of the permute.
8462 (define_insn "@aarch64_sve_trn1_conv<mode>"
8463 [(set (match_operand:VNx16BI 0 "register_operand" "=Upa")
8464 (unspec:VNx16BI [(match_operand:VNx16BI 1 "register_operand" "Upa")
8465 (match_operand:VNx16BI 2 "register_operand" "Upa")
8466 (match_operand:PRED_ALL 3 "aarch64_simd_imm_zero")]
8467 UNSPEC_TRN1_CONV))]
8468 "TARGET_SVE"
8469 "trn1\t%0.<PRED_ALL:Vetype>, %1.<PRED_ALL:Vetype>, %2.<PRED_ALL:Vetype>"
8470 )
8471
8472 ;; =========================================================================
8473 ;; == Conversions
8474 ;; =========================================================================
8475
8476 ;; -------------------------------------------------------------------------
8477 ;; ---- [INT<-INT] Packs
8478 ;; -------------------------------------------------------------------------
8479 ;; Includes:
8480 ;; - UZP1
8481 ;; -------------------------------------------------------------------------
8482
8483 ;; Integer pack. Use UZP1 on the narrower type, which discards
8484 ;; the high part of each wide element.
8485 (define_insn "vec_pack_trunc_<Vwide>"
8486 [(set (match_operand:SVE_FULL_BHSI 0 "register_operand" "=w")
8487 (unspec:SVE_FULL_BHSI
8488 [(match_operand:<VWIDE> 1 "register_operand" "w")
8489 (match_operand:<VWIDE> 2 "register_operand" "w")]
8490 UNSPEC_PACK))]
8491 "TARGET_SVE"
8492 "uzp1\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
8493 )
8494
8495 ;; -------------------------------------------------------------------------
8496 ;; ---- [INT<-INT] Unpacks
8497 ;; -------------------------------------------------------------------------
8498 ;; Includes:
8499 ;; - SUNPKHI
8500 ;; - SUNPKLO
8501 ;; - UUNPKHI
8502 ;; - UUNPKLO
8503 ;; -------------------------------------------------------------------------
8504
8505 ;; Unpack the low or high half of a vector, where "high" refers to
8506 ;; the low-numbered lanes for big-endian and the high-numbered lanes
8507 ;; for little-endian.
8508 (define_expand "vec_unpack<su>_<perm_hilo>_<SVE_FULL_BHSI:mode>"
8509 [(match_operand:<VWIDE> 0 "register_operand")
8510 (unspec:<VWIDE>
8511 [(match_operand:SVE_FULL_BHSI 1 "register_operand")] UNPACK)]
8512 "TARGET_SVE"
8513 {
8514 emit_insn ((<hi_lanes_optab>
8515 ? gen_aarch64_sve_<su>unpkhi_<SVE_FULL_BHSI:mode>
8516 : gen_aarch64_sve_<su>unpklo_<SVE_FULL_BHSI:mode>)
8517 (operands[0], operands[1]));
8518 DONE;
8519 }
8520 )
8521
8522 (define_insn "@aarch64_sve_<su>unpk<perm_hilo>_<SVE_FULL_BHSI:mode>"
8523 [(set (match_operand:<VWIDE> 0 "register_operand" "=w")
8524 (unspec:<VWIDE>
8525 [(match_operand:SVE_FULL_BHSI 1 "register_operand" "w")]
8526 UNPACK))]
8527 "TARGET_SVE"
8528 "<su>unpk<perm_hilo>\t%0.<Vewtype>, %1.<Vetype>"
8529 )
8530
8531 ;; -------------------------------------------------------------------------
8532 ;; ---- [INT<-FP] Conversions
8533 ;; -------------------------------------------------------------------------
8534 ;; Includes:
8535 ;; - FCVTZS
8536 ;; - FCVTZU
8537 ;; -------------------------------------------------------------------------
8538
8539 ;; Unpredicated conversion of floats to integers of the same size (HF to HI,
8540 ;; SF to SI or DF to DI).
8541 (define_expand "<optab><mode><v_int_equiv>2"
8542 [(set (match_operand:<V_INT_EQUIV> 0 "register_operand")
8543 (unspec:<V_INT_EQUIV>
8544 [(match_dup 2)
8545 (const_int SVE_RELAXED_GP)
8546 (match_operand:SVE_FULL_F 1 "register_operand")]
8547 SVE_COND_FCVTI))]
8548 "TARGET_SVE"
8549 {
8550 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
8551 }
8552 )
8553
8554 ;; Predicated float-to-integer conversion, either to the same width or wider.
8555 (define_insn "@aarch64_sve_<optab>_nontrunc<SVE_FULL_F:mode><SVE_FULL_HSDI:mode>"
8556 [(set (match_operand:SVE_FULL_HSDI 0 "register_operand" "=w, ?&w")
8557 (unspec:SVE_FULL_HSDI
8558 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand" "Upl, Upl")
8559 (match_operand:SI 3 "aarch64_sve_gp_strictness")
8560 (match_operand:SVE_FULL_F 2 "register_operand" "0, w")]
8561 SVE_COND_FCVTI))]
8562 "TARGET_SVE && <SVE_FULL_HSDI:elem_bits> >= <SVE_FULL_F:elem_bits>"
8563 "@
8564 fcvtz<su>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_F:Vetype>
8565 movprfx\t%0, %2\;fcvtz<su>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_F:Vetype>"
8566 [(set_attr "movprfx" "*,yes")]
8567 )
8568
8569 ;; Predicated narrowing float-to-integer conversion.
8570 (define_insn "@aarch64_sve_<optab>_trunc<VNx2DF_ONLY:mode><VNx4SI_ONLY:mode>"
8571 [(set (match_operand:VNx4SI_ONLY 0 "register_operand" "=w, ?&w")
8572 (unspec:VNx4SI_ONLY
8573 [(match_operand:VNx2BI 1 "register_operand" "Upl, Upl")
8574 (match_operand:SI 3 "aarch64_sve_gp_strictness")
8575 (match_operand:VNx2DF_ONLY 2 "register_operand" "0, w")]
8576 SVE_COND_FCVTI))]
8577 "TARGET_SVE"
8578 "@
8579 fcvtz<su>\t%0.<VNx4SI_ONLY:Vetype>, %1/m, %2.<VNx2DF_ONLY:Vetype>
8580 movprfx\t%0, %2\;fcvtz<su>\t%0.<VNx4SI_ONLY:Vetype>, %1/m, %2.<VNx2DF_ONLY:Vetype>"
8581 [(set_attr "movprfx" "*,yes")]
8582 )
8583
8584 ;; Predicated float-to-integer conversion with merging, either to the same
8585 ;; width or wider.
8586 (define_expand "@cond_<optab>_nontrunc<SVE_FULL_F:mode><SVE_FULL_HSDI:mode>"
8587 [(set (match_operand:SVE_FULL_HSDI 0 "register_operand")
8588 (unspec:SVE_FULL_HSDI
8589 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand")
8590 (unspec:SVE_FULL_HSDI
8591 [(match_dup 1)
8592 (const_int SVE_STRICT_GP)
8593 (match_operand:SVE_FULL_F 2 "register_operand")]
8594 SVE_COND_FCVTI)
8595 (match_operand:SVE_FULL_HSDI 3 "aarch64_simd_reg_or_zero")]
8596 UNSPEC_SEL))]
8597 "TARGET_SVE && <SVE_FULL_HSDI:elem_bits> >= <SVE_FULL_F:elem_bits>"
8598 )
8599
8600 ;; The first alternative doesn't need the earlyclobber, but the only case
8601 ;; it would help is the uninteresting one in which operands 2 and 3 are
8602 ;; the same register (despite having different modes). Making all the
8603 ;; alternatives earlyclobber makes things more consistent for the
8604 ;; register allocator.
8605 (define_insn_and_rewrite "*cond_<optab>_nontrunc<SVE_FULL_F:mode><SVE_FULL_HSDI:mode>_relaxed"
8606 [(set (match_operand:SVE_FULL_HSDI 0 "register_operand" "=&w, &w, ?&w")
8607 (unspec:SVE_FULL_HSDI
8608 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand" "Upl, Upl, Upl")
8609 (unspec:SVE_FULL_HSDI
8610 [(match_operand 4)
8611 (const_int SVE_RELAXED_GP)
8612 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w")]
8613 SVE_COND_FCVTI)
8614 (match_operand:SVE_FULL_HSDI 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
8615 UNSPEC_SEL))]
8616 "TARGET_SVE && <SVE_FULL_HSDI:elem_bits> >= <SVE_FULL_F:elem_bits>"
8617 "@
8618 fcvtz<su>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_F:Vetype>
8619 movprfx\t%0.<SVE_FULL_HSDI:Vetype>, %1/z, %2.<SVE_FULL_HSDI:Vetype>\;fcvtz<su>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_F:Vetype>
8620 movprfx\t%0, %3\;fcvtz<su>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_F:Vetype>"
8621 "&& !rtx_equal_p (operands[1], operands[4])"
8622 {
8623 operands[4] = copy_rtx (operands[1]);
8624 }
8625 [(set_attr "movprfx" "*,yes,yes")]
8626 )
8627
8628 (define_insn "*cond_<optab>_nontrunc<SVE_FULL_F:mode><SVE_FULL_HSDI:mode>_strict"
8629 [(set (match_operand:SVE_FULL_HSDI 0 "register_operand" "=&w, &w, ?&w")
8630 (unspec:SVE_FULL_HSDI
8631 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand" "Upl, Upl, Upl")
8632 (unspec:SVE_FULL_HSDI
8633 [(match_dup 1)
8634 (const_int SVE_STRICT_GP)
8635 (match_operand:SVE_FULL_F 2 "register_operand" "w, w, w")]
8636 SVE_COND_FCVTI)
8637 (match_operand:SVE_FULL_HSDI 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
8638 UNSPEC_SEL))]
8639 "TARGET_SVE && <SVE_FULL_HSDI:elem_bits> >= <SVE_FULL_F:elem_bits>"
8640 "@
8641 fcvtz<su>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_F:Vetype>
8642 movprfx\t%0.<SVE_FULL_HSDI:Vetype>, %1/z, %2.<SVE_FULL_HSDI:Vetype>\;fcvtz<su>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_F:Vetype>
8643 movprfx\t%0, %3\;fcvtz<su>\t%0.<SVE_FULL_HSDI:Vetype>, %1/m, %2.<SVE_FULL_F:Vetype>"
8644 [(set_attr "movprfx" "*,yes,yes")]
8645 )
8646
8647 ;; Predicated narrowing float-to-integer conversion with merging.
8648 (define_expand "@cond_<optab>_trunc<VNx2DF_ONLY:mode><VNx4SI_ONLY:mode>"
8649 [(set (match_operand:VNx4SI_ONLY 0 "register_operand")
8650 (unspec:VNx4SI_ONLY
8651 [(match_operand:VNx2BI 1 "register_operand")
8652 (unspec:VNx4SI_ONLY
8653 [(match_dup 1)
8654 (const_int SVE_STRICT_GP)
8655 (match_operand:VNx2DF_ONLY 2 "register_operand")]
8656 SVE_COND_FCVTI)
8657 (match_operand:VNx4SI_ONLY 3 "aarch64_simd_reg_or_zero")]
8658 UNSPEC_SEL))]
8659 "TARGET_SVE"
8660 )
8661
8662 (define_insn "*cond_<optab>_trunc<VNx2DF_ONLY:mode><VNx4SI_ONLY:mode>"
8663 [(set (match_operand:VNx4SI_ONLY 0 "register_operand" "=&w, &w, ?&w")
8664 (unspec:VNx4SI_ONLY
8665 [(match_operand:VNx2BI 1 "register_operand" "Upl, Upl, Upl")
8666 (unspec:VNx4SI_ONLY
8667 [(match_dup 1)
8668 (match_operand:SI 4 "aarch64_sve_gp_strictness")
8669 (match_operand:VNx2DF_ONLY 2 "register_operand" "w, w, w")]
8670 SVE_COND_FCVTI)
8671 (match_operand:VNx4SI_ONLY 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
8672 UNSPEC_SEL))]
8673 "TARGET_SVE"
8674 "@
8675 fcvtz<su>\t%0.<VNx4SI_ONLY:Vetype>, %1/m, %2.<VNx2DF_ONLY:Vetype>
8676 movprfx\t%0.<VNx2DF_ONLY:Vetype>, %1/z, %2.<VNx2DF_ONLY:Vetype>\;fcvtz<su>\t%0.<VNx4SI_ONLY:Vetype>, %1/m, %2.<VNx2DF_ONLY:Vetype>
8677 movprfx\t%0, %3\;fcvtz<su>\t%0.<VNx4SI_ONLY:Vetype>, %1/m, %2.<VNx2DF_ONLY:Vetype>"
8678 [(set_attr "movprfx" "*,yes,yes")]
8679 )
8680
8681 ;; -------------------------------------------------------------------------
8682 ;; ---- [INT<-FP] Packs
8683 ;; -------------------------------------------------------------------------
8684 ;; The patterns in this section are synthetic.
8685 ;; -------------------------------------------------------------------------
8686
8687 ;; Convert two vectors of DF to SI and pack the results into a single vector.
8688 (define_expand "vec_pack_<su>fix_trunc_vnx2df"
8689 [(set (match_dup 4)
8690 (unspec:VNx4SI
8691 [(match_dup 3)
8692 (const_int SVE_RELAXED_GP)
8693 (match_operand:VNx2DF 1 "register_operand")]
8694 SVE_COND_FCVTI))
8695 (set (match_dup 5)
8696 (unspec:VNx4SI
8697 [(match_dup 3)
8698 (const_int SVE_RELAXED_GP)
8699 (match_operand:VNx2DF 2 "register_operand")]
8700 SVE_COND_FCVTI))
8701 (set (match_operand:VNx4SI 0 "register_operand")
8702 (unspec:VNx4SI [(match_dup 4) (match_dup 5)] UNSPEC_UZP1))]
8703 "TARGET_SVE"
8704 {
8705 operands[3] = aarch64_ptrue_reg (VNx2BImode);
8706 operands[4] = gen_reg_rtx (VNx4SImode);
8707 operands[5] = gen_reg_rtx (VNx4SImode);
8708 }
8709 )
8710
8711 ;; -------------------------------------------------------------------------
8712 ;; ---- [INT<-FP] Unpacks
8713 ;; -------------------------------------------------------------------------
8714 ;; No patterns here yet!
8715 ;; -------------------------------------------------------------------------
8716
8717 ;; -------------------------------------------------------------------------
8718 ;; ---- [FP<-INT] Conversions
8719 ;; -------------------------------------------------------------------------
8720 ;; Includes:
8721 ;; - SCVTF
8722 ;; - UCVTF
8723 ;; -------------------------------------------------------------------------
8724
8725 ;; Unpredicated conversion of integers to floats of the same size
8726 ;; (HI to HF, SI to SF or DI to DF).
8727 (define_expand "<optab><v_int_equiv><mode>2"
8728 [(set (match_operand:SVE_FULL_F 0 "register_operand")
8729 (unspec:SVE_FULL_F
8730 [(match_dup 2)
8731 (const_int SVE_RELAXED_GP)
8732 (match_operand:<V_INT_EQUIV> 1 "register_operand")]
8733 SVE_COND_ICVTF))]
8734 "TARGET_SVE"
8735 {
8736 operands[2] = aarch64_ptrue_reg (<VPRED>mode);
8737 }
8738 )
8739
8740 ;; Predicated integer-to-float conversion, either to the same width or
8741 ;; narrower.
8742 (define_insn "@aarch64_sve_<optab>_nonextend<SVE_FULL_HSDI:mode><SVE_FULL_F:mode>"
8743 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=w, ?&w")
8744 (unspec:SVE_FULL_F
8745 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand" "Upl, Upl")
8746 (match_operand:SI 3 "aarch64_sve_gp_strictness")
8747 (match_operand:SVE_FULL_HSDI 2 "register_operand" "0, w")]
8748 SVE_COND_ICVTF))]
8749 "TARGET_SVE && <SVE_FULL_HSDI:elem_bits> >= <SVE_FULL_F:elem_bits>"
8750 "@
8751 <su>cvtf\t%0.<SVE_FULL_F:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>
8752 movprfx\t%0, %2\;<su>cvtf\t%0.<SVE_FULL_F:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>"
8753 [(set_attr "movprfx" "*,yes")]
8754 )
8755
8756 ;; Predicated widening integer-to-float conversion.
8757 (define_insn "@aarch64_sve_<optab>_extend<VNx4SI_ONLY:mode><VNx2DF_ONLY:mode>"
8758 [(set (match_operand:VNx2DF_ONLY 0 "register_operand" "=w, ?&w")
8759 (unspec:VNx2DF_ONLY
8760 [(match_operand:VNx2BI 1 "register_operand" "Upl, Upl")
8761 (match_operand:SI 3 "aarch64_sve_gp_strictness")
8762 (match_operand:VNx4SI_ONLY 2 "register_operand" "0, w")]
8763 SVE_COND_ICVTF))]
8764 "TARGET_SVE"
8765 "@
8766 <su>cvtf\t%0.<VNx2DF_ONLY:Vetype>, %1/m, %2.<VNx4SI_ONLY:Vetype>
8767 movprfx\t%0, %2\;<su>cvtf\t%0.<VNx2DF_ONLY:Vetype>, %1/m, %2.<VNx4SI_ONLY:Vetype>"
8768 [(set_attr "movprfx" "*,yes")]
8769 )
8770
8771 ;; Predicated integer-to-float conversion with merging, either to the same
8772 ;; width or narrower.
8773 (define_expand "@cond_<optab>_nonextend<SVE_FULL_HSDI:mode><SVE_FULL_F:mode>"
8774 [(set (match_operand:SVE_FULL_F 0 "register_operand")
8775 (unspec:SVE_FULL_F
8776 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand")
8777 (unspec:SVE_FULL_F
8778 [(match_dup 1)
8779 (const_int SVE_STRICT_GP)
8780 (match_operand:SVE_FULL_HSDI 2 "register_operand")]
8781 SVE_COND_ICVTF)
8782 (match_operand:SVE_FULL_F 3 "aarch64_simd_reg_or_zero")]
8783 UNSPEC_SEL))]
8784 "TARGET_SVE && <SVE_FULL_HSDI:elem_bits> >= <SVE_FULL_F:elem_bits>"
8785 )
8786
8787 ;; The first alternative doesn't need the earlyclobber, but the only case
8788 ;; it would help is the uninteresting one in which operands 2 and 3 are
8789 ;; the same register (despite having different modes). Making all the
8790 ;; alternatives earlyclobber makes things more consistent for the
8791 ;; register allocator.
8792 (define_insn_and_rewrite "*cond_<optab>_nonextend<SVE_FULL_HSDI:mode><SVE_FULL_F:mode>_relaxed"
8793 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, ?&w")
8794 (unspec:SVE_FULL_F
8795 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand" "Upl, Upl, Upl")
8796 (unspec:SVE_FULL_F
8797 [(match_operand 4)
8798 (const_int SVE_RELAXED_GP)
8799 (match_operand:SVE_FULL_HSDI 2 "register_operand" "w, w, w")]
8800 SVE_COND_ICVTF)
8801 (match_operand:SVE_FULL_F 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
8802 UNSPEC_SEL))]
8803 "TARGET_SVE && <SVE_FULL_HSDI:elem_bits> >= <SVE_FULL_F:elem_bits>"
8804 "@
8805 <su>cvtf\t%0.<SVE_FULL_F:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>
8806 movprfx\t%0.<SVE_FULL_HSDI:Vetype>, %1/z, %2.<SVE_FULL_HSDI:Vetype>\;<su>cvtf\t%0.<SVE_FULL_F:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>
8807 movprfx\t%0, %3\;<su>cvtf\t%0.<SVE_FULL_F:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>"
8808 "&& !rtx_equal_p (operands[1], operands[4])"
8809 {
8810 operands[4] = copy_rtx (operands[1]);
8811 }
8812 [(set_attr "movprfx" "*,yes,yes")]
8813 )
8814
8815 (define_insn "*cond_<optab>_nonextend<SVE_FULL_HSDI:mode><SVE_FULL_F:mode>_strict"
8816 [(set (match_operand:SVE_FULL_F 0 "register_operand" "=&w, &w, ?&w")
8817 (unspec:SVE_FULL_F
8818 [(match_operand:<SVE_FULL_HSDI:VPRED> 1 "register_operand" "Upl, Upl, Upl")
8819 (unspec:SVE_FULL_F
8820 [(match_dup 1)
8821 (const_int SVE_STRICT_GP)
8822 (match_operand:SVE_FULL_HSDI 2 "register_operand" "w, w, w")]
8823 SVE_COND_ICVTF)
8824 (match_operand:SVE_FULL_F 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
8825 UNSPEC_SEL))]
8826 "TARGET_SVE && <SVE_FULL_HSDI:elem_bits> >= <SVE_FULL_F:elem_bits>"
8827 "@
8828 <su>cvtf\t%0.<SVE_FULL_F:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>
8829 movprfx\t%0.<SVE_FULL_HSDI:Vetype>, %1/z, %2.<SVE_FULL_HSDI:Vetype>\;<su>cvtf\t%0.<SVE_FULL_F:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>
8830 movprfx\t%0, %3\;<su>cvtf\t%0.<SVE_FULL_F:Vetype>, %1/m, %2.<SVE_FULL_HSDI:Vetype>"
8831 [(set_attr "movprfx" "*,yes,yes")]
8832 )
8833
8834 ;; Predicated widening integer-to-float conversion with merging.
8835 (define_expand "@cond_<optab>_extend<VNx4SI_ONLY:mode><VNx2DF_ONLY:mode>"
8836 [(set (match_operand:VNx2DF_ONLY 0 "register_operand")
8837 (unspec:VNx2DF_ONLY
8838 [(match_operand:VNx2BI 1 "register_operand")
8839 (unspec:VNx2DF_ONLY
8840 [(match_dup 1)
8841 (const_int SVE_STRICT_GP)
8842 (match_operand:VNx4SI_ONLY 2 "register_operand")]
8843 SVE_COND_ICVTF)
8844 (match_operand:VNx2DF_ONLY 3 "aarch64_simd_reg_or_zero")]
8845 UNSPEC_SEL))]
8846 "TARGET_SVE"
8847 )
8848
8849 (define_insn "*cond_<optab>_extend<VNx4SI_ONLY:mode><VNx2DF_ONLY:mode>"
8850 [(set (match_operand:VNx2DF_ONLY 0 "register_operand" "=w, ?&w, ?&w")
8851 (unspec:VNx2DF_ONLY
8852 [(match_operand:VNx2BI 1 "register_operand" "Upl, Upl, Upl")
8853 (unspec:VNx2DF_ONLY
8854 [(match_dup 1)
8855 (match_operand:SI 4 "aarch64_sve_gp_strictness")
8856 (match_operand:VNx4SI_ONLY 2 "register_operand" "w, w, w")]
8857 SVE_COND_ICVTF)
8858 (match_operand:VNx2DF_ONLY 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
8859 UNSPEC_SEL))]
8860 "TARGET_SVE"
8861 "@
8862 <su>cvtf\t%0.<VNx2DF_ONLY:Vetype>, %1/m, %2.<VNx4SI_ONLY:Vetype>
8863 movprfx\t%0.<VNx2DF_ONLY:Vetype>, %1/z, %2.<VNx2DF_ONLY:Vetype>\;<su>cvtf\t%0.<VNx2DF_ONLY:Vetype>, %1/m, %2.<VNx4SI_ONLY:Vetype>
8864 movprfx\t%0, %3\;<su>cvtf\t%0.<VNx2DF_ONLY:Vetype>, %1/m, %2.<VNx4SI_ONLY:Vetype>"
8865 [(set_attr "movprfx" "*,yes,yes")]
8866 )
8867
8868 ;; -------------------------------------------------------------------------
8869 ;; ---- [FP<-INT] Packs
8870 ;; -------------------------------------------------------------------------
8871 ;; No patterns here yet!
8872 ;; -------------------------------------------------------------------------
8873
8874 ;; -------------------------------------------------------------------------
8875 ;; ---- [FP<-INT] Unpacks
8876 ;; -------------------------------------------------------------------------
8877 ;; The patterns in this section are synthetic.
8878 ;; -------------------------------------------------------------------------
8879
8880 ;; Unpack one half of a VNx4SI to VNx2DF. First unpack from VNx4SI
8881 ;; to VNx2DI, reinterpret the VNx2DI as a VNx4SI, then convert the
8882 ;; unpacked VNx4SI to VNx2DF.
8883 (define_expand "vec_unpack<su_optab>_float_<perm_hilo>_vnx4si"
8884 [(match_operand:VNx2DF 0 "register_operand")
8885 (FLOATUORS:VNx2DF
8886 (unspec:VNx2DI [(match_operand:VNx4SI 1 "register_operand")]
8887 UNPACK_UNSIGNED))]
8888 "TARGET_SVE"
8889 {
8890 /* Use ZIP to do the unpack, since we don't care about the upper halves
8891 and since it has the nice property of not needing any subregs.
8892 If using UUNPK* turns out to be preferable, we could model it as
8893 a ZIP whose first operand is zero. */
8894 rtx temp = gen_reg_rtx (VNx4SImode);
8895 emit_insn ((<hi_lanes_optab>
8896 ? gen_aarch64_sve_zip2vnx4si
8897 : gen_aarch64_sve_zip1vnx4si)
8898 (temp, operands[1], operands[1]));
8899 rtx ptrue = aarch64_ptrue_reg (VNx2BImode);
8900 rtx strictness = gen_int_mode (SVE_RELAXED_GP, SImode);
8901 emit_insn (gen_aarch64_sve_<FLOATUORS:optab>_extendvnx4sivnx2df
8902 (operands[0], ptrue, temp, strictness));
8903 DONE;
8904 }
8905 )
8906
8907 ;; -------------------------------------------------------------------------
8908 ;; ---- [FP<-FP] Packs
8909 ;; -------------------------------------------------------------------------
8910 ;; Includes:
8911 ;; - FCVT
8912 ;; -------------------------------------------------------------------------
8913
8914 ;; Convert two vectors of DF to SF, or two vectors of SF to HF, and pack
8915 ;; the results into a single vector.
8916 (define_expand "vec_pack_trunc_<Vwide>"
8917 [(set (match_dup 4)
8918 (unspec:SVE_FULL_HSF
8919 [(match_dup 3)
8920 (const_int SVE_RELAXED_GP)
8921 (match_operand:<VWIDE> 1 "register_operand")]
8922 UNSPEC_COND_FCVT))
8923 (set (match_dup 5)
8924 (unspec:SVE_FULL_HSF
8925 [(match_dup 3)
8926 (const_int SVE_RELAXED_GP)
8927 (match_operand:<VWIDE> 2 "register_operand")]
8928 UNSPEC_COND_FCVT))
8929 (set (match_operand:SVE_FULL_HSF 0 "register_operand")
8930 (unspec:SVE_FULL_HSF [(match_dup 4) (match_dup 5)] UNSPEC_UZP1))]
8931 "TARGET_SVE"
8932 {
8933 operands[3] = aarch64_ptrue_reg (<VWIDE_PRED>mode);
8934 operands[4] = gen_reg_rtx (<MODE>mode);
8935 operands[5] = gen_reg_rtx (<MODE>mode);
8936 }
8937 )
8938
8939 ;; Predicated float-to-float truncation.
8940 (define_insn "@aarch64_sve_<optab>_trunc<SVE_FULL_SDF:mode><SVE_FULL_HSF:mode>"
8941 [(set (match_operand:SVE_FULL_HSF 0 "register_operand" "=w, ?&w")
8942 (unspec:SVE_FULL_HSF
8943 [(match_operand:<SVE_FULL_SDF:VPRED> 1 "register_operand" "Upl, Upl")
8944 (match_operand:SI 3 "aarch64_sve_gp_strictness")
8945 (match_operand:SVE_FULL_SDF 2 "register_operand" "0, w")]
8946 SVE_COND_FCVT))]
8947 "TARGET_SVE && <SVE_FULL_SDF:elem_bits> > <SVE_FULL_HSF:elem_bits>"
8948 "@
8949 fcvt\t%0.<SVE_FULL_HSF:Vetype>, %1/m, %2.<SVE_FULL_SDF:Vetype>
8950 movprfx\t%0, %2\;fcvt\t%0.<SVE_FULL_HSF:Vetype>, %1/m, %2.<SVE_FULL_SDF:Vetype>"
8951 [(set_attr "movprfx" "*,yes")]
8952 )
8953
8954 ;; Predicated float-to-float truncation with merging.
8955 (define_expand "@cond_<optab>_trunc<SVE_FULL_SDF:mode><SVE_FULL_HSF:mode>"
8956 [(set (match_operand:SVE_FULL_HSF 0 "register_operand")
8957 (unspec:SVE_FULL_HSF
8958 [(match_operand:<SVE_FULL_SDF:VPRED> 1 "register_operand")
8959 (unspec:SVE_FULL_HSF
8960 [(match_dup 1)
8961 (const_int SVE_STRICT_GP)
8962 (match_operand:SVE_FULL_SDF 2 "register_operand")]
8963 SVE_COND_FCVT)
8964 (match_operand:SVE_FULL_HSF 3 "aarch64_simd_reg_or_zero")]
8965 UNSPEC_SEL))]
8966 "TARGET_SVE && <SVE_FULL_SDF:elem_bits> > <SVE_FULL_HSF:elem_bits>"
8967 )
8968
8969 (define_insn "*cond_<optab>_trunc<SVE_FULL_SDF:mode><SVE_FULL_HSF:mode>"
8970 [(set (match_operand:SVE_FULL_HSF 0 "register_operand" "=w, ?&w, ?&w")
8971 (unspec:SVE_FULL_HSF
8972 [(match_operand:<SVE_FULL_SDF:VPRED> 1 "register_operand" "Upl, Upl, Upl")
8973 (unspec:SVE_FULL_HSF
8974 [(match_dup 1)
8975 (match_operand:SI 4 "aarch64_sve_gp_strictness")
8976 (match_operand:SVE_FULL_SDF 2 "register_operand" "w, w, w")]
8977 SVE_COND_FCVT)
8978 (match_operand:SVE_FULL_HSF 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
8979 UNSPEC_SEL))]
8980 "TARGET_SVE && <SVE_FULL_SDF:elem_bits> > <SVE_FULL_HSF:elem_bits>"
8981 "@
8982 fcvt\t%0.<SVE_FULL_HSF:Vetype>, %1/m, %2.<SVE_FULL_SDF:Vetype>
8983 movprfx\t%0.<SVE_FULL_SDF:Vetype>, %1/z, %2.<SVE_FULL_SDF:Vetype>\;fcvt\t%0.<SVE_FULL_HSF:Vetype>, %1/m, %2.<SVE_FULL_SDF:Vetype>
8984 movprfx\t%0, %3\;fcvt\t%0.<SVE_FULL_HSF:Vetype>, %1/m, %2.<SVE_FULL_SDF:Vetype>"
8985 [(set_attr "movprfx" "*,yes,yes")]
8986 )
8987
8988 ;; -------------------------------------------------------------------------
8989 ;; ---- [FP<-FP] Packs (bfloat16)
8990 ;; -------------------------------------------------------------------------
8991 ;; Includes:
8992 ;; - BFCVT (BF16)
8993 ;; - BFCVTNT (BF16)
8994 ;; -------------------------------------------------------------------------
8995
8996 ;; Predicated BFCVT.
8997 (define_insn "@aarch64_sve_<optab>_trunc<VNx4SF_ONLY:mode><VNx8BF_ONLY:mode>"
8998 [(set (match_operand:VNx8BF_ONLY 0 "register_operand" "=w, ?&w")
8999 (unspec:VNx8BF_ONLY
9000 [(match_operand:VNx4BI 1 "register_operand" "Upl, Upl")
9001 (match_operand:SI 3 "aarch64_sve_gp_strictness")
9002 (match_operand:VNx4SF_ONLY 2 "register_operand" "0, w")]
9003 SVE_COND_FCVT))]
9004 "TARGET_SVE_BF16"
9005 "@
9006 bfcvt\t%0.h, %1/m, %2.s
9007 movprfx\t%0, %2\;bfcvt\t%0.h, %1/m, %2.s"
9008 [(set_attr "movprfx" "*,yes")]
9009 )
9010
9011 ;; Predicated BFCVT with merging.
9012 (define_expand "@cond_<optab>_trunc<VNx4SF_ONLY:mode><VNx8BF_ONLY:mode>"
9013 [(set (match_operand:VNx8BF_ONLY 0 "register_operand")
9014 (unspec:VNx8BF_ONLY
9015 [(match_operand:VNx4BI 1 "register_operand")
9016 (unspec:VNx8BF_ONLY
9017 [(match_dup 1)
9018 (const_int SVE_STRICT_GP)
9019 (match_operand:VNx4SF_ONLY 2 "register_operand")]
9020 SVE_COND_FCVT)
9021 (match_operand:VNx8BF_ONLY 3 "aarch64_simd_reg_or_zero")]
9022 UNSPEC_SEL))]
9023 "TARGET_SVE_BF16"
9024 )
9025
9026 (define_insn "*cond_<optab>_trunc<VNx4SF_ONLY:mode><VNx8BF_ONLY:mode>"
9027 [(set (match_operand:VNx8BF_ONLY 0 "register_operand" "=w, ?&w, ?&w")
9028 (unspec:VNx8BF_ONLY
9029 [(match_operand:VNx4BI 1 "register_operand" "Upl, Upl, Upl")
9030 (unspec:VNx8BF_ONLY
9031 [(match_dup 1)
9032 (match_operand:SI 4 "aarch64_sve_gp_strictness")
9033 (match_operand:VNx4SF_ONLY 2 "register_operand" "w, w, w")]
9034 SVE_COND_FCVT)
9035 (match_operand:VNx8BF_ONLY 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
9036 UNSPEC_SEL))]
9037 "TARGET_SVE_BF16"
9038 "@
9039 bfcvt\t%0.h, %1/m, %2.s
9040 movprfx\t%0.s, %1/z, %2.s\;bfcvt\t%0.h, %1/m, %2.s
9041 movprfx\t%0, %3\;bfcvt\t%0.h, %1/m, %2.s"
9042 [(set_attr "movprfx" "*,yes,yes")]
9043 )
9044
9045 ;; Predicated BFCVTNT. This doesn't give a natural aarch64_pred_*/cond_*
9046 ;; pair because the even elements always have to be supplied for active
9047 ;; elements, even if the inactive elements don't matter.
9048 ;;
9049 ;; This instructions does not take MOVPRFX.
9050 (define_insn "@aarch64_sve_cvtnt<mode>"
9051 [(set (match_operand:VNx8BF_ONLY 0 "register_operand" "=w")
9052 (unspec:VNx8BF_ONLY
9053 [(match_operand:VNx4BI 2 "register_operand" "Upl")
9054 (const_int SVE_STRICT_GP)
9055 (match_operand:VNx8BF_ONLY 1 "register_operand" "0")
9056 (match_operand:VNx4SF 3 "register_operand" "w")]
9057 UNSPEC_COND_FCVTNT))]
9058 "TARGET_SVE_BF16"
9059 "bfcvtnt\t%0.h, %2/m, %3.s"
9060 )
9061
9062 ;; -------------------------------------------------------------------------
9063 ;; ---- [FP<-FP] Unpacks
9064 ;; -------------------------------------------------------------------------
9065 ;; Includes:
9066 ;; - FCVT
9067 ;; -------------------------------------------------------------------------
9068
9069 ;; Unpack one half of a VNx4SF to VNx2DF, or one half of a VNx8HF to VNx4SF.
9070 ;; First unpack the source without conversion, then float-convert the
9071 ;; unpacked source.
9072 (define_expand "vec_unpacks_<perm_hilo>_<mode>"
9073 [(match_operand:<VWIDE> 0 "register_operand")
9074 (unspec:SVE_FULL_HSF
9075 [(match_operand:SVE_FULL_HSF 1 "register_operand")]
9076 UNPACK_UNSIGNED)]
9077 "TARGET_SVE"
9078 {
9079 /* Use ZIP to do the unpack, since we don't care about the upper halves
9080 and since it has the nice property of not needing any subregs.
9081 If using UUNPK* turns out to be preferable, we could model it as
9082 a ZIP whose first operand is zero. */
9083 rtx temp = gen_reg_rtx (<MODE>mode);
9084 emit_insn ((<hi_lanes_optab>
9085 ? gen_aarch64_sve_zip2<mode>
9086 : gen_aarch64_sve_zip1<mode>)
9087 (temp, operands[1], operands[1]));
9088 rtx ptrue = aarch64_ptrue_reg (<VWIDE_PRED>mode);
9089 rtx strictness = gen_int_mode (SVE_RELAXED_GP, SImode);
9090 emit_insn (gen_aarch64_sve_fcvt_nontrunc<mode><Vwide>
9091 (operands[0], ptrue, temp, strictness));
9092 DONE;
9093 }
9094 )
9095
9096 ;; Predicated float-to-float extension.
9097 (define_insn "@aarch64_sve_<optab>_nontrunc<SVE_FULL_HSF:mode><SVE_FULL_SDF:mode>"
9098 [(set (match_operand:SVE_FULL_SDF 0 "register_operand" "=w, ?&w")
9099 (unspec:SVE_FULL_SDF
9100 [(match_operand:<SVE_FULL_SDF:VPRED> 1 "register_operand" "Upl, Upl")
9101 (match_operand:SI 3 "aarch64_sve_gp_strictness")
9102 (match_operand:SVE_FULL_HSF 2 "register_operand" "0, w")]
9103 SVE_COND_FCVT))]
9104 "TARGET_SVE && <SVE_FULL_SDF:elem_bits> > <SVE_FULL_HSF:elem_bits>"
9105 "@
9106 fcvt\t%0.<SVE_FULL_SDF:Vetype>, %1/m, %2.<SVE_FULL_HSF:Vetype>
9107 movprfx\t%0, %2\;fcvt\t%0.<SVE_FULL_SDF:Vetype>, %1/m, %2.<SVE_FULL_HSF:Vetype>"
9108 [(set_attr "movprfx" "*,yes")]
9109 )
9110
9111 ;; Predicated float-to-float extension with merging.
9112 (define_expand "@cond_<optab>_nontrunc<SVE_FULL_HSF:mode><SVE_FULL_SDF:mode>"
9113 [(set (match_operand:SVE_FULL_SDF 0 "register_operand")
9114 (unspec:SVE_FULL_SDF
9115 [(match_operand:<SVE_FULL_SDF:VPRED> 1 "register_operand")
9116 (unspec:SVE_FULL_SDF
9117 [(match_dup 1)
9118 (const_int SVE_STRICT_GP)
9119 (match_operand:SVE_FULL_HSF 2 "register_operand")]
9120 SVE_COND_FCVT)
9121 (match_operand:SVE_FULL_SDF 3 "aarch64_simd_reg_or_zero")]
9122 UNSPEC_SEL))]
9123 "TARGET_SVE && <SVE_FULL_SDF:elem_bits> > <SVE_FULL_HSF:elem_bits>"
9124 )
9125
9126 (define_insn "*cond_<optab>_nontrunc<SVE_FULL_HSF:mode><SVE_FULL_SDF:mode>"
9127 [(set (match_operand:SVE_FULL_SDF 0 "register_operand" "=w, ?&w, ?&w")
9128 (unspec:SVE_FULL_SDF
9129 [(match_operand:<SVE_FULL_SDF:VPRED> 1 "register_operand" "Upl, Upl, Upl")
9130 (unspec:SVE_FULL_SDF
9131 [(match_dup 1)
9132 (match_operand:SI 4 "aarch64_sve_gp_strictness")
9133 (match_operand:SVE_FULL_HSF 2 "register_operand" "w, w, w")]
9134 SVE_COND_FCVT)
9135 (match_operand:SVE_FULL_SDF 3 "aarch64_simd_reg_or_zero" "0, Dz, w")]
9136 UNSPEC_SEL))]
9137 "TARGET_SVE && <SVE_FULL_SDF:elem_bits> > <SVE_FULL_HSF:elem_bits>"
9138 "@
9139 fcvt\t%0.<SVE_FULL_SDF:Vetype>, %1/m, %2.<SVE_FULL_HSF:Vetype>
9140 movprfx\t%0.<SVE_FULL_SDF:Vetype>, %1/z, %2.<SVE_FULL_SDF:Vetype>\;fcvt\t%0.<SVE_FULL_SDF:Vetype>, %1/m, %2.<SVE_FULL_HSF:Vetype>
9141 movprfx\t%0, %3\;fcvt\t%0.<SVE_FULL_SDF:Vetype>, %1/m, %2.<SVE_FULL_HSF:Vetype>"
9142 [(set_attr "movprfx" "*,yes,yes")]
9143 )
9144
9145 ;; -------------------------------------------------------------------------
9146 ;; ---- [PRED<-PRED] Packs
9147 ;; -------------------------------------------------------------------------
9148 ;; Includes:
9149 ;; - UZP1
9150 ;; -------------------------------------------------------------------------
9151
9152 ;; Predicate pack. Use UZP1 on the narrower type, which discards
9153 ;; the high part of each wide element.
9154 (define_insn "vec_pack_trunc_<Vwide>"
9155 [(set (match_operand:PRED_BHS 0 "register_operand" "=Upa")
9156 (unspec:PRED_BHS
9157 [(match_operand:<VWIDE> 1 "register_operand" "Upa")
9158 (match_operand:<VWIDE> 2 "register_operand" "Upa")]
9159 UNSPEC_PACK))]
9160 "TARGET_SVE"
9161 "uzp1\t%0.<Vetype>, %1.<Vetype>, %2.<Vetype>"
9162 )
9163
9164 ;; -------------------------------------------------------------------------
9165 ;; ---- [PRED<-PRED] Unpacks
9166 ;; -------------------------------------------------------------------------
9167 ;; Includes:
9168 ;; - PUNPKHI
9169 ;; - PUNPKLO
9170 ;; -------------------------------------------------------------------------
9171
9172 ;; Unpack the low or high half of a predicate, where "high" refers to
9173 ;; the low-numbered lanes for big-endian and the high-numbered lanes
9174 ;; for little-endian.
9175 (define_expand "vec_unpack<su>_<perm_hilo>_<mode>"
9176 [(match_operand:<VWIDE> 0 "register_operand")
9177 (unspec:<VWIDE> [(match_operand:PRED_BHS 1 "register_operand")]
9178 UNPACK)]
9179 "TARGET_SVE"
9180 {
9181 emit_insn ((<hi_lanes_optab>
9182 ? gen_aarch64_sve_punpkhi_<PRED_BHS:mode>
9183 : gen_aarch64_sve_punpklo_<PRED_BHS:mode>)
9184 (operands[0], operands[1]));
9185 DONE;
9186 }
9187 )
9188
9189 (define_insn "@aarch64_sve_punpk<perm_hilo>_<mode>"
9190 [(set (match_operand:<VWIDE> 0 "register_operand" "=Upa")
9191 (unspec:<VWIDE> [(match_operand:PRED_BHS 1 "register_operand" "Upa")]
9192 UNPACK_UNSIGNED))]
9193 "TARGET_SVE"
9194 "punpk<perm_hilo>\t%0.h, %1.b"
9195 )
9196
9197 ;; =========================================================================
9198 ;; == Vector partitioning
9199 ;; =========================================================================
9200
9201 ;; -------------------------------------------------------------------------
9202 ;; ---- [PRED] Unary partitioning
9203 ;; -------------------------------------------------------------------------
9204 ;; Includes:
9205 ;; - BRKA
9206 ;; - BRKAS
9207 ;; - BRKB
9208 ;; - BRKBS
9209 ;; -------------------------------------------------------------------------
9210
9211 ;; Note that unlike most other instructions that have both merging and
9212 ;; zeroing forms, these instructions don't operate elementwise and so
9213 ;; don't fit the IFN_COND model.
9214 (define_insn "@aarch64_brk<brk_op>"
9215 [(set (match_operand:VNx16BI 0 "register_operand" "=Upa, Upa")
9216 (unspec:VNx16BI
9217 [(match_operand:VNx16BI 1 "register_operand" "Upa, Upa")
9218 (match_operand:VNx16BI 2 "register_operand" "Upa, Upa")
9219 (match_operand:VNx16BI 3 "aarch64_simd_reg_or_zero" "Dz, 0")]
9220 SVE_BRK_UNARY))]
9221 "TARGET_SVE"
9222 "@
9223 brk<brk_op>\t%0.b, %1/z, %2.b
9224 brk<brk_op>\t%0.b, %1/m, %2.b"
9225 )
9226
9227 ;; Same, but also producing a flags result.
9228 (define_insn "*aarch64_brk<brk_op>_cc"
9229 [(set (reg:CC_NZC CC_REGNUM)
9230 (unspec:CC_NZC
9231 [(match_operand:VNx16BI 1 "register_operand" "Upa, Upa")
9232 (match_dup 1)
9233 (match_operand:SI 4 "aarch64_sve_ptrue_flag")
9234 (unspec:VNx16BI
9235 [(match_dup 1)
9236 (match_operand:VNx16BI 2 "register_operand" "Upa, Upa")
9237 (match_operand:VNx16BI 3 "aarch64_simd_reg_or_zero" "Dz, 0")]
9238 SVE_BRK_UNARY)]
9239 UNSPEC_PTEST))
9240 (set (match_operand:VNx16BI 0 "register_operand" "=Upa, Upa")
9241 (unspec:VNx16BI
9242 [(match_dup 1)
9243 (match_dup 2)
9244 (match_dup 3)]
9245 SVE_BRK_UNARY))]
9246 "TARGET_SVE"
9247 "@
9248 brk<brk_op>s\t%0.b, %1/z, %2.b
9249 brk<brk_op>s\t%0.b, %1/m, %2.b"
9250 )
9251
9252 ;; Same, but with only the flags result being interesting.
9253 (define_insn "*aarch64_brk<brk_op>_ptest"
9254 [(set (reg:CC_NZC CC_REGNUM)
9255 (unspec:CC_NZC
9256 [(match_operand:VNx16BI 1 "register_operand" "Upa, Upa")
9257 (match_dup 1)
9258 (match_operand:SI 4 "aarch64_sve_ptrue_flag")
9259 (unspec:VNx16BI
9260 [(match_dup 1)
9261 (match_operand:VNx16BI 2 "register_operand" "Upa, Upa")
9262 (match_operand:VNx16BI 3 "aarch64_simd_reg_or_zero" "Dz, 0")]
9263 SVE_BRK_UNARY)]
9264 UNSPEC_PTEST))
9265 (clobber (match_scratch:VNx16BI 0 "=Upa, Upa"))]
9266 "TARGET_SVE"
9267 "@
9268 brk<brk_op>s\t%0.b, %1/z, %2.b
9269 brk<brk_op>s\t%0.b, %1/m, %2.b"
9270 )
9271
9272 ;; -------------------------------------------------------------------------
9273 ;; ---- [PRED] Binary partitioning
9274 ;; -------------------------------------------------------------------------
9275 ;; Includes:
9276 ;; - BRKN
9277 ;; - BRKNS
9278 ;; - BRKPA
9279 ;; - BRKPAS
9280 ;; - BRKPB
9281 ;; - BRKPBS
9282 ;; -------------------------------------------------------------------------
9283
9284 ;; Binary BRKs (BRKN, BRKPA, BRKPB).
9285 (define_insn "@aarch64_brk<brk_op>"
9286 [(set (match_operand:VNx16BI 0 "register_operand" "=Upa")
9287 (unspec:VNx16BI
9288 [(match_operand:VNx16BI 1 "register_operand" "Upa")
9289 (match_operand:VNx16BI 2 "register_operand" "Upa")
9290 (match_operand:VNx16BI 3 "register_operand" "<brk_reg_con>")]
9291 SVE_BRK_BINARY))]
9292 "TARGET_SVE"
9293 "brk<brk_op>\t%0.b, %1/z, %2.b, %<brk_reg_opno>.b"
9294 )
9295
9296 ;; Same, but also producing a flags result.
9297 (define_insn "*aarch64_brk<brk_op>_cc"
9298 [(set (reg:CC_NZC CC_REGNUM)
9299 (unspec:CC_NZC
9300 [(match_operand:VNx16BI 1 "register_operand" "Upa")
9301 (match_dup 1)
9302 (match_operand:SI 4 "aarch64_sve_ptrue_flag")
9303 (unspec:VNx16BI
9304 [(match_dup 1)
9305 (match_operand:VNx16BI 2 "register_operand" "Upa")
9306 (match_operand:VNx16BI 3 "register_operand" "<brk_reg_con>")]
9307 SVE_BRK_BINARY)]
9308 UNSPEC_PTEST))
9309 (set (match_operand:VNx16BI 0 "register_operand" "=Upa")
9310 (unspec:VNx16BI
9311 [(match_dup 1)
9312 (match_dup 2)
9313 (match_dup 3)]
9314 SVE_BRK_BINARY))]
9315 "TARGET_SVE"
9316 "brk<brk_op>s\t%0.b, %1/z, %2.b, %<brk_reg_opno>.b"
9317 )
9318
9319 ;; Same, but with only the flags result being interesting.
9320 (define_insn "*aarch64_brk<brk_op>_ptest"
9321 [(set (reg:CC_NZC CC_REGNUM)
9322 (unspec:CC_NZC
9323 [(match_operand:VNx16BI 1 "register_operand" "Upa")
9324 (match_dup 1)
9325 (match_operand:SI 4 "aarch64_sve_ptrue_flag")
9326 (unspec:VNx16BI
9327 [(match_dup 1)
9328 (match_operand:VNx16BI 2 "register_operand" "Upa")
9329 (match_operand:VNx16BI 3 "register_operand" "<brk_reg_con>")]
9330 SVE_BRK_BINARY)]
9331 UNSPEC_PTEST))
9332 (clobber (match_scratch:VNx16BI 0 "=Upa"))]
9333 "TARGET_SVE"
9334 "brk<brk_op>s\t%0.b, %1/z, %2.b, %<brk_reg_opno>.b"
9335 )
9336
9337 ;; -------------------------------------------------------------------------
9338 ;; ---- [PRED] Scalarization
9339 ;; -------------------------------------------------------------------------
9340 ;; Includes:
9341 ;; - PFIRST
9342 ;; - PNEXT
9343 ;; -------------------------------------------------------------------------
9344
9345 (define_insn "@aarch64_sve_<sve_pred_op><mode>"
9346 [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
9347 (unspec:PRED_ALL
9348 [(match_operand:PRED_ALL 1 "register_operand" "Upa")
9349 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
9350 (match_operand:PRED_ALL 3 "register_operand" "0")]
9351 SVE_PITER))
9352 (clobber (reg:CC_NZC CC_REGNUM))]
9353 "TARGET_SVE && <max_elem_bits> >= <elem_bits>"
9354 "<sve_pred_op>\t%0.<Vetype>, %1, %0.<Vetype>"
9355 )
9356
9357 ;; Same, but also producing a flags result.
9358 (define_insn_and_rewrite "*aarch64_sve_<sve_pred_op><mode>_cc"
9359 [(set (reg:CC_NZC CC_REGNUM)
9360 (unspec:CC_NZC
9361 [(match_operand:VNx16BI 1 "register_operand" "Upa")
9362 (match_operand 2)
9363 (match_operand:SI 3 "aarch64_sve_ptrue_flag")
9364 (unspec:PRED_ALL
9365 [(match_operand 4)
9366 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
9367 (match_operand:PRED_ALL 6 "register_operand" "0")]
9368 SVE_PITER)]
9369 UNSPEC_PTEST))
9370 (set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
9371 (unspec:PRED_ALL
9372 [(match_dup 4)
9373 (match_dup 5)
9374 (match_dup 6)]
9375 SVE_PITER))]
9376 "TARGET_SVE
9377 && <max_elem_bits> >= <elem_bits>
9378 && aarch64_sve_same_pred_for_ptest_p (&operands[2], &operands[4])"
9379 "<sve_pred_op>\t%0.<Vetype>, %1, %0.<Vetype>"
9380 "&& !rtx_equal_p (operands[2], operands[4])"
9381 {
9382 operands[4] = operands[2];
9383 operands[5] = operands[3];
9384 }
9385 )
9386
9387 ;; Same, but with only the flags result being interesting.
9388 (define_insn_and_rewrite "*aarch64_sve_<sve_pred_op><mode>_ptest"
9389 [(set (reg:CC_NZC CC_REGNUM)
9390 (unspec:CC_NZC
9391 [(match_operand:VNx16BI 1 "register_operand" "Upa")
9392 (match_operand 2)
9393 (match_operand:SI 3 "aarch64_sve_ptrue_flag")
9394 (unspec:PRED_ALL
9395 [(match_operand 4)
9396 (match_operand:SI 5 "aarch64_sve_ptrue_flag")
9397 (match_operand:PRED_ALL 6 "register_operand" "0")]
9398 SVE_PITER)]
9399 UNSPEC_PTEST))
9400 (clobber (match_scratch:PRED_ALL 0 "=Upa"))]
9401 "TARGET_SVE
9402 && <max_elem_bits> >= <elem_bits>
9403 && aarch64_sve_same_pred_for_ptest_p (&operands[2], &operands[4])"
9404 "<sve_pred_op>\t%0.<Vetype>, %1, %0.<Vetype>"
9405 "&& !rtx_equal_p (operands[2], operands[4])"
9406 {
9407 operands[4] = operands[2];
9408 operands[5] = operands[3];
9409 }
9410 )
9411
9412 ;; =========================================================================
9413 ;; == Counting elements
9414 ;; =========================================================================
9415
9416 ;; -------------------------------------------------------------------------
9417 ;; ---- [INT] Count elements in a pattern (scalar)
9418 ;; -------------------------------------------------------------------------
9419 ;; Includes:
9420 ;; - CNTB
9421 ;; - CNTD
9422 ;; - CNTH
9423 ;; - CNTW
9424 ;; -------------------------------------------------------------------------
9425
9426 ;; Count the number of elements in an svpattern. Operand 1 is the pattern,
9427 ;; operand 2 is the number of elements that fit in a 128-bit block, and
9428 ;; operand 3 is a multiplier in the range [1, 16].
9429 ;;
9430 ;; Note that this pattern isn't used for SV_ALL (but would work for that too).
9431 (define_insn "aarch64_sve_cnt_pat"
9432 [(set (match_operand:DI 0 "register_operand" "=r")
9433 (zero_extend:DI
9434 (unspec:SI [(match_operand:DI 1 "const_int_operand")
9435 (match_operand:DI 2 "const_int_operand")
9436 (match_operand:DI 3 "const_int_operand")]
9437 UNSPEC_SVE_CNT_PAT)))]
9438 "TARGET_SVE"
9439 {
9440 return aarch64_output_sve_cnt_pat_immediate ("cnt", "%x0", operands + 1);
9441 }
9442 )
9443
9444 ;; -------------------------------------------------------------------------
9445 ;; ---- [INT] Increment by the number of elements in a pattern (scalar)
9446 ;; -------------------------------------------------------------------------
9447 ;; Includes:
9448 ;; - INC
9449 ;; - SQINC
9450 ;; - UQINC
9451 ;; -------------------------------------------------------------------------
9452
9453 ;; Increment a DImode register by the number of elements in an svpattern.
9454 ;; See aarch64_sve_cnt_pat for the counting behavior.
9455 (define_insn "@aarch64_sve_<inc_dec><mode>_pat"
9456 [(set (match_operand:DI 0 "register_operand" "=r")
9457 (ANY_PLUS:DI (zero_extend:DI
9458 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9459 (match_operand:DI 3 "const_int_operand")
9460 (match_operand:DI 4 "const_int_operand")]
9461 UNSPEC_SVE_CNT_PAT))
9462 (match_operand:DI_ONLY 1 "register_operand" "0")))]
9463 "TARGET_SVE"
9464 {
9465 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", "%x0",
9466 operands + 2);
9467 }
9468 )
9469
9470 ;; Increment an SImode register by the number of elements in an svpattern
9471 ;; using modular arithmetic. See aarch64_sve_cnt_pat for the counting
9472 ;; behavior.
9473 (define_insn "*aarch64_sve_incsi_pat"
9474 [(set (match_operand:SI 0 "register_operand" "=r")
9475 (plus:SI (unspec:SI [(match_operand:DI 2 "const_int_operand")
9476 (match_operand:DI 3 "const_int_operand")
9477 (match_operand:DI 4 "const_int_operand")]
9478 UNSPEC_SVE_CNT_PAT)
9479 (match_operand:SI 1 "register_operand" "0")))]
9480 "TARGET_SVE"
9481 {
9482 return aarch64_output_sve_cnt_pat_immediate ("inc", "%x0", operands + 2);
9483 }
9484 )
9485
9486 ;; Increment an SImode register by the number of elements in an svpattern
9487 ;; using saturating arithmetic, extending the result to 64 bits.
9488 ;;
9489 ;; See aarch64_sve_cnt_pat for the counting behavior.
9490 (define_insn "@aarch64_sve_<inc_dec><mode>_pat"
9491 [(set (match_operand:DI 0 "register_operand" "=r")
9492 (<paired_extend>:DI
9493 (SAT_PLUS:SI
9494 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9495 (match_operand:DI 3 "const_int_operand")
9496 (match_operand:DI 4 "const_int_operand")]
9497 UNSPEC_SVE_CNT_PAT)
9498 (match_operand:SI_ONLY 1 "register_operand" "0"))))]
9499 "TARGET_SVE"
9500 {
9501 const char *registers = (<CODE> == SS_PLUS ? "%x0, %w0" : "%w0");
9502 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", registers,
9503 operands + 2);
9504 }
9505 )
9506
9507 ;; -------------------------------------------------------------------------
9508 ;; ---- [INT] Increment by the number of elements in a pattern (vector)
9509 ;; -------------------------------------------------------------------------
9510 ;; Includes:
9511 ;; - INC
9512 ;; - SQINC
9513 ;; - UQINC
9514 ;; -------------------------------------------------------------------------
9515
9516 ;; Increment a vector of DIs by the number of elements in an svpattern.
9517 ;; See aarch64_sve_cnt_pat for the counting behavior.
9518 (define_insn "@aarch64_sve_<inc_dec><mode>_pat"
9519 [(set (match_operand:VNx2DI 0 "register_operand" "=w, ?&w")
9520 (ANY_PLUS:VNx2DI
9521 (vec_duplicate:VNx2DI
9522 (zero_extend:DI
9523 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9524 (match_operand:DI 3 "const_int_operand")
9525 (match_operand:DI 4 "const_int_operand")]
9526 UNSPEC_SVE_CNT_PAT)))
9527 (match_operand:VNx2DI_ONLY 1 "register_operand" "0, w")))]
9528 "TARGET_SVE"
9529 {
9530 if (which_alternative == 1)
9531 output_asm_insn ("movprfx\t%0, %1", operands);
9532 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", "%0.<Vetype>",
9533 operands + 2);
9534 }
9535 [(set_attr "movprfx" "*,yes")]
9536 )
9537
9538 ;; Increment a vector of SIs by the number of elements in an svpattern.
9539 ;; See aarch64_sve_cnt_pat for the counting behavior.
9540 (define_insn "@aarch64_sve_<inc_dec><mode>_pat"
9541 [(set (match_operand:VNx4SI 0 "register_operand" "=w, ?&w")
9542 (ANY_PLUS:VNx4SI
9543 (vec_duplicate:VNx4SI
9544 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9545 (match_operand:DI 3 "const_int_operand")
9546 (match_operand:DI 4 "const_int_operand")]
9547 UNSPEC_SVE_CNT_PAT))
9548 (match_operand:VNx4SI_ONLY 1 "register_operand" "0, w")))]
9549 "TARGET_SVE"
9550 {
9551 if (which_alternative == 1)
9552 output_asm_insn ("movprfx\t%0, %1", operands);
9553 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", "%0.<Vetype>",
9554 operands + 2);
9555 }
9556 [(set_attr "movprfx" "*,yes")]
9557 )
9558
9559 ;; Increment a vector of HIs by the number of elements in an svpattern.
9560 ;; See aarch64_sve_cnt_pat for the counting behavior.
9561 (define_expand "@aarch64_sve_<inc_dec><mode>_pat"
9562 [(set (match_operand:VNx8HI 0 "register_operand")
9563 (ANY_PLUS:VNx8HI
9564 (vec_duplicate:VNx8HI
9565 (truncate:HI
9566 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9567 (match_operand:DI 3 "const_int_operand")
9568 (match_operand:DI 4 "const_int_operand")]
9569 UNSPEC_SVE_CNT_PAT)))
9570 (match_operand:VNx8HI_ONLY 1 "register_operand")))]
9571 "TARGET_SVE"
9572 )
9573
9574 (define_insn "*aarch64_sve_<inc_dec><mode>_pat"
9575 [(set (match_operand:VNx8HI 0 "register_operand" "=w, ?&w")
9576 (ANY_PLUS:VNx8HI
9577 (vec_duplicate:VNx8HI
9578 (match_operator:HI 5 "subreg_lowpart_operator"
9579 [(unspec:SI [(match_operand:DI 2 "const_int_operand")
9580 (match_operand:DI 3 "const_int_operand")
9581 (match_operand:DI 4 "const_int_operand")]
9582 UNSPEC_SVE_CNT_PAT)]))
9583 (match_operand:VNx8HI_ONLY 1 "register_operand" "0, w")))]
9584 "TARGET_SVE"
9585 {
9586 if (which_alternative == 1)
9587 output_asm_insn ("movprfx\t%0, %1", operands);
9588 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", "%0.<Vetype>",
9589 operands + 2);
9590 }
9591 [(set_attr "movprfx" "*,yes")]
9592 )
9593
9594 ;; -------------------------------------------------------------------------
9595 ;; ---- [INT] Decrement by the number of elements in a pattern (scalar)
9596 ;; -------------------------------------------------------------------------
9597 ;; Includes:
9598 ;; - DEC
9599 ;; - SQDEC
9600 ;; - UQDEC
9601 ;; -------------------------------------------------------------------------
9602
9603 ;; Decrement a DImode register by the number of elements in an svpattern.
9604 ;; See aarch64_sve_cnt_pat for the counting behavior.
9605 (define_insn "@aarch64_sve_<inc_dec><mode>_pat"
9606 [(set (match_operand:DI 0 "register_operand" "=r")
9607 (ANY_MINUS:DI (match_operand:DI_ONLY 1 "register_operand" "0")
9608 (zero_extend:DI
9609 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9610 (match_operand:DI 3 "const_int_operand")
9611 (match_operand:DI 4 "const_int_operand")]
9612 UNSPEC_SVE_CNT_PAT))))]
9613 "TARGET_SVE"
9614 {
9615 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", "%x0",
9616 operands + 2);
9617 }
9618 )
9619
9620 ;; Decrement an SImode register by the number of elements in an svpattern
9621 ;; using modular arithmetic. See aarch64_sve_cnt_pat for the counting
9622 ;; behavior.
9623 (define_insn "*aarch64_sve_decsi_pat"
9624 [(set (match_operand:SI 0 "register_operand" "=r")
9625 (minus:SI (match_operand:SI 1 "register_operand" "0")
9626 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9627 (match_operand:DI 3 "const_int_operand")
9628 (match_operand:DI 4 "const_int_operand")]
9629 UNSPEC_SVE_CNT_PAT)))]
9630 "TARGET_SVE"
9631 {
9632 return aarch64_output_sve_cnt_pat_immediate ("dec", "%x0", operands + 2);
9633 }
9634 )
9635
9636 ;; Decrement an SImode register by the number of elements in an svpattern
9637 ;; using saturating arithmetic, extending the result to 64 bits.
9638 ;;
9639 ;; See aarch64_sve_cnt_pat for the counting behavior.
9640 (define_insn "@aarch64_sve_<inc_dec><mode>_pat"
9641 [(set (match_operand:DI 0 "register_operand" "=r")
9642 (<paired_extend>:DI
9643 (SAT_MINUS:SI
9644 (match_operand:SI_ONLY 1 "register_operand" "0")
9645 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9646 (match_operand:DI 3 "const_int_operand")
9647 (match_operand:DI 4 "const_int_operand")]
9648 UNSPEC_SVE_CNT_PAT))))]
9649 "TARGET_SVE"
9650 {
9651 const char *registers = (<CODE> == SS_MINUS ? "%x0, %w0" : "%w0");
9652 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", registers,
9653 operands + 2);
9654 }
9655 )
9656
9657 ;; -------------------------------------------------------------------------
9658 ;; ---- [INT] Decrement by the number of elements in a pattern (vector)
9659 ;; -------------------------------------------------------------------------
9660 ;; Includes:
9661 ;; - DEC
9662 ;; - SQDEC
9663 ;; - UQDEC
9664 ;; -------------------------------------------------------------------------
9665
9666 ;; Decrement a vector of DIs by the number of elements in an svpattern.
9667 ;; See aarch64_sve_cnt_pat for the counting behavior.
9668 (define_insn "@aarch64_sve_<inc_dec><mode>_pat"
9669 [(set (match_operand:VNx2DI 0 "register_operand" "=w, ?&w")
9670 (ANY_MINUS:VNx2DI
9671 (match_operand:VNx2DI_ONLY 1 "register_operand" "0, w")
9672 (vec_duplicate:VNx2DI
9673 (zero_extend:DI
9674 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9675 (match_operand:DI 3 "const_int_operand")
9676 (match_operand:DI 4 "const_int_operand")]
9677 UNSPEC_SVE_CNT_PAT)))))]
9678 "TARGET_SVE"
9679 {
9680 if (which_alternative == 1)
9681 output_asm_insn ("movprfx\t%0, %1", operands);
9682 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", "%0.<Vetype>",
9683 operands + 2);
9684 }
9685 [(set_attr "movprfx" "*,yes")]
9686 )
9687
9688 ;; Decrement a vector of SIs by the number of elements in an svpattern.
9689 ;; See aarch64_sve_cnt_pat for the counting behavior.
9690 (define_insn "@aarch64_sve_<inc_dec><mode>_pat"
9691 [(set (match_operand:VNx4SI 0 "register_operand" "=w, ?&w")
9692 (ANY_MINUS:VNx4SI
9693 (match_operand:VNx4SI_ONLY 1 "register_operand" "0, w")
9694 (vec_duplicate:VNx4SI
9695 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9696 (match_operand:DI 3 "const_int_operand")
9697 (match_operand:DI 4 "const_int_operand")]
9698 UNSPEC_SVE_CNT_PAT))))]
9699 "TARGET_SVE"
9700 {
9701 if (which_alternative == 1)
9702 output_asm_insn ("movprfx\t%0, %1", operands);
9703 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", "%0.<Vetype>",
9704 operands + 2);
9705 }
9706 [(set_attr "movprfx" "*,yes")]
9707 )
9708
9709 ;; Decrement a vector of HIs by the number of elements in an svpattern.
9710 ;; See aarch64_sve_cnt_pat for the counting behavior.
9711 (define_expand "@aarch64_sve_<inc_dec><mode>_pat"
9712 [(set (match_operand:VNx8HI 0 "register_operand")
9713 (ANY_MINUS:VNx8HI
9714 (match_operand:VNx8HI_ONLY 1 "register_operand")
9715 (vec_duplicate:VNx8HI
9716 (truncate:HI
9717 (unspec:SI [(match_operand:DI 2 "const_int_operand")
9718 (match_operand:DI 3 "const_int_operand")
9719 (match_operand:DI 4 "const_int_operand")]
9720 UNSPEC_SVE_CNT_PAT)))))]
9721 "TARGET_SVE"
9722 )
9723
9724 (define_insn "*aarch64_sve_<inc_dec><mode>_pat"
9725 [(set (match_operand:VNx8HI 0 "register_operand" "=w, ?&w")
9726 (ANY_MINUS:VNx8HI
9727 (match_operand:VNx8HI_ONLY 1 "register_operand" "0, w")
9728 (vec_duplicate:VNx8HI
9729 (match_operator:HI 5 "subreg_lowpart_operator"
9730 [(unspec:SI [(match_operand:DI 2 "const_int_operand")
9731 (match_operand:DI 3 "const_int_operand")
9732 (match_operand:DI 4 "const_int_operand")]
9733 UNSPEC_SVE_CNT_PAT)]))))]
9734 "TARGET_SVE"
9735 {
9736 if (which_alternative == 1)
9737 output_asm_insn ("movprfx\t%0, %1", operands);
9738 return aarch64_output_sve_cnt_pat_immediate ("<inc_dec>", "%0.<Vetype>",
9739 operands + 2);
9740 }
9741 [(set_attr "movprfx" "*,yes")]
9742 )
9743
9744 ;; -------------------------------------------------------------------------
9745 ;; ---- [INT] Count elements in a predicate (scalar)
9746 ;; -------------------------------------------------------------------------
9747 ;; Includes:
9748 ;; - CNTP
9749 ;; -------------------------------------------------------------------------
9750
9751 ;; Count the number of set bits in a predicate. Operand 3 is true if
9752 ;; operand 1 is known to be all-true.
9753 (define_insn "@aarch64_pred_cntp<mode>"
9754 [(set (match_operand:DI 0 "register_operand" "=r")
9755 (zero_extend:DI
9756 (unspec:SI [(match_operand:PRED_ALL 1 "register_operand" "Upl")
9757 (match_operand:SI 2 "aarch64_sve_ptrue_flag")
9758 (match_operand:PRED_ALL 3 "register_operand" "Upa")]
9759 UNSPEC_CNTP)))]
9760 "TARGET_SVE"
9761 "cntp\t%x0, %1, %3.<Vetype>")
9762
9763 ;; -------------------------------------------------------------------------
9764 ;; ---- [INT] Increment by the number of elements in a predicate (scalar)
9765 ;; -------------------------------------------------------------------------
9766 ;; Includes:
9767 ;; - INCP
9768 ;; - SQINCP
9769 ;; - UQINCP
9770 ;; -------------------------------------------------------------------------
9771
9772 ;; Increment a DImode register by the number of set bits in a predicate.
9773 ;; See aarch64_sve_cntp for a description of the operands.
9774 (define_expand "@aarch64_sve_<inc_dec><DI_ONLY:mode><PRED_ALL:mode>_cntp"
9775 [(set (match_operand:DI 0 "register_operand")
9776 (ANY_PLUS:DI
9777 (zero_extend:DI
9778 (unspec:SI [(match_dup 3)
9779 (const_int SVE_KNOWN_PTRUE)
9780 (match_operand:PRED_ALL 2 "register_operand")]
9781 UNSPEC_CNTP))
9782 (match_operand:DI_ONLY 1 "register_operand")))]
9783 "TARGET_SVE"
9784 {
9785 operands[3] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
9786 }
9787 )
9788
9789 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><DI_ONLY:mode><PRED_ALL:mode>_cntp"
9790 [(set (match_operand:DI 0 "register_operand" "=r")
9791 (ANY_PLUS:DI
9792 (zero_extend:DI
9793 (unspec:SI [(match_operand 3)
9794 (const_int SVE_KNOWN_PTRUE)
9795 (match_operand:PRED_ALL 2 "register_operand" "Upa")]
9796 UNSPEC_CNTP))
9797 (match_operand:DI_ONLY 1 "register_operand" "0")))]
9798 "TARGET_SVE"
9799 "<inc_dec>p\t%x0, %2.<PRED_ALL:Vetype>"
9800 "&& !CONSTANT_P (operands[3])"
9801 {
9802 operands[3] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
9803 }
9804 )
9805
9806 ;; Increment an SImode register by the number of set bits in a predicate
9807 ;; using modular arithmetic. See aarch64_sve_cntp for a description of
9808 ;; the operands.
9809 (define_insn_and_rewrite "*aarch64_incsi<mode>_cntp"
9810 [(set (match_operand:SI 0 "register_operand" "=r")
9811 (plus:SI
9812 (unspec:SI [(match_operand 3)
9813 (const_int SVE_KNOWN_PTRUE)
9814 (match_operand:PRED_ALL 2 "register_operand" "Upa")]
9815 UNSPEC_CNTP)
9816 (match_operand:SI 1 "register_operand" "0")))]
9817 "TARGET_SVE"
9818 "incp\t%x0, %2.<Vetype>"
9819 "&& !CONSTANT_P (operands[3])"
9820 {
9821 operands[3] = CONSTM1_RTX (<MODE>mode);
9822 }
9823 )
9824
9825 ;; Increment an SImode register by the number of set bits in a predicate
9826 ;; using saturating arithmetic, extending the result to 64 bits.
9827 ;;
9828 ;; See aarch64_sve_cntp for a description of the operands.
9829 (define_expand "@aarch64_sve_<inc_dec><SI_ONLY:mode><PRED_ALL:mode>_cntp"
9830 [(set (match_operand:DI 0 "register_operand")
9831 (<paired_extend>:DI
9832 (SAT_PLUS:SI
9833 (unspec:SI [(match_dup 3)
9834 (const_int SVE_KNOWN_PTRUE)
9835 (match_operand:PRED_ALL 2 "register_operand")]
9836 UNSPEC_CNTP)
9837 (match_operand:SI_ONLY 1 "register_operand"))))]
9838 "TARGET_SVE"
9839 {
9840 operands[3] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
9841 }
9842 )
9843
9844 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><SI_ONLY:mode><PRED_ALL:mode>_cntp"
9845 [(set (match_operand:DI 0 "register_operand" "=r")
9846 (<paired_extend>:DI
9847 (SAT_PLUS:SI
9848 (unspec:SI [(match_operand 3)
9849 (const_int SVE_KNOWN_PTRUE)
9850 (match_operand:PRED_ALL 2 "register_operand" "Upa")]
9851 UNSPEC_CNTP)
9852 (match_operand:SI_ONLY 1 "register_operand" "0"))))]
9853 "TARGET_SVE"
9854 {
9855 if (<CODE> == SS_PLUS)
9856 return "<inc_dec>p\t%x0, %2.<PRED_ALL:Vetype>, %w0";
9857 else
9858 return "<inc_dec>p\t%w0, %2.<PRED_ALL:Vetype>";
9859 }
9860 "&& !CONSTANT_P (operands[3])"
9861 {
9862 operands[3] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
9863 }
9864 )
9865
9866 ;; -------------------------------------------------------------------------
9867 ;; ---- [INT] Increment by the number of elements in a predicate (vector)
9868 ;; -------------------------------------------------------------------------
9869 ;; Includes:
9870 ;; - INCP
9871 ;; - SQINCP
9872 ;; - UQINCP
9873 ;; -------------------------------------------------------------------------
9874
9875 ;; Increment a vector of DIs by the number of set bits in a predicate.
9876 ;; See aarch64_sve_cntp for a description of the operands.
9877 (define_expand "@aarch64_sve_<inc_dec><mode>_cntp"
9878 [(set (match_operand:VNx2DI 0 "register_operand")
9879 (ANY_PLUS:VNx2DI
9880 (vec_duplicate:VNx2DI
9881 (zero_extend:DI
9882 (unspec:SI
9883 [(match_dup 3)
9884 (const_int SVE_KNOWN_PTRUE)
9885 (match_operand:<VPRED> 2 "register_operand")]
9886 UNSPEC_CNTP)))
9887 (match_operand:VNx2DI_ONLY 1 "register_operand")))]
9888 "TARGET_SVE"
9889 {
9890 operands[3] = CONSTM1_RTX (<VPRED>mode);
9891 }
9892 )
9893
9894 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><mode>_cntp"
9895 [(set (match_operand:VNx2DI 0 "register_operand" "=w, ?&w")
9896 (ANY_PLUS:VNx2DI
9897 (vec_duplicate:VNx2DI
9898 (zero_extend:DI
9899 (unspec:SI
9900 [(match_operand 3)
9901 (const_int SVE_KNOWN_PTRUE)
9902 (match_operand:<VPRED> 2 "register_operand" "Upa, Upa")]
9903 UNSPEC_CNTP)))
9904 (match_operand:VNx2DI_ONLY 1 "register_operand" "0, w")))]
9905 "TARGET_SVE"
9906 "@
9907 <inc_dec>p\t%0.d, %2
9908 movprfx\t%0, %1\;<inc_dec>p\t%0.d, %2"
9909 "&& !CONSTANT_P (operands[3])"
9910 {
9911 operands[3] = CONSTM1_RTX (<VPRED>mode);
9912 }
9913 [(set_attr "movprfx" "*,yes")]
9914 )
9915
9916 ;; Increment a vector of SIs by the number of set bits in a predicate.
9917 ;; See aarch64_sve_cntp for a description of the operands.
9918 (define_expand "@aarch64_sve_<inc_dec><mode>_cntp"
9919 [(set (match_operand:VNx4SI 0 "register_operand")
9920 (ANY_PLUS:VNx4SI
9921 (vec_duplicate:VNx4SI
9922 (unspec:SI
9923 [(match_dup 3)
9924 (const_int SVE_KNOWN_PTRUE)
9925 (match_operand:<VPRED> 2 "register_operand")]
9926 UNSPEC_CNTP))
9927 (match_operand:VNx4SI_ONLY 1 "register_operand")))]
9928 "TARGET_SVE"
9929 {
9930 operands[3] = CONSTM1_RTX (<VPRED>mode);
9931 }
9932 )
9933
9934 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><mode>_cntp"
9935 [(set (match_operand:VNx4SI 0 "register_operand" "=w, ?&w")
9936 (ANY_PLUS:VNx4SI
9937 (vec_duplicate:VNx4SI
9938 (unspec:SI
9939 [(match_operand 3)
9940 (const_int SVE_KNOWN_PTRUE)
9941 (match_operand:<VPRED> 2 "register_operand" "Upa, Upa")]
9942 UNSPEC_CNTP))
9943 (match_operand:VNx4SI_ONLY 1 "register_operand" "0, w")))]
9944 "TARGET_SVE"
9945 "@
9946 <inc_dec>p\t%0.s, %2
9947 movprfx\t%0, %1\;<inc_dec>p\t%0.s, %2"
9948 "&& !CONSTANT_P (operands[3])"
9949 {
9950 operands[3] = CONSTM1_RTX (<VPRED>mode);
9951 }
9952 [(set_attr "movprfx" "*,yes")]
9953 )
9954
9955 ;; Increment a vector of HIs by the number of set bits in a predicate.
9956 ;; See aarch64_sve_cntp for a description of the operands.
9957 (define_expand "@aarch64_sve_<inc_dec><mode>_cntp"
9958 [(set (match_operand:VNx8HI 0 "register_operand")
9959 (ANY_PLUS:VNx8HI
9960 (vec_duplicate:VNx8HI
9961 (truncate:HI
9962 (unspec:SI
9963 [(match_dup 3)
9964 (const_int SVE_KNOWN_PTRUE)
9965 (match_operand:<VPRED> 2 "register_operand")]
9966 UNSPEC_CNTP)))
9967 (match_operand:VNx8HI_ONLY 1 "register_operand")))]
9968 "TARGET_SVE"
9969 {
9970 operands[3] = CONSTM1_RTX (<VPRED>mode);
9971 }
9972 )
9973
9974 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><mode>_cntp"
9975 [(set (match_operand:VNx8HI 0 "register_operand" "=w, ?&w")
9976 (ANY_PLUS:VNx8HI
9977 (vec_duplicate:VNx8HI
9978 (match_operator:HI 3 "subreg_lowpart_operator"
9979 [(unspec:SI
9980 [(match_operand 4)
9981 (const_int SVE_KNOWN_PTRUE)
9982 (match_operand:<VPRED> 2 "register_operand" "Upa, Upa")]
9983 UNSPEC_CNTP)]))
9984 (match_operand:VNx8HI_ONLY 1 "register_operand" "0, w")))]
9985 "TARGET_SVE"
9986 "@
9987 <inc_dec>p\t%0.h, %2
9988 movprfx\t%0, %1\;<inc_dec>p\t%0.h, %2"
9989 "&& !CONSTANT_P (operands[4])"
9990 {
9991 operands[4] = CONSTM1_RTX (<VPRED>mode);
9992 }
9993 [(set_attr "movprfx" "*,yes")]
9994 )
9995
9996 ;; -------------------------------------------------------------------------
9997 ;; ---- [INT] Decrement by the number of elements in a predicate (scalar)
9998 ;; -------------------------------------------------------------------------
9999 ;; Includes:
10000 ;; - DECP
10001 ;; - SQDECP
10002 ;; - UQDECP
10003 ;; -------------------------------------------------------------------------
10004
10005 ;; Decrement a DImode register by the number of set bits in a predicate.
10006 ;; See aarch64_sve_cntp for a description of the operands.
10007 (define_expand "@aarch64_sve_<inc_dec><DI_ONLY:mode><PRED_ALL:mode>_cntp"
10008 [(set (match_operand:DI 0 "register_operand")
10009 (ANY_MINUS:DI
10010 (match_operand:DI_ONLY 1 "register_operand")
10011 (zero_extend:DI
10012 (unspec:SI [(match_dup 3)
10013 (const_int SVE_KNOWN_PTRUE)
10014 (match_operand:PRED_ALL 2 "register_operand")]
10015 UNSPEC_CNTP))))]
10016 "TARGET_SVE"
10017 {
10018 operands[3] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
10019 }
10020 )
10021
10022 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><DI_ONLY:mode><PRED_ALL:mode>_cntp"
10023 [(set (match_operand:DI 0 "register_operand" "=r")
10024 (ANY_MINUS:DI
10025 (match_operand:DI_ONLY 1 "register_operand" "0")
10026 (zero_extend:DI
10027 (unspec:SI [(match_operand 3)
10028 (const_int SVE_KNOWN_PTRUE)
10029 (match_operand:PRED_ALL 2 "register_operand" "Upa")]
10030 UNSPEC_CNTP))))]
10031 "TARGET_SVE"
10032 "<inc_dec>p\t%x0, %2.<PRED_ALL:Vetype>"
10033 "&& !CONSTANT_P (operands[3])"
10034 {
10035 operands[3] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
10036 }
10037 )
10038
10039 ;; Decrement an SImode register by the number of set bits in a predicate
10040 ;; using modular arithmetic. See aarch64_sve_cntp for a description of the
10041 ;; operands.
10042 (define_insn_and_rewrite "*aarch64_decsi<mode>_cntp"
10043 [(set (match_operand:SI 0 "register_operand" "=r")
10044 (minus:SI
10045 (match_operand:SI 1 "register_operand" "0")
10046 (unspec:SI [(match_operand 3)
10047 (const_int SVE_KNOWN_PTRUE)
10048 (match_operand:PRED_ALL 2 "register_operand" "Upa")]
10049 UNSPEC_CNTP)))]
10050 "TARGET_SVE"
10051 "decp\t%x0, %2.<Vetype>"
10052 "&& !CONSTANT_P (operands[3])"
10053 {
10054 operands[3] = CONSTM1_RTX (<MODE>mode);
10055 }
10056 )
10057
10058 ;; Decrement an SImode register by the number of set bits in a predicate
10059 ;; using saturating arithmetic, extending the result to 64 bits.
10060 ;;
10061 ;; See aarch64_sve_cntp for a description of the operands.
10062 (define_expand "@aarch64_sve_<inc_dec><SI_ONLY:mode><PRED_ALL:mode>_cntp"
10063 [(set (match_operand:DI 0 "register_operand")
10064 (<paired_extend>:DI
10065 (SAT_MINUS:SI
10066 (match_operand:SI_ONLY 1 "register_operand")
10067 (unspec:SI [(match_dup 3)
10068 (const_int SVE_KNOWN_PTRUE)
10069 (match_operand:PRED_ALL 2 "register_operand")]
10070 UNSPEC_CNTP))))]
10071 "TARGET_SVE"
10072 {
10073 operands[3] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
10074 }
10075 )
10076
10077 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><SI_ONLY:mode><PRED_ALL:mode>_cntp"
10078 [(set (match_operand:DI 0 "register_operand" "=r")
10079 (<paired_extend>:DI
10080 (SAT_MINUS:SI
10081 (match_operand:SI_ONLY 1 "register_operand" "0")
10082 (unspec:SI [(match_operand 3)
10083 (const_int SVE_KNOWN_PTRUE)
10084 (match_operand:PRED_ALL 2 "register_operand" "Upa")]
10085 UNSPEC_CNTP))))]
10086 "TARGET_SVE"
10087 {
10088 if (<CODE> == SS_MINUS)
10089 return "<inc_dec>p\t%x0, %2.<PRED_ALL:Vetype>, %w0";
10090 else
10091 return "<inc_dec>p\t%w0, %2.<PRED_ALL:Vetype>";
10092 }
10093 "&& !CONSTANT_P (operands[3])"
10094 {
10095 operands[3] = CONSTM1_RTX (<PRED_ALL:MODE>mode);
10096 }
10097 )
10098
10099 ;; -------------------------------------------------------------------------
10100 ;; ---- [INT] Decrement by the number of elements in a predicate (vector)
10101 ;; -------------------------------------------------------------------------
10102 ;; Includes:
10103 ;; - DECP
10104 ;; - SQDECP
10105 ;; - UQDECP
10106 ;; -------------------------------------------------------------------------
10107
10108 ;; Decrement a vector of DIs by the number of set bits in a predicate.
10109 ;; See aarch64_sve_cntp for a description of the operands.
10110 (define_expand "@aarch64_sve_<inc_dec><mode>_cntp"
10111 [(set (match_operand:VNx2DI 0 "register_operand")
10112 (ANY_MINUS:VNx2DI
10113 (match_operand:VNx2DI_ONLY 1 "register_operand")
10114 (vec_duplicate:VNx2DI
10115 (zero_extend:DI
10116 (unspec:SI
10117 [(match_dup 3)
10118 (const_int SVE_KNOWN_PTRUE)
10119 (match_operand:<VPRED> 2 "register_operand")]
10120 UNSPEC_CNTP)))))]
10121 "TARGET_SVE"
10122 {
10123 operands[3] = CONSTM1_RTX (<VPRED>mode);
10124 }
10125 )
10126
10127 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><mode>_cntp"
10128 [(set (match_operand:VNx2DI 0 "register_operand" "=w, ?&w")
10129 (ANY_MINUS:VNx2DI
10130 (match_operand:VNx2DI_ONLY 1 "register_operand" "0, w")
10131 (vec_duplicate:VNx2DI
10132 (zero_extend:DI
10133 (unspec:SI
10134 [(match_operand 3)
10135 (const_int SVE_KNOWN_PTRUE)
10136 (match_operand:<VPRED> 2 "register_operand" "Upa, Upa")]
10137 UNSPEC_CNTP)))))]
10138 "TARGET_SVE"
10139 "@
10140 <inc_dec>p\t%0.d, %2
10141 movprfx\t%0, %1\;<inc_dec>p\t%0.d, %2"
10142 "&& !CONSTANT_P (operands[3])"
10143 {
10144 operands[3] = CONSTM1_RTX (<VPRED>mode);
10145 }
10146 [(set_attr "movprfx" "*,yes")]
10147 )
10148
10149 ;; Decrement a vector of SIs by the number of set bits in a predicate.
10150 ;; See aarch64_sve_cntp for a description of the operands.
10151 (define_expand "@aarch64_sve_<inc_dec><mode>_cntp"
10152 [(set (match_operand:VNx4SI 0 "register_operand")
10153 (ANY_MINUS:VNx4SI
10154 (match_operand:VNx4SI_ONLY 1 "register_operand")
10155 (vec_duplicate:VNx4SI
10156 (unspec:SI
10157 [(match_dup 3)
10158 (const_int SVE_KNOWN_PTRUE)
10159 (match_operand:<VPRED> 2 "register_operand")]
10160 UNSPEC_CNTP))))]
10161 "TARGET_SVE"
10162 {
10163 operands[3] = CONSTM1_RTX (<VPRED>mode);
10164 }
10165 )
10166
10167 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><mode>_cntp"
10168 [(set (match_operand:VNx4SI 0 "register_operand" "=w, ?&w")
10169 (ANY_MINUS:VNx4SI
10170 (match_operand:VNx4SI_ONLY 1 "register_operand" "0, w")
10171 (vec_duplicate:VNx4SI
10172 (unspec:SI
10173 [(match_operand 3)
10174 (const_int SVE_KNOWN_PTRUE)
10175 (match_operand:<VPRED> 2 "register_operand" "Upa, Upa")]
10176 UNSPEC_CNTP))))]
10177 "TARGET_SVE"
10178 "@
10179 <inc_dec>p\t%0.s, %2
10180 movprfx\t%0, %1\;<inc_dec>p\t%0.s, %2"
10181 "&& !CONSTANT_P (operands[3])"
10182 {
10183 operands[3] = CONSTM1_RTX (<VPRED>mode);
10184 }
10185 [(set_attr "movprfx" "*,yes")]
10186 )
10187
10188 ;; Decrement a vector of HIs by the number of set bits in a predicate.
10189 ;; See aarch64_sve_cntp for a description of the operands.
10190 (define_expand "@aarch64_sve_<inc_dec><mode>_cntp"
10191 [(set (match_operand:VNx8HI 0 "register_operand")
10192 (ANY_MINUS:VNx8HI
10193 (match_operand:VNx8HI_ONLY 1 "register_operand")
10194 (vec_duplicate:VNx8HI
10195 (truncate:HI
10196 (unspec:SI
10197 [(match_dup 3)
10198 (const_int SVE_KNOWN_PTRUE)
10199 (match_operand:<VPRED> 2 "register_operand")]
10200 UNSPEC_CNTP)))))]
10201 "TARGET_SVE"
10202 {
10203 operands[3] = CONSTM1_RTX (<VPRED>mode);
10204 }
10205 )
10206
10207 (define_insn_and_rewrite "*aarch64_sve_<inc_dec><mode>_cntp"
10208 [(set (match_operand:VNx8HI 0 "register_operand" "=w, ?&w")
10209 (ANY_MINUS:VNx8HI
10210 (match_operand:VNx8HI_ONLY 1 "register_operand" "0, w")
10211 (vec_duplicate:VNx8HI
10212 (match_operator:HI 3 "subreg_lowpart_operator"
10213 [(unspec:SI
10214 [(match_operand 4)
10215 (const_int SVE_KNOWN_PTRUE)
10216 (match_operand:<VPRED> 2 "register_operand" "Upa, Upa")]
10217 UNSPEC_CNTP)]))))]
10218 "TARGET_SVE"
10219 "@
10220 <inc_dec>p\t%0.h, %2
10221 movprfx\t%0, %1\;<inc_dec>p\t%0.h, %2"
10222 "&& !CONSTANT_P (operands[4])"
10223 {
10224 operands[4] = CONSTM1_RTX (<VPRED>mode);
10225 }
10226 [(set_attr "movprfx" "*,yes")]
10227 )
10228