]>
Commit | Line | Data |
---|---|---|
43e9d192 | 1 | ;; Machine description for AArch64 architecture. |
7adcbafe | 2 | ;; Copyright (C) 2009-2022 Free Software Foundation, Inc. |
43e9d192 IB |
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 | ;; Register numbers | |
22 | (define_constants | |
23 | [ | |
24 | (R0_REGNUM 0) | |
25 | (R1_REGNUM 1) | |
26 | (R2_REGNUM 2) | |
27 | (R3_REGNUM 3) | |
28 | (R4_REGNUM 4) | |
29 | (R5_REGNUM 5) | |
30 | (R6_REGNUM 6) | |
31 | (R7_REGNUM 7) | |
32 | (R8_REGNUM 8) | |
33 | (R9_REGNUM 9) | |
34 | (R10_REGNUM 10) | |
35 | (R11_REGNUM 11) | |
36 | (R12_REGNUM 12) | |
37 | (R13_REGNUM 13) | |
38 | (R14_REGNUM 14) | |
39 | (R15_REGNUM 15) | |
40 | (R16_REGNUM 16) | |
43e9d192 | 41 | (R17_REGNUM 17) |
43e9d192 IB |
42 | (R18_REGNUM 18) |
43 | (R19_REGNUM 19) | |
44 | (R20_REGNUM 20) | |
45 | (R21_REGNUM 21) | |
46 | (R22_REGNUM 22) | |
47 | (R23_REGNUM 23) | |
48 | (R24_REGNUM 24) | |
49 | (R25_REGNUM 25) | |
50 | (R26_REGNUM 26) | |
51 | (R27_REGNUM 27) | |
52 | (R28_REGNUM 28) | |
53 | (R29_REGNUM 29) | |
54 | (R30_REGNUM 30) | |
43e9d192 IB |
55 | (SP_REGNUM 31) |
56 | (V0_REGNUM 32) | |
234c9000 AH |
57 | (V1_REGNUM 33) |
58 | (V2_REGNUM 34) | |
59 | (V3_REGNUM 35) | |
80c13ac5 | 60 | (V4_REGNUM 36) |
234c9000 AH |
61 | (V5_REGNUM 37) |
62 | (V6_REGNUM 38) | |
63 | (V7_REGNUM 39) | |
80c13ac5 | 64 | (V8_REGNUM 40) |
234c9000 AH |
65 | (V9_REGNUM 41) |
66 | (V10_REGNUM 42) | |
67 | (V11_REGNUM 43) | |
80c13ac5 | 68 | (V12_REGNUM 44) |
234c9000 AH |
69 | (V13_REGNUM 45) |
70 | (V14_REGNUM 46) | |
43e9d192 | 71 | (V15_REGNUM 47) |
80c13ac5 | 72 | (V16_REGNUM 48) |
234c9000 AH |
73 | (V17_REGNUM 49) |
74 | (V18_REGNUM 50) | |
75 | (V19_REGNUM 51) | |
80c13ac5 | 76 | (V20_REGNUM 52) |
234c9000 AH |
77 | (V21_REGNUM 53) |
78 | (V22_REGNUM 54) | |
79 | (V23_REGNUM 55) | |
80c13ac5 | 80 | (V24_REGNUM 56) |
234c9000 AH |
81 | (V25_REGNUM 57) |
82 | (V26_REGNUM 58) | |
83 | (V27_REGNUM 59) | |
80c13ac5 | 84 | (V28_REGNUM 60) |
234c9000 AH |
85 | (V29_REGNUM 61) |
86 | (V30_REGNUM 62) | |
43e9d192 IB |
87 | (V31_REGNUM 63) |
88 | (SFP_REGNUM 64) | |
89 | (AP_REGNUM 65) | |
90 | (CC_REGNUM 66) | |
43cacb12 RS |
91 | ;; Defined only to make the DWARF description simpler. |
92 | (VG_REGNUM 67) | |
93 | (P0_REGNUM 68) | |
80c13ac5 RS |
94 | (P1_REGNUM 69) |
95 | (P2_REGNUM 70) | |
96 | (P3_REGNUM 71) | |
97 | (P4_REGNUM 72) | |
98 | (P5_REGNUM 73) | |
99 | (P6_REGNUM 74) | |
43cacb12 | 100 | (P7_REGNUM 75) |
80c13ac5 RS |
101 | (P8_REGNUM 76) |
102 | (P9_REGNUM 77) | |
103 | (P10_REGNUM 78) | |
104 | (P11_REGNUM 79) | |
105 | (P12_REGNUM 80) | |
106 | (P13_REGNUM 81) | |
107 | (P14_REGNUM 82) | |
43cacb12 | 108 | (P15_REGNUM 83) |
c600df9a | 109 | (LAST_SAVED_REGNUM 83) |
183bfdaf RS |
110 | (FFR_REGNUM 84) |
111 | ;; "FFR token": a fake register used for representing the scheduling | |
112 | ;; restrictions on FFR-related operations. | |
113 | (FFRT_REGNUM 85) | |
5773855c OH |
114 | ;; The pair of scratch registers used for stack probing with -fstack-check. |
115 | ;; Leave R9 alone as a possible choice for the static chain. | |
116 | ;; Note that the use of these registers is mutually exclusive with the use | |
117 | ;; of STACK_CLASH_SVE_CFA_REGNUM, which is for -fstack-clash-protection | |
118 | ;; rather than -fstack-check. | |
119 | (PROBE_STACK_FIRST_REGNUM 10) | |
120 | (PROBE_STACK_SECOND_REGNUM 11) | |
143d3b15 TC |
121 | ;; Scratch register used by stack clash protection to calculate |
122 | ;; SVE CFA offsets during probing. | |
123 | (STACK_CLASH_SVE_CFA_REGNUM 11) | |
124 | ;; Scratch registers for prologue/epilogue use. | |
125 | (EP0_REGNUM 12) | |
126 | (EP1_REGNUM 13) | |
3751345d RE |
127 | ;; A couple of call-clobbered registers that we need to reserve when |
128 | ;; tracking speculation this is not ABI, so is subject to change. | |
3751345d | 129 | (SPECULATION_SCRATCH_REGNUM 14) |
143d3b15 TC |
130 | (SPECULATION_TRACKER_REGNUM 15) |
131 | ;; Scratch registers used in frame layout. | |
132 | (IP0_REGNUM 16) | |
133 | (IP1_REGNUM 17) | |
0e510d18 | 134 | (FP_REGNUM 29) |
143d3b15 | 135 | (LR_REGNUM 30) |
43e9d192 IB |
136 | ] |
137 | ) | |
138 | ||
139 | (define_c_enum "unspec" [ | |
8fc16d72 ST |
140 | UNSPEC_AUTIA1716 |
141 | UNSPEC_AUTIB1716 | |
142 | UNSPEC_AUTIASP | |
143 | UNSPEC_AUTIBSP | |
08cc4d92 | 144 | UNSPEC_CALLEE_ABI |
43e9d192 | 145 | UNSPEC_CASESI |
0caf592d | 146 | UNSPEC_CPYMEM |
5d357f26 KT |
147 | UNSPEC_CRC32B |
148 | UNSPEC_CRC32CB | |
149 | UNSPEC_CRC32CH | |
150 | UNSPEC_CRC32CW | |
151 | UNSPEC_CRC32CX | |
152 | UNSPEC_CRC32H | |
153 | UNSPEC_CRC32W | |
154 | UNSPEC_CRC32X | |
3f598afe JW |
155 | UNSPEC_FCVTZS |
156 | UNSPEC_FCVTZU | |
e1d5d19e | 157 | UNSPEC_FJCVTZS |
10bd1d96 KT |
158 | UNSPEC_FRINT32Z |
159 | UNSPEC_FRINT32X | |
160 | UNSPEC_FRINT64Z | |
161 | UNSPEC_FRINT64X | |
58a3bd25 | 162 | UNSPEC_URECPE |
0050faf8 JG |
163 | UNSPEC_FRECPE |
164 | UNSPEC_FRECPS | |
165 | UNSPEC_FRECPX | |
43e9d192 IB |
166 | UNSPEC_FRINTA |
167 | UNSPEC_FRINTI | |
168 | UNSPEC_FRINTM | |
0659ce6f | 169 | UNSPEC_FRINTN |
43e9d192 IB |
170 | UNSPEC_FRINTP |
171 | UNSPEC_FRINTX | |
172 | UNSPEC_FRINTZ | |
173 | UNSPEC_GOTSMALLPIC | |
1b1e81f8 | 174 | UNSPEC_GOTSMALLPIC28K |
43e9d192 | 175 | UNSPEC_GOTSMALLTLS |
87dd8ab0 | 176 | UNSPEC_GOTTINYPIC |
5ae7caad | 177 | UNSPEC_GOTTINYTLS |
89b4515c | 178 | UNSPEC_LD1 |
43e9d192 | 179 | UNSPEC_LD2 |
fca7d0a4 | 180 | UNSPEC_LD2_DREG |
77efea31 | 181 | UNSPEC_LD2_DUP |
43e9d192 | 182 | UNSPEC_LD3 |
fca7d0a4 | 183 | UNSPEC_LD3_DREG |
77efea31 | 184 | UNSPEC_LD3_DUP |
43e9d192 | 185 | UNSPEC_LD4 |
fca7d0a4 | 186 | UNSPEC_LD4_DREG |
77efea31 | 187 | UNSPEC_LD4_DUP |
3ec1be97 CB |
188 | UNSPEC_LD2_LANE |
189 | UNSPEC_LD3_LANE | |
190 | UNSPEC_LD4_LANE | |
fdcddba8 PW |
191 | UNSPEC_LD64B |
192 | UNSPEC_ST64B | |
193 | UNSPEC_ST64BV | |
194 | UNSPEC_ST64BV_RET | |
195 | UNSPEC_ST64BV0 | |
196 | UNSPEC_ST64BV0_RET | |
43e9d192 | 197 | UNSPEC_MB |
bb768f8b | 198 | UNSPEC_MOVMEM |
43e9d192 | 199 | UNSPEC_NOP |
8fc16d72 ST |
200 | UNSPEC_PACIA1716 |
201 | UNSPEC_PACIB1716 | |
202 | UNSPEC_PACIASP | |
203 | UNSPEC_PACIBSP | |
43e9d192 | 204 | UNSPEC_PRLG_STK |
43cacb12 | 205 | UNSPEC_REV |
43e9d192 | 206 | UNSPEC_RBIT |
75add2d0 | 207 | UNSPEC_SABAL |
9f499a86 | 208 | UNSPEC_SABAL2 |
6b2034c4 | 209 | UNSPEC_SABDL |
75add2d0 KT |
210 | UNSPEC_SABDL2 |
211 | UNSPEC_SADALP | |
3f598afe | 212 | UNSPEC_SCVTF |
d3bd985e | 213 | UNSPEC_SETMEM |
2e100703 VP |
214 | UNSPEC_SISD_NEG |
215 | UNSPEC_SISD_SSHL | |
216 | UNSPEC_SISD_USHL | |
217 | UNSPEC_SSHL_2S | |
89b4515c | 218 | UNSPEC_ST1 |
43e9d192 IB |
219 | UNSPEC_ST2 |
220 | UNSPEC_ST3 | |
221 | UNSPEC_ST4 | |
ba081b77 JG |
222 | UNSPEC_ST2_LANE |
223 | UNSPEC_ST3_LANE | |
224 | UNSPEC_ST4_LANE | |
43e9d192 IB |
225 | UNSPEC_TLS |
226 | UNSPEC_TLSDESC | |
cbf5629e JW |
227 | UNSPEC_TLSLE12 |
228 | UNSPEC_TLSLE24 | |
229 | UNSPEC_TLSLE32 | |
230 | UNSPEC_TLSLE48 | |
75add2d0 | 231 | UNSPEC_UABAL |
9f499a86 | 232 | UNSPEC_UABAL2 |
6b2034c4 | 233 | UNSPEC_UABDL |
75add2d0 KT |
234 | UNSPEC_UABDL2 |
235 | UNSPEC_UADALP | |
3f598afe | 236 | UNSPEC_UCVTF |
2e100703 | 237 | UNSPEC_USHL_2S |
43e9d192 | 238 | UNSPEC_VSTRUCTDUMMY |
cd0b2d36 | 239 | UNSPEC_SSP_SYSREG |
36e17020 VK |
240 | UNSPEC_SP_SET |
241 | UNSPEC_SP_TEST | |
850e5878 | 242 | UNSPEC_RSHRN |
a6fc00da BH |
243 | UNSPEC_RSQRT |
244 | UNSPEC_RSQRTE | |
245 | UNSPEC_RSQRTS | |
d9c981c8 | 246 | UNSPEC_NZCV |
db58fd89 | 247 | UNSPEC_XPACLRI |
43cacb12 RS |
248 | UNSPEC_LD1_SVE |
249 | UNSPEC_ST1_SVE | |
624d0f07 RS |
250 | UNSPEC_LDNT1_SVE |
251 | UNSPEC_STNT1_SVE | |
43cacb12 | 252 | UNSPEC_LD1RQ |
bfaa08b7 | 253 | UNSPEC_LD1_GATHER |
624d0f07 | 254 | UNSPEC_LDFF1_GATHER |
0a09a948 | 255 | UNSPEC_LDNT1_GATHER |
f307441a | 256 | UNSPEC_ST1_SCATTER |
0a09a948 | 257 | UNSPEC_STNT1_SCATTER |
06308276 | 258 | UNSPEC_PRED_X |
00fa90d9 | 259 | UNSPEC_PRED_Z |
34467289 | 260 | UNSPEC_PTEST |
624d0f07 | 261 | UNSPEC_PTRUE |
43cacb12 RS |
262 | UNSPEC_UNPACKSHI |
263 | UNSPEC_UNPACKUHI | |
264 | UNSPEC_UNPACKSLO | |
265 | UNSPEC_UNPACKULO | |
266 | UNSPEC_PACK | |
0a09a948 RS |
267 | UNSPEC_WHILEGE |
268 | UNSPEC_WHILEGT | |
269 | UNSPEC_WHILEHI | |
270 | UNSPEC_WHILEHS | |
6ad9571b RS |
271 | UNSPEC_WHILELE |
272 | UNSPEC_WHILELO | |
273 | UNSPEC_WHILELS | |
274 | UNSPEC_WHILELT | |
58c036c8 RS |
275 | UNSPEC_WHILERW |
276 | UNSPEC_WHILEWR | |
9f4cbab8 RS |
277 | UNSPEC_LDN |
278 | UNSPEC_STN | |
f1739b48 | 279 | UNSPEC_INSR |
624d0f07 | 280 | UNSPEC_CLASTA |
bb6c2b68 | 281 | UNSPEC_CLASTB |
b781a135 | 282 | UNSPEC_FADDA |
002092be | 283 | UNSPEC_REV_SUBREG |
4aeb1ba7 | 284 | UNSPEC_REINTERPRET |
3751345d | 285 | UNSPEC_SPECULATION_TRACKER |
2214053f | 286 | UNSPEC_SPECULATION_TRACKER_REV |
315fdae8 | 287 | UNSPEC_COPYSIGN |
89626179 | 288 | UNSPEC_TTEST ; Represent transaction test. |
624d0f07 RS |
289 | UNSPEC_UPDATE_FFR |
290 | UNSPEC_UPDATE_FFRT | |
291 | UNSPEC_RDFFR | |
292 | UNSPEC_WRFFR | |
293 | ;; Represents an SVE-style lane index, in which the indexing applies | |
294 | ;; within the containing 128-bit block. | |
295 | UNSPEC_SVE_LANE_SELECT | |
296 | UNSPEC_SVE_CNT_PAT | |
297 | UNSPEC_SVE_PREFETCH | |
298 | UNSPEC_SVE_PREFETCH_GATHER | |
299 | UNSPEC_SVE_COMPACT | |
300 | UNSPEC_SVE_SPLICE | |
ef01e6bb DZ |
301 | UNSPEC_GEN_TAG ; Generate a 4-bit MTE tag. |
302 | UNSPEC_GEN_TAG_RND ; Generate a random 4-bit MTE tag. | |
303 | UNSPEC_TAG_SPACE ; Translate address to MTE tag address space. | |
9ceec73f | 304 | UNSPEC_LD1RO |
74b27d8e | 305 | UNSPEC_SALT_ADDR |
43e9d192 IB |
306 | ]) |
307 | ||
308 | (define_c_enum "unspecv" [ | |
309 | UNSPECV_EH_RETURN ; Represent EH_RETURN | |
aa87aced KV |
310 | UNSPECV_GET_FPCR ; Represent fetch of FPCR content. |
311 | UNSPECV_SET_FPCR ; Represent assign of FPCR content. | |
312 | UNSPECV_GET_FPSR ; Represent fetch of FPSR content. | |
313 | UNSPECV_SET_FPSR ; Represent assign of FPSR content. | |
a3eb8a52 | 314 | UNSPECV_BLOCKAGE ; Represent a blockage |
f58101cf | 315 | UNSPECV_PROBE_STACK_RANGE ; Represent stack range probing. |
def745b2 | 316 | UNSPECV_SPECULATION_BARRIER ; Represent speculation barrier. |
b5f794b4 SD |
317 | UNSPECV_BTI_NOARG ; Represent BTI. |
318 | UNSPECV_BTI_C ; Represent BTI c. | |
319 | UNSPECV_BTI_J ; Represent BTI j. | |
320 | UNSPECV_BTI_JC ; Represent BTI jc. | |
89626179 SD |
321 | UNSPECV_TSTART ; Represent transaction start. |
322 | UNSPECV_TCOMMIT ; Represent transaction commit. | |
323 | UNSPECV_TCANCEL ; Represent transaction cancel. | |
c5dc215d KT |
324 | UNSPEC_RNDR ; Represent RNDR |
325 | UNSPEC_RNDRRS ; Represent RNDRRS | |
43e9d192 IB |
326 | ] |
327 | ) | |
328 | ||
34467289 RS |
329 | ;; These constants are used as a const_int in various SVE unspecs |
330 | ;; to indicate whether the governing predicate is known to be a PTRUE. | |
331 | (define_constants | |
332 | [; Indicates that the predicate might not be a PTRUE. | |
333 | (SVE_MAYBE_NOT_PTRUE 0) | |
334 | ||
335 | ; Indicates that the predicate is known to be a PTRUE. | |
336 | (SVE_KNOWN_PTRUE 1)]) | |
337 | ||
c9c5a809 RS |
338 | ;; These constants are used as a const_int in predicated SVE FP arithmetic |
339 | ;; to indicate whether the operation is allowed to make additional lanes | |
340 | ;; active without worrying about the effect on faulting behavior. | |
341 | (define_constants | |
342 | [; Indicates either that all lanes are active or that the instruction may | |
343 | ; operate on inactive inputs even if doing so could induce a fault. | |
344 | (SVE_RELAXED_GP 0) | |
345 | ||
346 | ; Indicates that some lanes might be inactive and that the instruction | |
347 | ; must not operate on inactive inputs if doing so could induce a fault. | |
348 | (SVE_STRICT_GP 1)]) | |
349 | ||
43e9d192 IB |
350 | ;; If further include files are added the defintion of MD_INCLUDES |
351 | ;; must be updated. | |
352 | ||
353 | (include "constraints.md") | |
354 | (include "predicates.md") | |
355 | (include "iterators.md") | |
356 | ||
43e9d192 IB |
357 | ;; ------------------------------------------------------------------- |
358 | ;; Instruction types and attributes | |
359 | ;; ------------------------------------------------------------------- | |
360 | ||
026c3cfd | 361 | ; The "type" attribute is included here from AArch32 backend to be able |
34b512f5 SN |
362 | ; to share pipeline descriptions. |
363 | (include "../arm/types.md") | |
43e9d192 | 364 | |
a636bb40 KT |
365 | ;; It is important to set the fp or simd attributes to yes when a pattern |
366 | ;; alternative uses the FP or SIMD register files, usually signified by use of | |
367 | ;; the 'w' constraint. This will ensure that the alternative will be | |
368 | ;; disabled when compiling with -mgeneral-regs-only or with the +nofp/+nosimd | |
369 | ;; architecture extensions. If all the alternatives in a pattern use the | |
370 | ;; FP or SIMD registers then the pattern predicate should include TARGET_FLOAT | |
371 | ;; or TARGET_SIMD. | |
372 | ||
488461d8 MM |
373 | ;; Attributes of the architecture required to support the instruction (or |
374 | ;; alternative). This attribute is used to compute attribute "enabled", use type | |
375 | ;; "any" to enable an alternative in all cases. | |
43e9d192 | 376 | |
3c5af608 | 377 | (define_enum "arches" [ any rcpc8_4 fp simd sve fp16]) |
90e6443f | 378 | |
488461d8 | 379 | (define_enum_attr "arch" "arches" (const_string "any")) |
43e9d192 | 380 | |
488461d8 MM |
381 | ;; [For compatibility with Arm in pipeline models] |
382 | ;; Attribute that specifies whether or not the instruction touches fp | |
383 | ;; registers. | |
384 | ;; Note that this attribute is not used anywhere in either the arm or aarch64 | |
385 | ;; backends except in the scheduling description for xgene1. In that | |
386 | ;; scheduling description this attribute is used to subclass the load_4 and | |
387 | ;; load_8 types. | |
388 | (define_attr "fp" "no,yes" | |
389 | (if_then_else | |
390 | (eq_attr "arch" "fp") | |
391 | (const_string "yes") | |
392 | (const_string "no"))) | |
393 | ||
394 | (define_attr "arch_enabled" "no,yes" | |
395 | (if_then_else | |
396 | (ior | |
397 | (eq_attr "arch" "any") | |
398 | ||
3c5af608 MM |
399 | (and (eq_attr "arch" "rcpc8_4") |
400 | (match_test "AARCH64_ISA_RCPC8_4")) | |
401 | ||
488461d8 MM |
402 | (and (eq_attr "arch" "fp") |
403 | (match_test "TARGET_FLOAT")) | |
404 | ||
405 | (and (eq_attr "arch" "simd") | |
406 | (match_test "TARGET_SIMD")) | |
407 | ||
408 | (and (eq_attr "arch" "fp16") | |
409 | (match_test "TARGET_FP_F16INST")) | |
410 | ||
411 | (and (eq_attr "arch" "sve") | |
412 | (match_test "TARGET_SVE"))) | |
413 | (const_string "yes") | |
414 | (const_string "no"))) | |
43cacb12 | 415 | |
43e9d192 IB |
416 | ;; Attribute that controls whether an alternative is enabled or not. |
417 | ;; Currently it is only used to disable alternatives which touch fp or simd | |
488461d8 MM |
418 | ;; registers when -mgeneral-regs-only is specified or to require a special |
419 | ;; architecture support. | |
420 | (define_attr "enabled" "no,yes" (attr "arch_enabled")) | |
43e9d192 | 421 | |
973d2e01 TP |
422 | ;; Attribute that specifies whether we are dealing with a branch to a |
423 | ;; label that is far away, i.e. further away than the maximum/minimum | |
424 | ;; representable in a signed 21-bits number. | |
425 | ;; 0 :=: no | |
426 | ;; 1 :=: yes | |
427 | (define_attr "far_branch" "" (const_int 0)) | |
428 | ||
a08acce8 RH |
429 | ;; Attribute that specifies whether the alternative uses MOVPRFX. |
430 | (define_attr "movprfx" "no,yes" (const_string "no")) | |
431 | ||
be178ecd MM |
432 | ;; Attribute to specify that an alternative has the length of a single |
433 | ;; instruction plus a speculation barrier. | |
434 | (define_attr "sls_length" "none,retbr,casesi" (const_string "none")) | |
435 | ||
a08acce8 RH |
436 | (define_attr "length" "" |
437 | (cond [(eq_attr "movprfx" "yes") | |
438 | (const_int 8) | |
be178ecd MM |
439 | |
440 | (eq_attr "sls_length" "retbr") | |
441 | (cond [(match_test "!aarch64_harden_sls_retbr_p ()") (const_int 4) | |
442 | (match_test "TARGET_SB") (const_int 8)] | |
443 | (const_int 12)) | |
444 | ||
445 | (eq_attr "sls_length" "casesi") | |
446 | (cond [(match_test "!aarch64_harden_sls_retbr_p ()") (const_int 16) | |
447 | (match_test "TARGET_SB") (const_int 20)] | |
448 | (const_int 24)) | |
449 | ] | |
450 | (const_int 4))) | |
a08acce8 | 451 | |
4adcec71 EM |
452 | ;; Strictly for compatibility with AArch32 in pipeline models, since AArch64 has |
453 | ;; no predicated insns. | |
454 | (define_attr "predicated" "yes,no" (const_string "no")) | |
455 | ||
3751345d RE |
456 | ;; Set to true on an insn that requires the speculation tracking state to be |
457 | ;; in the tracking register before the insn issues. Otherwise the compiler | |
458 | ;; may chose to hold the tracking state encoded in SP. | |
459 | (define_attr "speculation_barrier" "true,false" (const_string "false")) | |
460 | ||
43e9d192 IB |
461 | ;; ------------------------------------------------------------------- |
462 | ;; Pipeline descriptions and scheduling | |
463 | ;; ------------------------------------------------------------------- | |
464 | ||
465 | ;; Processor types. | |
466 | (include "aarch64-tune.md") | |
467 | ||
468 | ;; Scheduling | |
c451f4d6 | 469 | (include "../arm/cortex-a53.md") |
e0ae0288 | 470 | (include "../arm/cortex-a57.md") |
b2ca46df | 471 | (include "../arm/exynos-m1.md") |
825ada35 | 472 | (include "falkor.md") |
2e9d500e | 473 | (include "saphira.md") |
2d41ed58 | 474 | (include "thunderx.md") |
a0f06fc9 | 475 | (include "../arm/xgene1.md") |
f192030d | 476 | (include "thunderx2t99.md") |
8108dfde | 477 | (include "tsv110.md") |
fa477e45 | 478 | (include "thunderx3t110.md") |
43e9d192 IB |
479 | |
480 | ;; ------------------------------------------------------------------- | |
481 | ;; Jumps and other miscellaneous insns | |
482 | ;; ------------------------------------------------------------------- | |
483 | ||
484 | (define_insn "indirect_jump" | |
485 | [(set (pc) (match_operand:DI 0 "register_operand" "r"))] | |
486 | "" | |
be178ecd MM |
487 | { |
488 | output_asm_insn ("br\\t%0", operands); | |
489 | return aarch64_sls_barrier (aarch64_harden_sls_retbr_p ()); | |
490 | } | |
491 | [(set_attr "type" "branch") | |
492 | (set_attr "sls_length" "retbr")] | |
43e9d192 IB |
493 | ) |
494 | ||
495 | (define_insn "jump" | |
496 | [(set (pc) (label_ref (match_operand 0 "" "")))] | |
497 | "" | |
498 | "b\\t%l0" | |
c1dfcec7 | 499 | [(set_attr "type" "branch")] |
43e9d192 IB |
500 | ) |
501 | ||
502 | (define_expand "cbranch<mode>4" | |
503 | [(set (pc) (if_then_else (match_operator 0 "aarch64_comparison_operator" | |
1bbffb87 DZ |
504 | [(match_operand:GPI 1 "register_operand") |
505 | (match_operand:GPI 2 "aarch64_plus_operand")]) | |
43e9d192 IB |
506 | (label_ref (match_operand 3 "" "")) |
507 | (pc)))] | |
508 | "" | |
509 | " | |
510 | operands[1] = aarch64_gen_compare_reg (GET_CODE (operands[0]), operands[1], | |
511 | operands[2]); | |
512 | operands[2] = const0_rtx; | |
513 | " | |
514 | ) | |
515 | ||
516 | (define_expand "cbranch<mode>4" | |
517 | [(set (pc) (if_then_else (match_operator 0 "aarch64_comparison_operator" | |
1bbffb87 DZ |
518 | [(match_operand:GPF 1 "register_operand") |
519 | (match_operand:GPF 2 "aarch64_fp_compare_operand")]) | |
43e9d192 IB |
520 | (label_ref (match_operand 3 "" "")) |
521 | (pc)))] | |
522 | "" | |
523 | " | |
524 | operands[1] = aarch64_gen_compare_reg (GET_CODE (operands[0]), operands[1], | |
525 | operands[2]); | |
526 | operands[2] = const0_rtx; | |
527 | " | |
528 | ) | |
529 | ||
cf670503 ZC |
530 | (define_expand "cbranchcc4" |
531 | [(set (pc) (if_then_else | |
532 | (match_operator 0 "aarch64_comparison_operator" | |
1bbffb87 | 533 | [(match_operand 1 "cc_register") |
cf670503 ZC |
534 | (match_operand 2 "const0_operand")]) |
535 | (label_ref (match_operand 3 "" "")) | |
536 | (pc)))] | |
537 | "" | |
538 | "") | |
539 | ||
865257c4 RS |
540 | (define_insn "@ccmp<CC_ONLY:mode><GPI:mode>" |
541 | [(set (match_operand:CC_ONLY 1 "cc_register" "") | |
542 | (if_then_else:CC_ONLY | |
cf670503 | 543 | (match_operator 4 "aarch64_comparison_operator" |
c8012fbc | 544 | [(match_operand 0 "cc_register" "") |
cf670503 | 545 | (const_int 0)]) |
865257c4 | 546 | (compare:CC_ONLY |
c8012fbc WD |
547 | (match_operand:GPI 2 "register_operand" "r,r,r") |
548 | (match_operand:GPI 3 "aarch64_ccmp_operand" "r,Uss,Usn")) | |
865257c4 RS |
549 | (unspec:CC_ONLY |
550 | [(match_operand 5 "immediate_operand")] | |
551 | UNSPEC_NZCV)))] | |
c8012fbc | 552 | "" |
cf670503 ZC |
553 | "@ |
554 | ccmp\\t%<w>2, %<w>3, %k5, %m4 | |
e2b691c4 | 555 | ccmp\\t%<w>2, %3, %k5, %m4 |
cf670503 ZC |
556 | ccmn\\t%<w>2, #%n3, %k5, %m4" |
557 | [(set_attr "type" "alus_sreg,alus_imm,alus_imm")] | |
558 | ) | |
559 | ||
865257c4 RS |
560 | (define_insn "@ccmp<CCFP_CCFPE:mode><GPF:mode>" |
561 | [(set (match_operand:CCFP_CCFPE 1 "cc_register" "") | |
562 | (if_then_else:CCFP_CCFPE | |
786e3c06 WD |
563 | (match_operator 4 "aarch64_comparison_operator" |
564 | [(match_operand 0 "cc_register" "") | |
565 | (const_int 0)]) | |
865257c4 | 566 | (compare:CCFP_CCFPE |
786e3c06 WD |
567 | (match_operand:GPF 2 "register_operand" "w") |
568 | (match_operand:GPF 3 "register_operand" "w")) | |
865257c4 RS |
569 | (unspec:CCFP_CCFPE |
570 | [(match_operand 5 "immediate_operand")] | |
571 | UNSPEC_NZCV)))] | |
786e3c06 | 572 | "TARGET_FLOAT" |
865257c4 | 573 | "fccmp<e>\\t%<s>2, %<s>3, %k5, %m4" |
c297d256 | 574 | [(set_attr "type" "fccmp<s>")] |
786e3c06 WD |
575 | ) |
576 | ||
865257c4 RS |
577 | (define_insn "@ccmp<CC_ONLY:mode><GPI:mode>_rev" |
578 | [(set (match_operand:CC_ONLY 1 "cc_register" "") | |
579 | (if_then_else:CC_ONLY | |
786e3c06 WD |
580 | (match_operator 4 "aarch64_comparison_operator" |
581 | [(match_operand 0 "cc_register" "") | |
865257c4 RS |
582 | (const_int 0)]) |
583 | (unspec:CC_ONLY | |
584 | [(match_operand 5 "immediate_operand")] | |
585 | UNSPEC_NZCV) | |
586 | (compare:CC_ONLY | |
587 | (match_operand:GPI 2 "register_operand" "r,r,r") | |
588 | (match_operand:GPI 3 "aarch64_ccmp_operand" "r,Uss,Usn"))))] | |
589 | "" | |
590 | "@ | |
591 | ccmp\\t%<w>2, %<w>3, %k5, %M4 | |
592 | ccmp\\t%<w>2, %3, %k5, %M4 | |
593 | ccmn\\t%<w>2, #%n3, %k5, %M4" | |
594 | [(set_attr "type" "alus_sreg,alus_imm,alus_imm")] | |
595 | ) | |
596 | ||
597 | (define_insn "@ccmp<CCFP_CCFPE:mode><GPF:mode>_rev" | |
598 | [(set (match_operand:CCFP_CCFPE 1 "cc_register" "") | |
599 | (if_then_else:CCFP_CCFPE | |
600 | (match_operator 4 "aarch64_comparison_operator" | |
601 | [(match_operand 0 "cc_register" "") | |
602 | (const_int 0)]) | |
603 | (unspec:CCFP_CCFPE | |
604 | [(match_operand 5 "immediate_operand")] | |
605 | UNSPEC_NZCV) | |
606 | (compare:CCFP_CCFPE | |
786e3c06 | 607 | (match_operand:GPF 2 "register_operand" "w") |
865257c4 | 608 | (match_operand:GPF 3 "register_operand" "w"))))] |
786e3c06 | 609 | "TARGET_FLOAT" |
865257c4 | 610 | "fccmp<e>\\t%<s>2, %<s>3, %k5, %M4" |
c297d256 | 611 | [(set_attr "type" "fccmp<s>")] |
786e3c06 WD |
612 | ) |
613 | ||
4f58fe36 KT |
614 | ;; Expansion of signed mod by a power of 2 using CSNEG. |
615 | ;; For x0 % n where n is a power of 2 produce: | |
616 | ;; negs x1, x0 | |
617 | ;; and x0, x0, #(n - 1) | |
618 | ;; and x1, x1, #(n - 1) | |
619 | ;; csneg x0, x0, x1, mi | |
620 | ||
621 | (define_expand "mod<mode>3" | |
1bbffb87 DZ |
622 | [(match_operand:GPI 0 "register_operand") |
623 | (match_operand:GPI 1 "register_operand") | |
624 | (match_operand:GPI 2 "const_int_operand")] | |
4f58fe36 KT |
625 | "" |
626 | { | |
627 | HOST_WIDE_INT val = INTVAL (operands[2]); | |
628 | ||
629 | if (val <= 0 | |
630 | || exact_log2 (val) <= 0 | |
631 | || !aarch64_bitmask_imm (val - 1, <MODE>mode)) | |
632 | FAIL; | |
633 | ||
634 | rtx mask = GEN_INT (val - 1); | |
635 | ||
636 | /* In the special case of x0 % 2 we can do the even shorter: | |
637 | cmp x0, xzr | |
638 | and x0, x0, 1 | |
639 | cneg x0, x0, lt. */ | |
640 | if (val == 2) | |
641 | { | |
642 | rtx masked = gen_reg_rtx (<MODE>mode); | |
643 | rtx ccreg = aarch64_gen_compare_reg (LT, operands[1], const0_rtx); | |
644 | emit_insn (gen_and<mode>3 (masked, operands[1], mask)); | |
645 | rtx x = gen_rtx_LT (VOIDmode, ccreg, const0_rtx); | |
646 | emit_insn (gen_csneg3<mode>_insn (operands[0], x, masked, masked)); | |
647 | DONE; | |
648 | } | |
649 | ||
650 | rtx neg_op = gen_reg_rtx (<MODE>mode); | |
651 | rtx_insn *insn = emit_insn (gen_neg<mode>2_compare0 (neg_op, operands[1])); | |
652 | ||
653 | /* Extract the condition register and mode. */ | |
654 | rtx cmp = XVECEXP (PATTERN (insn), 0, 0); | |
655 | rtx cc_reg = SET_DEST (cmp); | |
656 | rtx cond = gen_rtx_GE (VOIDmode, cc_reg, const0_rtx); | |
657 | ||
658 | rtx masked_pos = gen_reg_rtx (<MODE>mode); | |
659 | emit_insn (gen_and<mode>3 (masked_pos, operands[1], mask)); | |
660 | ||
661 | rtx masked_neg = gen_reg_rtx (<MODE>mode); | |
662 | emit_insn (gen_and<mode>3 (masked_neg, neg_op, mask)); | |
663 | ||
664 | emit_insn (gen_csneg3<mode>_insn (operands[0], cond, | |
665 | masked_neg, masked_pos)); | |
666 | DONE; | |
667 | } | |
668 | ) | |
669 | ||
714e1b3b | 670 | (define_insn "condjump" |
43e9d192 IB |
671 | [(set (pc) (if_then_else (match_operator 0 "aarch64_comparison_operator" |
672 | [(match_operand 1 "cc_register" "") (const_int 0)]) | |
673 | (label_ref (match_operand 2 "" "")) | |
674 | (pc)))] | |
675 | "" | |
973d2e01 | 676 | { |
57d6f4d0 RS |
677 | /* GCC's traditional style has been to use "beq" instead of "b.eq", etc., |
678 | but the "." is required for SVE conditions. */ | |
679 | bool use_dot_p = GET_MODE (operands[1]) == CC_NZCmode; | |
973d2e01 | 680 | if (get_attr_length (insn) == 8) |
57d6f4d0 RS |
681 | return aarch64_gen_far_branch (operands, 2, "Lbcond", |
682 | use_dot_p ? "b.%M0\\t" : "b%M0\\t"); | |
973d2e01 | 683 | else |
57d6f4d0 | 684 | return use_dot_p ? "b.%m0\\t%l2" : "b%m0\\t%l2"; |
973d2e01 TP |
685 | } |
686 | [(set_attr "type" "branch") | |
687 | (set (attr "length") | |
688 | (if_then_else (and (ge (minus (match_dup 2) (pc)) (const_int -1048576)) | |
689 | (lt (minus (match_dup 2) (pc)) (const_int 1048572))) | |
690 | (const_int 4) | |
691 | (const_int 8))) | |
692 | (set (attr "far_branch") | |
693 | (if_then_else (and (ge (minus (match_dup 2) (pc)) (const_int -1048576)) | |
694 | (lt (minus (match_dup 2) (pc)) (const_int 1048572))) | |
695 | (const_int 0) | |
696 | (const_int 1)))] | |
43e9d192 IB |
697 | ) |
698 | ||
714e1b3b KT |
699 | ;; For a 24-bit immediate CST we can optimize the compare for equality |
700 | ;; and branch sequence from: | |
701 | ;; mov x0, #imm1 | |
702 | ;; movk x0, #imm2, lsl 16 /* x0 contains CST. */ | |
703 | ;; cmp x1, x0 | |
704 | ;; b<ne,eq> .Label | |
705 | ;; into the shorter: | |
706 | ;; sub x0, x1, #(CST & 0xfff000) | |
707 | ;; subs x0, x0, #(CST & 0x000fff) | |
708 | ;; b<ne,eq> .Label | |
e7ba492a | 709 | (define_insn_and_split "*compare_condjump<GPI:mode>" |
714e1b3b KT |
710 | [(set (pc) (if_then_else (EQL |
711 | (match_operand:GPI 0 "register_operand" "r") | |
712 | (match_operand:GPI 1 "aarch64_imm24" "n")) | |
713 | (label_ref:P (match_operand 2 "" "")) | |
714 | (pc)))] | |
e7ba492a RS |
715 | "!aarch64_move_imm (INTVAL (operands[1]), <GPI:MODE>mode) |
716 | && !aarch64_plus_operand (operands[1], <GPI:MODE>mode) | |
714e1b3b KT |
717 | && !reload_completed" |
718 | "#" | |
719 | "&& true" | |
720 | [(const_int 0)] | |
721 | { | |
722 | HOST_WIDE_INT lo_imm = UINTVAL (operands[1]) & 0xfff; | |
723 | HOST_WIDE_INT hi_imm = UINTVAL (operands[1]) & 0xfff000; | |
e7ba492a RS |
724 | rtx tmp = gen_reg_rtx (<GPI:MODE>mode); |
725 | emit_insn (gen_add<GPI:mode>3 (tmp, operands[0], GEN_INT (-hi_imm))); | |
726 | emit_insn (gen_add<GPI:mode>3_compare0 (tmp, tmp, GEN_INT (-lo_imm))); | |
714e1b3b | 727 | rtx cc_reg = gen_rtx_REG (CC_NZmode, CC_REGNUM); |
e7ba492a RS |
728 | rtx cmp_rtx = gen_rtx_fmt_ee (<EQL:CMP>, <GPI:MODE>mode, |
729 | cc_reg, const0_rtx); | |
714e1b3b KT |
730 | emit_jump_insn (gen_condjump (cmp_rtx, cc_reg, operands[2])); |
731 | DONE; | |
732 | } | |
733 | ) | |
734 | ||
43e9d192 | 735 | (define_expand "casesi" |
1bbffb87 DZ |
736 | [(match_operand:SI 0 "register_operand") ; Index |
737 | (match_operand:SI 1 "const_int_operand") ; Lower bound | |
738 | (match_operand:SI 2 "const_int_operand") ; Total range | |
43e9d192 IB |
739 | (match_operand:DI 3 "" "") ; Table label |
740 | (match_operand:DI 4 "" "")] ; Out of range label | |
741 | "" | |
742 | { | |
743 | if (operands[1] != const0_rtx) | |
744 | { | |
745 | rtx reg = gen_reg_rtx (SImode); | |
746 | ||
747 | /* Canonical RTL says that if you have: | |
748 | ||
749 | (minus (X) (CONST)) | |
750 | ||
751 | then this should be emitted as: | |
752 | ||
753 | (plus (X) (-CONST)) | |
754 | ||
755 | The use of trunc_int_for_mode ensures that the resulting | |
756 | constant can be represented in SImode, this is important | |
757 | for the corner case where operand[1] is INT_MIN. */ | |
758 | ||
618ae596 JJ |
759 | operands[1] |
760 | = GEN_INT (trunc_int_for_mode (-UINTVAL (operands[1]), SImode)); | |
43e9d192 IB |
761 | |
762 | if (!(*insn_data[CODE_FOR_addsi3].operand[2].predicate) | |
763 | (operands[1], SImode)) | |
764 | operands[1] = force_reg (SImode, operands[1]); | |
765 | emit_insn (gen_addsi3 (reg, operands[0], operands[1])); | |
766 | operands[0] = reg; | |
767 | } | |
768 | ||
769 | if (!aarch64_plus_operand (operands[2], SImode)) | |
770 | operands[2] = force_reg (SImode, operands[2]); | |
771 | emit_jump_insn (gen_cbranchsi4 (gen_rtx_GTU (SImode, const0_rtx, | |
772 | const0_rtx), | |
773 | operands[0], operands[2], operands[4])); | |
774 | ||
4033e62c | 775 | operands[2] = force_reg (DImode, gen_rtx_LABEL_REF (DImode, operands[3])); |
ec302910 JJ |
776 | operands[2] |
777 | = gen_rtx_UNSPEC (Pmode, gen_rtvec (2, operands[2], operands[0]), | |
778 | UNSPEC_CASESI); | |
779 | operands[2] = gen_rtx_MEM (DImode, operands[2]); | |
780 | MEM_READONLY_P (operands[2]) = 1; | |
781 | MEM_NOTRAP_P (operands[2]) = 1; | |
782 | emit_jump_insn (gen_casesi_dispatch (operands[2], operands[3])); | |
43e9d192 IB |
783 | DONE; |
784 | } | |
785 | ) | |
786 | ||
ec302910 JJ |
787 | (define_expand "casesi_dispatch" |
788 | [(parallel | |
789 | [(set (pc) (match_operand:DI 0 "")) | |
790 | (clobber (reg:CC CC_REGNUM)) | |
791 | (clobber (match_scratch:DI 2)) | |
792 | (clobber (match_scratch:DI 3)) | |
793 | (use (label_ref:DI (match_operand 1 "")))])] | |
794 | "") | |
795 | ||
796 | (define_insn "*casesi_dispatch" | |
43e9d192 IB |
797 | [(parallel |
798 | [(set (pc) | |
799 | (mem:DI (unspec [(match_operand:DI 0 "register_operand" "r") | |
800 | (match_operand:SI 1 "register_operand" "r")] | |
801 | UNSPEC_CASESI))) | |
802 | (clobber (reg:CC CC_REGNUM)) | |
803 | (clobber (match_scratch:DI 3 "=r")) | |
804 | (clobber (match_scratch:DI 4 "=r")) | |
ec302910 | 805 | (use (label_ref:DI (match_operand 2 "" "")))])] |
43e9d192 IB |
806 | "" |
807 | "* | |
808 | return aarch64_output_casesi (operands); | |
809 | " | |
be178ecd | 810 | [(set_attr "sls_length" "casesi") |
34b512f5 | 811 | (set_attr "type" "branch")] |
43e9d192 IB |
812 | ) |
813 | ||
814 | (define_insn "nop" | |
815 | [(unspec[(const_int 0)] UNSPEC_NOP)] | |
816 | "" | |
817 | "nop" | |
c1dfcec7 | 818 | [(set_attr "type" "no_insn")] |
43e9d192 IB |
819 | ) |
820 | ||
79c6bb5f | 821 | (define_insn "prefetch" |
9bf2f779 | 822 | [(prefetch (match_operand:DI 0 "aarch64_prefetch_operand" "Dp") |
79c6bb5f GG |
823 | (match_operand:QI 1 "const_int_operand" "") |
824 | (match_operand:QI 2 "const_int_operand" ""))] | |
825 | "" | |
826 | { | |
9bf2f779 | 827 | const char * pftype[2][4] = |
79c6bb5f | 828 | { |
9bf2f779 KT |
829 | {"prfm\\tPLDL1STRM, %0", |
830 | "prfm\\tPLDL3KEEP, %0", | |
831 | "prfm\\tPLDL2KEEP, %0", | |
832 | "prfm\\tPLDL1KEEP, %0"}, | |
833 | {"prfm\\tPSTL1STRM, %0", | |
834 | "prfm\\tPSTL3KEEP, %0", | |
835 | "prfm\\tPSTL2KEEP, %0", | |
836 | "prfm\\tPSTL1KEEP, %0"}, | |
79c6bb5f GG |
837 | }; |
838 | ||
839 | int locality = INTVAL (operands[2]); | |
840 | ||
841 | gcc_assert (IN_RANGE (locality, 0, 3)); | |
842 | ||
9bf2f779 KT |
843 | /* PRFM accepts the same addresses as a 64-bit LDR so wrap |
844 | the address into a DImode MEM so that aarch64_print_operand knows | |
845 | how to print it. */ | |
846 | operands[0] = gen_rtx_MEM (DImode, operands[0]); | |
79c6bb5f GG |
847 | return pftype[INTVAL(operands[1])][locality]; |
848 | } | |
89b2133e | 849 | [(set_attr "type" "load_4")] |
79c6bb5f GG |
850 | ) |
851 | ||
cd93c60e AP |
852 | (define_insn "trap" |
853 | [(trap_if (const_int 1) (const_int 8))] | |
854 | "" | |
855 | "brk #1000" | |
856 | [(set_attr "type" "trap")]) | |
857 | ||
43e9d192 IB |
858 | (define_expand "prologue" |
859 | [(clobber (const_int 0))] | |
860 | "" | |
861 | " | |
862 | aarch64_expand_prologue (); | |
863 | DONE; | |
864 | " | |
865 | ) | |
866 | ||
867 | (define_expand "epilogue" | |
868 | [(clobber (const_int 0))] | |
869 | "" | |
870 | " | |
871 | aarch64_expand_epilogue (false); | |
872 | DONE; | |
873 | " | |
874 | ) | |
875 | ||
876 | (define_expand "sibcall_epilogue" | |
877 | [(clobber (const_int 0))] | |
878 | "" | |
879 | " | |
880 | aarch64_expand_epilogue (true); | |
881 | DONE; | |
882 | " | |
883 | ) | |
884 | ||
885 | (define_insn "*do_return" | |
886 | [(return)] | |
887 | "" | |
db58fd89 | 888 | { |
be178ecd | 889 | const char *ret = NULL; |
db58fd89 | 890 | if (aarch64_return_address_signing_enabled () |
14d31404 | 891 | && (TARGET_PAUTH)) |
8fc16d72 ST |
892 | { |
893 | if (aarch64_ra_sign_key == AARCH64_KEY_B) | |
be178ecd | 894 | ret = "retab"; |
8fc16d72 | 895 | else |
be178ecd | 896 | ret = "retaa"; |
8fc16d72 | 897 | } |
be178ecd MM |
898 | else |
899 | ret = "ret"; | |
900 | output_asm_insn (ret, operands); | |
901 | return aarch64_sls_barrier (aarch64_harden_sls_retbr_p ()); | |
db58fd89 | 902 | } |
be178ecd MM |
903 | [(set_attr "type" "branch") |
904 | (set_attr "sls_length" "retbr")] | |
43e9d192 IB |
905 | ) |
906 | ||
4f942779 RL |
907 | (define_expand "return" |
908 | [(simple_return)] | |
909 | "aarch64_use_return_insn_p ()" | |
910 | "" | |
911 | ) | |
912 | ||
913 | (define_insn "simple_return" | |
914 | [(simple_return)] | |
ce9d2a37 | 915 | "" |
be178ecd MM |
916 | { |
917 | output_asm_insn ("ret", operands); | |
918 | return aarch64_sls_barrier (aarch64_harden_sls_retbr_p ()); | |
919 | } | |
920 | [(set_attr "type" "branch") | |
921 | (set_attr "sls_length" "retbr")] | |
4f942779 RL |
922 | ) |
923 | ||
43e9d192 IB |
924 | (define_insn "*cb<optab><mode>1" |
925 | [(set (pc) (if_then_else (EQL (match_operand:GPI 0 "register_operand" "r") | |
926 | (const_int 0)) | |
927 | (label_ref (match_operand 1 "" "")) | |
928 | (pc)))] | |
6e1eaca9 | 929 | "!aarch64_track_speculation" |
973d2e01 TP |
930 | { |
931 | if (get_attr_length (insn) == 8) | |
932 | return aarch64_gen_far_branch (operands, 1, "Lcb", "<inv_cb>\\t%<w>0, "); | |
933 | else | |
934 | return "<cbz>\\t%<w>0, %l1"; | |
935 | } | |
936 | [(set_attr "type" "branch") | |
937 | (set (attr "length") | |
938 | (if_then_else (and (ge (minus (match_dup 1) (pc)) (const_int -1048576)) | |
939 | (lt (minus (match_dup 1) (pc)) (const_int 1048572))) | |
940 | (const_int 4) | |
941 | (const_int 8))) | |
942 | (set (attr "far_branch") | |
943 | (if_then_else (and (ge (minus (match_dup 2) (pc)) (const_int -1048576)) | |
944 | (lt (minus (match_dup 2) (pc)) (const_int 1048572))) | |
945 | (const_int 0) | |
946 | (const_int 1)))] | |
43e9d192 IB |
947 | ) |
948 | ||
949 | (define_insn "*tb<optab><mode>1" | |
950 | [(set (pc) (if_then_else | |
951 | (EQL (zero_extract:DI (match_operand:GPI 0 "register_operand" "r") | |
952 | (const_int 1) | |
e27bfda2 KT |
953 | (match_operand 1 |
954 | "aarch64_simd_shift_imm_<mode>" "n")) | |
43e9d192 IB |
955 | (const_int 0)) |
956 | (label_ref (match_operand 2 "" "")) | |
957 | (pc))) | |
096e8448 | 958 | (clobber (reg:CC CC_REGNUM))] |
6e1eaca9 | 959 | "!aarch64_track_speculation" |
096e8448 JW |
960 | { |
961 | if (get_attr_length (insn) == 8) | |
962 | { | |
973d2e01 TP |
963 | if (get_attr_far_branch (insn) == 1) |
964 | return aarch64_gen_far_branch (operands, 2, "Ltb", | |
965 | "<inv_tb>\\t%<w>0, %1, "); | |
966 | else | |
967 | { | |
968 | operands[1] = GEN_INT (HOST_WIDE_INT_1U << UINTVAL (operands[1])); | |
969 | return "tst\t%<w>0, %1\;<bcond>\t%l2"; | |
970 | } | |
096e8448 JW |
971 | } |
972 | else | |
973 | return "<tbz>\t%<w>0, %1, %l2"; | |
974 | } | |
c1dfcec7 | 975 | [(set_attr "type" "branch") |
43e9d192 IB |
976 | (set (attr "length") |
977 | (if_then_else (and (ge (minus (match_dup 2) (pc)) (const_int -32768)) | |
978 | (lt (minus (match_dup 2) (pc)) (const_int 32764))) | |
979 | (const_int 4) | |
973d2e01 TP |
980 | (const_int 8))) |
981 | (set (attr "far_branch") | |
982 | (if_then_else (and (ge (minus (match_dup 2) (pc)) (const_int -1048576)) | |
983 | (lt (minus (match_dup 2) (pc)) (const_int 1048572))) | |
984 | (const_int 0) | |
985 | (const_int 1)))] | |
986 | ||
43e9d192 IB |
987 | ) |
988 | ||
989 | (define_insn "*cb<optab><mode>1" | |
990 | [(set (pc) (if_then_else (LTGE (match_operand:ALLI 0 "register_operand" "r") | |
991 | (const_int 0)) | |
992 | (label_ref (match_operand 1 "" "")) | |
993 | (pc))) | |
096e8448 | 994 | (clobber (reg:CC CC_REGNUM))] |
6e1eaca9 | 995 | "!aarch64_track_speculation" |
096e8448 JW |
996 | { |
997 | if (get_attr_length (insn) == 8) | |
998 | { | |
973d2e01 TP |
999 | if (get_attr_far_branch (insn) == 1) |
1000 | return aarch64_gen_far_branch (operands, 1, "Ltb", | |
1001 | "<inv_tb>\\t%<w>0, <sizem1>, "); | |
1002 | else | |
1003 | { | |
1004 | char buf[64]; | |
1005 | uint64_t val = ((uint64_t) 1) | |
1006 | << (GET_MODE_SIZE (<MODE>mode) * BITS_PER_UNIT - 1); | |
1007 | sprintf (buf, "tst\t%%<w>0, %" PRId64, val); | |
1008 | output_asm_insn (buf, operands); | |
1009 | return "<bcond>\t%l1"; | |
1010 | } | |
096e8448 JW |
1011 | } |
1012 | else | |
1013 | return "<tbz>\t%<w>0, <sizem1>, %l1"; | |
1014 | } | |
c1dfcec7 | 1015 | [(set_attr "type" "branch") |
43e9d192 IB |
1016 | (set (attr "length") |
1017 | (if_then_else (and (ge (minus (match_dup 1) (pc)) (const_int -32768)) | |
1018 | (lt (minus (match_dup 1) (pc)) (const_int 32764))) | |
1019 | (const_int 4) | |
973d2e01 TP |
1020 | (const_int 8))) |
1021 | (set (attr "far_branch") | |
1022 | (if_then_else (and (ge (minus (match_dup 1) (pc)) (const_int -1048576)) | |
1023 | (lt (minus (match_dup 1) (pc)) (const_int 1048572))) | |
1024 | (const_int 0) | |
1025 | (const_int 1)))] | |
43e9d192 IB |
1026 | ) |
1027 | ||
1028 | ;; ------------------------------------------------------------------- | |
1029 | ;; Subroutine calls and sibcalls | |
1030 | ;; ------------------------------------------------------------------- | |
1031 | ||
1032 | (define_expand "call" | |
08cc4d92 RS |
1033 | [(parallel |
1034 | [(call (match_operand 0 "memory_operand") | |
1035 | (match_operand 1 "general_operand")) | |
1036 | (unspec:DI [(match_operand 2 "const_int_operand")] UNSPEC_CALLEE_ABI) | |
1037 | (clobber (reg:DI LR_REGNUM))])] | |
43e9d192 IB |
1038 | "" |
1039 | " | |
1040 | { | |
08cc4d92 | 1041 | aarch64_expand_call (NULL_RTX, operands[0], operands[2], false); |
78607708 | 1042 | DONE; |
43e9d192 IB |
1043 | }" |
1044 | ) | |
1045 | ||
47210a04 | 1046 | (define_insn "*call_insn" |
96b7f495 | 1047 | [(call (mem:DI (match_operand:DI 0 "aarch64_call_insn_operand" "Ucr, Usf")) |
43e9d192 | 1048 | (match_operand 1 "" "")) |
08cc4d92 | 1049 | (unspec:DI [(match_operand:DI 2 "const_int_operand")] UNSPEC_CALLEE_ABI) |
43e9d192 IB |
1050 | (clobber (reg:DI LR_REGNUM))] |
1051 | "" | |
47210a04 | 1052 | "@ |
96b7f495 | 1053 | * return aarch64_indirect_call_asm (operands[0]); |
d6591257 | 1054 | bl\\t%c0" |
96b7f495 | 1055 | [(set_attr "type" "call, call")]) |
43e9d192 IB |
1056 | |
1057 | (define_expand "call_value" | |
08cc4d92 RS |
1058 | [(parallel |
1059 | [(set (match_operand 0 "") | |
1060 | (call (match_operand 1 "memory_operand") | |
1061 | (match_operand 2 "general_operand"))) | |
1062 | (unspec:DI [(match_operand 3 "const_int_operand")] UNSPEC_CALLEE_ABI) | |
1063 | (clobber (reg:DI LR_REGNUM))])] | |
43e9d192 IB |
1064 | "" |
1065 | " | |
1066 | { | |
08cc4d92 | 1067 | aarch64_expand_call (operands[0], operands[1], operands[3], false); |
78607708 | 1068 | DONE; |
43e9d192 IB |
1069 | }" |
1070 | ) | |
1071 | ||
47210a04 | 1072 | (define_insn "*call_value_insn" |
43e9d192 | 1073 | [(set (match_operand 0 "" "") |
96b7f495 | 1074 | (call (mem:DI (match_operand:DI 1 "aarch64_call_insn_operand" "Ucr, Usf")) |
43e9d192 | 1075 | (match_operand 2 "" ""))) |
08cc4d92 | 1076 | (unspec:DI [(match_operand:DI 3 "const_int_operand")] UNSPEC_CALLEE_ABI) |
43e9d192 IB |
1077 | (clobber (reg:DI LR_REGNUM))] |
1078 | "" | |
47210a04 | 1079 | "@ |
96b7f495 | 1080 | * return aarch64_indirect_call_asm (operands[1]); |
d6591257 | 1081 | bl\\t%c1" |
47210a04 | 1082 | [(set_attr "type" "call, call")] |
43e9d192 IB |
1083 | ) |
1084 | ||
1085 | (define_expand "sibcall" | |
08cc4d92 RS |
1086 | [(parallel |
1087 | [(call (match_operand 0 "memory_operand") | |
1088 | (match_operand 1 "general_operand")) | |
1089 | (unspec:DI [(match_operand 2 "const_int_operand")] UNSPEC_CALLEE_ABI) | |
1090 | (return)])] | |
43e9d192 IB |
1091 | "" |
1092 | { | |
08cc4d92 | 1093 | aarch64_expand_call (NULL_RTX, operands[0], operands[2], true); |
78607708 | 1094 | DONE; |
43e9d192 IB |
1095 | } |
1096 | ) | |
1097 | ||
1098 | (define_expand "sibcall_value" | |
08cc4d92 RS |
1099 | [(parallel |
1100 | [(set (match_operand 0 "") | |
1101 | (call (match_operand 1 "memory_operand") | |
1102 | (match_operand 2 "general_operand"))) | |
1103 | (unspec:DI [(match_operand 3 "const_int_operand")] UNSPEC_CALLEE_ABI) | |
1104 | (return)])] | |
43e9d192 IB |
1105 | "" |
1106 | { | |
08cc4d92 | 1107 | aarch64_expand_call (operands[0], operands[1], operands[3], true); |
78607708 | 1108 | DONE; |
43e9d192 IB |
1109 | } |
1110 | ) | |
1111 | ||
1112 | (define_insn "*sibcall_insn" | |
fee9ba42 | 1113 | [(call (mem:DI (match_operand:DI 0 "aarch64_call_insn_operand" "Ucs, Usf")) |
08cc4d92 RS |
1114 | (match_operand 1 "")) |
1115 | (unspec:DI [(match_operand:DI 2 "const_int_operand")] UNSPEC_CALLEE_ABI) | |
47210a04 | 1116 | (return)] |
fee9ba42 | 1117 | "SIBLING_CALL_P (insn)" |
be178ecd MM |
1118 | { |
1119 | if (which_alternative == 0) | |
1120 | { | |
1121 | output_asm_insn ("br\\t%0", operands); | |
1122 | return aarch64_sls_barrier (aarch64_harden_sls_retbr_p ()); | |
1123 | } | |
1124 | return "b\\t%c0"; | |
1125 | } | |
1126 | [(set_attr "type" "branch, branch") | |
1127 | (set_attr "sls_length" "retbr,none")] | |
43e9d192 IB |
1128 | ) |
1129 | ||
1130 | (define_insn "*sibcall_value_insn" | |
08cc4d92 | 1131 | [(set (match_operand 0 "") |
aaea00aa JG |
1132 | (call (mem:DI |
1133 | (match_operand:DI 1 "aarch64_call_insn_operand" "Ucs, Usf")) | |
08cc4d92 RS |
1134 | (match_operand 2 ""))) |
1135 | (unspec:DI [(match_operand:DI 3 "const_int_operand")] UNSPEC_CALLEE_ABI) | |
47210a04 | 1136 | (return)] |
fee9ba42 | 1137 | "SIBLING_CALL_P (insn)" |
be178ecd MM |
1138 | { |
1139 | if (which_alternative == 0) | |
1140 | { | |
1141 | output_asm_insn ("br\\t%1", operands); | |
1142 | return aarch64_sls_barrier (aarch64_harden_sls_retbr_p ()); | |
1143 | } | |
1144 | return "b\\t%c1"; | |
1145 | } | |
1146 | [(set_attr "type" "branch, branch") | |
1147 | (set_attr "sls_length" "retbr,none")] | |
43e9d192 IB |
1148 | ) |
1149 | ||
1150 | ;; Call subroutine returning any type. | |
1151 | ||
1152 | (define_expand "untyped_call" | |
1153 | [(parallel [(call (match_operand 0 "") | |
1154 | (const_int 0)) | |
1155 | (match_operand 1 "") | |
1156 | (match_operand 2 "")])] | |
1157 | "" | |
1158 | { | |
1159 | int i; | |
1160 | ||
08cc4d92 RS |
1161 | /* Untyped calls always use the default ABI. It's only possible to use |
1162 | ABI variants if we know the type of the target function. */ | |
1163 | emit_call_insn (gen_call (operands[0], const0_rtx, const0_rtx)); | |
43e9d192 IB |
1164 | |
1165 | for (i = 0; i < XVECLEN (operands[2], 0); i++) | |
1166 | { | |
1167 | rtx set = XVECEXP (operands[2], 0, i); | |
1168 | emit_move_insn (SET_DEST (set), SET_SRC (set)); | |
1169 | } | |
1170 | ||
1171 | /* The optimizer does not know that the call sets the function value | |
1172 | registers we stored in the result block. We avoid problems by | |
1173 | claiming that all hard registers are used and clobbered at this | |
1174 | point. */ | |
1175 | emit_insn (gen_blockage ()); | |
1176 | DONE; | |
1177 | }) | |
1178 | ||
1179 | ;; ------------------------------------------------------------------- | |
1180 | ;; Moves | |
1181 | ;; ------------------------------------------------------------------- | |
1182 | ||
1183 | (define_expand "mov<mode>" | |
1bbffb87 DZ |
1184 | [(set (match_operand:SHORT 0 "nonimmediate_operand") |
1185 | (match_operand:SHORT 1 "general_operand"))] | |
43e9d192 IB |
1186 | "" |
1187 | " | |
1188 | if (GET_CODE (operands[0]) == MEM && operands[1] != const0_rtx) | |
1189 | operands[1] = force_reg (<MODE>mode, operands[1]); | |
43cacb12 RS |
1190 | |
1191 | if (GET_CODE (operands[1]) == CONST_POLY_INT) | |
1192 | { | |
1193 | aarch64_expand_mov_immediate (operands[0], operands[1]); | |
1194 | DONE; | |
1195 | } | |
43e9d192 IB |
1196 | " |
1197 | ) | |
1198 | ||
1199 | (define_insn "*mov<mode>_aarch64" | |
43d0a8ee WD |
1200 | [(set (match_operand:SHORT 0 "nonimmediate_operand" "=r,r, w,r ,r,w, m,m,r,w,w") |
1201 | (match_operand:SHORT 1 "aarch64_mov_operand" " r,M,D<hq>,Usv,m,m,rZ,w,w,r,w"))] | |
43e9d192 IB |
1202 | "(register_operand (operands[0], <MODE>mode) |
1203 | || aarch64_reg_or_zero (operands[1], <MODE>mode))" | |
b7342d25 IB |
1204 | { |
1205 | switch (which_alternative) | |
1206 | { | |
1207 | case 0: | |
1208 | return "mov\t%w0, %w1"; | |
1209 | case 1: | |
1210 | return "mov\t%w0, %1"; | |
1211 | case 2: | |
1212 | return aarch64_output_scalar_simd_mov_immediate (operands[1], | |
1213 | <MODE>mode); | |
1214 | case 3: | |
43cacb12 | 1215 | return aarch64_output_sve_cnt_immediate (\"cnt\", \"%x0\", operands[1]); |
b7342d25 | 1216 | case 4: |
43cacb12 | 1217 | return "ldr<size>\t%w0, %1"; |
b7342d25 | 1218 | case 5: |
43cacb12 | 1219 | return "ldr\t%<size>0, %1"; |
b7342d25 | 1220 | case 6: |
43cacb12 | 1221 | return "str<size>\t%w1, %0"; |
b7342d25 | 1222 | case 7: |
43cacb12 | 1223 | return "str\t%<size>1, %0"; |
b7342d25 | 1224 | case 8: |
43cacb12 | 1225 | return "umov\t%w0, %1.<v>[0]"; |
b7342d25 | 1226 | case 9: |
43cacb12 RS |
1227 | return "dup\t%0.<Vallxd>, %w1"; |
1228 | case 10: | |
21216998 | 1229 | return "dup\t%<Vetype>0, %1.<v>[0]"; |
b7342d25 IB |
1230 | default: |
1231 | gcc_unreachable (); | |
1232 | } | |
1233 | } | |
43cacb12 RS |
1234 | ;; The "mov_imm" type for CNT is just a placeholder. |
1235 | [(set_attr "type" "mov_reg,mov_imm,neon_move,mov_imm,load_4,load_4,store_4, | |
1236 | store_4,neon_to_gp<q>,neon_from_gp<q>,neon_dup") | |
488461d8 | 1237 | (set_attr "arch" "*,*,simd,sve,*,*,*,*,simd,simd,simd")] |
43e9d192 IB |
1238 | ) |
1239 | ||
1240 | (define_expand "mov<mode>" | |
1bbffb87 DZ |
1241 | [(set (match_operand:GPI 0 "nonimmediate_operand") |
1242 | (match_operand:GPI 1 "general_operand"))] | |
43e9d192 IB |
1243 | "" |
1244 | " | |
01b9402c KT |
1245 | if (MEM_P (operands[0]) && !MEM_VOLATILE_P (operands[0]) |
1246 | && CONST_INT_P (operands[1]) && <MODE>mode == DImode | |
141a3ccf KT |
1247 | && aarch64_split_dimode_const_store (operands[0], operands[1])) |
1248 | DONE; | |
1249 | ||
43e9d192 IB |
1250 | if (GET_CODE (operands[0]) == MEM && operands[1] != const0_rtx) |
1251 | operands[1] = force_reg (<MODE>mode, operands[1]); | |
1252 | ||
48c79f05 RS |
1253 | /* Lower moves of symbolic constants into individual instructions. |
1254 | Doing this now is sometimes necessary for correctness, since some | |
1255 | sequences require temporary pseudo registers. Lowering now is also | |
1256 | often better for optimization, since more RTL passes get the | |
1257 | chance to optimize the individual instructions. | |
1258 | ||
1259 | When called after RA, also split multi-instruction moves into | |
1260 | smaller pieces now, since we can't be sure that sure that there | |
1261 | will be a following split pass. */ | |
1262 | if (CONST_INT_P (operands[1]) | |
1263 | ? (reload_completed | |
1264 | && !aarch64_mov_imm_operand (operands[1], <MODE>mode)) | |
1265 | : CONSTANT_P (operands[1])) | |
82614948 RR |
1266 | { |
1267 | aarch64_expand_mov_immediate (operands[0], operands[1]); | |
1268 | DONE; | |
1269 | } | |
43e9d192 IB |
1270 | " |
1271 | ) | |
1272 | ||
82614948 | 1273 | (define_insn_and_split "*movsi_aarch64" |
a195c727 WD |
1274 | [(set (match_operand:SI 0 "nonimmediate_operand" "=r,k,r,r,r,r, r,w, m, m, r, r, r, w,r,w, w") |
1275 | (match_operand:SI 1 "aarch64_mov_operand" " r,r,k,M,n,Usv,m,m,rZ,w,Usw,Usa,Ush,rZ,w,w,Ds"))] | |
43e9d192 IB |
1276 | "(register_operand (operands[0], SImode) |
1277 | || aarch64_reg_or_zero (operands[1], SImode))" | |
1278 | "@ | |
28514dda YZ |
1279 | mov\\t%w0, %w1 |
1280 | mov\\t%w0, %w1 | |
43e9d192 IB |
1281 | mov\\t%w0, %w1 |
1282 | mov\\t%w0, %1 | |
82614948 | 1283 | # |
43cacb12 | 1284 | * return aarch64_output_sve_cnt_immediate (\"cnt\", \"%x0\", operands[1]); |
43e9d192 | 1285 | ldr\\t%w0, %1 |
0b064172 | 1286 | ldr\\t%s0, %1 |
43e9d192 | 1287 | str\\t%w1, %0 |
0b064172 | 1288 | str\\t%s1, %0 |
a195c727 | 1289 | adrp\\t%x0, %A1\;ldr\\t%w0, [%x0, %L1] |
d6591257 | 1290 | adr\\t%x0, %c1 |
28514dda | 1291 | adrp\\t%x0, %A1 |
43e9d192 IB |
1292 | fmov\\t%s0, %w1 |
1293 | fmov\\t%w0, %s1 | |
a2170965 TC |
1294 | fmov\\t%s0, %s1 |
1295 | * return aarch64_output_scalar_simd_mov_immediate (operands[1], SImode);" | |
1296 | "CONST_INT_P (operands[1]) && !aarch64_move_imm (INTVAL (operands[1]), SImode) | |
7263fa9f | 1297 | && REG_P (operands[0]) && GP_REGNUM_P (REGNO (operands[0]))" |
82614948 RR |
1298 | [(const_int 0)] |
1299 | "{ | |
1300 | aarch64_expand_mov_immediate (operands[0], operands[1]); | |
1301 | DONE; | |
1302 | }" | |
43cacb12 RS |
1303 | ;; The "mov_imm" type for CNT is just a placeholder. |
1304 | [(set_attr "type" "mov_reg,mov_reg,mov_reg,mov_imm,mov_imm,mov_imm,load_4, | |
a195c727 WD |
1305 | load_4,store_4,store_4,load_4,adr,adr,f_mcr,f_mrc,fmov,neon_move") |
1306 | (set_attr "arch" "*,*,*,*,*,sve,*,fp,*,fp,*,*,*,fp,fp,fp,simd") | |
1307 | (set_attr "length" "4,4,4,4,*, 4,4, 4,4, 4,8,4,4, 4, 4, 4, 4") | |
1308 | ] | |
43e9d192 IB |
1309 | ) |
1310 | ||
82614948 | 1311 | (define_insn_and_split "*movdi_aarch64" |
a195c727 WD |
1312 | [(set (match_operand:DI 0 "nonimmediate_operand" "=r,k,r,r,r,r,r, r,w, m,m, r, r, r, w,r,w, w") |
1313 | (match_operand:DI 1 "aarch64_mov_operand" " r,r,k,N,M,n,Usv,m,m,rZ,w,Usw,Usa,Ush,rZ,w,w,Dd"))] | |
43e9d192 IB |
1314 | "(register_operand (operands[0], DImode) |
1315 | || aarch64_reg_or_zero (operands[1], DImode))" | |
1316 | "@ | |
1317 | mov\\t%x0, %x1 | |
1318 | mov\\t%0, %x1 | |
1319 | mov\\t%x0, %1 | |
1320 | mov\\t%x0, %1 | |
9de00935 | 1321 | mov\\t%w0, %1 |
82614948 | 1322 | # |
43cacb12 | 1323 | * return aarch64_output_sve_cnt_immediate (\"cnt\", \"%x0\", operands[1]); |
43e9d192 | 1324 | ldr\\t%x0, %1 |
0b064172 | 1325 | ldr\\t%d0, %1 |
43e9d192 | 1326 | str\\t%x1, %0 |
0b064172 | 1327 | str\\t%d1, %0 |
a195c727 | 1328 | * return TARGET_ILP32 ? \"adrp\\t%0, %A1\;ldr\\t%w0, [%0, %L1]\" : \"adrp\\t%0, %A1\;ldr\\t%0, [%0, %L1]\"; |
d6591257 | 1329 | adr\\t%x0, %c1 |
43e9d192 IB |
1330 | adrp\\t%x0, %A1 |
1331 | fmov\\t%d0, %x1 | |
1332 | fmov\\t%x0, %d1 | |
1333 | fmov\\t%d0, %d1 | |
a2170965 | 1334 | * return aarch64_output_scalar_simd_mov_immediate (operands[1], DImode);" |
a195c727 | 1335 | "CONST_INT_P (operands[1]) && !aarch64_move_imm (INTVAL (operands[1]), DImode) |
7263fa9f | 1336 | && REG_P (operands[0]) && GP_REGNUM_P (REGNO (operands[0]))" |
82614948 RR |
1337 | [(const_int 0)] |
1338 | "{ | |
1339 | aarch64_expand_mov_immediate (operands[0], operands[1]); | |
1340 | DONE; | |
1341 | }" | |
43cacb12 RS |
1342 | ;; The "mov_imm" type for CNTD is just a placeholder. |
1343 | [(set_attr "type" "mov_reg,mov_reg,mov_reg,mov_imm,mov_imm,mov_imm,mov_imm, | |
a195c727 WD |
1344 | load_8,load_8,store_8,store_8,load_8,adr,adr,f_mcr,f_mrc, |
1345 | fmov,neon_move") | |
1346 | (set_attr "arch" "*,*,*,*,*,*,sve,*,fp,*,fp,*,*,*,fp,fp,fp,simd") | |
1347 | (set_attr "length" "4,4,4,4,4,*, 4,4, 4,4, 4,8,4,4, 4, 4, 4, 4")] | |
43e9d192 IB |
1348 | ) |
1349 | ||
1350 | (define_insn "insv_imm<mode>" | |
1351 | [(set (zero_extract:GPI (match_operand:GPI 0 "register_operand" "+r") | |
1352 | (const_int 16) | |
0b81f26c JG |
1353 | (match_operand:GPI 1 "const_int_operand" "n")) |
1354 | (match_operand:GPI 2 "const_int_operand" "n"))] | |
50d38551 IB |
1355 | "UINTVAL (operands[1]) < GET_MODE_BITSIZE (<MODE>mode) |
1356 | && UINTVAL (operands[1]) % 16 == 0" | |
a05c0ddf | 1357 | "movk\\t%<w>0, %X2, lsl %1" |
99e6af15 | 1358 | [(set_attr "type" "mov_imm")] |
43e9d192 IB |
1359 | ) |
1360 | ||
bba0c624 RS |
1361 | ;; Match MOVK as a normal AND and IOR operation. |
1362 | (define_insn "aarch64_movk<mode>" | |
1363 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
1364 | (ior:GPI (and:GPI (match_operand:GPI 1 "register_operand" "0") | |
1365 | (match_operand:GPI 2 "const_int_operand")) | |
1366 | (match_operand:GPI 3 "const_int_operand")))] | |
1367 | "aarch64_movk_shift (rtx_mode_t (operands[2], <MODE>mode), | |
1368 | rtx_mode_t (operands[3], <MODE>mode)) >= 0" | |
1369 | { | |
1370 | int shift = aarch64_movk_shift (rtx_mode_t (operands[2], <MODE>mode), | |
1371 | rtx_mode_t (operands[3], <MODE>mode)); | |
1372 | operands[2] = gen_int_mode (UINTVAL (operands[3]) >> shift, SImode); | |
1373 | operands[3] = gen_int_mode (shift, SImode); | |
1374 | return "movk\\t%<w>0, #%X2, lsl %3"; | |
1375 | } | |
1ccdd460 | 1376 | [(set_attr "type" "mov_imm")] |
bba0c624 RS |
1377 | ) |
1378 | ||
43e9d192 | 1379 | (define_expand "movti" |
1bbffb87 DZ |
1380 | [(set (match_operand:TI 0 "nonimmediate_operand") |
1381 | (match_operand:TI 1 "general_operand"))] | |
43e9d192 IB |
1382 | "" |
1383 | " | |
1384 | if (GET_CODE (operands[0]) == MEM && operands[1] != const0_rtx) | |
1385 | operands[1] = force_reg (TImode, operands[1]); | |
43cacb12 RS |
1386 | |
1387 | if (GET_CODE (operands[1]) == CONST_POLY_INT) | |
1388 | { | |
1389 | emit_move_insn (gen_lowpart (DImode, operands[0]), | |
1390 | gen_lowpart (DImode, operands[1])); | |
1391 | emit_move_insn (gen_highpart (DImode, operands[0]), const0_rtx); | |
1392 | DONE; | |
1393 | } | |
43e9d192 IB |
1394 | " |
1395 | ) | |
1396 | ||
1397 | (define_insn "*movti_aarch64" | |
1398 | [(set (match_operand:TI 0 | |
d4f9e819 | 1399 | "nonimmediate_operand" "= r,w,w, r,w,r,m,m,w,m") |
43e9d192 | 1400 | (match_operand:TI 1 |
d4f9e819 | 1401 | "aarch64_movti_operand" " rUti,Z,r, w,w,m,r,Z,m,w"))] |
43e9d192 IB |
1402 | "(register_operand (operands[0], TImode) |
1403 | || aarch64_reg_or_zero (operands[1], TImode))" | |
1404 | "@ | |
1405 | # | |
d4f9e819 | 1406 | movi\\t%0.2d, #0 |
43e9d192 IB |
1407 | # |
1408 | # | |
9fe9816b | 1409 | mov\\t%0.16b, %1.16b |
43e9d192 IB |
1410 | ldp\\t%0, %H0, %1 |
1411 | stp\\t%1, %H1, %0 | |
1412 | stp\\txzr, xzr, %0 | |
1413 | ldr\\t%q0, %1 | |
1414 | str\\t%q1, %0" | |
d4f9e819 | 1415 | [(set_attr "type" "multiple,neon_move,f_mcr,f_mrc,neon_logic_q, \ |
db46a2e6 JG |
1416 | load_16,store_16,store_16,\ |
1417 | load_16,store_16") | |
d4f9e819 RS |
1418 | (set_attr "length" "8,4,8,8,4,4,4,4,4,4") |
1419 | (set_attr "arch" "*,simd,*,*,simd,*,*,*,fp,fp")] | |
78ec3036 | 1420 | ) |
43e9d192 IB |
1421 | |
1422 | ;; Split a TImode register-register or register-immediate move into | |
1423 | ;; its component DImode pieces, taking care to handle overlapping | |
1424 | ;; source and dest registers. | |
1425 | (define_split | |
1426 | [(set (match_operand:TI 0 "register_operand" "") | |
1427 | (match_operand:TI 1 "aarch64_reg_or_imm" ""))] | |
1428 | "reload_completed && aarch64_split_128bit_move_p (operands[0], operands[1])" | |
1429 | [(const_int 0)] | |
1430 | { | |
1431 | aarch64_split_128bit_move (operands[0], operands[1]); | |
1432 | DONE; | |
1433 | }) | |
1434 | ||
1435 | (define_expand "mov<mode>" | |
abbe1ed2 SMW |
1436 | [(set (match_operand:GPF_TF_F16_MOV 0 "nonimmediate_operand") |
1437 | (match_operand:GPF_TF_F16_MOV 1 "general_operand"))] | |
43e9d192 | 1438 | "" |
261fb553 | 1439 | { |
43e9d192 | 1440 | if (!TARGET_FLOAT) |
261fb553 | 1441 | { |
fc29dfc9 | 1442 | aarch64_err_no_fpadvsimd (<MODE>mode); |
248e357f FY |
1443 | machine_mode intmode |
1444 | = int_mode_for_size (GET_MODE_BITSIZE (<MODE>mode), 0).require (); | |
1445 | emit_move_insn (gen_lowpart (intmode, operands[0]), | |
1446 | gen_lowpart (intmode, operands[1])); | |
1447 | DONE; | |
261fb553 | 1448 | } |
43e9d192 | 1449 | |
490948ca JW |
1450 | if (GET_CODE (operands[0]) == MEM |
1451 | && ! (GET_CODE (operands[1]) == CONST_DOUBLE | |
1452 | && aarch64_float_const_zero_rtx_p (operands[1]))) | |
43e9d192 | 1453 | operands[1] = force_reg (<MODE>mode, operands[1]); |
261fb553 | 1454 | } |
43e9d192 IB |
1455 | ) |
1456 | ||
abbe1ed2 SMW |
1457 | (define_insn "*mov<mode>_aarch64" |
1458 | [(set (match_operand:HFBF 0 "nonimmediate_operand" "=w,w , w,?r,w,w ,w ,w,m,r,m ,r") | |
1459 | (match_operand:HFBF 1 "general_operand" "Y ,?rY,?r, w,w,Ufc,Uvi,m,w,m,rY,r"))] | |
1460 | "TARGET_FLOAT && (register_operand (operands[0], <MODE>mode) | |
1461 | || aarch64_reg_or_fp_zero (operands[1], <MODE>mode))" | |
c2ec330c | 1462 | "@ |
523d7207 | 1463 | movi\\t%0.4h, #0 |
90e6443f | 1464 | fmov\\t%h0, %w1 |
d89fe63e | 1465 | dup\\t%w0.4h, %w1 |
c2ec330c AL |
1466 | umov\\t%w0, %1.h[0] |
1467 | mov\\t%0.h[0], %1.h[0] | |
90e6443f | 1468 | fmov\\t%h0, %1 |
980902fb | 1469 | * return aarch64_output_scalar_simd_mov_immediate (operands[1], HImode); |
c2ec330c AL |
1470 | ldr\\t%h0, %1 |
1471 | str\\t%h1, %0 | |
1472 | ldrh\\t%w0, %1 | |
1473 | strh\\t%w1, %0 | |
1474 | mov\\t%w0, %w1" | |
d89fe63e | 1475 | [(set_attr "type" "neon_move,f_mcr,neon_move,neon_to_gp, neon_move,fconsts, \ |
89b2133e | 1476 | neon_move,f_loads,f_stores,load_4,store_4,mov_reg") |
488461d8 | 1477 | (set_attr "arch" "simd,fp16,simd,simd,simd,fp16,simd,*,*,*,*,*")] |
c2ec330c AL |
1478 | ) |
1479 | ||
43e9d192 | 1480 | (define_insn "*movsf_aarch64" |
90e6443f TC |
1481 | [(set (match_operand:SF 0 "nonimmediate_operand" "=w,w ,?r,w,w ,w ,w,m,r,m ,r,r") |
1482 | (match_operand:SF 1 "general_operand" "Y ,?rY, w,w,Ufc,Uvi,m,w,m,rY,r,M"))] | |
43e9d192 | 1483 | "TARGET_FLOAT && (register_operand (operands[0], SFmode) |
701170a4 | 1484 | || aarch64_reg_or_fp_zero (operands[1], SFmode))" |
43e9d192 | 1485 | "@ |
523d7207 | 1486 | movi\\t%0.2s, #0 |
43e9d192 IB |
1487 | fmov\\t%s0, %w1 |
1488 | fmov\\t%w0, %s1 | |
1489 | fmov\\t%s0, %s1 | |
3520f7cc | 1490 | fmov\\t%s0, %1 |
90e6443f | 1491 | * return aarch64_output_scalar_simd_mov_immediate (operands[1], SImode); |
43e9d192 IB |
1492 | ldr\\t%s0, %1 |
1493 | str\\t%s1, %0 | |
1494 | ldr\\t%w0, %1 | |
1495 | str\\t%w1, %0 | |
90e6443f TC |
1496 | mov\\t%w0, %w1 |
1497 | mov\\t%w0, %1" | |
1498 | [(set_attr "type" "neon_move,f_mcr,f_mrc,fmov,fconsts,neon_move,\ | |
89b2133e | 1499 | f_loads,f_stores,load_4,store_4,mov_reg,\ |
90e6443f | 1500 | fconsts") |
488461d8 | 1501 | (set_attr "arch" "simd,*,*,*,*,simd,*,*,*,*,*,*")] |
43e9d192 IB |
1502 | ) |
1503 | ||
1504 | (define_insn "*movdf_aarch64" | |
90e6443f TC |
1505 | [(set (match_operand:DF 0 "nonimmediate_operand" "=w, w ,?r,w,w ,w ,w,m,r,m ,r,r") |
1506 | (match_operand:DF 1 "general_operand" "Y , ?rY, w,w,Ufc,Uvi,m,w,m,rY,r,N"))] | |
43e9d192 | 1507 | "TARGET_FLOAT && (register_operand (operands[0], DFmode) |
701170a4 | 1508 | || aarch64_reg_or_fp_zero (operands[1], DFmode))" |
43e9d192 | 1509 | "@ |
523d7207 | 1510 | movi\\t%d0, #0 |
43e9d192 IB |
1511 | fmov\\t%d0, %x1 |
1512 | fmov\\t%x0, %d1 | |
1513 | fmov\\t%d0, %d1 | |
3520f7cc | 1514 | fmov\\t%d0, %1 |
90e6443f | 1515 | * return aarch64_output_scalar_simd_mov_immediate (operands[1], DImode); |
43e9d192 IB |
1516 | ldr\\t%d0, %1 |
1517 | str\\t%d1, %0 | |
1518 | ldr\\t%x0, %1 | |
1519 | str\\t%x1, %0 | |
90e6443f TC |
1520 | mov\\t%x0, %x1 |
1521 | mov\\t%x0, %1" | |
1522 | [(set_attr "type" "neon_move,f_mcr,f_mrc,fmov,fconstd,neon_move,\ | |
db46a2e6 | 1523 | f_loadd,f_stored,load_8,store_8,mov_reg,\ |
90e6443f | 1524 | fconstd") |
488461d8 | 1525 | (set_attr "arch" "simd,*,*,*,*,simd,*,*,*,*,*,*")] |
90e6443f TC |
1526 | ) |
1527 | ||
1528 | (define_split | |
1529 | [(set (match_operand:GPF_HF 0 "nonimmediate_operand") | |
1530 | (match_operand:GPF_HF 1 "general_operand"))] | |
1531 | "can_create_pseudo_p () | |
1532 | && !aarch64_can_const_movi_rtx_p (operands[1], <MODE>mode) | |
1533 | && !aarch64_float_const_representable_p (operands[1]) | |
1534 | && aarch64_float_const_rtx_p (operands[1])" | |
1535 | [(const_int 0)] | |
1536 | { | |
1537 | unsigned HOST_WIDE_INT ival; | |
1538 | if (!aarch64_reinterpret_float_as_int (operands[1], &ival)) | |
1539 | FAIL; | |
1540 | ||
1541 | rtx tmp = gen_reg_rtx (<FCVT_TARGET>mode); | |
1542 | emit_move_insn (tmp, gen_int_mode (ival, <FCVT_TARGET>mode)); | |
1543 | emit_move_insn (operands[0], gen_lowpart (<MODE>mode, tmp)); | |
1544 | DONE; | |
1545 | } | |
43e9d192 IB |
1546 | ) |
1547 | ||
43e9d192 IB |
1548 | (define_insn "*movtf_aarch64" |
1549 | [(set (match_operand:TF 0 | |
d4f9e819 | 1550 | "nonimmediate_operand" "=w,?r ,w ,?r,w,?w,w,m,?r,m ,m") |
43e9d192 | 1551 | (match_operand:TF 1 |
d4f9e819 | 1552 | "general_operand" " w,?rY,?r,w ,Y,Y ,m,w,m ,?r,Y"))] |
43e9d192 | 1553 | "TARGET_FLOAT && (register_operand (operands[0], TFmode) |
6a0f8c01 | 1554 | || aarch64_reg_or_fp_zero (operands[1], TFmode))" |
43e9d192 | 1555 | "@ |
9fe9816b | 1556 | mov\\t%0.16b, %1.16b |
12dc6974 SN |
1557 | # |
1558 | # | |
1559 | # | |
43e9d192 IB |
1560 | movi\\t%0.2d, #0 |
1561 | fmov\\t%s0, wzr | |
1562 | ldr\\t%q0, %1 | |
1563 | str\\t%q1, %0 | |
1564 | ldp\\t%0, %H0, %1 | |
6a0f8c01 JW |
1565 | stp\\t%1, %H1, %0 |
1566 | stp\\txzr, xzr, %0" | |
04e34811 | 1567 | [(set_attr "type" "logic_reg,multiple,f_mcr,f_mrc,neon_move_q,f_mcr,\ |
db46a2e6 | 1568 | f_loadd,f_stored,load_16,store_16,store_16") |
6a0f8c01 | 1569 | (set_attr "length" "4,8,8,8,4,4,4,4,4,4,4") |
488461d8 | 1570 | (set_attr "arch" "simd,*,*,*,simd,*,*,*,*,*,*")] |
43e9d192 IB |
1571 | ) |
1572 | ||
12dc6974 SN |
1573 | (define_split |
1574 | [(set (match_operand:TF 0 "register_operand" "") | |
d4f9e819 | 1575 | (match_operand:TF 1 "nonmemory_operand" ""))] |
12dc6974 SN |
1576 | "reload_completed && aarch64_split_128bit_move_p (operands[0], operands[1])" |
1577 | [(const_int 0)] | |
1578 | { | |
1579 | aarch64_split_128bit_move (operands[0], operands[1]); | |
1580 | DONE; | |
1581 | } | |
1582 | ) | |
1583 | ||
0caf592d KT |
1584 | (define_insn "aarch64_cpymemdi" |
1585 | [(parallel [ | |
1586 | (set (match_operand:DI 2 "register_operand" "+&r") (const_int 0)) | |
1587 | (clobber (match_operand:DI 0 "register_operand" "+&r")) | |
1588 | (clobber (match_operand:DI 1 "register_operand" "+&r")) | |
1589 | (set (mem:BLK (match_dup 0)) | |
1590 | (unspec:BLK [(mem:BLK (match_dup 1)) (match_dup 2)] UNSPEC_CPYMEM))])] | |
1591 | "TARGET_MOPS" | |
1592 | "cpyfp\t[%x0]!, [%x1]!, %x2!\;cpyfm\t[%x0]!, [%x1]!, %x2!\;cpyfe\t[%x0]!, [%x1]!, %x2!" | |
1593 | [(set_attr "length" "12")] | |
1594 | ) | |
1595 | ||
e2c75eea JG |
1596 | ;; 0 is dst |
1597 | ;; 1 is src | |
76715c32 | 1598 | ;; 2 is size of copy in bytes |
e2c75eea JG |
1599 | ;; 3 is alignment |
1600 | ||
76715c32 | 1601 | (define_expand "cpymemdi" |
e2c75eea JG |
1602 | [(match_operand:BLK 0 "memory_operand") |
1603 | (match_operand:BLK 1 "memory_operand") | |
0caf592d | 1604 | (match_operand:DI 2 "general_operand") |
e2c75eea | 1605 | (match_operand:DI 3 "immediate_operand")] |
0caf592d | 1606 | "!STRICT_ALIGNMENT || TARGET_MOPS" |
e2c75eea | 1607 | { |
76715c32 | 1608 | if (aarch64_expand_cpymem (operands)) |
e2c75eea JG |
1609 | DONE; |
1610 | FAIL; | |
1611 | } | |
1612 | ) | |
1613 | ||
bb768f8b KT |
1614 | (define_insn "aarch64_movmemdi" |
1615 | [(parallel [ | |
1616 | (set (match_operand:DI 2 "register_operand" "+&r") (const_int 0)) | |
1617 | (clobber (match_operand:DI 0 "register_operand" "+&r")) | |
1618 | (clobber (match_operand:DI 1 "register_operand" "+&r")) | |
1619 | (set (mem:BLK (match_dup 0)) | |
1620 | (unspec:BLK [(mem:BLK (match_dup 1)) (match_dup 2)] UNSPEC_MOVMEM))])] | |
1621 | "TARGET_MOPS" | |
1622 | "cpyp\t[%x0]!, [%x1]!, %x2!\;cpym\t[%x0]!, [%x1]!, %x2!\;cpye\t[%x0]!, [%x1]!, %x2!" | |
1623 | [(set_attr "length" "12")] | |
1624 | ) | |
1625 | ||
1626 | ;; 0 is dst | |
1627 | ;; 1 is src | |
1628 | ;; 2 is size of copy in bytes | |
1629 | ;; 3 is alignment | |
1630 | ||
1631 | (define_expand "movmemdi" | |
1632 | [(match_operand:BLK 0 "memory_operand") | |
1633 | (match_operand:BLK 1 "memory_operand") | |
1634 | (match_operand:DI 2 "general_operand") | |
1635 | (match_operand:DI 3 "immediate_operand")] | |
1636 | "TARGET_MOPS" | |
1637 | { | |
1638 | rtx sz_reg = operands[2]; | |
1639 | /* For constant-sized memmoves check the threshold. | |
1640 | FIXME: We should add a non-MOPS memmove expansion for smaller, | |
1641 | constant-sized memmove to avoid going to a libcall. */ | |
1642 | if (CONST_INT_P (sz_reg) | |
1643 | && INTVAL (sz_reg) < aarch64_mops_memmove_size_threshold) | |
1644 | FAIL; | |
1645 | ||
1646 | rtx addr_dst = XEXP (operands[0], 0); | |
1647 | rtx addr_src = XEXP (operands[1], 0); | |
1648 | ||
1649 | if (!REG_P (sz_reg)) | |
1650 | sz_reg = force_reg (DImode, sz_reg); | |
1651 | if (!REG_P (addr_dst)) | |
1652 | addr_dst = force_reg (DImode, addr_dst); | |
1653 | if (!REG_P (addr_src)) | |
1654 | addr_src = force_reg (DImode, addr_src); | |
1655 | emit_insn (gen_aarch64_movmemdi (addr_dst, addr_src, sz_reg)); | |
1656 | DONE; | |
1657 | } | |
1658 | ) | |
1659 | ||
d3bd985e KT |
1660 | (define_insn "aarch64_setmemdi" |
1661 | [(parallel [ | |
1662 | (set (match_operand:DI 2 "register_operand" "+&r") (const_int 0)) | |
1663 | (clobber (match_operand:DI 0 "register_operand" "+&r")) | |
1664 | (set (mem:BLK (match_dup 0)) | |
1665 | (unspec:BLK [(match_operand:QI 1 "aarch64_reg_or_zero" "rZ") | |
1666 | (match_dup 2)] UNSPEC_SETMEM))])] | |
1667 | "TARGET_MOPS" | |
1668 | "setp\t[%x0]!, %x2!, %x1\;setm\t[%x0]!, %x2!, %x1\;sete\t[%x0]!, %x2!, %x1" | |
1669 | [(set_attr "length" "12")] | |
1670 | ) | |
1671 | ||
54bbde55 SD |
1672 | ;; 0 is dst |
1673 | ;; 1 is val | |
1674 | ;; 2 is size of copy in bytes | |
1675 | ;; 3 is alignment | |
54bbde55 SD |
1676 | (define_expand "setmemdi" |
1677 | [(set (match_operand:BLK 0 "memory_operand") ;; Dest | |
1678 | (match_operand:QI 2 "nonmemory_operand")) ;; Value | |
d3bd985e | 1679 | (use (match_operand:DI 1 "general_operand")) ;; Length |
54bbde55 | 1680 | (match_operand 3 "immediate_operand")] ;; Align |
d3bd985e KT |
1681 | "TARGET_SIMD || TARGET_MOPS" |
1682 | { | |
54bbde55 SD |
1683 | if (aarch64_expand_setmem (operands)) |
1684 | DONE; | |
1685 | ||
1686 | FAIL; | |
1687 | }) | |
1688 | ||
43e9d192 IB |
1689 | ;; Operands 1 and 3 are tied together by the final condition; so we allow |
1690 | ;; fairly lax checking on the second memory operation. | |
dfe1da23 JW |
1691 | (define_insn "load_pair_sw_<SX:mode><SX2:mode>" |
1692 | [(set (match_operand:SX 0 "register_operand" "=r,w") | |
1693 | (match_operand:SX 1 "aarch64_mem_pair_operand" "Ump,Ump")) | |
1694 | (set (match_operand:SX2 2 "register_operand" "=r,w") | |
1695 | (match_operand:SX2 3 "memory_operand" "m,m"))] | |
1696 | "rtx_equal_p (XEXP (operands[3], 0), | |
1697 | plus_constant (Pmode, | |
1698 | XEXP (operands[1], 0), | |
1699 | GET_MODE_SIZE (<SX:MODE>mode)))" | |
350013bc | 1700 | "@ |
2c272091 RS |
1701 | ldp\\t%w0, %w2, %z1 |
1702 | ldp\\t%s0, %s2, %z1" | |
89b2133e | 1703 | [(set_attr "type" "load_8,neon_load1_2reg") |
488461d8 | 1704 | (set_attr "arch" "*,fp")] |
43e9d192 IB |
1705 | ) |
1706 | ||
dfe1da23 JW |
1707 | ;; Storing different modes that can still be merged |
1708 | (define_insn "load_pair_dw_<DX:mode><DX2:mode>" | |
1709 | [(set (match_operand:DX 0 "register_operand" "=r,w") | |
1710 | (match_operand:DX 1 "aarch64_mem_pair_operand" "Ump,Ump")) | |
1711 | (set (match_operand:DX2 2 "register_operand" "=r,w") | |
1712 | (match_operand:DX2 3 "memory_operand" "m,m"))] | |
1713 | "rtx_equal_p (XEXP (operands[3], 0), | |
1714 | plus_constant (Pmode, | |
1715 | XEXP (operands[1], 0), | |
1716 | GET_MODE_SIZE (<DX:MODE>mode)))" | |
350013bc | 1717 | "@ |
2c272091 RS |
1718 | ldp\\t%x0, %x2, %z1 |
1719 | ldp\\t%d0, %d2, %z1" | |
db46a2e6 | 1720 | [(set_attr "type" "load_16,neon_load1_2reg") |
488461d8 | 1721 | (set_attr "arch" "*,fp")] |
350013bc BC |
1722 | ) |
1723 | ||
a0d0b980 SE |
1724 | (define_insn "load_pair_dw_tftf" |
1725 | [(set (match_operand:TF 0 "register_operand" "=w") | |
1726 | (match_operand:TF 1 "aarch64_mem_pair_operand" "Ump")) | |
1727 | (set (match_operand:TF 2 "register_operand" "=w") | |
1728 | (match_operand:TF 3 "memory_operand" "m"))] | |
1729 | "TARGET_SIMD | |
1730 | && rtx_equal_p (XEXP (operands[3], 0), | |
1731 | plus_constant (Pmode, | |
1732 | XEXP (operands[1], 0), | |
1733 | GET_MODE_SIZE (TFmode)))" | |
2c272091 | 1734 | "ldp\\t%q0, %q2, %z1" |
a0d0b980 SE |
1735 | [(set_attr "type" "neon_ldp_q") |
1736 | (set_attr "fp" "yes")] | |
1737 | ) | |
1738 | ||
43e9d192 IB |
1739 | ;; Operands 0 and 2 are tied together by the final condition; so we allow |
1740 | ;; fairly lax checking on the second memory operation. | |
dfe1da23 JW |
1741 | (define_insn "store_pair_sw_<SX:mode><SX2:mode>" |
1742 | [(set (match_operand:SX 0 "aarch64_mem_pair_operand" "=Ump,Ump") | |
1743 | (match_operand:SX 1 "aarch64_reg_zero_or_fp_zero" "rYZ,w")) | |
1744 | (set (match_operand:SX2 2 "memory_operand" "=m,m") | |
1745 | (match_operand:SX2 3 "aarch64_reg_zero_or_fp_zero" "rYZ,w"))] | |
1746 | "rtx_equal_p (XEXP (operands[2], 0), | |
1747 | plus_constant (Pmode, | |
1748 | XEXP (operands[0], 0), | |
1749 | GET_MODE_SIZE (<SX:MODE>mode)))" | |
350013bc | 1750 | "@ |
2c272091 RS |
1751 | stp\\t%w1, %w3, %z0 |
1752 | stp\\t%s1, %s3, %z0" | |
89b2133e | 1753 | [(set_attr "type" "store_8,neon_store1_2reg") |
488461d8 | 1754 | (set_attr "arch" "*,fp")] |
350013bc BC |
1755 | ) |
1756 | ||
dfe1da23 JW |
1757 | ;; Storing different modes that can still be merged |
1758 | (define_insn "store_pair_dw_<DX:mode><DX2:mode>" | |
1759 | [(set (match_operand:DX 0 "aarch64_mem_pair_operand" "=Ump,Ump") | |
1760 | (match_operand:DX 1 "aarch64_reg_zero_or_fp_zero" "rYZ,w")) | |
1761 | (set (match_operand:DX2 2 "memory_operand" "=m,m") | |
1762 | (match_operand:DX2 3 "aarch64_reg_zero_or_fp_zero" "rYZ,w"))] | |
1763 | "rtx_equal_p (XEXP (operands[2], 0), | |
1764 | plus_constant (Pmode, | |
1765 | XEXP (operands[0], 0), | |
1766 | GET_MODE_SIZE (<DX:MODE>mode)))" | |
350013bc | 1767 | "@ |
2c272091 RS |
1768 | stp\\t%x1, %x3, %z0 |
1769 | stp\\t%d1, %d3, %z0" | |
db46a2e6 | 1770 | [(set_attr "type" "store_16,neon_store1_2reg") |
488461d8 | 1771 | (set_attr "arch" "*,fp")] |
43e9d192 IB |
1772 | ) |
1773 | ||
a0d0b980 SE |
1774 | (define_insn "store_pair_dw_tftf" |
1775 | [(set (match_operand:TF 0 "aarch64_mem_pair_operand" "=Ump") | |
1776 | (match_operand:TF 1 "register_operand" "w")) | |
1777 | (set (match_operand:TF 2 "memory_operand" "=m") | |
1778 | (match_operand:TF 3 "register_operand" "w"))] | |
1779 | "TARGET_SIMD && | |
1780 | rtx_equal_p (XEXP (operands[2], 0), | |
1781 | plus_constant (Pmode, | |
1782 | XEXP (operands[0], 0), | |
1783 | GET_MODE_SIZE (TFmode)))" | |
2c272091 | 1784 | "stp\\t%q1, %q3, %z0" |
a0d0b980 SE |
1785 | [(set_attr "type" "neon_stp_q") |
1786 | (set_attr "fp" "yes")] | |
1787 | ) | |
1788 | ||
3e322b3f JW |
1789 | ;; Load pair with post-index writeback. This is primarily used in function |
1790 | ;; epilogues. | |
28514dda | 1791 | (define_insn "loadwb_pair<GPI:mode>_<P:mode>" |
43e9d192 | 1792 | [(parallel |
28514dda YZ |
1793 | [(set (match_operand:P 0 "register_operand" "=k") |
1794 | (plus:P (match_operand:P 1 "register_operand" "0") | |
44707478 | 1795 | (match_operand:P 4 "aarch64_mem_pair_offset" "n"))) |
43e9d192 | 1796 | (set (match_operand:GPI 2 "register_operand" "=r") |
3e322b3f | 1797 | (mem:GPI (match_dup 1))) |
43e9d192 | 1798 | (set (match_operand:GPI 3 "register_operand" "=r") |
28514dda YZ |
1799 | (mem:GPI (plus:P (match_dup 1) |
1800 | (match_operand:P 5 "const_int_operand" "n"))))])] | |
3e322b3f | 1801 | "INTVAL (operands[5]) == GET_MODE_SIZE (<GPI:MODE>mode)" |
e7ba492a RS |
1802 | "ldp\\t%<GPI:w>2, %<GPI:w>3, [%1], %4" |
1803 | [(set_attr "type" "load_<GPI:ldpstp_sz>")] | |
43e9d192 IB |
1804 | ) |
1805 | ||
159313d9 JW |
1806 | (define_insn "loadwb_pair<GPF:mode>_<P:mode>" |
1807 | [(parallel | |
1808 | [(set (match_operand:P 0 "register_operand" "=k") | |
1809 | (plus:P (match_operand:P 1 "register_operand" "0") | |
44707478 | 1810 | (match_operand:P 4 "aarch64_mem_pair_offset" "n"))) |
159313d9 | 1811 | (set (match_operand:GPF 2 "register_operand" "=w") |
3e322b3f | 1812 | (mem:GPF (match_dup 1))) |
159313d9 JW |
1813 | (set (match_operand:GPF 3 "register_operand" "=w") |
1814 | (mem:GPF (plus:P (match_dup 1) | |
1815 | (match_operand:P 5 "const_int_operand" "n"))))])] | |
3e322b3f | 1816 | "INTVAL (operands[5]) == GET_MODE_SIZE (<GPF:MODE>mode)" |
e7ba492a | 1817 | "ldp\\t%<GPF:w>2, %<GPF:w>3, [%1], %4" |
159313d9 JW |
1818 | [(set_attr "type" "neon_load1_2reg")] |
1819 | ) | |
1820 | ||
a0d0b980 SE |
1821 | (define_insn "loadwb_pair<TX:mode>_<P:mode>" |
1822 | [(parallel | |
1823 | [(set (match_operand:P 0 "register_operand" "=k") | |
1824 | (plus:P (match_operand:P 1 "register_operand" "0") | |
1825 | (match_operand:P 4 "aarch64_mem_pair_offset" "n"))) | |
1826 | (set (match_operand:TX 2 "register_operand" "=w") | |
1827 | (mem:TX (match_dup 1))) | |
1828 | (set (match_operand:TX 3 "register_operand" "=w") | |
1829 | (mem:TX (plus:P (match_dup 1) | |
1830 | (match_operand:P 5 "const_int_operand" "n"))))])] | |
1831 | "TARGET_SIMD && INTVAL (operands[5]) == GET_MODE_SIZE (<TX:MODE>mode)" | |
1832 | "ldp\\t%q2, %q3, [%1], %4" | |
1833 | [(set_attr "type" "neon_ldp_q")] | |
1834 | ) | |
1835 | ||
3e322b3f JW |
1836 | ;; Store pair with pre-index writeback. This is primarily used in function |
1837 | ;; prologues. | |
28514dda | 1838 | (define_insn "storewb_pair<GPI:mode>_<P:mode>" |
43e9d192 | 1839 | [(parallel |
28514dda YZ |
1840 | [(set (match_operand:P 0 "register_operand" "=&k") |
1841 | (plus:P (match_operand:P 1 "register_operand" "0") | |
44707478 | 1842 | (match_operand:P 4 "aarch64_mem_pair_offset" "n"))) |
28514dda | 1843 | (set (mem:GPI (plus:P (match_dup 0) |
43e9d192 IB |
1844 | (match_dup 4))) |
1845 | (match_operand:GPI 2 "register_operand" "r")) | |
28514dda YZ |
1846 | (set (mem:GPI (plus:P (match_dup 0) |
1847 | (match_operand:P 5 "const_int_operand" "n"))) | |
43e9d192 IB |
1848 | (match_operand:GPI 3 "register_operand" "r"))])] |
1849 | "INTVAL (operands[5]) == INTVAL (operands[4]) + GET_MODE_SIZE (<GPI:MODE>mode)" | |
e7ba492a RS |
1850 | "stp\\t%<GPI:w>2, %<GPI:w>3, [%0, %4]!" |
1851 | [(set_attr "type" "store_<GPI:ldpstp_sz>")] | |
80c11907 JW |
1852 | ) |
1853 | ||
1854 | (define_insn "storewb_pair<GPF:mode>_<P:mode>" | |
1855 | [(parallel | |
1856 | [(set (match_operand:P 0 "register_operand" "=&k") | |
1857 | (plus:P (match_operand:P 1 "register_operand" "0") | |
44707478 | 1858 | (match_operand:P 4 "aarch64_mem_pair_offset" "n"))) |
80c11907 JW |
1859 | (set (mem:GPF (plus:P (match_dup 0) |
1860 | (match_dup 4))) | |
1861 | (match_operand:GPF 2 "register_operand" "w")) | |
1862 | (set (mem:GPF (plus:P (match_dup 0) | |
1863 | (match_operand:P 5 "const_int_operand" "n"))) | |
1864 | (match_operand:GPF 3 "register_operand" "w"))])] | |
1865 | "INTVAL (operands[5]) == INTVAL (operands[4]) + GET_MODE_SIZE (<GPF:MODE>mode)" | |
e7ba492a | 1866 | "stp\\t%<GPF:w>2, %<GPF:w>3, [%0, %4]!" |
80c11907 | 1867 | [(set_attr "type" "neon_store1_2reg<q>")] |
43e9d192 IB |
1868 | ) |
1869 | ||
a0d0b980 SE |
1870 | (define_insn "storewb_pair<TX:mode>_<P:mode>" |
1871 | [(parallel | |
1872 | [(set (match_operand:P 0 "register_operand" "=&k") | |
1873 | (plus:P (match_operand:P 1 "register_operand" "0") | |
1874 | (match_operand:P 4 "aarch64_mem_pair_offset" "n"))) | |
1875 | (set (mem:TX (plus:P (match_dup 0) | |
1876 | (match_dup 4))) | |
1877 | (match_operand:TX 2 "register_operand" "w")) | |
1878 | (set (mem:TX (plus:P (match_dup 0) | |
1879 | (match_operand:P 5 "const_int_operand" "n"))) | |
1880 | (match_operand:TX 3 "register_operand" "w"))])] | |
1881 | "TARGET_SIMD | |
1882 | && INTVAL (operands[5]) | |
1883 | == INTVAL (operands[4]) + GET_MODE_SIZE (<TX:MODE>mode)" | |
1884 | "stp\\t%q2, %q3, [%0, %4]!" | |
1885 | [(set_attr "type" "neon_stp_q")] | |
1886 | ) | |
1887 | ||
43e9d192 IB |
1888 | ;; ------------------------------------------------------------------- |
1889 | ;; Sign/Zero extension | |
1890 | ;; ------------------------------------------------------------------- | |
1891 | ||
1892 | (define_expand "<optab>sidi2" | |
1893 | [(set (match_operand:DI 0 "register_operand") | |
1894 | (ANY_EXTEND:DI (match_operand:SI 1 "nonimmediate_operand")))] | |
1895 | "" | |
1896 | ) | |
1897 | ||
1898 | (define_insn "*extendsidi2_aarch64" | |
1899 | [(set (match_operand:DI 0 "register_operand" "=r,r") | |
1900 | (sign_extend:DI (match_operand:SI 1 "nonimmediate_operand" "r,m")))] | |
1901 | "" | |
1902 | "@ | |
1903 | sxtw\t%0, %w1 | |
1904 | ldrsw\t%0, %1" | |
89b2133e | 1905 | [(set_attr "type" "extend,load_4")] |
43e9d192 IB |
1906 | ) |
1907 | ||
350013bc BC |
1908 | (define_insn "*load_pair_extendsidi2_aarch64" |
1909 | [(set (match_operand:DI 0 "register_operand" "=r") | |
1910 | (sign_extend:DI (match_operand:SI 1 "aarch64_mem_pair_operand" "Ump"))) | |
1911 | (set (match_operand:DI 2 "register_operand" "=r") | |
1912 | (sign_extend:DI (match_operand:SI 3 "memory_operand" "m")))] | |
1913 | "rtx_equal_p (XEXP (operands[3], 0), | |
1914 | plus_constant (Pmode, | |
1915 | XEXP (operands[1], 0), | |
1916 | GET_MODE_SIZE (SImode)))" | |
2c272091 | 1917 | "ldpsw\\t%0, %2, %z1" |
89b2133e | 1918 | [(set_attr "type" "load_8")] |
350013bc BC |
1919 | ) |
1920 | ||
43e9d192 | 1921 | (define_insn "*zero_extendsidi2_aarch64" |
0cfc095c WD |
1922 | [(set (match_operand:DI 0 "register_operand" "=r,r,w,w,r,w") |
1923 | (zero_extend:DI (match_operand:SI 1 "nonimmediate_operand" "r,m,r,m,w,w")))] | |
43e9d192 IB |
1924 | "" |
1925 | "@ | |
1926 | uxtw\t%0, %w1 | |
0cfc095c WD |
1927 | ldr\t%w0, %1 |
1928 | fmov\t%s0, %w1 | |
1929 | ldr\t%s0, %1 | |
1930 | fmov\t%w0, %s1 | |
1931 | fmov\t%s0, %s1" | |
e6b85350 | 1932 | [(set_attr "type" "mov_reg,load_4,f_mcr,f_loads,f_mrc,fmov") |
0cfc095c | 1933 | (set_attr "arch" "*,*,fp,fp,fp,fp")] |
43e9d192 IB |
1934 | ) |
1935 | ||
350013bc | 1936 | (define_insn "*load_pair_zero_extendsidi2_aarch64" |
0cfc095c WD |
1937 | [(set (match_operand:DI 0 "register_operand" "=r,w") |
1938 | (zero_extend:DI (match_operand:SI 1 "aarch64_mem_pair_operand" "Ump,Ump"))) | |
1939 | (set (match_operand:DI 2 "register_operand" "=r,w") | |
1940 | (zero_extend:DI (match_operand:SI 3 "memory_operand" "m,m")))] | |
350013bc BC |
1941 | "rtx_equal_p (XEXP (operands[3], 0), |
1942 | plus_constant (Pmode, | |
1943 | XEXP (operands[1], 0), | |
1944 | GET_MODE_SIZE (SImode)))" | |
0cfc095c | 1945 | "@ |
2c272091 RS |
1946 | ldp\t%w0, %w2, %z1 |
1947 | ldp\t%s0, %s2, %z1" | |
0cfc095c WD |
1948 | [(set_attr "type" "load_8,neon_load1_2reg") |
1949 | (set_attr "arch" "*,fp")] | |
350013bc BC |
1950 | ) |
1951 | ||
43e9d192 IB |
1952 | (define_expand "<ANY_EXTEND:optab><SHORT:mode><GPI:mode>2" |
1953 | [(set (match_operand:GPI 0 "register_operand") | |
1954 | (ANY_EXTEND:GPI (match_operand:SHORT 1 "nonimmediate_operand")))] | |
1955 | "" | |
1956 | ) | |
1957 | ||
1958 | (define_insn "*extend<SHORT:mode><GPI:mode>2_aarch64" | |
8b06ccb2 KT |
1959 | [(set (match_operand:GPI 0 "register_operand" "=r,r,r") |
1960 | (sign_extend:GPI (match_operand:SHORT 1 "nonimmediate_operand" "r,m,w")))] | |
43e9d192 IB |
1961 | "" |
1962 | "@ | |
1963 | sxt<SHORT:size>\t%<GPI:w>0, %w1 | |
8b06ccb2 | 1964 | ldrs<SHORT:size>\t%<GPI:w>0, %1 |
a5e78ee6 | 1965 | smov\t%<GPI:w>0, %1.<SHORT:size>[0]" |
8b06ccb2 KT |
1966 | [(set_attr "type" "extend,load_4,neon_to_gp") |
1967 | (set_attr "arch" "*,*,fp")] | |
43e9d192 IB |
1968 | ) |
1969 | ||
1970 | (define_insn "*zero_extend<SHORT:mode><GPI:mode>2_aarch64" | |
8695bf78 JW |
1971 | [(set (match_operand:GPI 0 "register_operand" "=r,r,w,r") |
1972 | (zero_extend:GPI (match_operand:SHORT 1 "nonimmediate_operand" "r,m,m,w")))] | |
43e9d192 IB |
1973 | "" |
1974 | "@ | |
63715e5e | 1975 | and\t%<GPI:w>0, %<GPI:w>1, <SHORT:short_mask> |
c902d3c8 | 1976 | ldr<SHORT:size>\t%w0, %1 |
8695bf78 JW |
1977 | ldr\t%<SHORT:size>0, %1 |
1978 | umov\t%w0, %1.<SHORT:size>[0]" | |
1979 | [(set_attr "type" "logic_imm,load_4,f_loads,neon_to_gp") | |
1980 | (set_attr "arch" "*,*,fp,fp")] | |
43e9d192 IB |
1981 | ) |
1982 | ||
1983 | (define_expand "<optab>qihi2" | |
1984 | [(set (match_operand:HI 0 "register_operand") | |
1985 | (ANY_EXTEND:HI (match_operand:QI 1 "nonimmediate_operand")))] | |
1986 | "" | |
1987 | ) | |
1988 | ||
63715e5e | 1989 | (define_insn "*extendqihi2_aarch64" |
43e9d192 | 1990 | [(set (match_operand:HI 0 "register_operand" "=r,r") |
63715e5e | 1991 | (sign_extend:HI (match_operand:QI 1 "nonimmediate_operand" "r,m")))] |
43e9d192 IB |
1992 | "" |
1993 | "@ | |
63715e5e WD |
1994 | sxtb\t%w0, %w1 |
1995 | ldrsb\t%w0, %1" | |
89b2133e | 1996 | [(set_attr "type" "extend,load_4")] |
43e9d192 IB |
1997 | ) |
1998 | ||
63715e5e WD |
1999 | (define_insn "*zero_extendqihi2_aarch64" |
2000 | [(set (match_operand:HI 0 "register_operand" "=r,r") | |
2001 | (zero_extend:HI (match_operand:QI 1 "nonimmediate_operand" "r,m")))] | |
2002 | "" | |
2003 | "@ | |
2004 | and\t%w0, %w1, 255 | |
2005 | ldrb\t%w0, %1" | |
89b2133e | 2006 | [(set_attr "type" "logic_imm,load_4")] |
63715e5e WD |
2007 | ) |
2008 | ||
43e9d192 IB |
2009 | ;; ------------------------------------------------------------------- |
2010 | ;; Simple arithmetic | |
2011 | ;; ------------------------------------------------------------------- | |
2012 | ||
2013 | (define_expand "add<mode>3" | |
2014 | [(set | |
1bbffb87 DZ |
2015 | (match_operand:GPI 0 "register_operand") |
2016 | (plus:GPI (match_operand:GPI 1 "register_operand") | |
2017 | (match_operand:GPI 2 "aarch64_pluslong_or_poly_operand")))] | |
43e9d192 | 2018 | "" |
89f0a62e | 2019 | { |
35323bd8 KT |
2020 | /* If operands[1] is a subreg extract the inner RTX. */ |
2021 | rtx op1 = REG_P (operands[1]) ? operands[1] : SUBREG_REG (operands[1]); | |
2022 | ||
45a6c1e3 WD |
2023 | /* If the constant is too large for a single instruction and isn't frame |
2024 | based, split off the immediate so it is available for CSE. */ | |
2025 | if (!aarch64_plus_immediate (operands[2], <MODE>mode) | |
0fdc30bc | 2026 | && !(TARGET_SVE && aarch64_sve_plus_immediate (operands[2], <MODE>mode)) |
45a6c1e3 | 2027 | && can_create_pseudo_p () |
35323bd8 KT |
2028 | && (!REG_P (op1) |
2029 | || !REGNO_PTR_FRAME_P (REGNO (op1)))) | |
45a6c1e3 | 2030 | operands[2] = force_reg (<MODE>mode, operands[2]); |
a65b9ad8 KT |
2031 | /* Some tunings prefer to avoid VL-based operations. |
2032 | Split off the poly immediate here. The rtx costs hook will reject attempts | |
2033 | to combine them back. */ | |
2034 | else if (GET_CODE (operands[2]) == CONST_POLY_INT | |
2035 | && can_create_pseudo_p () | |
2036 | && (aarch64_tune_params.extra_tuning_flags | |
2037 | & AARCH64_EXTRA_TUNE_CSE_SVE_VL_CONSTANTS)) | |
2038 | operands[2] = force_reg (<MODE>mode, operands[2]); | |
43cacb12 RS |
2039 | /* Expand polynomial additions now if the destination is the stack |
2040 | pointer, since we don't want to use that as a temporary. */ | |
2041 | else if (operands[0] == stack_pointer_rtx | |
2042 | && aarch64_split_add_offset_immediate (operands[2], <MODE>mode)) | |
2043 | { | |
2044 | aarch64_split_add_offset (<MODE>mode, operands[0], operands[1], | |
2045 | operands[2], NULL_RTX, NULL_RTX); | |
2046 | DONE; | |
2047 | } | |
89f0a62e | 2048 | }) |
43e9d192 | 2049 | |
89f0a62e | 2050 | (define_insn "*add<mode>3_aarch64" |
43e9d192 | 2051 | [(set |
0fdc30bc | 2052 | (match_operand:GPI 0 "register_operand" "=rk,rk,w,rk,r,r,rk") |
89f0a62e | 2053 | (plus:GPI |
0fdc30bc RS |
2054 | (match_operand:GPI 1 "register_operand" "%rk,rk,w,rk,rk,0,rk") |
2055 | (match_operand:GPI 2 "aarch64_pluslong_operand" "I,r,w,J,Uaa,Uai,Uav")))] | |
43e9d192 IB |
2056 | "" |
2057 | "@ | |
89f0a62e RH |
2058 | add\\t%<w>0, %<w>1, %2 |
2059 | add\\t%<w>0, %<w>1, %<w>2 | |
2060 | add\\t%<rtn>0<vas>, %<rtn>1<vas>, %<rtn>2<vas> | |
2061 | sub\\t%<w>0, %<w>1, #%n2 | |
43cacb12 | 2062 | # |
0fdc30bc RS |
2063 | * return aarch64_output_sve_scalar_inc_dec (operands[2]); |
2064 | * return aarch64_output_sve_addvl_addpl (operands[2]);" | |
2065 | ;; The "alu_imm" types for INC/DEC and ADDVL/ADDPL are just placeholders. | |
2066 | [(set_attr "type" "alu_imm,alu_sreg,neon_add,alu_imm,multiple,alu_imm,alu_imm") | |
2067 | (set_attr "arch" "*,*,simd,*,*,sve,sve")] | |
43e9d192 IB |
2068 | ) |
2069 | ||
3b585938 IB |
2070 | ;; zero_extend version of above |
2071 | (define_insn "*addsi3_aarch64_uxtw" | |
2072 | [(set | |
89f0a62e | 2073 | (match_operand:DI 0 "register_operand" "=rk,rk,rk,r") |
3b585938 | 2074 | (zero_extend:DI |
89f0a62e | 2075 | (plus:SI (match_operand:SI 1 "register_operand" "%rk,rk,rk,rk") |
ff1335e2 | 2076 | (match_operand:SI 2 "aarch64_pluslong_operand" "I,r,J,Uaa"))))] |
3b585938 IB |
2077 | "" |
2078 | "@ | |
2079 | add\\t%w0, %w1, %2 | |
2080 | add\\t%w0, %w1, %w2 | |
89f0a62e RH |
2081 | sub\\t%w0, %w1, #%n2 |
2082 | #" | |
2083 | [(set_attr "type" "alu_imm,alu_sreg,alu_imm,multiple")] | |
2084 | ) | |
2085 | ||
2086 | ;; If there's a free register, and we can load the constant with a | |
2087 | ;; single instruction, do so. This has a chance to improve scheduling. | |
2088 | (define_peephole2 | |
2089 | [(match_scratch:GPI 3 "r") | |
2090 | (set (match_operand:GPI 0 "register_operand") | |
2091 | (plus:GPI | |
2092 | (match_operand:GPI 1 "register_operand") | |
2093 | (match_operand:GPI 2 "aarch64_pluslong_strict_immedate")))] | |
2094 | "aarch64_move_imm (INTVAL (operands[2]), <MODE>mode)" | |
2095 | [(set (match_dup 3) (match_dup 2)) | |
2096 | (set (match_dup 0) (plus:GPI (match_dup 1) (match_dup 3)))] | |
2097 | ) | |
2098 | ||
2099 | (define_peephole2 | |
2100 | [(match_scratch:SI 3 "r") | |
2101 | (set (match_operand:DI 0 "register_operand") | |
2102 | (zero_extend:DI | |
2103 | (plus:SI | |
2104 | (match_operand:SI 1 "register_operand") | |
2105 | (match_operand:SI 2 "aarch64_pluslong_strict_immedate"))))] | |
2106 | "aarch64_move_imm (INTVAL (operands[2]), SImode)" | |
2107 | [(set (match_dup 3) (match_dup 2)) | |
2108 | (set (match_dup 0) (zero_extend:DI (plus:SI (match_dup 1) (match_dup 3))))] | |
2109 | ) | |
2110 | ||
2111 | ;; After peephole2 has had a chance to run, split any remaining long | |
2112 | ;; additions into two add immediates. | |
2113 | (define_split | |
2114 | [(set (match_operand:GPI 0 "register_operand") | |
2115 | (plus:GPI | |
2116 | (match_operand:GPI 1 "register_operand") | |
2117 | (match_operand:GPI 2 "aarch64_pluslong_strict_immedate")))] | |
2118 | "epilogue_completed" | |
2119 | [(set (match_dup 0) (plus:GPI (match_dup 1) (match_dup 3))) | |
2120 | (set (match_dup 0) (plus:GPI (match_dup 0) (match_dup 4)))] | |
2121 | { | |
2122 | HOST_WIDE_INT i = INTVAL (operands[2]); | |
2123 | HOST_WIDE_INT s = (i >= 0 ? i & 0xfff : -(-i & 0xfff)); | |
2124 | operands[3] = GEN_INT (i - s); | |
2125 | operands[4] = GEN_INT (s); | |
2126 | } | |
3b585938 IB |
2127 | ) |
2128 | ||
43cacb12 RS |
2129 | ;; Match addition of polynomial offsets that require one temporary, for which |
2130 | ;; we can use the early-clobbered destination register. This is a separate | |
2131 | ;; pattern so that the early clobber doesn't affect register allocation | |
2132 | ;; for other forms of addition. However, we still need to provide an | |
2133 | ;; all-register alternative, in case the offset goes out of range after | |
2134 | ;; elimination. For completeness we might as well provide all GPR-based | |
2135 | ;; alternatives from the main pattern. | |
2136 | ;; | |
2137 | ;; We don't have a pattern for additions requiring two temporaries since at | |
2138 | ;; present LRA doesn't allow new scratches to be added during elimination. | |
2139 | ;; Such offsets should be rare anyway. | |
2140 | ;; | |
2141 | ;; ??? But if we added LRA support for new scratches, much of the ugliness | |
2142 | ;; here would go away. We could just handle all polynomial constants in | |
2143 | ;; this pattern. | |
2144 | (define_insn_and_split "*add<mode>3_poly_1" | |
2145 | [(set | |
0fdc30bc | 2146 | (match_operand:GPI 0 "register_operand" "=r,r,r,r,r,r,&r") |
43cacb12 | 2147 | (plus:GPI |
c001c194 JJ |
2148 | (match_operand:GPI 1 "register_operand" "%rk,rk,rk,rk,0,rk,rk") |
2149 | (match_operand:GPI 2 "aarch64_pluslong_or_poly_operand" "I,r,J,Uaa,Uai,Uav,Uat")))] | |
43cacb12 RS |
2150 | "TARGET_SVE && operands[0] != stack_pointer_rtx" |
2151 | "@ | |
2152 | add\\t%<w>0, %<w>1, %2 | |
2153 | add\\t%<w>0, %<w>1, %<w>2 | |
2154 | sub\\t%<w>0, %<w>1, #%n2 | |
2155 | # | |
0fdc30bc RS |
2156 | * return aarch64_output_sve_scalar_inc_dec (operands[2]); |
2157 | * return aarch64_output_sve_addvl_addpl (operands[2]); | |
43cacb12 RS |
2158 | #" |
2159 | "&& epilogue_completed | |
2160 | && !reg_overlap_mentioned_p (operands[0], operands[1]) | |
2161 | && aarch64_split_add_offset_immediate (operands[2], <MODE>mode)" | |
2162 | [(const_int 0)] | |
2163 | { | |
2164 | aarch64_split_add_offset (<MODE>mode, operands[0], operands[1], | |
2165 | operands[2], operands[0], NULL_RTX); | |
2166 | DONE; | |
2167 | } | |
0fdc30bc RS |
2168 | ;; The "alu_imm" types for INC/DEC and ADDVL/ADDPL are just placeholders. |
2169 | [(set_attr "type" "alu_imm,alu_sreg,alu_imm,multiple,alu_imm,alu_imm,multiple")] | |
43cacb12 RS |
2170 | ) |
2171 | ||
89f0a62e RH |
2172 | (define_split |
2173 | [(set (match_operand:DI 0 "register_operand") | |
2174 | (zero_extend:DI | |
2175 | (plus:SI | |
2176 | (match_operand:SI 1 "register_operand") | |
2177 | (match_operand:SI 2 "aarch64_pluslong_strict_immedate"))))] | |
2178 | "epilogue_completed" | |
2179 | [(set (match_dup 5) (plus:SI (match_dup 1) (match_dup 3))) | |
2180 | (set (match_dup 0) (zero_extend:DI (plus:SI (match_dup 5) (match_dup 4))))] | |
2181 | { | |
2182 | HOST_WIDE_INT i = INTVAL (operands[2]); | |
2183 | HOST_WIDE_INT s = (i >= 0 ? i & 0xfff : -(-i & 0xfff)); | |
2184 | operands[3] = GEN_INT (i - s); | |
2185 | operands[4] = GEN_INT (s); | |
2186 | operands[5] = gen_lowpart (SImode, operands[0]); | |
2187 | } | |
43e9d192 IB |
2188 | ) |
2189 | ||
30c46053 MC |
2190 | (define_expand "addv<mode>4" |
2191 | [(match_operand:GPI 0 "register_operand") | |
2192 | (match_operand:GPI 1 "register_operand") | |
a58fe3c5 | 2193 | (match_operand:GPI 2 "aarch64_plus_operand") |
30c46053 MC |
2194 | (label_ref (match_operand 3 "" ""))] |
2195 | "" | |
2196 | { | |
a58fe3c5 RE |
2197 | if (CONST_INT_P (operands[2])) |
2198 | emit_insn (gen_add<mode>3_compareV_imm (operands[0], operands[1], | |
2199 | operands[2])); | |
2200 | else | |
2201 | emit_insn (gen_add<mode>3_compareV (operands[0], operands[1], operands[2])); | |
30c46053 MC |
2202 | aarch64_gen_unlikely_cbranch (NE, CC_Vmode, operands[3]); |
2203 | ||
2204 | DONE; | |
2205 | }) | |
2206 | ||
2207 | (define_expand "uaddv<mode>4" | |
2208 | [(match_operand:GPI 0 "register_operand") | |
2209 | (match_operand:GPI 1 "register_operand") | |
2210 | (match_operand:GPI 2 "register_operand") | |
2211 | (label_ref (match_operand 3 "" ""))] | |
2212 | "" | |
2213 | { | |
2214 | emit_insn (gen_add<mode>3_compareC (operands[0], operands[1], operands[2])); | |
f7343f20 | 2215 | aarch64_gen_unlikely_cbranch (LTU, CC_Cmode, operands[3]); |
30c46053 MC |
2216 | |
2217 | DONE; | |
2218 | }) | |
2219 | ||
2195867f | 2220 | (define_expand "addti3" |
1bbffb87 DZ |
2221 | [(set (match_operand:TI 0 "register_operand") |
2222 | (plus:TI (match_operand:TI 1 "register_operand") | |
2223 | (match_operand:TI 2 "aarch64_reg_or_imm")))] | |
30c46053 MC |
2224 | "" |
2225 | { | |
2226 | rtx low_dest, op1_low, op2_low, high_dest, op1_high, op2_high; | |
2227 | ||
2228 | aarch64_addti_scratch_regs (operands[1], operands[2], | |
2229 | &low_dest, &op1_low, &op2_low, | |
2230 | &high_dest, &op1_high, &op2_high); | |
2231 | ||
2232 | if (op2_low == const0_rtx) | |
2233 | { | |
2234 | low_dest = op1_low; | |
2235 | if (!aarch64_pluslong_operand (op2_high, DImode)) | |
2236 | op2_high = force_reg (DImode, op2_high); | |
2237 | emit_insn (gen_adddi3 (high_dest, op1_high, op2_high)); | |
2238 | } | |
2239 | else | |
2240 | { | |
2241 | emit_insn (gen_adddi3_compareC (low_dest, op1_low, | |
2242 | force_reg (DImode, op2_low))); | |
2243 | emit_insn (gen_adddi3_carryin (high_dest, op1_high, | |
2244 | force_reg (DImode, op2_high))); | |
2245 | } | |
2246 | ||
2247 | emit_move_insn (gen_lowpart (DImode, operands[0]), low_dest); | |
2248 | emit_move_insn (gen_highpart (DImode, operands[0]), high_dest); | |
2249 | ||
2250 | DONE; | |
2251 | }) | |
2252 | ||
2253 | (define_expand "addvti4" | |
1bbffb87 DZ |
2254 | [(match_operand:TI 0 "register_operand") |
2255 | (match_operand:TI 1 "register_operand") | |
2256 | (match_operand:TI 2 "aarch64_reg_or_imm") | |
30c46053 | 2257 | (label_ref (match_operand 3 "" ""))] |
2195867f RH |
2258 | "" |
2259 | { | |
30c46053 | 2260 | rtx low_dest, op1_low, op2_low, high_dest, op1_high, op2_high; |
2195867f | 2261 | |
30c46053 MC |
2262 | aarch64_addti_scratch_regs (operands[1], operands[2], |
2263 | &low_dest, &op1_low, &op2_low, | |
2264 | &high_dest, &op1_high, &op2_high); | |
2195867f | 2265 | |
30c46053 MC |
2266 | if (op2_low == const0_rtx) |
2267 | { | |
2268 | low_dest = op1_low; | |
2269 | emit_insn (gen_adddi3_compareV (high_dest, op1_high, | |
2270 | force_reg (DImode, op2_high))); | |
2271 | } | |
2272 | else | |
2273 | { | |
2274 | emit_insn (gen_adddi3_compareC (low_dest, op1_low, | |
2275 | force_reg (DImode, op2_low))); | |
2276 | emit_insn (gen_adddi3_carryinV (high_dest, op1_high, | |
2277 | force_reg (DImode, op2_high))); | |
2278 | } | |
2279 | ||
2280 | emit_move_insn (gen_lowpart (DImode, operands[0]), low_dest); | |
2281 | emit_move_insn (gen_highpart (DImode, operands[0]), high_dest); | |
2282 | ||
2283 | aarch64_gen_unlikely_cbranch (NE, CC_Vmode, operands[3]); | |
2195867f RH |
2284 | DONE; |
2285 | }) | |
2286 | ||
30c46053 | 2287 | (define_expand "uaddvti4" |
1bbffb87 DZ |
2288 | [(match_operand:TI 0 "register_operand") |
2289 | (match_operand:TI 1 "register_operand") | |
2290 | (match_operand:TI 2 "aarch64_reg_or_imm") | |
30c46053 MC |
2291 | (label_ref (match_operand 3 "" ""))] |
2292 | "" | |
2293 | { | |
2294 | rtx low_dest, op1_low, op2_low, high_dest, op1_high, op2_high; | |
2295 | ||
2296 | aarch64_addti_scratch_regs (operands[1], operands[2], | |
2297 | &low_dest, &op1_low, &op2_low, | |
2298 | &high_dest, &op1_high, &op2_high); | |
2299 | ||
2300 | if (op2_low == const0_rtx) | |
2301 | { | |
2302 | low_dest = op1_low; | |
2303 | emit_insn (gen_adddi3_compareC (high_dest, op1_high, | |
2304 | force_reg (DImode, op2_high))); | |
2305 | } | |
2306 | else | |
2307 | { | |
2308 | emit_insn (gen_adddi3_compareC (low_dest, op1_low, | |
2309 | force_reg (DImode, op2_low))); | |
2310 | emit_insn (gen_adddi3_carryinC (high_dest, op1_high, | |
2311 | force_reg (DImode, op2_high))); | |
2312 | } | |
2313 | ||
2314 | emit_move_insn (gen_lowpart (DImode, operands[0]), low_dest); | |
2315 | emit_move_insn (gen_highpart (DImode, operands[0]), high_dest); | |
2316 | ||
f7343f20 | 2317 | aarch64_gen_unlikely_cbranch (GEU, CC_ADCmode, operands[3]); |
30c46053 MC |
2318 | DONE; |
2319 | }) | |
2320 | ||
2195867f | 2321 | (define_insn "add<mode>3_compare0" |
43e9d192 IB |
2322 | [(set (reg:CC_NZ CC_REGNUM) |
2323 | (compare:CC_NZ | |
341fca1b | 2324 | (plus:GPI (match_operand:GPI 1 "register_operand" "%rk,rk,rk") |
34b512f5 | 2325 | (match_operand:GPI 2 "aarch64_plus_operand" "r,I,J")) |
43e9d192 | 2326 | (const_int 0))) |
34b512f5 | 2327 | (set (match_operand:GPI 0 "register_operand" "=r,r,r") |
43e9d192 IB |
2328 | (plus:GPI (match_dup 1) (match_dup 2)))] |
2329 | "" | |
2330 | "@ | |
2331 | adds\\t%<w>0, %<w>1, %<w>2 | |
e2b691c4 | 2332 | adds\\t%<w>0, %<w>1, %2 |
43e9d192 | 2333 | subs\\t%<w>0, %<w>1, #%n2" |
1d61feeb | 2334 | [(set_attr "type" "alus_sreg,alus_imm,alus_imm")] |
43e9d192 IB |
2335 | ) |
2336 | ||
3b585938 IB |
2337 | ;; zero_extend version of above |
2338 | (define_insn "*addsi3_compare0_uxtw" | |
2339 | [(set (reg:CC_NZ CC_REGNUM) | |
2340 | (compare:CC_NZ | |
341fca1b | 2341 | (plus:SI (match_operand:SI 1 "register_operand" "%rk,rk,rk") |
34b512f5 | 2342 | (match_operand:SI 2 "aarch64_plus_operand" "r,I,J")) |
3b585938 | 2343 | (const_int 0))) |
34b512f5 | 2344 | (set (match_operand:DI 0 "register_operand" "=r,r,r") |
3b585938 IB |
2345 | (zero_extend:DI (plus:SI (match_dup 1) (match_dup 2))))] |
2346 | "" | |
2347 | "@ | |
2348 | adds\\t%w0, %w1, %w2 | |
e2b691c4 | 2349 | adds\\t%w0, %w1, %2 |
3b585938 | 2350 | subs\\t%w0, %w1, #%n2" |
1d61feeb | 2351 | [(set_attr "type" "alus_sreg,alus_imm,alus_imm")] |
3b585938 IB |
2352 | ) |
2353 | ||
ef22810a RH |
2354 | (define_insn "*add<mode>3_compareC_cconly" |
2355 | [(set (reg:CC_C CC_REGNUM) | |
f7343f20 RE |
2356 | (compare:CC_C |
2357 | (plus:GPI | |
2358 | (match_operand:GPI 0 "register_operand" "r,r,r") | |
2359 | (match_operand:GPI 1 "aarch64_plus_operand" "r,I,J")) | |
2360 | (match_dup 0)))] | |
ef22810a | 2361 | "" |
ef22810a | 2362 | "@ |
f7343f20 RE |
2363 | cmn\\t%<w>0, %<w>1 |
2364 | cmn\\t%<w>0, %1 | |
2365 | cmp\\t%<w>0, #%n1" | |
2366 | [(set_attr "type" "alus_sreg,alus_imm,alus_imm")] | |
ef22810a | 2367 | ) |
954224d7 | 2368 | |
ef22810a RH |
2369 | (define_insn "add<mode>3_compareC" |
2370 | [(set (reg:CC_C CC_REGNUM) | |
30c46053 | 2371 | (compare:CC_C |
f7343f20 | 2372 | (plus:GPI |
341fca1b | 2373 | (match_operand:GPI 1 "register_operand" "rk,rk,rk") |
f7343f20 RE |
2374 | (match_operand:GPI 2 "aarch64_plus_operand" "r,I,J")) |
2375 | (match_dup 1))) | |
2376 | (set (match_operand:GPI 0 "register_operand" "=r,r,r") | |
ef22810a RH |
2377 | (plus:GPI (match_dup 1) (match_dup 2)))] |
2378 | "" | |
f7343f20 RE |
2379 | "@ |
2380 | adds\\t%<w>0, %<w>1, %<w>2 | |
2381 | adds\\t%<w>0, %<w>1, %2 | |
2382 | subs\\t%<w>0, %<w>1, #%n2" | |
2383 | [(set_attr "type" "alus_sreg,alus_imm,alus_imm")] | |
ef22810a RH |
2384 | ) |
2385 | ||
30c46053 MC |
2386 | (define_insn "*add<mode>3_compareV_cconly_imm" |
2387 | [(set (reg:CC_V CC_REGNUM) | |
2388 | (compare:CC_V | |
2389 | (plus:<DWI> | |
2390 | (sign_extend:<DWI> (match_operand:GPI 0 "register_operand" "r,r")) | |
2391 | (match_operand:<DWI> 1 "const_scalar_int_operand" "")) | |
2392 | (sign_extend:<DWI> | |
2393 | (plus:GPI | |
2394 | (match_dup 0) | |
2395 | (match_operand:GPI 2 "aarch64_plus_immediate" "I,J")))))] | |
2396 | "INTVAL (operands[1]) == INTVAL (operands[2])" | |
2397 | "@ | |
2398 | cmn\\t%<w>0, %<w>1 | |
2399 | cmp\\t%<w>0, #%n1" | |
2400 | [(set_attr "type" "alus_imm")] | |
2401 | ) | |
2402 | ||
2403 | (define_insn "*add<mode>3_compareV_cconly" | |
2404 | [(set (reg:CC_V CC_REGNUM) | |
2405 | (compare:CC_V | |
2406 | (plus:<DWI> | |
2407 | (sign_extend:<DWI> (match_operand:GPI 0 "register_operand" "r")) | |
2408 | (sign_extend:<DWI> (match_operand:GPI 1 "register_operand" "r"))) | |
2409 | (sign_extend:<DWI> (plus:GPI (match_dup 0) (match_dup 1)))))] | |
2410 | "" | |
2411 | "cmn\\t%<w>0, %<w>1" | |
2412 | [(set_attr "type" "alus_sreg")] | |
2413 | ) | |
2414 | ||
a58fe3c5 | 2415 | (define_insn "add<mode>3_compareV_imm" |
30c46053 MC |
2416 | [(set (reg:CC_V CC_REGNUM) |
2417 | (compare:CC_V | |
2418 | (plus:<DWI> | |
2419 | (sign_extend:<DWI> | |
341fca1b | 2420 | (match_operand:GPI 1 "register_operand" "rk,rk")) |
30c46053 MC |
2421 | (match_operand:GPI 2 "aarch64_plus_immediate" "I,J")) |
2422 | (sign_extend:<DWI> | |
2423 | (plus:GPI (match_dup 1) (match_dup 2))))) | |
2424 | (set (match_operand:GPI 0 "register_operand" "=r,r") | |
2425 | (plus:GPI (match_dup 1) (match_dup 2)))] | |
2426 | "" | |
2427 | "@ | |
2428 | adds\\t%<w>0, %<w>1, %<w>2 | |
2429 | subs\\t%<w>0, %<w>1, #%n2" | |
2430 | [(set_attr "type" "alus_imm,alus_imm")] | |
2431 | ) | |
2432 | ||
2433 | (define_insn "add<mode>3_compareV" | |
2434 | [(set (reg:CC_V CC_REGNUM) | |
2435 | (compare:CC_V | |
2436 | (plus:<DWI> | |
341fca1b | 2437 | (sign_extend:<DWI> (match_operand:GPI 1 "register_operand" "rk")) |
30c46053 MC |
2438 | (sign_extend:<DWI> (match_operand:GPI 2 "register_operand" "r"))) |
2439 | (sign_extend:<DWI> (plus:GPI (match_dup 1) (match_dup 2))))) | |
2440 | (set (match_operand:GPI 0 "register_operand" "=r") | |
2441 | (plus:GPI (match_dup 1) (match_dup 2)))] | |
2442 | "" | |
2443 | "adds\\t%<w>0, %<w>1, %<w>2" | |
2444 | [(set_attr "type" "alus_sreg")] | |
2445 | ) | |
2446 | ||
dd1bdce4 VK |
2447 | (define_insn "*adds_shift_imm_<mode>" |
2448 | [(set (reg:CC_NZ CC_REGNUM) | |
2449 | (compare:CC_NZ | |
30c46053 | 2450 | (plus:GPI (ASHIFT:GPI |
dd1bdce4 VK |
2451 | (match_operand:GPI 1 "register_operand" "r") |
2452 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n")) | |
2453 | (match_operand:GPI 3 "register_operand" "r")) | |
2454 | (const_int 0))) | |
2455 | (set (match_operand:GPI 0 "register_operand" "=r") | |
2456 | (plus:GPI (ASHIFT:GPI (match_dup 1) (match_dup 2)) | |
2457 | (match_dup 3)))] | |
2458 | "" | |
2459 | "adds\\t%<w>0, %<w>3, %<w>1, <shift> %2" | |
2460 | [(set_attr "type" "alus_shift_imm")] | |
2461 | ) | |
2462 | ||
2463 | (define_insn "*subs_shift_imm_<mode>" | |
2464 | [(set (reg:CC_NZ CC_REGNUM) | |
2465 | (compare:CC_NZ | |
2466 | (minus:GPI (match_operand:GPI 1 "register_operand" "r") | |
2467 | (ASHIFT:GPI | |
2468 | (match_operand:GPI 2 "register_operand" "r") | |
2469 | (match_operand:QI 3 "aarch64_shift_imm_<mode>" "n"))) | |
2470 | (const_int 0))) | |
2471 | (set (match_operand:GPI 0 "register_operand" "=r") | |
2472 | (minus:GPI (match_dup 1) | |
2473 | (ASHIFT:GPI (match_dup 2) (match_dup 3))))] | |
2474 | "" | |
2475 | "subs\\t%<w>0, %<w>1, %<w>2, <shift> %3" | |
2476 | [(set_attr "type" "alus_shift_imm")] | |
2477 | ) | |
2478 | ||
f3d096b0 N |
2479 | (define_insn "*adds_<optab><ALLX:mode>_<GPI:mode>" |
2480 | [(set (reg:CC_NZ CC_REGNUM) | |
2481 | (compare:CC_NZ | |
2482 | (plus:GPI | |
2483 | (ANY_EXTEND:GPI (match_operand:ALLX 1 "register_operand" "r")) | |
341fca1b | 2484 | (match_operand:GPI 2 "register_operand" "rk")) |
f3d096b0 N |
2485 | (const_int 0))) |
2486 | (set (match_operand:GPI 0 "register_operand" "=r") | |
2487 | (plus:GPI (ANY_EXTEND:GPI (match_dup 1)) (match_dup 2)))] | |
2488 | "" | |
d4febc75 | 2489 | "adds\\t%<GPI:w>0, %<GPI:w>2, %w1, <su>xt<ALLX:size>" |
99e6af15 | 2490 | [(set_attr "type" "alus_ext")] |
f3d096b0 N |
2491 | ) |
2492 | ||
2493 | (define_insn "*subs_<optab><ALLX:mode>_<GPI:mode>" | |
2494 | [(set (reg:CC_NZ CC_REGNUM) | |
2495 | (compare:CC_NZ | |
341fca1b | 2496 | (minus:GPI (match_operand:GPI 1 "register_operand" "rk") |
f3d096b0 N |
2497 | (ANY_EXTEND:GPI |
2498 | (match_operand:ALLX 2 "register_operand" "r"))) | |
2499 | (const_int 0))) | |
2500 | (set (match_operand:GPI 0 "register_operand" "=r") | |
2501 | (minus:GPI (match_dup 1) (ANY_EXTEND:GPI (match_dup 2))))] | |
2502 | "" | |
d4febc75 | 2503 | "subs\\t%<GPI:w>0, %<GPI:w>1, %w2, <su>xt<ALLX:size>" |
99e6af15 | 2504 | [(set_attr "type" "alus_ext")] |
f3d096b0 N |
2505 | ) |
2506 | ||
dd1bdce4 VK |
2507 | (define_insn "*adds_<optab><ALLX:mode>_shift_<GPI:mode>" |
2508 | [(set (reg:CC_NZ CC_REGNUM) | |
2509 | (compare:CC_NZ | |
2510 | (plus:GPI (ashift:GPI | |
2511 | (ANY_EXTEND:GPI | |
2512 | (match_operand:ALLX 1 "register_operand" "r")) | |
2513 | (match_operand 2 "aarch64_imm3" "Ui3")) | |
341fca1b | 2514 | (match_operand:GPI 3 "register_operand" "rk")) |
dd1bdce4 VK |
2515 | (const_int 0))) |
2516 | (set (match_operand:GPI 0 "register_operand" "=rk") | |
2517 | (plus:GPI (ashift:GPI (ANY_EXTEND:GPI (match_dup 1)) | |
2518 | (match_dup 2)) | |
2519 | (match_dup 3)))] | |
2520 | "" | |
d4febc75 | 2521 | "adds\\t%<GPI:w>0, %<GPI:w>3, %w1, <su>xt<ALLX:size> %2" |
dd1bdce4 VK |
2522 | [(set_attr "type" "alus_ext")] |
2523 | ) | |
2524 | ||
2525 | (define_insn "*subs_<optab><ALLX:mode>_shift_<GPI:mode>" | |
2526 | [(set (reg:CC_NZ CC_REGNUM) | |
2527 | (compare:CC_NZ | |
341fca1b | 2528 | (minus:GPI (match_operand:GPI 1 "register_operand" "rk") |
dd1bdce4 VK |
2529 | (ashift:GPI |
2530 | (ANY_EXTEND:GPI | |
2531 | (match_operand:ALLX 2 "register_operand" "r")) | |
2532 | (match_operand 3 "aarch64_imm3" "Ui3"))) | |
2533 | (const_int 0))) | |
2534 | (set (match_operand:GPI 0 "register_operand" "=rk") | |
2535 | (minus:GPI (match_dup 1) | |
2536 | (ashift:GPI (ANY_EXTEND:GPI (match_dup 2)) | |
2537 | (match_dup 3))))] | |
2538 | "" | |
d4febc75 | 2539 | "subs\\t%<GPI:w>0, %<GPI:w>1, %w2, <su>xt<ALLX:size> %3" |
dd1bdce4 VK |
2540 | [(set_attr "type" "alus_ext")] |
2541 | ) | |
2542 | ||
43e9d192 IB |
2543 | (define_insn "*add<mode>3nr_compare0" |
2544 | [(set (reg:CC_NZ CC_REGNUM) | |
2545 | (compare:CC_NZ | |
34b512f5 SN |
2546 | (plus:GPI (match_operand:GPI 0 "register_operand" "%r,r,r") |
2547 | (match_operand:GPI 1 "aarch64_plus_operand" "r,I,J")) | |
43e9d192 IB |
2548 | (const_int 0)))] |
2549 | "" | |
2550 | "@ | |
2551 | cmn\\t%<w>0, %<w>1 | |
e2b691c4 | 2552 | cmn\\t%<w>0, %1 |
43e9d192 | 2553 | cmp\\t%<w>0, #%n1" |
1d61feeb | 2554 | [(set_attr "type" "alus_sreg,alus_imm,alus_imm")] |
43e9d192 IB |
2555 | ) |
2556 | ||
2f65ab2e MC |
2557 | (define_insn "aarch64_sub<mode>_compare0" |
2558 | [(set (reg:CC_NZ CC_REGNUM) | |
2559 | (compare:CC_NZ | |
2560 | (minus:GPI (match_operand:GPI 0 "register_operand" "r") | |
2561 | (match_operand:GPI 1 "aarch64_plus_operand" "r")) | |
2562 | (const_int 0)))] | |
2563 | "" | |
2564 | "cmp\\t%<w>0, %<w>1" | |
2565 | [(set_attr "type" "alus_sreg")] | |
2566 | ) | |
2567 | ||
f5313c63 | 2568 | (define_insn "*compare_neg<mode>" |
1c992d1e RE |
2569 | [(set (reg:CC_Z CC_REGNUM) |
2570 | (compare:CC_Z | |
274b2532 KT |
2571 | (neg:GPI (match_operand:GPI 0 "register_operand" "r")) |
2572 | (match_operand:GPI 1 "register_operand" "r")))] | |
f5313c63 | 2573 | "" |
274b2532 | 2574 | "cmn\\t%<w>1, %<w>0" |
1d61feeb | 2575 | [(set_attr "type" "alus_sreg")] |
f5313c63 IB |
2576 | ) |
2577 | ||
43e9d192 | 2578 | (define_insn "*add_<shift>_<mode>" |
24c56925 | 2579 | [(set (match_operand:GPI 0 "register_operand" "=r") |
43e9d192 IB |
2580 | (plus:GPI (ASHIFT:GPI (match_operand:GPI 1 "register_operand" "r") |
2581 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n")) | |
2582 | (match_operand:GPI 3 "register_operand" "r")))] | |
2583 | "" | |
2584 | "add\\t%<w>0, %<w>3, %<w>1, <shift> %2" | |
ae27ce51 | 2585 | [(set_attr "autodetect_type" "alu_shift_<shift>_op2")] |
43e9d192 IB |
2586 | ) |
2587 | ||
3b585938 IB |
2588 | ;; zero_extend version of above |
2589 | (define_insn "*add_<shift>_si_uxtw" | |
24c56925 | 2590 | [(set (match_operand:DI 0 "register_operand" "=r") |
3b585938 IB |
2591 | (zero_extend:DI |
2592 | (plus:SI (ASHIFT:SI (match_operand:SI 1 "register_operand" "r") | |
2a9704d0 MS |
2593 | (match_operand:QI 2 "aarch64_shift_imm_si" "n")) |
2594 | (match_operand:SI 3 "register_operand" "r"))))] | |
3b585938 IB |
2595 | "" |
2596 | "add\\t%w0, %w3, %w1, <shift> %2" | |
ae27ce51 | 2597 | [(set_attr "autodetect_type" "alu_shift_<shift>_op2")] |
3b585938 IB |
2598 | ) |
2599 | ||
43e9d192 IB |
2600 | (define_insn "*add_<optab><ALLX:mode>_<GPI:mode>" |
2601 | [(set (match_operand:GPI 0 "register_operand" "=rk") | |
2602 | (plus:GPI (ANY_EXTEND:GPI (match_operand:ALLX 1 "register_operand" "r")) | |
2603 | (match_operand:GPI 2 "register_operand" "r")))] | |
2604 | "" | |
d4febc75 | 2605 | "add\\t%<GPI:w>0, %<GPI:w>2, %w1, <su>xt<ALLX:size>" |
99e6af15 | 2606 | [(set_attr "type" "alu_ext")] |
43e9d192 IB |
2607 | ) |
2608 | ||
3b585938 IB |
2609 | ;; zero_extend version of above |
2610 | (define_insn "*add_<optab><SHORT:mode>_si_uxtw" | |
2611 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
2612 | (zero_extend:DI | |
2613 | (plus:SI (ANY_EXTEND:SI (match_operand:SHORT 1 "register_operand" "r")) | |
2614 | (match_operand:GPI 2 "register_operand" "r"))))] | |
2615 | "" | |
2616 | "add\\t%w0, %w2, %w1, <su>xt<SHORT:size>" | |
99e6af15 | 2617 | [(set_attr "type" "alu_ext")] |
3b585938 IB |
2618 | ) |
2619 | ||
43e9d192 IB |
2620 | (define_insn "*add_<optab><ALLX:mode>_shft_<GPI:mode>" |
2621 | [(set (match_operand:GPI 0 "register_operand" "=rk") | |
2622 | (plus:GPI (ashift:GPI (ANY_EXTEND:GPI | |
2623 | (match_operand:ALLX 1 "register_operand" "r")) | |
2624 | (match_operand 2 "aarch64_imm3" "Ui3")) | |
2625 | (match_operand:GPI 3 "register_operand" "r")))] | |
2626 | "" | |
d4febc75 | 2627 | "add\\t%<GPI:w>0, %<GPI:w>3, %w1, <su>xt<ALLX:size> %2" |
99e6af15 | 2628 | [(set_attr "type" "alu_ext")] |
43e9d192 IB |
2629 | ) |
2630 | ||
3b585938 IB |
2631 | ;; zero_extend version of above |
2632 | (define_insn "*add_<optab><SHORT:mode>_shft_si_uxtw" | |
2633 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
2634 | (zero_extend:DI | |
2635 | (plus:SI (ashift:SI (ANY_EXTEND:SI | |
2636 | (match_operand:SHORT 1 "register_operand" "r")) | |
2637 | (match_operand 2 "aarch64_imm3" "Ui3")) | |
2638 | (match_operand:SI 3 "register_operand" "r"))))] | |
2639 | "" | |
2640 | "add\\t%w0, %w3, %w1, <su>xt<SHORT:size> %2" | |
99e6af15 | 2641 | [(set_attr "type" "alu_ext")] |
3b585938 IB |
2642 | ) |
2643 | ||
ef22810a RH |
2644 | (define_expand "add<mode>3_carryin" |
2645 | [(set (match_operand:GPI 0 "register_operand") | |
2646 | (plus:GPI | |
2647 | (plus:GPI | |
f7343f20 | 2648 | (ltu:GPI (reg:CC_C CC_REGNUM) (const_int 0)) |
ef22810a RH |
2649 | (match_operand:GPI 1 "aarch64_reg_or_zero")) |
2650 | (match_operand:GPI 2 "aarch64_reg_or_zero")))] | |
3b585938 | 2651 | "" |
43e9d192 | 2652 | "" |
43e9d192 IB |
2653 | ) |
2654 | ||
ef22810a RH |
2655 | ;; Note that add with carry with two zero inputs is matched by cset, |
2656 | ;; and that add with carry with one zero input is matched by cinc. | |
3b585938 | 2657 | |
ef22810a RH |
2658 | (define_insn "*add<mode>3_carryin" |
2659 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
2660 | (plus:GPI | |
2661 | (plus:GPI | |
2662 | (match_operand:GPI 3 "aarch64_carry_operation" "") | |
2663 | (match_operand:GPI 1 "register_operand" "r")) | |
2664 | (match_operand:GPI 2 "register_operand" "r")))] | |
43e9d192 IB |
2665 | "" |
2666 | "adc\\t%<w>0, %<w>1, %<w>2" | |
99e6af15 | 2667 | [(set_attr "type" "adc_reg")] |
43e9d192 IB |
2668 | ) |
2669 | ||
3b585938 | 2670 | ;; zero_extend version of above |
ef22810a RH |
2671 | (define_insn "*addsi3_carryin_uxtw" |
2672 | [(set (match_operand:DI 0 "register_operand" "=r") | |
2673 | (zero_extend:DI | |
2674 | (plus:SI | |
2675 | (plus:SI | |
2676 | (match_operand:SI 3 "aarch64_carry_operation" "") | |
2677 | (match_operand:SI 1 "register_operand" "r")) | |
2678 | (match_operand:SI 2 "register_operand" "r"))))] | |
3b585938 IB |
2679 | "" |
2680 | "adc\\t%w0, %w1, %w2" | |
99e6af15 | 2681 | [(set_attr "type" "adc_reg")] |
3b585938 IB |
2682 | ) |
2683 | ||
30c46053 MC |
2684 | (define_expand "add<mode>3_carryinC" |
2685 | [(parallel | |
2686 | [(set (match_dup 3) | |
f7343f20 | 2687 | (compare:CC_ADC |
30c46053 MC |
2688 | (plus:<DWI> |
2689 | (plus:<DWI> | |
2690 | (match_dup 4) | |
2691 | (zero_extend:<DWI> | |
1bbffb87 | 2692 | (match_operand:GPI 1 "register_operand"))) |
30c46053 | 2693 | (zero_extend:<DWI> |
1bbffb87 | 2694 | (match_operand:GPI 2 "register_operand"))) |
f7343f20 | 2695 | (match_dup 6))) |
30c46053 MC |
2696 | (set (match_operand:GPI 0 "register_operand") |
2697 | (plus:GPI | |
2698 | (plus:GPI (match_dup 5) (match_dup 1)) | |
2699 | (match_dup 2)))])] | |
2700 | "" | |
2701 | { | |
f7343f20 RE |
2702 | operands[3] = gen_rtx_REG (CC_ADCmode, CC_REGNUM); |
2703 | rtx ccin = gen_rtx_REG (CC_Cmode, CC_REGNUM); | |
2704 | operands[4] = gen_rtx_LTU (<DWI>mode, ccin, const0_rtx); | |
2705 | operands[5] = gen_rtx_LTU (<MODE>mode, ccin, const0_rtx); | |
2706 | operands[6] = immed_wide_int_const (wi::shwi (1, <DWI>mode) | |
2707 | << GET_MODE_BITSIZE (<MODE>mode), | |
2708 | TImode); | |
30c46053 MC |
2709 | }) |
2710 | ||
2711 | (define_insn "*add<mode>3_carryinC_zero" | |
f7343f20 RE |
2712 | [(set (reg:CC_ADC CC_REGNUM) |
2713 | (compare:CC_ADC | |
30c46053 MC |
2714 | (plus:<DWI> |
2715 | (match_operand:<DWI> 2 "aarch64_carry_operation" "") | |
2716 | (zero_extend:<DWI> (match_operand:GPI 1 "register_operand" "r"))) | |
f7343f20 | 2717 | (match_operand 4 "const_scalar_int_operand" ""))) |
16621f0d | 2718 | (set (match_operand:GPI 0 "register_operand" "=r") |
f7343f20 RE |
2719 | (plus:GPI (match_operand:GPI 3 "aarch64_carry_operation" "") |
2720 | (match_dup 1)))] | |
2721 | "rtx_mode_t (operands[4], <DWI>mode) | |
2722 | == (wi::shwi (1, <DWI>mode) << (unsigned) GET_MODE_BITSIZE (<MODE>mode))" | |
30c46053 MC |
2723 | "adcs\\t%<w>0, %<w>1, <w>zr" |
2724 | [(set_attr "type" "adc_reg")] | |
2725 | ) | |
2726 | ||
2727 | (define_insn "*add<mode>3_carryinC" | |
f7343f20 RE |
2728 | [(set (reg:CC_ADC CC_REGNUM) |
2729 | (compare:CC_ADC | |
30c46053 MC |
2730 | (plus:<DWI> |
2731 | (plus:<DWI> | |
2732 | (match_operand:<DWI> 3 "aarch64_carry_operation" "") | |
2733 | (zero_extend:<DWI> (match_operand:GPI 1 "register_operand" "r"))) | |
2734 | (zero_extend:<DWI> (match_operand:GPI 2 "register_operand" "r"))) | |
f7343f20 | 2735 | (match_operand 5 "const_scalar_int_operand" ""))) |
16621f0d | 2736 | (set (match_operand:GPI 0 "register_operand" "=r") |
30c46053 | 2737 | (plus:GPI |
f7343f20 RE |
2738 | (plus:GPI (match_operand:GPI 4 "aarch64_carry_operation" "") |
2739 | (match_dup 1)) | |
30c46053 | 2740 | (match_dup 2)))] |
f7343f20 RE |
2741 | "rtx_mode_t (operands[5], <DWI>mode) |
2742 | == (wi::shwi (1, <DWI>mode) << (unsigned) GET_MODE_BITSIZE (<MODE>mode))" | |
30c46053 MC |
2743 | "adcs\\t%<w>0, %<w>1, %<w>2" |
2744 | [(set_attr "type" "adc_reg")] | |
2745 | ) | |
2746 | ||
2747 | (define_expand "add<mode>3_carryinV" | |
2748 | [(parallel | |
2749 | [(set (reg:CC_V CC_REGNUM) | |
2750 | (compare:CC_V | |
2751 | (plus:<DWI> | |
2752 | (plus:<DWI> | |
2753 | (match_dup 3) | |
2754 | (sign_extend:<DWI> | |
1bbffb87 | 2755 | (match_operand:GPI 1 "register_operand"))) |
30c46053 | 2756 | (sign_extend:<DWI> |
1bbffb87 | 2757 | (match_operand:GPI 2 "register_operand"))) |
30c46053 MC |
2758 | (sign_extend:<DWI> |
2759 | (plus:GPI | |
2760 | (plus:GPI (match_dup 4) (match_dup 1)) | |
2761 | (match_dup 2))))) | |
2762 | (set (match_operand:GPI 0 "register_operand") | |
2763 | (plus:GPI | |
2764 | (plus:GPI (match_dup 4) (match_dup 1)) | |
2765 | (match_dup 2)))])] | |
2766 | "" | |
2767 | { | |
2768 | rtx cc = gen_rtx_REG (CC_Cmode, CC_REGNUM); | |
f7343f20 RE |
2769 | operands[3] = gen_rtx_LTU (<DWI>mode, cc, const0_rtx); |
2770 | operands[4] = gen_rtx_LTU (<MODE>mode, cc, const0_rtx); | |
30c46053 MC |
2771 | }) |
2772 | ||
2773 | (define_insn "*add<mode>3_carryinV_zero" | |
2774 | [(set (reg:CC_V CC_REGNUM) | |
2775 | (compare:CC_V | |
2776 | (plus:<DWI> | |
2777 | (match_operand:<DWI> 2 "aarch64_carry_operation" "") | |
2778 | (sign_extend:<DWI> (match_operand:GPI 1 "register_operand" "r"))) | |
2779 | (sign_extend:<DWI> | |
2780 | (plus:GPI | |
2781 | (match_operand:GPI 3 "aarch64_carry_operation" "") | |
2782 | (match_dup 1))))) | |
16621f0d | 2783 | (set (match_operand:GPI 0 "register_operand" "=r") |
30c46053 MC |
2784 | (plus:GPI (match_dup 3) (match_dup 1)))] |
2785 | "" | |
2786 | "adcs\\t%<w>0, %<w>1, <w>zr" | |
2787 | [(set_attr "type" "adc_reg")] | |
2788 | ) | |
2789 | ||
2790 | (define_insn "*add<mode>3_carryinV" | |
2791 | [(set (reg:CC_V CC_REGNUM) | |
2792 | (compare:CC_V | |
2793 | (plus:<DWI> | |
2794 | (plus:<DWI> | |
2795 | (match_operand:<DWI> 3 "aarch64_carry_operation" "") | |
2796 | (sign_extend:<DWI> (match_operand:GPI 1 "register_operand" "r"))) | |
2797 | (sign_extend:<DWI> (match_operand:GPI 2 "register_operand" "r"))) | |
2798 | (sign_extend:<DWI> | |
2799 | (plus:GPI | |
2800 | (plus:GPI | |
2801 | (match_operand:GPI 4 "aarch64_carry_operation" "") | |
2802 | (match_dup 1)) | |
2803 | (match_dup 2))))) | |
16621f0d | 2804 | (set (match_operand:GPI 0 "register_operand" "=r") |
30c46053 MC |
2805 | (plus:GPI |
2806 | (plus:GPI (match_dup 4) (match_dup 1)) | |
2807 | (match_dup 2)))] | |
2808 | "" | |
2809 | "adcs\\t%<w>0, %<w>1, %<w>2" | |
2810 | [(set_attr "type" "adc_reg")] | |
2811 | ) | |
2812 | ||
dd1bdce4 VK |
2813 | (define_insn "*add_uxt<mode>_shift2" |
2814 | [(set (match_operand:GPI 0 "register_operand" "=rk") | |
2815 | (plus:GPI (and:GPI | |
2816 | (ashift:GPI (match_operand:GPI 1 "register_operand" "r") | |
2817 | (match_operand 2 "aarch64_imm3" "Ui3")) | |
2818 | (match_operand 3 "const_int_operand" "n")) | |
2819 | (match_operand:GPI 4 "register_operand" "r")))] | |
2820 | "aarch64_uxt_size (INTVAL (operands[2]), INTVAL (operands[3])) != 0" | |
2821 | "* | |
2822 | operands[3] = GEN_INT (aarch64_uxt_size (INTVAL(operands[2]), | |
2823 | INTVAL (operands[3]))); | |
d4febc75 | 2824 | return \"add\t%<w>0, %<w>4, %w1, uxt%e3 %2\";" |
dd1bdce4 VK |
2825 | [(set_attr "type" "alu_ext")] |
2826 | ) | |
2827 | ||
2828 | ;; zero_extend version of above | |
2829 | (define_insn "*add_uxtsi_shift2_uxtw" | |
2830 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
2831 | (zero_extend:DI | |
2832 | (plus:SI (and:SI | |
2833 | (ashift:SI (match_operand:SI 1 "register_operand" "r") | |
2834 | (match_operand 2 "aarch64_imm3" "Ui3")) | |
2835 | (match_operand 3 "const_int_operand" "n")) | |
2836 | (match_operand:SI 4 "register_operand" "r"))))] | |
2837 | "aarch64_uxt_size (INTVAL (operands[2]), INTVAL (operands[3])) != 0" | |
2838 | "* | |
2839 | operands[3] = GEN_INT (aarch64_uxt_size (INTVAL (operands[2]), | |
2840 | INTVAL (operands[3]))); | |
2841 | return \"add\t%w0, %w4, %w1, uxt%e3 %2\";" | |
2842 | [(set_attr "type" "alu_ext")] | |
2843 | ) | |
2844 | ||
43e9d192 IB |
2845 | (define_insn "subsi3" |
2846 | [(set (match_operand:SI 0 "register_operand" "=rk") | |
7eb8fb77 RS |
2847 | (minus:SI (match_operand:SI 1 "register_operand" "rk") |
2848 | (match_operand:SI 2 "register_operand" "r")))] | |
43e9d192 IB |
2849 | "" |
2850 | "sub\\t%w0, %w1, %w2" | |
1d61feeb | 2851 | [(set_attr "type" "alu_sreg")] |
43e9d192 IB |
2852 | ) |
2853 | ||
3b585938 IB |
2854 | ;; zero_extend version of above |
2855 | (define_insn "*subsi3_uxtw" | |
2856 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
2857 | (zero_extend:DI | |
7eb8fb77 | 2858 | (minus:SI (match_operand:SI 1 "register_operand" "rk") |
3b585938 IB |
2859 | (match_operand:SI 2 "register_operand" "r"))))] |
2860 | "" | |
2861 | "sub\\t%w0, %w1, %w2" | |
1d61feeb | 2862 | [(set_attr "type" "alu_sreg")] |
3b585938 IB |
2863 | ) |
2864 | ||
43e9d192 | 2865 | (define_insn "subdi3" |
4f2962fd | 2866 | [(set (match_operand:DI 0 "register_operand" "=rk,w") |
7eb8fb77 RS |
2867 | (minus:DI (match_operand:DI 1 "register_operand" "rk,w") |
2868 | (match_operand:DI 2 "register_operand" "r,w")))] | |
43e9d192 IB |
2869 | "" |
2870 | "@ | |
2871 | sub\\t%x0, %x1, %x2 | |
2872 | sub\\t%d0, %d1, %d2" | |
1d61feeb | 2873 | [(set_attr "type" "alu_sreg, neon_sub") |
488461d8 | 2874 | (set_attr "arch" "*,simd")] |
43e9d192 IB |
2875 | ) |
2876 | ||
a58fe3c5 | 2877 | (define_expand "subv<GPI:mode>4" |
30c46053 | 2878 | [(match_operand:GPI 0 "register_operand") |
a58fe3c5 RE |
2879 | (match_operand:GPI 1 "register_operand") |
2880 | (match_operand:GPI 2 "aarch64_plus_operand") | |
30c46053 MC |
2881 | (label_ref (match_operand 3 "" ""))] |
2882 | "" | |
2883 | { | |
a58fe3c5 RE |
2884 | if (CONST_INT_P (operands[2])) |
2885 | emit_insn (gen_subv<mode>_imm (operands[0], operands[1], operands[2])); | |
2886 | else | |
2887 | emit_insn (gen_subv<mode>_insn (operands[0], operands[1], operands[2])); | |
30c46053 MC |
2888 | aarch64_gen_unlikely_cbranch (NE, CC_Vmode, operands[3]); |
2889 | ||
2890 | DONE; | |
2891 | }) | |
2892 | ||
a58fe3c5 RE |
2893 | (define_insn "subv<GPI:mode>_insn" |
2894 | [(set (reg:CC_V CC_REGNUM) | |
2895 | (compare:CC_V | |
2896 | (sign_extend:<DWI> | |
2897 | (minus:GPI | |
2898 | (match_operand:GPI 1 "register_operand" "rk") | |
2899 | (match_operand:GPI 2 "register_operand" "r"))) | |
2900 | (minus:<DWI> (sign_extend:<DWI> (match_dup 1)) | |
2901 | (sign_extend:<DWI> (match_dup 2))))) | |
2902 | (set (match_operand:GPI 0 "register_operand" "=r") | |
2903 | (minus:GPI (match_dup 1) (match_dup 2)))] | |
2904 | "" | |
2905 | "subs\\t%<w>0, %<w>1, %<w>2" | |
2906 | [(set_attr "type" "alus_sreg")] | |
2907 | ) | |
2908 | ||
2909 | (define_insn "subv<GPI:mode>_imm" | |
2910 | [(set (reg:CC_V CC_REGNUM) | |
2911 | (compare:CC_V | |
2912 | (sign_extend:<DWI> | |
2913 | (minus:GPI | |
2914 | (match_operand:GPI 1 "register_operand" "rk,rk") | |
2915 | (match_operand:GPI 2 "aarch64_plus_immediate" "I,J"))) | |
2916 | (minus:<DWI> (sign_extend:<DWI> (match_dup 1)) | |
2917 | (match_dup 2)))) | |
2918 | (set (match_operand:GPI 0 "register_operand" "=r,r") | |
2919 | (minus:GPI (match_dup 1) (match_dup 2)))] | |
2920 | "" | |
2921 | "@ | |
2922 | subs\\t%<w>0, %<w>1, %2 | |
2923 | adds\\t%<w>0, %<w>1, #%n2" | |
2924 | [(set_attr "type" "alus_sreg")] | |
2925 | ) | |
2926 | ||
2927 | (define_expand "negv<GPI:mode>3" | |
2928 | [(match_operand:GPI 0 "register_operand") | |
2929 | (match_operand:GPI 1 "register_operand") | |
2930 | (label_ref (match_operand 2 "" ""))] | |
2931 | "" | |
2932 | { | |
2933 | emit_insn (gen_negv<mode>_insn (operands[0], operands[1])); | |
2934 | aarch64_gen_unlikely_cbranch (NE, CC_Vmode, operands[2]); | |
2935 | ||
2936 | DONE; | |
2937 | } | |
2938 | ) | |
2939 | ||
2940 | (define_insn "negv<GPI:mode>_insn" | |
2941 | [(set (reg:CC_V CC_REGNUM) | |
2942 | (compare:CC_V | |
2943 | (sign_extend:<DWI> | |
2944 | (neg:GPI (match_operand:GPI 1 "register_operand" "r"))) | |
2945 | (neg:<DWI> (sign_extend:<DWI> (match_dup 1))))) | |
2946 | (set (match_operand:GPI 0 "register_operand" "=r") | |
2947 | (neg:GPI (match_dup 1)))] | |
2948 | "" | |
2949 | "negs\\t%<w>0, %<w>1" | |
2950 | [(set_attr "type" "alus_sreg")] | |
2951 | ) | |
2952 | ||
2953 | (define_insn "negv<GPI:mode>_cmp_only" | |
2954 | [(set (reg:CC_V CC_REGNUM) | |
2955 | (compare:CC_V | |
2956 | (sign_extend:<DWI> | |
2957 | (neg:GPI (match_operand:GPI 0 "register_operand" "r"))) | |
2958 | (neg:<DWI> (sign_extend:<DWI> (match_dup 0)))))] | |
2959 | "" | |
2960 | "negs\\t%<w>zr, %<w>0" | |
2961 | [(set_attr "type" "alus_sreg")] | |
2962 | ) | |
2963 | ||
2964 | (define_insn "*cmpv<GPI:mode>_insn" | |
2965 | [(set (reg:CC_V CC_REGNUM) | |
2966 | (compare:CC_V | |
2967 | (sign_extend:<DWI> | |
2968 | (minus:GPI (match_operand:GPI 0 "register_operand" "r,r,r") | |
2969 | (match_operand:GPI 1 "aarch64_plus_operand" "r,I,J"))) | |
2970 | (minus:<DWI> (sign_extend:<DWI> (match_dup 0)) | |
2971 | (sign_extend:<DWI> (match_dup 1)))))] | |
2972 | "" | |
2973 | "@ | |
2974 | cmp\\t%<w>0, %<w>1 | |
2975 | cmp\\t%<w>0, %1 | |
2976 | cmp\\t%<w>0, #%n1" | |
2977 | [(set_attr "type" "alus_sreg")] | |
2978 | ) | |
2979 | ||
30c46053 MC |
2980 | (define_expand "usubv<mode>4" |
2981 | [(match_operand:GPI 0 "register_operand") | |
2982 | (match_operand:GPI 1 "aarch64_reg_or_zero") | |
2983 | (match_operand:GPI 2 "aarch64_reg_or_zero") | |
2984 | (label_ref (match_operand 3 "" ""))] | |
2985 | "" | |
2986 | { | |
2987 | emit_insn (gen_sub<mode>3_compare1 (operands[0], operands[1], operands[2])); | |
2988 | aarch64_gen_unlikely_cbranch (LTU, CCmode, operands[3]); | |
2989 | ||
2990 | DONE; | |
2991 | }) | |
2992 | ||
2195867f | 2993 | (define_expand "subti3" |
1bbffb87 DZ |
2994 | [(set (match_operand:TI 0 "register_operand") |
2995 | (minus:TI (match_operand:TI 1 "aarch64_reg_or_zero") | |
2996 | (match_operand:TI 2 "register_operand")))] | |
2195867f RH |
2997 | "" |
2998 | { | |
30c46053 MC |
2999 | rtx low_dest, op1_low, op2_low, high_dest, op1_high, op2_high; |
3000 | ||
3001 | aarch64_subvti_scratch_regs (operands[1], operands[2], | |
3002 | &low_dest, &op1_low, &op2_low, | |
3003 | &high_dest, &op1_high, &op2_high); | |
2195867f | 3004 | |
30c46053 MC |
3005 | emit_insn (gen_subdi3_compare1 (low_dest, op1_low, op2_low)); |
3006 | emit_insn (gen_subdi3_carryin (high_dest, op1_high, op2_high)); | |
2195867f | 3007 | |
30c46053 MC |
3008 | emit_move_insn (gen_lowpart (DImode, operands[0]), low_dest); |
3009 | emit_move_insn (gen_highpart (DImode, operands[0]), high_dest); | |
3010 | DONE; | |
3011 | }) | |
3012 | ||
3013 | (define_expand "subvti4" | |
3014 | [(match_operand:TI 0 "register_operand") | |
a58fe3c5 | 3015 | (match_operand:TI 1 "register_operand") |
30c46053 MC |
3016 | (match_operand:TI 2 "aarch64_reg_or_imm") |
3017 | (label_ref (match_operand 3 "" ""))] | |
3018 | "" | |
3019 | { | |
3020 | rtx low_dest, op1_low, op2_low, high_dest, op1_high, op2_high; | |
3021 | ||
3022 | aarch64_subvti_scratch_regs (operands[1], operands[2], | |
3023 | &low_dest, &op1_low, &op2_low, | |
3024 | &high_dest, &op1_high, &op2_high); | |
3025 | aarch64_expand_subvti (operands[0], low_dest, op1_low, op2_low, | |
a58fe3c5 | 3026 | high_dest, op1_high, op2_high, false); |
30c46053 MC |
3027 | |
3028 | aarch64_gen_unlikely_cbranch (NE, CC_Vmode, operands[3]); | |
3029 | DONE; | |
3030 | }) | |
3031 | ||
3032 | (define_expand "usubvti4" | |
3033 | [(match_operand:TI 0 "register_operand") | |
a58fe3c5 | 3034 | (match_operand:TI 1 "register_operand") |
30c46053 MC |
3035 | (match_operand:TI 2 "aarch64_reg_or_imm") |
3036 | (label_ref (match_operand 3 "" ""))] | |
3037 | "" | |
3038 | { | |
3039 | rtx low_dest, op1_low, op2_low, high_dest, op1_high, op2_high; | |
3040 | ||
3041 | aarch64_subvti_scratch_regs (operands[1], operands[2], | |
3042 | &low_dest, &op1_low, &op2_low, | |
3043 | &high_dest, &op1_high, &op2_high); | |
3044 | aarch64_expand_subvti (operands[0], low_dest, op1_low, op2_low, | |
a58fe3c5 | 3045 | high_dest, op1_high, op2_high, true); |
30c46053 MC |
3046 | |
3047 | aarch64_gen_unlikely_cbranch (LTU, CCmode, operands[3]); | |
2195867f RH |
3048 | DONE; |
3049 | }) | |
43e9d192 | 3050 | |
a58fe3c5 RE |
3051 | (define_expand "negvti3" |
3052 | [(match_operand:TI 0 "register_operand") | |
3053 | (match_operand:TI 1 "register_operand") | |
3054 | (label_ref (match_operand 2 "" ""))] | |
3055 | "" | |
3056 | { | |
3057 | emit_insn (gen_negdi_carryout (gen_lowpart (DImode, operands[0]), | |
3058 | gen_lowpart (DImode, operands[1]))); | |
3059 | emit_insn (gen_negvdi_carryinV (gen_highpart (DImode, operands[0]), | |
3060 | gen_highpart (DImode, operands[1]))); | |
3061 | aarch64_gen_unlikely_cbranch (NE, CC_Vmode, operands[2]); | |
3062 | ||
3063 | DONE; | |
3064 | } | |
3065 | ) | |
3066 | ||
3067 | (define_insn "negdi_carryout" | |
3068 | [(set (reg:CC CC_REGNUM) | |
3069 | (compare:CC | |
3070 | (const_int 0) (match_operand:DI 1 "register_operand" "r"))) | |
3071 | (set (match_operand:DI 0 "register_operand" "=r") | |
3072 | (neg:DI (match_dup 1)))] | |
3073 | "" | |
3074 | "negs\\t%0, %1" | |
3075 | [(set_attr "type" "alus_sreg")] | |
3076 | ) | |
3077 | ||
3078 | (define_insn "negvdi_carryinV" | |
3079 | [(set (reg:CC_V CC_REGNUM) | |
3080 | (compare:CC_V | |
3081 | (neg:TI (plus:TI | |
3082 | (ltu:TI (reg:CC CC_REGNUM) (const_int 0)) | |
3083 | (sign_extend:TI (match_operand:DI 1 "register_operand" "r")))) | |
3084 | (sign_extend:TI | |
3085 | (neg:DI (plus:DI (ltu:DI (reg:CC CC_REGNUM) (const_int 0)) | |
3086 | (match_dup 1)))))) | |
3087 | (set (match_operand:DI 0 "register_operand" "=r") | |
3088 | (neg:DI (plus:DI (ltu:DI (reg:CC CC_REGNUM) (const_int 0)) | |
3089 | (match_dup 1))))] | |
3090 | "" | |
3091 | "ngcs\\t%0, %1" | |
3092 | [(set_attr "type" "alus_sreg")] | |
3093 | ) | |
3094 | ||
ef22810a | 3095 | (define_insn "*sub<mode>3_compare0" |
43e9d192 | 3096 | [(set (reg:CC_NZ CC_REGNUM) |
341fca1b | 3097 | (compare:CC_NZ (minus:GPI (match_operand:GPI 1 "register_operand" "rk") |
43e9d192 IB |
3098 | (match_operand:GPI 2 "register_operand" "r")) |
3099 | (const_int 0))) | |
3100 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3101 | (minus:GPI (match_dup 1) (match_dup 2)))] | |
3102 | "" | |
3103 | "subs\\t%<w>0, %<w>1, %<w>2" | |
1d61feeb | 3104 | [(set_attr "type" "alus_sreg")] |
43e9d192 IB |
3105 | ) |
3106 | ||
3b585938 IB |
3107 | ;; zero_extend version of above |
3108 | (define_insn "*subsi3_compare0_uxtw" | |
3109 | [(set (reg:CC_NZ CC_REGNUM) | |
341fca1b | 3110 | (compare:CC_NZ (minus:SI (match_operand:SI 1 "register_operand" "rk") |
3b585938 IB |
3111 | (match_operand:SI 2 "register_operand" "r")) |
3112 | (const_int 0))) | |
3113 | (set (match_operand:DI 0 "register_operand" "=r") | |
3114 | (zero_extend:DI (minus:SI (match_dup 1) (match_dup 2))))] | |
3115 | "" | |
3116 | "subs\\t%w0, %w1, %w2" | |
1d61feeb | 3117 | [(set_attr "type" "alus_sreg")] |
3b585938 IB |
3118 | ) |
3119 | ||
a58fe3c5 | 3120 | (define_insn "sub<mode>3_compare1_imm" |
30c46053 MC |
3121 | [(set (reg:CC CC_REGNUM) |
3122 | (compare:CC | |
19199a6f | 3123 | (match_operand:GPI 1 "register_operand" "rk,rk") |
30c46053 MC |
3124 | (match_operand:GPI 2 "aarch64_plus_immediate" "I,J"))) |
3125 | (set (match_operand:GPI 0 "register_operand" "=r,r") | |
3126 | (plus:GPI | |
3127 | (match_dup 1) | |
3128 | (match_operand:GPI 3 "aarch64_plus_immediate" "J,I")))] | |
3129 | "UINTVAL (operands[2]) == -UINTVAL (operands[3])" | |
3130 | "@ | |
a58fe3c5 RE |
3131 | subs\\t%<w>0, %<w>1, %2 |
3132 | adds\\t%<w>0, %<w>1, #%n2" | |
30c46053 MC |
3133 | [(set_attr "type" "alus_imm")] |
3134 | ) | |
3135 | ||
ef22810a RH |
3136 | (define_insn "sub<mode>3_compare1" |
3137 | [(set (reg:CC CC_REGNUM) | |
3138 | (compare:CC | |
341fca1b | 3139 | (match_operand:GPI 1 "aarch64_reg_or_zero" "rkZ") |
ef22810a RH |
3140 | (match_operand:GPI 2 "aarch64_reg_or_zero" "rZ"))) |
3141 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3142 | (minus:GPI (match_dup 1) (match_dup 2)))] | |
3143 | "" | |
3144 | "subs\\t%<w>0, %<w>1, %<w>2" | |
3145 | [(set_attr "type" "alus_sreg")] | |
3146 | ) | |
3147 | ||
e49d8b68 | 3148 | (define_peephole2 |
341fca1b | 3149 | [(set (match_operand:GPI 0 "aarch64_general_reg") |
e49d8b68 KT |
3150 | (minus:GPI (match_operand:GPI 1 "aarch64_reg_or_zero") |
3151 | (match_operand:GPI 2 "aarch64_reg_or_zero"))) | |
3152 | (set (reg:CC CC_REGNUM) | |
3153 | (compare:CC | |
3154 | (match_dup 1) | |
3155 | (match_dup 2)))] | |
3156 | "!reg_overlap_mentioned_p (operands[0], operands[1]) | |
3157 | && !reg_overlap_mentioned_p (operands[0], operands[2])" | |
3158 | [(const_int 0)] | |
3159 | { | |
3160 | emit_insn (gen_sub<mode>3_compare1 (operands[0], operands[1], | |
3161 | operands[2])); | |
3162 | DONE; | |
3163 | } | |
3164 | ) | |
3165 | ||
de47f61f KT |
3166 | ;; Same as the above peephole but with the compare and minus in |
3167 | ;; swapped order. The restriction on overlap between operand 0 | |
3168 | ;; and operands 1 and 2 doesn't apply here. | |
3169 | (define_peephole2 | |
3170 | [(set (reg:CC CC_REGNUM) | |
3171 | (compare:CC | |
3172 | (match_operand:GPI 1 "aarch64_reg_or_zero") | |
3173 | (match_operand:GPI 2 "aarch64_reg_or_zero"))) | |
341fca1b | 3174 | (set (match_operand:GPI 0 "aarch64_general_reg") |
de47f61f KT |
3175 | (minus:GPI (match_dup 1) |
3176 | (match_dup 2)))] | |
3177 | "" | |
3178 | [(const_int 0)] | |
3179 | { | |
3180 | emit_insn (gen_sub<mode>3_compare1 (operands[0], operands[1], | |
3181 | operands[2])); | |
3182 | DONE; | |
3183 | } | |
3184 | ) | |
3185 | ||
279dc4b4 | 3186 | (define_peephole2 |
341fca1b | 3187 | [(set (match_operand:GPI 0 "aarch64_general_reg") |
279dc4b4 | 3188 | (plus:GPI (match_operand:GPI 1 "register_operand") |
a58fe3c5 | 3189 | (match_operand:GPI 2 "aarch64_plus_immediate"))) |
279dc4b4 KT |
3190 | (set (reg:CC CC_REGNUM) |
3191 | (compare:CC | |
3192 | (match_dup 1) | |
3193 | (match_operand:GPI 3 "const_int_operand")))] | |
3194 | "!reg_overlap_mentioned_p (operands[0], operands[1]) | |
3195 | && INTVAL (operands[3]) == -INTVAL (operands[2])" | |
3196 | [(const_int 0)] | |
3197 | { | |
3198 | emit_insn (gen_sub<mode>3_compare1_imm (operands[0], operands[1], | |
a58fe3c5 | 3199 | operands[3], operands[2])); |
279dc4b4 KT |
3200 | DONE; |
3201 | } | |
3202 | ) | |
3203 | ||
de47f61f KT |
3204 | ;; Same as the above peephole but with the compare and minus in |
3205 | ;; swapped order. The restriction on overlap between operand 0 | |
3206 | ;; and operands 1 doesn't apply here. | |
3207 | (define_peephole2 | |
3208 | [(set (reg:CC CC_REGNUM) | |
3209 | (compare:CC | |
3210 | (match_operand:GPI 1 "register_operand") | |
3211 | (match_operand:GPI 3 "const_int_operand"))) | |
341fca1b | 3212 | (set (match_operand:GPI 0 "aarch64_general_reg") |
de47f61f | 3213 | (plus:GPI (match_dup 1) |
a58fe3c5 | 3214 | (match_operand:GPI 2 "aarch64_plus_immediate")))] |
de47f61f KT |
3215 | "INTVAL (operands[3]) == -INTVAL (operands[2])" |
3216 | [(const_int 0)] | |
3217 | { | |
3218 | emit_insn (gen_sub<mode>3_compare1_imm (operands[0], operands[1], | |
a58fe3c5 | 3219 | operands[3], operands[2])); |
de47f61f KT |
3220 | DONE; |
3221 | } | |
3222 | ) | |
3223 | ||
43e9d192 | 3224 | (define_insn "*sub_<shift>_<mode>" |
24c56925 | 3225 | [(set (match_operand:GPI 0 "register_operand" "=r") |
43e9d192 IB |
3226 | (minus:GPI (match_operand:GPI 3 "register_operand" "r") |
3227 | (ASHIFT:GPI | |
3228 | (match_operand:GPI 1 "register_operand" "r") | |
3229 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))))] | |
3230 | "" | |
3231 | "sub\\t%<w>0, %<w>3, %<w>1, <shift> %2" | |
ae27ce51 | 3232 | [(set_attr "autodetect_type" "alu_shift_<shift>_op2")] |
43e9d192 IB |
3233 | ) |
3234 | ||
3b585938 IB |
3235 | ;; zero_extend version of above |
3236 | (define_insn "*sub_<shift>_si_uxtw" | |
24c56925 | 3237 | [(set (match_operand:DI 0 "register_operand" "=r") |
3b585938 IB |
3238 | (zero_extend:DI |
3239 | (minus:SI (match_operand:SI 3 "register_operand" "r") | |
3240 | (ASHIFT:SI | |
3241 | (match_operand:SI 1 "register_operand" "r") | |
3242 | (match_operand:QI 2 "aarch64_shift_imm_si" "n")))))] | |
3243 | "" | |
3244 | "sub\\t%w0, %w3, %w1, <shift> %2" | |
ae27ce51 | 3245 | [(set_attr "autodetect_type" "alu_shift_<shift>_op2")] |
3b585938 IB |
3246 | ) |
3247 | ||
43e9d192 IB |
3248 | (define_insn "*sub_<optab><ALLX:mode>_<GPI:mode>" |
3249 | [(set (match_operand:GPI 0 "register_operand" "=rk") | |
7eb8fb77 | 3250 | (minus:GPI (match_operand:GPI 1 "register_operand" "rk") |
43e9d192 IB |
3251 | (ANY_EXTEND:GPI |
3252 | (match_operand:ALLX 2 "register_operand" "r"))))] | |
3253 | "" | |
d4febc75 | 3254 | "sub\\t%<GPI:w>0, %<GPI:w>1, %w2, <su>xt<ALLX:size>" |
99e6af15 | 3255 | [(set_attr "type" "alu_ext")] |
43e9d192 IB |
3256 | ) |
3257 | ||
3b585938 IB |
3258 | ;; zero_extend version of above |
3259 | (define_insn "*sub_<optab><SHORT:mode>_si_uxtw" | |
3260 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
3261 | (zero_extend:DI | |
7eb8fb77 | 3262 | (minus:SI (match_operand:SI 1 "register_operand" "rk") |
3b585938 IB |
3263 | (ANY_EXTEND:SI |
3264 | (match_operand:SHORT 2 "register_operand" "r")))))] | |
3265 | "" | |
3266 | "sub\\t%w0, %w1, %w2, <su>xt<SHORT:size>" | |
99e6af15 | 3267 | [(set_attr "type" "alu_ext")] |
3b585938 IB |
3268 | ) |
3269 | ||
43e9d192 IB |
3270 | (define_insn "*sub_<optab><ALLX:mode>_shft_<GPI:mode>" |
3271 | [(set (match_operand:GPI 0 "register_operand" "=rk") | |
7eb8fb77 | 3272 | (minus:GPI (match_operand:GPI 1 "register_operand" "rk") |
43e9d192 IB |
3273 | (ashift:GPI (ANY_EXTEND:GPI |
3274 | (match_operand:ALLX 2 "register_operand" "r")) | |
3275 | (match_operand 3 "aarch64_imm3" "Ui3"))))] | |
3276 | "" | |
d4febc75 | 3277 | "sub\\t%<GPI:w>0, %<GPI:w>1, %w2, <su>xt<ALLX:size> %3" |
99e6af15 | 3278 | [(set_attr "type" "alu_ext")] |
43e9d192 IB |
3279 | ) |
3280 | ||
3b585938 IB |
3281 | ;; zero_extend version of above |
3282 | (define_insn "*sub_<optab><SHORT:mode>_shft_si_uxtw" | |
3283 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
3284 | (zero_extend:DI | |
7eb8fb77 | 3285 | (minus:SI (match_operand:SI 1 "register_operand" "rk") |
3b585938 IB |
3286 | (ashift:SI (ANY_EXTEND:SI |
3287 | (match_operand:SHORT 2 "register_operand" "r")) | |
3288 | (match_operand 3 "aarch64_imm3" "Ui3")))))] | |
3289 | "" | |
3290 | "sub\\t%w0, %w1, %w2, <su>xt<SHORT:size> %3" | |
99e6af15 | 3291 | [(set_attr "type" "alu_ext")] |
3b585938 IB |
3292 | ) |
3293 | ||
ef22810a RH |
3294 | ;; The hardware description is op1 + ~op2 + C. |
3295 | ;; = op1 + (-op2 + 1) + (1 - !C) | |
3296 | ;; = op1 - op2 - 1 + 1 - !C | |
3297 | ;; = op1 - op2 - !C. | |
3298 | ;; We describe the latter. | |
3299 | ||
3300 | (define_insn "*sub<mode>3_carryin0" | |
3301 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
3302 | (minus:GPI | |
3303 | (match_operand:GPI 1 "aarch64_reg_or_zero" "rZ") | |
3304 | (match_operand:GPI 2 "aarch64_borrow_operation" "")))] | |
3305 | "" | |
3306 | "sbc\\t%<w>0, %<w>1, <w>zr" | |
3307 | [(set_attr "type" "adc_reg")] | |
3308 | ) | |
3309 | ||
3310 | ;; zero_extend version of the above | |
3311 | (define_insn "*subsi3_carryin_uxtw" | |
3312 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3313 | (zero_extend:DI | |
3314 | (minus:SI | |
3315 | (match_operand:SI 1 "aarch64_reg_or_zero" "rZ") | |
3316 | (match_operand:SI 2 "aarch64_borrow_operation" ""))))] | |
3317 | "" | |
3318 | "sbc\\t%w0, %w1, wzr" | |
3319 | [(set_attr "type" "adc_reg")] | |
3320 | ) | |
3321 | ||
3322 | (define_expand "sub<mode>3_carryin" | |
3323 | [(set (match_operand:GPI 0 "register_operand") | |
3324 | (minus:GPI | |
3325 | (minus:GPI | |
3326 | (match_operand:GPI 1 "aarch64_reg_or_zero") | |
3327 | (match_operand:GPI 2 "register_operand")) | |
3328 | (ltu:GPI (reg:CC CC_REGNUM) (const_int 0))))] | |
3329 | "" | |
3330 | "" | |
3331 | ) | |
3332 | ||
3333 | (define_insn "*sub<mode>3_carryin" | |
3334 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
3335 | (minus:GPI | |
3336 | (minus:GPI | |
3337 | (match_operand:GPI 1 "aarch64_reg_or_zero" "rZ") | |
3338 | (match_operand:GPI 2 "register_operand" "r")) | |
3339 | (match_operand:GPI 3 "aarch64_borrow_operation" "")))] | |
3340 | ||
0b8cdc58 IB |
3341 | "" |
3342 | "sbc\\t%<w>0, %<w>1, %<w>2" | |
99e6af15 | 3343 | [(set_attr "type" "adc_reg")] |
0b8cdc58 IB |
3344 | ) |
3345 | ||
3346 | ;; zero_extend version of the above | |
3347 | (define_insn "*subsi3_carryin_uxtw" | |
ef22810a RH |
3348 | [(set (match_operand:DI 0 "register_operand" "=r") |
3349 | (zero_extend:DI | |
3350 | (minus:SI | |
3351 | (minus:SI | |
3352 | (match_operand:SI 1 "aarch64_reg_or_zero" "rZ") | |
3353 | (match_operand:SI 2 "register_operand" "r")) | |
3354 | (match_operand:SI 3 "aarch64_borrow_operation" ""))))] | |
3355 | ||
3356 | "" | |
3357 | "sbc\\t%w0, %w1, %w2" | |
3358 | [(set_attr "type" "adc_reg")] | |
3359 | ) | |
3360 | ||
3361 | (define_insn "*sub<mode>3_carryin_alt" | |
3362 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
3363 | (minus:GPI | |
3364 | (minus:GPI | |
3365 | (match_operand:GPI 1 "aarch64_reg_or_zero" "rZ") | |
3366 | (match_operand:GPI 3 "aarch64_borrow_operation" "")) | |
3367 | (match_operand:GPI 2 "register_operand" "r")))] | |
3368 | "" | |
3369 | "sbc\\t%<w>0, %<w>1, %<w>2" | |
3370 | [(set_attr "type" "adc_reg")] | |
3371 | ) | |
3372 | ||
3373 | ;; zero_extend version of the above | |
3374 | (define_insn "*subsi3_carryin_alt_uxtw" | |
3375 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3376 | (zero_extend:DI | |
3377 | (minus:SI | |
3378 | (minus:SI | |
3379 | (match_operand:SI 1 "aarch64_reg_or_zero" "rZ") | |
3380 | (match_operand:SI 3 "aarch64_borrow_operation" "")) | |
3381 | (match_operand:SI 2 "register_operand" "r"))))] | |
0b8cdc58 IB |
3382 | "" |
3383 | "sbc\\t%w0, %w1, %w2" | |
99e6af15 | 3384 | [(set_attr "type" "adc_reg")] |
0b8cdc58 IB |
3385 | ) |
3386 | ||
a58fe3c5 | 3387 | (define_expand "usub<GPI:mode>3_carryinC" |
30c46053 MC |
3388 | [(parallel |
3389 | [(set (reg:CC CC_REGNUM) | |
3390 | (compare:CC | |
a58fe3c5 | 3391 | (zero_extend:<DWI> |
1bbffb87 | 3392 | (match_operand:GPI 1 "aarch64_reg_or_zero")) |
30c46053 | 3393 | (plus:<DWI> |
a58fe3c5 | 3394 | (zero_extend:<DWI> |
1bbffb87 | 3395 | (match_operand:GPI 2 "register_operand")) |
30c46053 | 3396 | (ltu:<DWI> (reg:CC CC_REGNUM) (const_int 0))))) |
1bbffb87 | 3397 | (set (match_operand:GPI 0 "register_operand") |
30c46053 MC |
3398 | (minus:GPI |
3399 | (minus:GPI (match_dup 1) (match_dup 2)) | |
3400 | (ltu:GPI (reg:CC CC_REGNUM) (const_int 0))))])] | |
3401 | "" | |
3402 | ) | |
3403 | ||
a58fe3c5 | 3404 | (define_insn "*usub<GPI:mode>3_carryinC_z1" |
30c46053 MC |
3405 | [(set (reg:CC CC_REGNUM) |
3406 | (compare:CC | |
3407 | (const_int 0) | |
3408 | (plus:<DWI> | |
a58fe3c5 | 3409 | (zero_extend:<DWI> |
30c46053 MC |
3410 | (match_operand:GPI 1 "register_operand" "r")) |
3411 | (match_operand:<DWI> 2 "aarch64_borrow_operation" "")))) | |
3412 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3413 | (minus:GPI | |
3414 | (neg:GPI (match_dup 1)) | |
3415 | (match_operand:GPI 3 "aarch64_borrow_operation" "")))] | |
3416 | "" | |
3417 | "sbcs\\t%<w>0, <w>zr, %<w>1" | |
3418 | [(set_attr "type" "adc_reg")] | |
3419 | ) | |
3420 | ||
a58fe3c5 | 3421 | (define_insn "*usub<GPI:mode>3_carryinC_z2" |
30c46053 MC |
3422 | [(set (reg:CC CC_REGNUM) |
3423 | (compare:CC | |
a58fe3c5 | 3424 | (zero_extend:<DWI> |
30c46053 MC |
3425 | (match_operand:GPI 1 "register_operand" "r")) |
3426 | (match_operand:<DWI> 2 "aarch64_borrow_operation" ""))) | |
3427 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3428 | (minus:GPI | |
3429 | (match_dup 1) | |
3430 | (match_operand:GPI 3 "aarch64_borrow_operation" "")))] | |
3431 | "" | |
3432 | "sbcs\\t%<w>0, %<w>1, <w>zr" | |
3433 | [(set_attr "type" "adc_reg")] | |
3434 | ) | |
3435 | ||
a58fe3c5 | 3436 | (define_insn "*usub<GPI:mode>3_carryinC" |
30c46053 MC |
3437 | [(set (reg:CC CC_REGNUM) |
3438 | (compare:CC | |
a58fe3c5 | 3439 | (zero_extend:<DWI> |
30c46053 MC |
3440 | (match_operand:GPI 1 "register_operand" "r")) |
3441 | (plus:<DWI> | |
a58fe3c5 | 3442 | (zero_extend:<DWI> |
30c46053 MC |
3443 | (match_operand:GPI 2 "register_operand" "r")) |
3444 | (match_operand:<DWI> 3 "aarch64_borrow_operation" "")))) | |
3445 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3446 | (minus:GPI | |
3447 | (minus:GPI (match_dup 1) (match_dup 2)) | |
3448 | (match_operand:GPI 4 "aarch64_borrow_operation" "")))] | |
3449 | "" | |
3450 | "sbcs\\t%<w>0, %<w>1, %<w>2" | |
3451 | [(set_attr "type" "adc_reg")] | |
3452 | ) | |
3453 | ||
a58fe3c5 RE |
3454 | (define_expand "sub<GPI:mode>3_carryinV" |
3455 | [(parallel | |
3456 | [(set (reg:CC_V CC_REGNUM) | |
3457 | (compare:CC_V | |
3458 | (minus:<DWI> | |
3459 | (sign_extend:<DWI> | |
1bbffb87 | 3460 | (match_operand:GPI 1 "aarch64_reg_or_zero")) |
a58fe3c5 RE |
3461 | (plus:<DWI> |
3462 | (sign_extend:<DWI> | |
1bbffb87 | 3463 | (match_operand:GPI 2 "register_operand")) |
a58fe3c5 RE |
3464 | (ltu:<DWI> (reg:CC CC_REGNUM) (const_int 0)))) |
3465 | (sign_extend:<DWI> | |
3466 | (minus:GPI (match_dup 1) | |
3467 | (plus:GPI (ltu:GPI (reg:CC CC_REGNUM) (const_int 0)) | |
3468 | (match_dup 2)))))) | |
1bbffb87 | 3469 | (set (match_operand:GPI 0 "register_operand") |
a58fe3c5 RE |
3470 | (minus:GPI |
3471 | (minus:GPI (match_dup 1) (match_dup 2)) | |
3472 | (ltu:GPI (reg:CC CC_REGNUM) (const_int 0))))])] | |
3473 | "" | |
3474 | ) | |
3475 | ||
3476 | (define_insn "*sub<mode>3_carryinV_z2" | |
3477 | [(set (reg:CC_V CC_REGNUM) | |
3478 | (compare:CC_V | |
3479 | (minus:<DWI> | |
3480 | (sign_extend:<DWI> (match_operand:GPI 1 "register_operand" "r")) | |
3481 | (match_operand:<DWI> 2 "aarch64_borrow_operation" "")) | |
3482 | (sign_extend:<DWI> | |
3483 | (minus:GPI (match_dup 1) | |
3484 | (match_operand:GPI 3 "aarch64_borrow_operation" ""))))) | |
3485 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3486 | (minus:GPI | |
3487 | (match_dup 1) (match_dup 3)))] | |
3488 | "" | |
3489 | "sbcs\\t%<w>0, %<w>1, <w>zr" | |
3490 | [(set_attr "type" "adc_reg")] | |
3491 | ) | |
3492 | ||
3493 | (define_insn "*sub<mode>3_carryinV" | |
3494 | [(set (reg:CC_V CC_REGNUM) | |
3495 | (compare:CC_V | |
3496 | (minus:<DWI> | |
3497 | (sign_extend:<DWI> | |
3498 | (match_operand:GPI 1 "register_operand" "r")) | |
3499 | (plus:<DWI> | |
3500 | (sign_extend:<DWI> | |
3501 | (match_operand:GPI 2 "register_operand" "r")) | |
3502 | (match_operand:<DWI> 3 "aarch64_borrow_operation" ""))) | |
3503 | (sign_extend:<DWI> | |
3504 | (minus:GPI | |
3505 | (match_dup 1) | |
3506 | (plus:GPI (match_operand:GPI 4 "aarch64_borrow_operation" "") | |
3507 | (match_dup 2)))))) | |
3508 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3509 | (minus:GPI | |
3510 | (minus:GPI (match_dup 1) (match_dup 2)) | |
3511 | (match_dup 4)))] | |
3512 | "" | |
3513 | "sbcs\\t%<w>0, %<w>1, %<w>2" | |
3514 | [(set_attr "type" "adc_reg")] | |
3515 | ) | |
3516 | ||
dd1bdce4 VK |
3517 | (define_insn "*sub_uxt<mode>_shift2" |
3518 | [(set (match_operand:GPI 0 "register_operand" "=rk") | |
3519 | (minus:GPI (match_operand:GPI 4 "register_operand" "rk") | |
3520 | (and:GPI | |
3521 | (ashift:GPI (match_operand:GPI 1 "register_operand" "r") | |
3522 | (match_operand 2 "aarch64_imm3" "Ui3")) | |
3523 | (match_operand 3 "const_int_operand" "n"))))] | |
3524 | "aarch64_uxt_size (INTVAL (operands[2]),INTVAL (operands[3])) != 0" | |
3525 | "* | |
3526 | operands[3] = GEN_INT (aarch64_uxt_size (INTVAL (operands[2]), | |
3527 | INTVAL (operands[3]))); | |
d4febc75 | 3528 | return \"sub\t%<w>0, %<w>4, %w1, uxt%e3 %2\";" |
dd1bdce4 VK |
3529 | [(set_attr "type" "alu_ext")] |
3530 | ) | |
3531 | ||
3532 | ;; zero_extend version of above | |
3533 | (define_insn "*sub_uxtsi_shift2_uxtw" | |
3534 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
3535 | (zero_extend:DI | |
3536 | (minus:SI (match_operand:SI 4 "register_operand" "rk") | |
3537 | (and:SI | |
3538 | (ashift:SI (match_operand:SI 1 "register_operand" "r") | |
3539 | (match_operand 2 "aarch64_imm3" "Ui3")) | |
3540 | (match_operand 3 "const_int_operand" "n")))))] | |
3541 | "aarch64_uxt_size (INTVAL (operands[2]),INTVAL (operands[3])) != 0" | |
3542 | "* | |
3543 | operands[3] = GEN_INT (aarch64_uxt_size (INTVAL (operands[2]), | |
3544 | INTVAL (operands[3]))); | |
3545 | return \"sub\t%w0, %w4, %w1, uxt%e3 %2\";" | |
3546 | [(set_attr "type" "alu_ext")] | |
3547 | ) | |
3548 | ||
fc8b587d | 3549 | (define_expand "abs<mode>2" |
1bbffb87 DZ |
3550 | [(match_operand:GPI 0 "register_operand") |
3551 | (match_operand:GPI 1 "register_operand")] | |
26366d28 | 3552 | "" |
26366d28 | 3553 | { |
fc8b587d WD |
3554 | rtx ccreg = aarch64_gen_compare_reg (LT, operands[1], const0_rtx); |
3555 | rtx x = gen_rtx_LT (VOIDmode, ccreg, const0_rtx); | |
3556 | emit_insn (gen_csneg3<mode>_insn (operands[0], x, operands[1], operands[1])); | |
26366d28 IB |
3557 | DONE; |
3558 | } | |
26366d28 IB |
3559 | ) |
3560 | ||
43e9d192 | 3561 | (define_insn "neg<mode>2" |
040d8a1c IB |
3562 | [(set (match_operand:GPI 0 "register_operand" "=r,w") |
3563 | (neg:GPI (match_operand:GPI 1 "register_operand" "r,w")))] | |
43e9d192 | 3564 | "" |
040d8a1c IB |
3565 | "@ |
3566 | neg\\t%<w>0, %<w>1 | |
3567 | neg\\t%<rtn>0<vas>, %<rtn>1<vas>" | |
1d61feeb | 3568 | [(set_attr "type" "alu_sreg, neon_neg<q>") |
488461d8 | 3569 | (set_attr "arch" "*,simd")] |
43e9d192 IB |
3570 | ) |
3571 | ||
3b585938 IB |
3572 | ;; zero_extend version of above |
3573 | (define_insn "*negsi2_uxtw" | |
3574 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3575 | (zero_extend:DI (neg:SI (match_operand:SI 1 "register_operand" "r"))))] | |
3576 | "" | |
3577 | "neg\\t%w0, %w1" | |
1d61feeb | 3578 | [(set_attr "type" "alu_sreg")] |
3b585938 IB |
3579 | ) |
3580 | ||
7e0228bf N |
3581 | (define_insn "*ngc<mode>" |
3582 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
ef22810a RH |
3583 | (minus:GPI |
3584 | (neg:GPI (match_operand:GPI 2 "aarch64_borrow_operation" "")) | |
3585 | (match_operand:GPI 1 "register_operand" "r")))] | |
7e0228bf N |
3586 | "" |
3587 | "ngc\\t%<w>0, %<w>1" | |
99e6af15 | 3588 | [(set_attr "type" "adc_reg")] |
7e0228bf N |
3589 | ) |
3590 | ||
3591 | (define_insn "*ngcsi_uxtw" | |
3592 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3593 | (zero_extend:DI | |
ef22810a RH |
3594 | (minus:SI |
3595 | (neg:SI (match_operand:SI 2 "aarch64_borrow_operation" "")) | |
3596 | (match_operand:SI 1 "register_operand" "r"))))] | |
7e0228bf N |
3597 | "" |
3598 | "ngc\\t%w0, %w1" | |
99e6af15 | 3599 | [(set_attr "type" "adc_reg")] |
7e0228bf N |
3600 | ) |
3601 | ||
4f58fe36 | 3602 | (define_insn "neg<mode>2_compare0" |
43e9d192 IB |
3603 | [(set (reg:CC_NZ CC_REGNUM) |
3604 | (compare:CC_NZ (neg:GPI (match_operand:GPI 1 "register_operand" "r")) | |
3605 | (const_int 0))) | |
3606 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3607 | (neg:GPI (match_dup 1)))] | |
3608 | "" | |
3609 | "negs\\t%<w>0, %<w>1" | |
1d61feeb | 3610 | [(set_attr "type" "alus_sreg")] |
43e9d192 IB |
3611 | ) |
3612 | ||
3b585938 IB |
3613 | ;; zero_extend version of above |
3614 | (define_insn "*negsi2_compare0_uxtw" | |
3615 | [(set (reg:CC_NZ CC_REGNUM) | |
3616 | (compare:CC_NZ (neg:SI (match_operand:SI 1 "register_operand" "r")) | |
3617 | (const_int 0))) | |
3618 | (set (match_operand:DI 0 "register_operand" "=r") | |
3619 | (zero_extend:DI (neg:SI (match_dup 1))))] | |
3620 | "" | |
3621 | "negs\\t%w0, %w1" | |
1d61feeb | 3622 | [(set_attr "type" "alus_sreg")] |
3b585938 IB |
3623 | ) |
3624 | ||
b056c910 N |
3625 | (define_insn "*neg_<shift><mode>3_compare0" |
3626 | [(set (reg:CC_NZ CC_REGNUM) | |
3627 | (compare:CC_NZ | |
3628 | (neg:GPI (ASHIFT:GPI | |
3629 | (match_operand:GPI 1 "register_operand" "r") | |
3630 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))) | |
3631 | (const_int 0))) | |
3632 | (set (match_operand:GPI 0 "register_operand" "=r") | |
3633 | (neg:GPI (ASHIFT:GPI (match_dup 1) (match_dup 2))))] | |
3634 | "" | |
3635 | "negs\\t%<w>0, %<w>1, <shift> %2" | |
99e6af15 | 3636 | [(set_attr "type" "alus_shift_imm")] |
b056c910 N |
3637 | ) |
3638 | ||
43e9d192 IB |
3639 | (define_insn "*neg_<shift>_<mode>2" |
3640 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
3641 | (neg:GPI (ASHIFT:GPI | |
3642 | (match_operand:GPI 1 "register_operand" "r") | |
3643 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))))] | |
3644 | "" | |
3645 | "neg\\t%<w>0, %<w>1, <shift> %2" | |
ae27ce51 | 3646 | [(set_attr "autodetect_type" "alu_shift_<shift>_op2")] |
43e9d192 IB |
3647 | ) |
3648 | ||
3b585938 IB |
3649 | ;; zero_extend version of above |
3650 | (define_insn "*neg_<shift>_si2_uxtw" | |
3651 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3652 | (zero_extend:DI | |
3653 | (neg:SI (ASHIFT:SI | |
3654 | (match_operand:SI 1 "register_operand" "r") | |
3655 | (match_operand:QI 2 "aarch64_shift_imm_si" "n")))))] | |
3656 | "" | |
3657 | "neg\\t%w0, %w1, <shift> %2" | |
ae27ce51 | 3658 | [(set_attr "autodetect_type" "alu_shift_<shift>_op2")] |
3b585938 IB |
3659 | ) |
3660 | ||
47f42744 JJ |
3661 | (define_insn "*neg_asr_si2_extr" |
3662 | [(set (match_operand:SI 0 "register_operand" "=r") | |
3663 | (neg:SI (match_operator:SI 4 "subreg_lowpart_operator" | |
3664 | [(sign_extract:DI | |
3665 | (match_operand:DI 1 "register_operand" "r") | |
3666 | (match_operand 3 "aarch64_simd_shift_imm_offset_si" "n") | |
3667 | (match_operand 2 "aarch64_simd_shift_imm_offset_si" "n"))])))] | |
3668 | "INTVAL (operands[2]) + INTVAL (operands[3]) == 32" | |
3669 | "neg\\t%w0, %w1, asr %2" | |
3670 | [(set_attr "autodetect_type" "alu_shift_asr_op2")] | |
3671 | ) | |
3672 | ||
43e9d192 IB |
3673 | (define_insn "mul<mode>3" |
3674 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
3675 | (mult:GPI (match_operand:GPI 1 "register_operand" "r") | |
3676 | (match_operand:GPI 2 "register_operand" "r")))] | |
3677 | "" | |
3678 | "mul\\t%<w>0, %<w>1, %<w>2" | |
99e6af15 | 3679 | [(set_attr "type" "mul")] |
43e9d192 IB |
3680 | ) |
3681 | ||
3b585938 IB |
3682 | ;; zero_extend version of above |
3683 | (define_insn "*mulsi3_uxtw" | |
3684 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3685 | (zero_extend:DI | |
3686 | (mult:SI (match_operand:SI 1 "register_operand" "r") | |
3687 | (match_operand:SI 2 "register_operand" "r"))))] | |
3688 | "" | |
3689 | "mul\\t%w0, %w1, %w2" | |
99e6af15 | 3690 | [(set_attr "type" "mul")] |
3b585938 IB |
3691 | ) |
3692 | ||
2195867f | 3693 | (define_insn "madd<mode>" |
43e9d192 IB |
3694 | [(set (match_operand:GPI 0 "register_operand" "=r") |
3695 | (plus:GPI (mult:GPI (match_operand:GPI 1 "register_operand" "r") | |
3696 | (match_operand:GPI 2 "register_operand" "r")) | |
3697 | (match_operand:GPI 3 "register_operand" "r")))] | |
3698 | "" | |
3699 | "madd\\t%<w>0, %<w>1, %<w>2, %<w>3" | |
99e6af15 | 3700 | [(set_attr "type" "mla")] |
43e9d192 IB |
3701 | ) |
3702 | ||
3b585938 IB |
3703 | ;; zero_extend version of above |
3704 | (define_insn "*maddsi_uxtw" | |
3705 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3706 | (zero_extend:DI | |
3707 | (plus:SI (mult:SI (match_operand:SI 1 "register_operand" "r") | |
3708 | (match_operand:SI 2 "register_operand" "r")) | |
3709 | (match_operand:SI 3 "register_operand" "r"))))] | |
3710 | "" | |
3711 | "madd\\t%w0, %w1, %w2, %w3" | |
99e6af15 | 3712 | [(set_attr "type" "mla")] |
3b585938 IB |
3713 | ) |
3714 | ||
43e9d192 IB |
3715 | (define_insn "*msub<mode>" |
3716 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
3717 | (minus:GPI (match_operand:GPI 3 "register_operand" "r") | |
3718 | (mult:GPI (match_operand:GPI 1 "register_operand" "r") | |
3719 | (match_operand:GPI 2 "register_operand" "r"))))] | |
3720 | ||
3721 | "" | |
3722 | "msub\\t%<w>0, %<w>1, %<w>2, %<w>3" | |
99e6af15 | 3723 | [(set_attr "type" "mla")] |
43e9d192 IB |
3724 | ) |
3725 | ||
3b585938 IB |
3726 | ;; zero_extend version of above |
3727 | (define_insn "*msubsi_uxtw" | |
3728 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3729 | (zero_extend:DI | |
3730 | (minus:SI (match_operand:SI 3 "register_operand" "r") | |
3731 | (mult:SI (match_operand:SI 1 "register_operand" "r") | |
3732 | (match_operand:SI 2 "register_operand" "r")))))] | |
3733 | ||
3734 | "" | |
3735 | "msub\\t%w0, %w1, %w2, %w3" | |
99e6af15 | 3736 | [(set_attr "type" "mla")] |
3b585938 IB |
3737 | ) |
3738 | ||
43e9d192 IB |
3739 | (define_insn "*mul<mode>_neg" |
3740 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
3741 | (mult:GPI (neg:GPI (match_operand:GPI 1 "register_operand" "r")) | |
3742 | (match_operand:GPI 2 "register_operand" "r")))] | |
3743 | ||
3744 | "" | |
3745 | "mneg\\t%<w>0, %<w>1, %<w>2" | |
99e6af15 | 3746 | [(set_attr "type" "mul")] |
43e9d192 IB |
3747 | ) |
3748 | ||
3b585938 IB |
3749 | ;; zero_extend version of above |
3750 | (define_insn "*mulsi_neg_uxtw" | |
3751 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3752 | (zero_extend:DI | |
3753 | (mult:SI (neg:SI (match_operand:SI 1 "register_operand" "r")) | |
3754 | (match_operand:SI 2 "register_operand" "r"))))] | |
3755 | ||
3756 | "" | |
3757 | "mneg\\t%w0, %w1, %w2" | |
99e6af15 | 3758 | [(set_attr "type" "mul")] |
3b585938 IB |
3759 | ) |
3760 | ||
43e9d192 IB |
3761 | (define_insn "<su_optab>mulsidi3" |
3762 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3763 | (mult:DI (ANY_EXTEND:DI (match_operand:SI 1 "register_operand" "r")) | |
3764 | (ANY_EXTEND:DI (match_operand:SI 2 "register_operand" "r"))))] | |
3765 | "" | |
3766 | "<su>mull\\t%0, %w1, %w2" | |
99e6af15 | 3767 | [(set_attr "type" "<su>mull")] |
43e9d192 IB |
3768 | ) |
3769 | ||
3770 | (define_insn "<su_optab>maddsidi4" | |
3771 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3772 | (plus:DI (mult:DI | |
3773 | (ANY_EXTEND:DI (match_operand:SI 1 "register_operand" "r")) | |
3774 | (ANY_EXTEND:DI (match_operand:SI 2 "register_operand" "r"))) | |
3775 | (match_operand:DI 3 "register_operand" "r")))] | |
3776 | "" | |
3777 | "<su>maddl\\t%0, %w1, %w2, %3" | |
99e6af15 | 3778 | [(set_attr "type" "<su>mlal")] |
43e9d192 IB |
3779 | ) |
3780 | ||
3781 | (define_insn "<su_optab>msubsidi4" | |
3782 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3783 | (minus:DI | |
3784 | (match_operand:DI 3 "register_operand" "r") | |
3785 | (mult:DI (ANY_EXTEND:DI (match_operand:SI 1 "register_operand" "r")) | |
3786 | (ANY_EXTEND:DI | |
3787 | (match_operand:SI 2 "register_operand" "r")))))] | |
3788 | "" | |
3789 | "<su>msubl\\t%0, %w1, %w2, %3" | |
99e6af15 | 3790 | [(set_attr "type" "<su>mlal")] |
43e9d192 IB |
3791 | ) |
3792 | ||
3793 | (define_insn "*<su_optab>mulsidi_neg" | |
3794 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3795 | (mult:DI (neg:DI | |
3796 | (ANY_EXTEND:DI (match_operand:SI 1 "register_operand" "r"))) | |
3797 | (ANY_EXTEND:DI (match_operand:SI 2 "register_operand" "r"))))] | |
3798 | "" | |
3799 | "<su>mnegl\\t%0, %w1, %w2" | |
99e6af15 | 3800 | [(set_attr "type" "<su>mull")] |
43e9d192 IB |
3801 | ) |
3802 | ||
2195867f RH |
3803 | (define_expand "<su_optab>mulditi3" |
3804 | [(set (match_operand:TI 0 "register_operand") | |
3805 | (mult:TI (ANY_EXTEND:TI (match_operand:DI 1 "register_operand")) | |
3806 | (ANY_EXTEND:TI (match_operand:DI 2 "register_operand"))))] | |
3807 | "" | |
3808 | { | |
3809 | rtx low = gen_reg_rtx (DImode); | |
3810 | emit_insn (gen_muldi3 (low, operands[1], operands[2])); | |
3811 | ||
3812 | rtx high = gen_reg_rtx (DImode); | |
3813 | emit_insn (gen_<su>muldi3_highpart (high, operands[1], operands[2])); | |
3814 | ||
3815 | emit_move_insn (gen_lowpart (DImode, operands[0]), low); | |
3816 | emit_move_insn (gen_highpart (DImode, operands[0]), high); | |
3817 | DONE; | |
3818 | }) | |
3819 | ||
3820 | ;; The default expansion of multi3 using umuldi3_highpart will perform | |
3821 | ;; the additions in an order that fails to combine into two madd insns. | |
3822 | (define_expand "multi3" | |
3823 | [(set (match_operand:TI 0 "register_operand") | |
3824 | (mult:TI (match_operand:TI 1 "register_operand") | |
3825 | (match_operand:TI 2 "register_operand")))] | |
3826 | "" | |
3827 | { | |
3828 | rtx l0 = gen_reg_rtx (DImode); | |
3829 | rtx l1 = gen_lowpart (DImode, operands[1]); | |
3830 | rtx l2 = gen_lowpart (DImode, operands[2]); | |
3831 | rtx h0 = gen_reg_rtx (DImode); | |
3832 | rtx h1 = gen_highpart (DImode, operands[1]); | |
3833 | rtx h2 = gen_highpart (DImode, operands[2]); | |
3834 | ||
3835 | emit_insn (gen_muldi3 (l0, l1, l2)); | |
3836 | emit_insn (gen_umuldi3_highpart (h0, l1, l2)); | |
3837 | emit_insn (gen_madddi (h0, h1, l2, h0)); | |
3838 | emit_insn (gen_madddi (h0, l1, h2, h0)); | |
3839 | ||
3840 | emit_move_insn (gen_lowpart (DImode, operands[0]), l0); | |
3841 | emit_move_insn (gen_highpart (DImode, operands[0]), h0); | |
3842 | DONE; | |
3843 | }) | |
3844 | ||
43e9d192 IB |
3845 | (define_insn "<su>muldi3_highpart" |
3846 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3847 | (truncate:DI | |
3848 | (lshiftrt:TI | |
3849 | (mult:TI | |
3850 | (ANY_EXTEND:TI (match_operand:DI 1 "register_operand" "r")) | |
3851 | (ANY_EXTEND:TI (match_operand:DI 2 "register_operand" "r"))) | |
3852 | (const_int 64))))] | |
3853 | "" | |
3854 | "<su>mulh\\t%0, %1, %2" | |
99e6af15 | 3855 | [(set_attr "type" "<su>mull")] |
43e9d192 IB |
3856 | ) |
3857 | ||
3858 | (define_insn "<su_optab>div<mode>3" | |
3859 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
3860 | (ANY_DIV:GPI (match_operand:GPI 1 "register_operand" "r") | |
3861 | (match_operand:GPI 2 "register_operand" "r")))] | |
3862 | "" | |
3863 | "<su>div\\t%<w>0, %<w>1, %<w>2" | |
99e6af15 | 3864 | [(set_attr "type" "<su>div")] |
43e9d192 IB |
3865 | ) |
3866 | ||
3b585938 IB |
3867 | ;; zero_extend version of above |
3868 | (define_insn "*<su_optab>divsi3_uxtw" | |
3869 | [(set (match_operand:DI 0 "register_operand" "=r") | |
3870 | (zero_extend:DI | |
3871 | (ANY_DIV:SI (match_operand:SI 1 "register_operand" "r") | |
3872 | (match_operand:SI 2 "register_operand" "r"))))] | |
3873 | "" | |
3874 | "<su>div\\t%w0, %w1, %w2" | |
99e6af15 | 3875 | [(set_attr "type" "<su>div")] |
3b585938 IB |
3876 | ) |
3877 | ||
43e9d192 IB |
3878 | ;; ------------------------------------------------------------------- |
3879 | ;; Comparison insns | |
3880 | ;; ------------------------------------------------------------------- | |
3881 | ||
c8012fbc | 3882 | (define_insn "cmp<mode>" |
43e9d192 | 3883 | [(set (reg:CC CC_REGNUM) |
3751345d | 3884 | (compare:CC (match_operand:GPI 0 "register_operand" "rk,rk,rk") |
34b512f5 | 3885 | (match_operand:GPI 1 "aarch64_plus_operand" "r,I,J")))] |
43e9d192 IB |
3886 | "" |
3887 | "@ | |
34b512f5 | 3888 | cmp\\t%<w>0, %<w>1 |
e2b691c4 | 3889 | cmp\\t%<w>0, %1 |
43e9d192 | 3890 | cmn\\t%<w>0, #%n1" |
1d61feeb | 3891 | [(set_attr "type" "alus_sreg,alus_imm,alus_imm")] |
43e9d192 IB |
3892 | ) |
3893 | ||
786e3c06 | 3894 | (define_insn "fcmp<mode>" |
43e9d192 IB |
3895 | [(set (reg:CCFP CC_REGNUM) |
3896 | (compare:CCFP (match_operand:GPF 0 "register_operand" "w,w") | |
3897 | (match_operand:GPF 1 "aarch64_fp_compare_operand" "Y,w")))] | |
3898 | "TARGET_FLOAT" | |
3899 | "@ | |
3900 | fcmp\\t%<s>0, #0.0 | |
3901 | fcmp\\t%<s>0, %<s>1" | |
99e6af15 | 3902 | [(set_attr "type" "fcmp<s>")] |
43e9d192 IB |
3903 | ) |
3904 | ||
786e3c06 | 3905 | (define_insn "fcmpe<mode>" |
43e9d192 IB |
3906 | [(set (reg:CCFPE CC_REGNUM) |
3907 | (compare:CCFPE (match_operand:GPF 0 "register_operand" "w,w") | |
3908 | (match_operand:GPF 1 "aarch64_fp_compare_operand" "Y,w")))] | |
3909 | "TARGET_FLOAT" | |
3910 | "@ | |
3911 | fcmpe\\t%<s>0, #0.0 | |
3912 | fcmpe\\t%<s>0, %<s>1" | |
99e6af15 | 3913 | [(set_attr "type" "fcmp<s>")] |
43e9d192 IB |
3914 | ) |
3915 | ||
3916 | (define_insn "*cmp_swp_<shift>_reg<mode>" | |
3917 | [(set (reg:CC_SWP CC_REGNUM) | |
3918 | (compare:CC_SWP (ASHIFT:GPI | |
3919 | (match_operand:GPI 0 "register_operand" "r") | |
3920 | (match_operand:QI 1 "aarch64_shift_imm_<mode>" "n")) | |
3921 | (match_operand:GPI 2 "aarch64_reg_or_zero" "rZ")))] | |
3922 | "" | |
3923 | "cmp\\t%<w>2, %<w>0, <shift> %1" | |
99e6af15 | 3924 | [(set_attr "type" "alus_shift_imm")] |
43e9d192 IB |
3925 | ) |
3926 | ||
3927 | (define_insn "*cmp_swp_<optab><ALLX:mode>_reg<GPI:mode>" | |
3928 | [(set (reg:CC_SWP CC_REGNUM) | |
3929 | (compare:CC_SWP (ANY_EXTEND:GPI | |
3930 | (match_operand:ALLX 0 "register_operand" "r")) | |
3931 | (match_operand:GPI 1 "register_operand" "r")))] | |
3932 | "" | |
d4febc75 | 3933 | "cmp\\t%<GPI:w>1, %w0, <su>xt<ALLX:size>" |
99e6af15 | 3934 | [(set_attr "type" "alus_ext")] |
43e9d192 IB |
3935 | ) |
3936 | ||
58cff58c N |
3937 | (define_insn "*cmp_swp_<optab><ALLX:mode>_shft_<GPI:mode>" |
3938 | [(set (reg:CC_SWP CC_REGNUM) | |
3939 | (compare:CC_SWP (ashift:GPI | |
3940 | (ANY_EXTEND:GPI | |
3941 | (match_operand:ALLX 0 "register_operand" "r")) | |
4095f9fa | 3942 | (match_operand 1 "aarch64_imm3" "Ui3")) |
58cff58c N |
3943 | (match_operand:GPI 2 "register_operand" "r")))] |
3944 | "" | |
d4febc75 | 3945 | "cmp\\t%<GPI:w>2, %w0, <su>xt<ALLX:size> %1" |
99e6af15 | 3946 | [(set_attr "type" "alus_ext")] |
58cff58c | 3947 | ) |
43e9d192 IB |
3948 | |
3949 | ;; ------------------------------------------------------------------- | |
3950 | ;; Store-flag and conditional select insns | |
3951 | ;; ------------------------------------------------------------------- | |
3952 | ||
3953 | (define_expand "cstore<mode>4" | |
1bbffb87 | 3954 | [(set (match_operand:SI 0 "register_operand") |
43e9d192 | 3955 | (match_operator:SI 1 "aarch64_comparison_operator" |
1bbffb87 DZ |
3956 | [(match_operand:GPI 2 "register_operand") |
3957 | (match_operand:GPI 3 "aarch64_plus_operand")]))] | |
43e9d192 IB |
3958 | "" |
3959 | " | |
3960 | operands[2] = aarch64_gen_compare_reg (GET_CODE (operands[1]), operands[2], | |
3961 | operands[3]); | |
3962 | operands[3] = const0_rtx; | |
3963 | " | |
3964 | ) | |
3965 | ||
cf670503 ZC |
3966 | (define_expand "cstorecc4" |
3967 | [(set (match_operand:SI 0 "register_operand") | |
a44e0ebb | 3968 | (match_operator 1 "aarch64_comparison_operator_mode" |
c8012fbc | 3969 | [(match_operand 2 "cc_register") |
cf670503 ZC |
3970 | (match_operand 3 "const0_operand")]))] |
3971 | "" | |
3972 | "{ | |
f7df4a84 | 3973 | emit_insn (gen_rtx_SET (operands[0], operands[1])); |
cf670503 ZC |
3974 | DONE; |
3975 | }") | |
3976 | ||
3977 | ||
43e9d192 | 3978 | (define_expand "cstore<mode>4" |
1bbffb87 | 3979 | [(set (match_operand:SI 0 "register_operand") |
a44e0ebb | 3980 | (match_operator:SI 1 "aarch64_comparison_operator_mode" |
1bbffb87 DZ |
3981 | [(match_operand:GPF 2 "register_operand") |
3982 | (match_operand:GPF 3 "aarch64_fp_compare_operand")]))] | |
43e9d192 IB |
3983 | "" |
3984 | " | |
3985 | operands[2] = aarch64_gen_compare_reg (GET_CODE (operands[1]), operands[2], | |
3986 | operands[3]); | |
3987 | operands[3] = const0_rtx; | |
3988 | " | |
3989 | ) | |
3990 | ||
714e1b3b | 3991 | (define_insn "aarch64_cstore<mode>" |
43e9d192 | 3992 | [(set (match_operand:ALLI 0 "register_operand" "=r") |
a44e0ebb | 3993 | (match_operator:ALLI 1 "aarch64_comparison_operator_mode" |
43e9d192 IB |
3994 | [(match_operand 2 "cc_register" "") (const_int 0)]))] |
3995 | "" | |
3996 | "cset\\t%<w>0, %m1" | |
99e6af15 | 3997 | [(set_attr "type" "csel")] |
43e9d192 IB |
3998 | ) |
3999 | ||
714e1b3b KT |
4000 | ;; For a 24-bit immediate CST we can optimize the compare for equality |
4001 | ;; and branch sequence from: | |
4002 | ;; mov x0, #imm1 | |
4003 | ;; movk x0, #imm2, lsl 16 /* x0 contains CST. */ | |
4004 | ;; cmp x1, x0 | |
4005 | ;; cset x2, <ne,eq> | |
4006 | ;; into the shorter: | |
4007 | ;; sub x0, x1, #(CST & 0xfff000) | |
4008 | ;; subs x0, x0, #(CST & 0x000fff) | |
4009 | ;; cset x2, <ne, eq>. | |
4010 | (define_insn_and_split "*compare_cstore<mode>_insn" | |
4011 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4012 | (EQL:GPI (match_operand:GPI 1 "register_operand" "r") | |
c8574943 KT |
4013 | (match_operand:GPI 2 "aarch64_imm24" "n"))) |
4014 | (clobber (reg:CC CC_REGNUM))] | |
714e1b3b KT |
4015 | "!aarch64_move_imm (INTVAL (operands[2]), <MODE>mode) |
4016 | && !aarch64_plus_operand (operands[2], <MODE>mode) | |
4017 | && !reload_completed" | |
4018 | "#" | |
4019 | "&& true" | |
4020 | [(const_int 0)] | |
4021 | { | |
4022 | HOST_WIDE_INT lo_imm = UINTVAL (operands[2]) & 0xfff; | |
4023 | HOST_WIDE_INT hi_imm = UINTVAL (operands[2]) & 0xfff000; | |
4024 | rtx tmp = gen_reg_rtx (<MODE>mode); | |
4025 | emit_insn (gen_add<mode>3 (tmp, operands[1], GEN_INT (-hi_imm))); | |
4026 | emit_insn (gen_add<mode>3_compare0 (tmp, tmp, GEN_INT (-lo_imm))); | |
4027 | rtx cc_reg = gen_rtx_REG (CC_NZmode, CC_REGNUM); | |
4028 | rtx cmp_rtx = gen_rtx_fmt_ee (<EQL:CMP>, <MODE>mode, cc_reg, const0_rtx); | |
4029 | emit_insn (gen_aarch64_cstore<mode> (operands[0], cmp_rtx, cc_reg)); | |
4030 | DONE; | |
4031 | } | |
4032 | [(set_attr "type" "csel")] | |
4033 | ) | |
4034 | ||
d0b6bb1b IB |
4035 | ;; zero_extend version of the above |
4036 | (define_insn "*cstoresi_insn_uxtw" | |
4037 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4038 | (zero_extend:DI | |
a44e0ebb | 4039 | (match_operator:SI 1 "aarch64_comparison_operator_mode" |
d0b6bb1b IB |
4040 | [(match_operand 2 "cc_register" "") (const_int 0)])))] |
4041 | "" | |
4042 | "cset\\t%w0, %m1" | |
99e6af15 | 4043 | [(set_attr "type" "csel")] |
d0b6bb1b IB |
4044 | ) |
4045 | ||
889b9412 | 4046 | (define_insn "cstore<mode>_neg" |
43e9d192 | 4047 | [(set (match_operand:ALLI 0 "register_operand" "=r") |
a44e0ebb | 4048 | (neg:ALLI (match_operator:ALLI 1 "aarch64_comparison_operator_mode" |
43e9d192 IB |
4049 | [(match_operand 2 "cc_register" "") (const_int 0)])))] |
4050 | "" | |
4051 | "csetm\\t%<w>0, %m1" | |
99e6af15 | 4052 | [(set_attr "type" "csel")] |
43e9d192 IB |
4053 | ) |
4054 | ||
d0b6bb1b IB |
4055 | ;; zero_extend version of the above |
4056 | (define_insn "*cstoresi_neg_uxtw" | |
4057 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4058 | (zero_extend:DI | |
a44e0ebb | 4059 | (neg:SI (match_operator:SI 1 "aarch64_comparison_operator_mode" |
d0b6bb1b IB |
4060 | [(match_operand 2 "cc_register" "") (const_int 0)]))))] |
4061 | "" | |
4062 | "csetm\\t%w0, %m1" | |
99e6af15 | 4063 | [(set_attr "type" "csel")] |
d0b6bb1b IB |
4064 | ) |
4065 | ||
43e9d192 | 4066 | (define_expand "cmov<mode>6" |
1bbffb87 | 4067 | [(set (match_operand:GPI 0 "register_operand") |
43e9d192 IB |
4068 | (if_then_else:GPI |
4069 | (match_operator 1 "aarch64_comparison_operator" | |
1bbffb87 DZ |
4070 | [(match_operand:GPI 2 "register_operand") |
4071 | (match_operand:GPI 3 "aarch64_plus_operand")]) | |
4072 | (match_operand:GPI 4 "register_operand") | |
4073 | (match_operand:GPI 5 "register_operand")))] | |
43e9d192 IB |
4074 | "" |
4075 | " | |
4076 | operands[2] = aarch64_gen_compare_reg (GET_CODE (operands[1]), operands[2], | |
4077 | operands[3]); | |
4078 | operands[3] = const0_rtx; | |
4079 | " | |
4080 | ) | |
4081 | ||
4082 | (define_expand "cmov<mode>6" | |
1bbffb87 | 4083 | [(set (match_operand:GPF 0 "register_operand") |
43e9d192 IB |
4084 | (if_then_else:GPF |
4085 | (match_operator 1 "aarch64_comparison_operator" | |
1bbffb87 DZ |
4086 | [(match_operand:GPF 2 "register_operand") |
4087 | (match_operand:GPF 3 "aarch64_fp_compare_operand")]) | |
4088 | (match_operand:GPF 4 "register_operand") | |
4089 | (match_operand:GPF 5 "register_operand")))] | |
43e9d192 IB |
4090 | "" |
4091 | " | |
4092 | operands[2] = aarch64_gen_compare_reg (GET_CODE (operands[1]), operands[2], | |
4093 | operands[3]); | |
4094 | operands[3] = const0_rtx; | |
4095 | " | |
4096 | ) | |
4097 | ||
4098 | (define_insn "*cmov<mode>_insn" | |
06b1198f | 4099 | [(set (match_operand:ALLI 0 "register_operand" "=r,r,r,r,r,r,r") |
43e9d192 IB |
4100 | (if_then_else:ALLI |
4101 | (match_operator 1 "aarch64_comparison_operator" | |
4102 | [(match_operand 2 "cc_register" "") (const_int 0)]) | |
06b1198f IB |
4103 | (match_operand:ALLI 3 "aarch64_reg_zero_or_m1_or_1" "rZ,rZ,UsM,rZ,Ui1,UsM,Ui1") |
4104 | (match_operand:ALLI 4 "aarch64_reg_zero_or_m1_or_1" "rZ,UsM,rZ,Ui1,rZ,UsM,Ui1")))] | |
4105 | "!((operands[3] == const1_rtx && operands[4] == constm1_rtx) | |
4106 | || (operands[3] == constm1_rtx && operands[4] == const1_rtx))" | |
4107 | ;; Final two alternatives should be unreachable, but included for completeness | |
43e9d192 IB |
4108 | "@ |
4109 | csel\\t%<w>0, %<w>3, %<w>4, %m1 | |
4110 | csinv\\t%<w>0, %<w>3, <w>zr, %m1 | |
4111 | csinv\\t%<w>0, %<w>4, <w>zr, %M1 | |
06b1198f IB |
4112 | csinc\\t%<w>0, %<w>3, <w>zr, %m1 |
4113 | csinc\\t%<w>0, %<w>4, <w>zr, %M1 | |
4114 | mov\\t%<w>0, -1 | |
4115 | mov\\t%<w>0, 1" | |
04e34811 | 4116 | [(set_attr "type" "csel, csel, csel, csel, csel, mov_imm, mov_imm")] |
43e9d192 IB |
4117 | ) |
4118 | ||
d0b6bb1b IB |
4119 | ;; zero_extend version of above |
4120 | (define_insn "*cmovsi_insn_uxtw" | |
4121 | [(set (match_operand:DI 0 "register_operand" "=r,r,r,r,r,r,r") | |
4122 | (zero_extend:DI | |
4123 | (if_then_else:SI | |
4124 | (match_operator 1 "aarch64_comparison_operator" | |
4125 | [(match_operand 2 "cc_register" "") (const_int 0)]) | |
4126 | (match_operand:SI 3 "aarch64_reg_zero_or_m1_or_1" "rZ,rZ,UsM,rZ,Ui1,UsM,Ui1") | |
4127 | (match_operand:SI 4 "aarch64_reg_zero_or_m1_or_1" "rZ,UsM,rZ,Ui1,rZ,UsM,Ui1"))))] | |
4128 | "!((operands[3] == const1_rtx && operands[4] == constm1_rtx) | |
4129 | || (operands[3] == constm1_rtx && operands[4] == const1_rtx))" | |
4130 | ;; Final two alternatives should be unreachable, but included for completeness | |
4131 | "@ | |
4132 | csel\\t%w0, %w3, %w4, %m1 | |
4133 | csinv\\t%w0, %w3, wzr, %m1 | |
4134 | csinv\\t%w0, %w4, wzr, %M1 | |
4135 | csinc\\t%w0, %w3, wzr, %m1 | |
4136 | csinc\\t%w0, %w4, wzr, %M1 | |
4137 | mov\\t%w0, -1 | |
4138 | mov\\t%w0, 1" | |
04e34811 | 4139 | [(set_attr "type" "csel, csel, csel, csel, csel, mov_imm, mov_imm")] |
d0b6bb1b IB |
4140 | ) |
4141 | ||
1d3e6ed3 RL |
4142 | (define_insn "*cmovdi_insn_uxtw" |
4143 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4144 | (if_then_else:DI | |
4145 | (match_operator 1 "aarch64_comparison_operator" | |
4146 | [(match_operand 2 "cc_register" "") (const_int 0)]) | |
4147 | (zero_extend:DI (match_operand:SI 3 "register_operand" "r")) | |
4148 | (zero_extend:DI (match_operand:SI 4 "register_operand" "r"))))] | |
4149 | "" | |
4150 | "csel\\t%w0, %w3, %w4, %m1" | |
4151 | [(set_attr "type" "csel")] | |
4152 | ) | |
4153 | ||
43e9d192 IB |
4154 | (define_insn "*cmov<mode>_insn" |
4155 | [(set (match_operand:GPF 0 "register_operand" "=w") | |
4156 | (if_then_else:GPF | |
4157 | (match_operator 1 "aarch64_comparison_operator" | |
4158 | [(match_operand 2 "cc_register" "") (const_int 0)]) | |
4159 | (match_operand:GPF 3 "register_operand" "w") | |
4160 | (match_operand:GPF 4 "register_operand" "w")))] | |
4161 | "TARGET_FLOAT" | |
4162 | "fcsel\\t%<s>0, %<s>3, %<s>4, %m1" | |
99e6af15 | 4163 | [(set_attr "type" "fcsel")] |
43e9d192 IB |
4164 | ) |
4165 | ||
4166 | (define_expand "mov<mode>cc" | |
1bbffb87 DZ |
4167 | [(set (match_operand:ALLI 0 "register_operand") |
4168 | (if_then_else:ALLI (match_operand 1 "aarch64_comparison_operator") | |
4169 | (match_operand:ALLI 2 "register_operand") | |
4170 | (match_operand:ALLI 3 "register_operand")))] | |
43e9d192 IB |
4171 | "" |
4172 | { | |
c8012fbc | 4173 | rtx ccreg; |
43e9d192 IB |
4174 | enum rtx_code code = GET_CODE (operands[1]); |
4175 | ||
4176 | if (code == UNEQ || code == LTGT) | |
4177 | FAIL; | |
4178 | ||
c8012fbc WD |
4179 | ccreg = aarch64_gen_compare_reg (code, XEXP (operands[1], 0), |
4180 | XEXP (operands[1], 1)); | |
4181 | operands[1] = gen_rtx_fmt_ee (code, VOIDmode, ccreg, const0_rtx); | |
43e9d192 IB |
4182 | } |
4183 | ) | |
4184 | ||
4185 | (define_expand "mov<GPF:mode><GPI:mode>cc" | |
1bbffb87 DZ |
4186 | [(set (match_operand:GPI 0 "register_operand") |
4187 | (if_then_else:GPI (match_operand 1 "aarch64_comparison_operator") | |
4188 | (match_operand:GPF 2 "register_operand") | |
4189 | (match_operand:GPF 3 "register_operand")))] | |
43e9d192 IB |
4190 | "" |
4191 | { | |
4192 | rtx ccreg; | |
4193 | enum rtx_code code = GET_CODE (operands[1]); | |
4194 | ||
4195 | if (code == UNEQ || code == LTGT) | |
4196 | FAIL; | |
4197 | ||
4198 | ccreg = aarch64_gen_compare_reg (code, XEXP (operands[1], 0), | |
4199 | XEXP (operands[1], 1)); | |
4200 | operands[1] = gen_rtx_fmt_ee (code, VOIDmode, ccreg, const0_rtx); | |
4201 | } | |
4202 | ) | |
4203 | ||
d3a3182b | 4204 | (define_expand "mov<mode>cc" |
1bbffb87 DZ |
4205 | [(set (match_operand:GPF 0 "register_operand") |
4206 | (if_then_else:GPF (match_operand 1 "aarch64_comparison_operator") | |
4207 | (match_operand:GPF 2 "register_operand") | |
4208 | (match_operand:GPF 3 "register_operand")))] | |
d3a3182b ZC |
4209 | "" |
4210 | { | |
4211 | rtx ccreg; | |
4212 | enum rtx_code code = GET_CODE (operands[1]); | |
4213 | ||
4214 | if (code == UNEQ || code == LTGT) | |
4215 | FAIL; | |
4216 | ||
4217 | ccreg = aarch64_gen_compare_reg (code, XEXP (operands[1], 0), | |
4218 | XEXP (operands[1], 1)); | |
4219 | operands[1] = gen_rtx_fmt_ee (code, VOIDmode, ccreg, const0_rtx); | |
4220 | } | |
4221 | ) | |
4222 | ||
3204ac98 | 4223 | (define_expand "<neg_not_op><mode>cc" |
1bbffb87 DZ |
4224 | [(set (match_operand:GPI 0 "register_operand") |
4225 | (if_then_else:GPI (match_operand 1 "aarch64_comparison_operator") | |
4226 | (NEG_NOT:GPI (match_operand:GPI 2 "register_operand")) | |
4227 | (match_operand:GPI 3 "register_operand")))] | |
3204ac98 KT |
4228 | "" |
4229 | { | |
4230 | rtx ccreg; | |
4231 | enum rtx_code code = GET_CODE (operands[1]); | |
4232 | ||
4233 | if (code == UNEQ || code == LTGT) | |
4234 | FAIL; | |
4235 | ||
4236 | ccreg = aarch64_gen_compare_reg (code, XEXP (operands[1], 0), | |
4237 | XEXP (operands[1], 1)); | |
4238 | operands[1] = gen_rtx_fmt_ee (code, VOIDmode, ccreg, const0_rtx); | |
4239 | } | |
4240 | ) | |
5d357f26 KT |
4241 | |
4242 | ;; CRC32 instructions. | |
4243 | (define_insn "aarch64_<crc_variant>" | |
4244 | [(set (match_operand:SI 0 "register_operand" "=r") | |
4245 | (unspec:SI [(match_operand:SI 1 "register_operand" "r") | |
4246 | (match_operand:<crc_mode> 2 "register_operand" "r")] | |
4247 | CRC))] | |
4248 | "TARGET_CRC32" | |
4249 | { | |
6a70badb | 4250 | if (GET_MODE_BITSIZE (<crc_mode>mode) >= 64) |
5d357f26 KT |
4251 | return "<crc_variant>\\t%w0, %w1, %x2"; |
4252 | else | |
4253 | return "<crc_variant>\\t%w0, %w1, %w2"; | |
4254 | } | |
4255 | [(set_attr "type" "crc")] | |
4256 | ) | |
4257 | ||
43e9d192 IB |
4258 | (define_insn "*csinc2<mode>_insn" |
4259 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
cd5660ab KT |
4260 | (plus:GPI (match_operand 2 "aarch64_comparison_operation" "") |
4261 | (match_operand:GPI 1 "register_operand" "r")))] | |
43e9d192 | 4262 | "" |
49f88e43 | 4263 | "cinc\\t%<w>0, %<w>1, %m2" |
99e6af15 JG |
4264 | [(set_attr "type" "csel")] |
4265 | ) | |
43e9d192 IB |
4266 | |
4267 | (define_insn "csinc3<mode>_insn" | |
4268 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4269 | (if_then_else:GPI | |
cd5660ab KT |
4270 | (match_operand 1 "aarch64_comparison_operation" "") |
4271 | (plus:GPI (match_operand:GPI 2 "register_operand" "r") | |
43e9d192 | 4272 | (const_int 1)) |
cd5660ab | 4273 | (match_operand:GPI 3 "aarch64_reg_or_zero" "rZ")))] |
43e9d192 | 4274 | "" |
cd5660ab | 4275 | "csinc\\t%<w>0, %<w>3, %<w>2, %M1" |
99e6af15 | 4276 | [(set_attr "type" "csel")] |
43e9d192 IB |
4277 | ) |
4278 | ||
4279 | (define_insn "*csinv3<mode>_insn" | |
4280 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4281 | (if_then_else:GPI | |
cd5660ab KT |
4282 | (match_operand 1 "aarch64_comparison_operation" "") |
4283 | (not:GPI (match_operand:GPI 2 "register_operand" "r")) | |
4284 | (match_operand:GPI 3 "aarch64_reg_or_zero" "rZ")))] | |
43e9d192 | 4285 | "" |
cd5660ab | 4286 | "csinv\\t%<w>0, %<w>3, %<w>2, %M1" |
99e6af15 JG |
4287 | [(set_attr "type" "csel")] |
4288 | ) | |
43e9d192 | 4289 | |
db95ac77 | 4290 | (define_insn "*cs<neg_not_cs>3_uxtw_insn4" |
0f6ca79c RL |
4291 | [(set (match_operand:DI 0 "register_operand" "=r") |
4292 | (zero_extend:DI | |
4293 | (if_then_else:SI | |
4294 | (match_operand 1 "aarch64_comparison_operation" "") | |
db95ac77 | 4295 | (NEG_NOT:SI (match_operand:SI 2 "register_operand" "r")) |
0f6ca79c RL |
4296 | (match_operand:SI 3 "aarch64_reg_or_zero" "rZ"))))] |
4297 | "" | |
db95ac77 | 4298 | "cs<neg_not_cs>\\t%w0, %w3, %w2, %M1" |
0f6ca79c RL |
4299 | [(set_attr "type" "csel")] |
4300 | ) | |
4301 | ||
fc8b587d | 4302 | (define_insn "csneg3<mode>_insn" |
43e9d192 IB |
4303 | [(set (match_operand:GPI 0 "register_operand" "=r") |
4304 | (if_then_else:GPI | |
cd5660ab KT |
4305 | (match_operand 1 "aarch64_comparison_operation" "") |
4306 | (neg:GPI (match_operand:GPI 2 "register_operand" "r")) | |
4307 | (match_operand:GPI 3 "aarch64_reg_or_zero" "rZ")))] | |
43e9d192 | 4308 | "" |
cd5660ab | 4309 | "csneg\\t%<w>0, %<w>3, %<w>2, %M1" |
99e6af15 | 4310 | [(set_attr "type" "csel")] |
d572ad49 AC |
4311 | ) |
4312 | ||
4313 | (define_insn "*csinv3_uxtw_insn1" | |
4314 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4315 | (if_then_else:DI | |
4316 | (match_operand 1 "aarch64_comparison_operation" "") | |
4317 | (zero_extend:DI | |
4318 | (match_operand:SI 2 "register_operand" "r")) | |
4319 | (zero_extend:DI | |
4320 | (NEG_NOT:SI (match_operand:SI 3 "register_operand" "r")))))] | |
4321 | "" | |
4322 | "cs<neg_not_cs>\\t%w0, %w2, %w3, %m1" | |
4323 | [(set_attr "type" "csel")] | |
4324 | ) | |
4325 | ||
4326 | (define_insn "*csinv3_uxtw_insn2" | |
4327 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4328 | (if_then_else:DI | |
4329 | (match_operand 1 "aarch64_comparison_operation" "") | |
4330 | (zero_extend:DI | |
4331 | (NEG_NOT:SI (match_operand:SI 2 "register_operand" "r"))) | |
4332 | (zero_extend:DI | |
4333 | (match_operand:SI 3 "register_operand" "r"))))] | |
4334 | "" | |
4335 | "cs<neg_not_cs>\\t%w0, %w3, %w2, %M1" | |
4336 | [(set_attr "type" "csel")] | |
4337 | ) | |
4338 | ||
4339 | (define_insn "*csinv3_uxtw_insn3" | |
4340 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4341 | (if_then_else:DI | |
4342 | (match_operand 1 "aarch64_comparison_operation" "") | |
4343 | (zero_extend:DI | |
4344 | (NEG_NOT:SI (match_operand:SI 2 "register_operand" "r"))) | |
4345 | (const_int 0)))] | |
4346 | "" | |
4347 | "cs<neg_not_cs>\\t%w0, wzr, %w2, %M1" | |
4348 | [(set_attr "type" "csel")] | |
99e6af15 | 4349 | ) |
43e9d192 | 4350 | |
7cfb4d93 RS |
4351 | ;; If X can be loaded by a single CNT[BHWD] instruction, |
4352 | ;; | |
4353 | ;; A = UMAX (B, X) | |
4354 | ;; | |
4355 | ;; is equivalent to: | |
4356 | ;; | |
4357 | ;; TMP = UQDEC[BHWD] (B, X) | |
4358 | ;; A = TMP + X | |
4359 | ;; | |
4360 | ;; Defining the pattern this way means that: | |
4361 | ;; | |
4362 | ;; A = UMAX (B, X) - X | |
4363 | ;; | |
4364 | ;; becomes: | |
4365 | ;; | |
4366 | ;; TMP1 = UQDEC[BHWD] (B, X) | |
4367 | ;; TMP2 = TMP1 + X | |
4368 | ;; A = TMP2 - X | |
4369 | ;; | |
4370 | ;; which combine can optimize to: | |
4371 | ;; | |
4372 | ;; A = UQDEC[BHWD] (B, X) | |
4373 | ;; | |
4374 | ;; We don't use match_operand predicates because the order of the operands | |
4375 | ;; can vary: the CNT[BHWD] constant will come first if the other operand is | |
4376 | ;; a simpler constant (such as a CONST_INT), otherwise it will come second. | |
4377 | (define_expand "umax<mode>3" | |
4378 | [(set (match_operand:GPI 0 "register_operand") | |
4379 | (umax:GPI (match_operand:GPI 1 "") | |
4380 | (match_operand:GPI 2 "")))] | |
4381 | "TARGET_SVE" | |
4382 | { | |
4383 | if (aarch64_sve_cnt_immediate (operands[1], <MODE>mode)) | |
4384 | std::swap (operands[1], operands[2]); | |
4385 | else if (!aarch64_sve_cnt_immediate (operands[2], <MODE>mode)) | |
4386 | FAIL; | |
4387 | rtx temp = gen_reg_rtx (<MODE>mode); | |
4388 | operands[1] = force_reg (<MODE>mode, operands[1]); | |
4389 | emit_insn (gen_aarch64_uqdec<mode> (temp, operands[1], operands[2])); | |
4390 | emit_insn (gen_add<mode>3 (operands[0], temp, operands[2])); | |
4391 | DONE; | |
4392 | } | |
4393 | ) | |
4394 | ||
4395 | ;; Saturating unsigned subtraction of a CNT[BHWD] immediate. | |
4396 | (define_insn "aarch64_uqdec<mode>" | |
4397 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4398 | (minus:GPI | |
4399 | (umax:GPI (match_operand:GPI 1 "register_operand" "0") | |
4400 | (match_operand:GPI 2 "aarch64_sve_cnt_immediate" "Usv")) | |
4401 | (match_dup 2)))] | |
4402 | "TARGET_SVE" | |
4403 | { | |
4404 | return aarch64_output_sve_cnt_immediate ("uqdec", "%<w>0", operands[2]); | |
4405 | } | |
4406 | ) | |
4407 | ||
43e9d192 IB |
4408 | ;; ------------------------------------------------------------------- |
4409 | ;; Logical operations | |
4410 | ;; ------------------------------------------------------------------- | |
4411 | ||
43fd192f MC |
4412 | |
4413 | (define_insn_and_split "*aarch64_and<mode>_imm2" | |
4414 | [(set (match_operand:GPI 0 "register_operand" "=rk") | |
4415 | (and:GPI (match_operand:GPI 1 "register_operand" "%r") | |
4416 | (match_operand:GPI 2 "aarch64_logical_and_immediate" "<lconst2>")))] | |
4417 | "" | |
4418 | "#" | |
4419 | "true" | |
4420 | [(const_int 0)] | |
4421 | { | |
4422 | HOST_WIDE_INT val = INTVAL (operands[2]); | |
4423 | rtx imm1 = GEN_INT (aarch64_and_split_imm1 (val)); | |
4424 | rtx imm2 = GEN_INT (aarch64_and_split_imm2 (val)); | |
4425 | ||
4426 | emit_insn (gen_and<mode>3 (operands[0], operands[1], imm1)); | |
4427 | emit_insn (gen_and<mode>3 (operands[0], operands[0], imm2)); | |
4428 | DONE; | |
4429 | } | |
4430 | ) | |
4431 | ||
43e9d192 | 4432 | (define_insn "<optab><mode>3" |
fe82d1f2 AL |
4433 | [(set (match_operand:GPI 0 "register_operand" "=r,rk,w") |
4434 | (LOGICAL:GPI (match_operand:GPI 1 "register_operand" "%r,r,w") | |
4435 | (match_operand:GPI 2 "aarch64_logical_operand" "r,<lconst>,w")))] | |
43e9d192 | 4436 | "" |
fe82d1f2 AL |
4437 | "@ |
4438 | <logical>\\t%<w>0, %<w>1, %<w>2 | |
e2b691c4 | 4439 | <logical>\\t%<w>0, %<w>1, %2 |
fe82d1f2 AL |
4440 | <logical>\\t%0.<Vbtype>, %1.<Vbtype>, %2.<Vbtype>" |
4441 | [(set_attr "type" "logic_reg,logic_imm,neon_logic") | |
488461d8 | 4442 | (set_attr "arch" "*,*,simd")] |
99e6af15 | 4443 | ) |
43e9d192 | 4444 | |
d0b6bb1b IB |
4445 | ;; zero_extend version of above |
4446 | (define_insn "*<optab>si3_uxtw" | |
4447 | [(set (match_operand:DI 0 "register_operand" "=r,rk") | |
4448 | (zero_extend:DI | |
4449 | (LOGICAL:SI (match_operand:SI 1 "register_operand" "%r,r") | |
4450 | (match_operand:SI 2 "aarch64_logical_operand" "r,K"))))] | |
4451 | "" | |
954224d7 WD |
4452 | "@ |
4453 | <logical>\\t%w0, %w1, %w2 | |
4454 | <logical>\\t%w0, %w1, %2" | |
99e6af15 JG |
4455 | [(set_attr "type" "logic_reg,logic_imm")] |
4456 | ) | |
d0b6bb1b | 4457 | |
2a1d1991 N |
4458 | (define_insn "*and<mode>3_compare0" |
4459 | [(set (reg:CC_NZ CC_REGNUM) | |
4460 | (compare:CC_NZ | |
4461 | (and:GPI (match_operand:GPI 1 "register_operand" "%r,r") | |
4462 | (match_operand:GPI 2 "aarch64_logical_operand" "r,<lconst>")) | |
4463 | (const_int 0))) | |
4464 | (set (match_operand:GPI 0 "register_operand" "=r,r") | |
4465 | (and:GPI (match_dup 1) (match_dup 2)))] | |
4466 | "" | |
954224d7 WD |
4467 | "@ |
4468 | ands\\t%<w>0, %<w>1, %<w>2 | |
4469 | ands\\t%<w>0, %<w>1, %2" | |
99e6af15 | 4470 | [(set_attr "type" "logics_reg,logics_imm")] |
2a1d1991 N |
4471 | ) |
4472 | ||
4473 | ;; zero_extend version of above | |
4474 | (define_insn "*andsi3_compare0_uxtw" | |
4475 | [(set (reg:CC_NZ CC_REGNUM) | |
4476 | (compare:CC_NZ | |
4477 | (and:SI (match_operand:SI 1 "register_operand" "%r,r") | |
4478 | (match_operand:SI 2 "aarch64_logical_operand" "r,K")) | |
4479 | (const_int 0))) | |
4480 | (set (match_operand:DI 0 "register_operand" "=r,r") | |
4481 | (zero_extend:DI (and:SI (match_dup 1) (match_dup 2))))] | |
4482 | "" | |
954224d7 WD |
4483 | "@ |
4484 | ands\\t%w0, %w1, %w2 | |
4485 | ands\\t%w0, %w1, %2" | |
99e6af15 | 4486 | [(set_attr "type" "logics_reg,logics_imm")] |
2a1d1991 N |
4487 | ) |
4488 | ||
4489 | (define_insn "*and_<SHIFT:optab><mode>3_compare0" | |
4490 | [(set (reg:CC_NZ CC_REGNUM) | |
4491 | (compare:CC_NZ | |
4492 | (and:GPI (SHIFT:GPI | |
4493 | (match_operand:GPI 1 "register_operand" "r") | |
4494 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n")) | |
4495 | (match_operand:GPI 3 "register_operand" "r")) | |
4496 | (const_int 0))) | |
4497 | (set (match_operand:GPI 0 "register_operand" "=r") | |
4498 | (and:GPI (SHIFT:GPI (match_dup 1) (match_dup 2)) (match_dup 3)))] | |
4499 | "" | |
4500 | "ands\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2" | |
99e6af15 | 4501 | [(set_attr "type" "logics_shift_imm")] |
2a1d1991 N |
4502 | ) |
4503 | ||
4504 | ;; zero_extend version of above | |
4505 | (define_insn "*and_<SHIFT:optab>si3_compare0_uxtw" | |
4506 | [(set (reg:CC_NZ CC_REGNUM) | |
4507 | (compare:CC_NZ | |
4508 | (and:SI (SHIFT:SI | |
4509 | (match_operand:SI 1 "register_operand" "r") | |
4510 | (match_operand:QI 2 "aarch64_shift_imm_si" "n")) | |
4511 | (match_operand:SI 3 "register_operand" "r")) | |
4512 | (const_int 0))) | |
4513 | (set (match_operand:DI 0 "register_operand" "=r") | |
4514 | (zero_extend:DI (and:SI (SHIFT:SI (match_dup 1) (match_dup 2)) | |
4515 | (match_dup 3))))] | |
4516 | "" | |
4517 | "ands\\t%w0, %w3, %w1, <SHIFT:shift> %2" | |
99e6af15 | 4518 | [(set_attr "type" "logics_shift_imm")] |
2a1d1991 N |
4519 | ) |
4520 | ||
43e9d192 IB |
4521 | (define_insn "*<LOGICAL:optab>_<SHIFT:optab><mode>3" |
4522 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4523 | (LOGICAL:GPI (SHIFT:GPI | |
4524 | (match_operand:GPI 1 "register_operand" "r") | |
4525 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n")) | |
4526 | (match_operand:GPI 3 "register_operand" "r")))] | |
4527 | "" | |
4528 | "<LOGICAL:logical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2" | |
99e6af15 JG |
4529 | [(set_attr "type" "logic_shift_imm")] |
4530 | ) | |
43e9d192 | 4531 | |
39d23b79 JJ |
4532 | (define_split |
4533 | [(set (match_operand:GPI 0 "register_operand") | |
4534 | (LOGICAL:GPI | |
4535 | (and:GPI (ashift:GPI (match_operand:GPI 1 "register_operand") | |
4536 | (match_operand:QI 2 "aarch64_shift_imm_<mode>")) | |
4537 | (match_operand:GPI 3 "const_int_operand")) | |
4538 | (zero_extend:GPI (match_operand 4 "register_operand"))))] | |
4539 | "can_create_pseudo_p () | |
4540 | && ((paradoxical_subreg_p (operands[1]) | |
4541 | && rtx_equal_p (SUBREG_REG (operands[1]), operands[4])) | |
4542 | || (REG_P (operands[1]) | |
4543 | && REG_P (operands[4]) | |
4544 | && REGNO (operands[1]) == REGNO (operands[4]))) | |
4545 | && (trunc_int_for_mode (GET_MODE_MASK (GET_MODE (operands[4])) | |
4546 | << INTVAL (operands[2]), <MODE>mode) | |
4547 | == INTVAL (operands[3]))" | |
4548 | [(set (match_dup 5) (zero_extend:GPI (match_dup 4))) | |
4549 | (set (match_dup 0) (LOGICAL:GPI (ashift:GPI (match_dup 5) (match_dup 2)) | |
4550 | (match_dup 5)))] | |
4551 | "operands[5] = gen_reg_rtx (<MODE>mode);" | |
4552 | ) | |
4553 | ||
4554 | (define_split | |
4555 | [(set (match_operand:GPI 0 "register_operand") | |
4556 | (LOGICAL:GPI | |
4557 | (and:GPI (ashift:GPI (match_operand:GPI 1 "register_operand") | |
4558 | (match_operand:QI 2 "aarch64_shift_imm_<mode>")) | |
4559 | (match_operand:GPI 4 "const_int_operand")) | |
4560 | (and:GPI (match_dup 1) (match_operand:GPI 3 "const_int_operand"))))] | |
4561 | "can_create_pseudo_p () | |
4562 | && pow2_or_zerop (UINTVAL (operands[3]) + 1) | |
4563 | && (trunc_int_for_mode (UINTVAL (operands[3]) | |
4564 | << INTVAL (operands[2]), <MODE>mode) | |
4565 | == INTVAL (operands[4]))" | |
4566 | [(set (match_dup 5) (and:GPI (match_dup 1) (match_dup 3))) | |
4567 | (set (match_dup 0) (LOGICAL:GPI (ashift:GPI (match_dup 5) (match_dup 2)) | |
4568 | (match_dup 5)))] | |
4569 | "operands[5] = gen_reg_rtx (<MODE>mode);" | |
4570 | ) | |
4571 | ||
4572 | (define_split | |
4573 | [(set (match_operand:GPI 0 "register_operand") | |
4574 | (LOGICAL:GPI | |
4575 | (ashift:GPI (sign_extend:GPI (match_operand 1 "register_operand")) | |
4576 | (match_operand:QI 2 "aarch64_shift_imm_<mode>")) | |
4577 | (sign_extend:GPI (match_dup 1))))] | |
4578 | "can_create_pseudo_p ()" | |
4579 | [(set (match_dup 3) (sign_extend:GPI (match_dup 1))) | |
4580 | (set (match_dup 0) (LOGICAL:GPI (ashift:GPI (match_dup 3) (match_dup 2)) | |
4581 | (match_dup 3)))] | |
4582 | "operands[3] = gen_reg_rtx (<MODE>mode);" | |
4583 | ) | |
4584 | ||
57b77d46 RE |
4585 | (define_insn "*<optab>_rol<mode>3" |
4586 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4587 | (LOGICAL:GPI (rotate:GPI | |
4588 | (match_operand:GPI 1 "register_operand" "r") | |
4589 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n")) | |
4590 | (match_operand:GPI 3 "register_operand" "r")))] | |
4591 | "" | |
5051c121 | 4592 | "<logical>\\t%<w>0, %<w>3, %<w>1, ror #(<sizen> - %2)" |
57b77d46 RE |
4593 | [(set_attr "type" "logic_shift_imm")] |
4594 | ) | |
4595 | ||
4596 | ;; zero_extend versions of above | |
d0b6bb1b IB |
4597 | (define_insn "*<LOGICAL:optab>_<SHIFT:optab>si3_uxtw" |
4598 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4599 | (zero_extend:DI | |
4600 | (LOGICAL:SI (SHIFT:SI | |
4601 | (match_operand:SI 1 "register_operand" "r") | |
4602 | (match_operand:QI 2 "aarch64_shift_imm_si" "n")) | |
4603 | (match_operand:SI 3 "register_operand" "r"))))] | |
4604 | "" | |
4605 | "<LOGICAL:logical>\\t%w0, %w3, %w1, <SHIFT:shift> %2" | |
99e6af15 JG |
4606 | [(set_attr "type" "logic_shift_imm")] |
4607 | ) | |
d0b6bb1b | 4608 | |
57b77d46 RE |
4609 | (define_insn "*<optab>_rolsi3_uxtw" |
4610 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4611 | (zero_extend:DI | |
4612 | (LOGICAL:SI (rotate:SI | |
4613 | (match_operand:SI 1 "register_operand" "r") | |
4614 | (match_operand:QI 2 "aarch64_shift_imm_si" "n")) | |
4615 | (match_operand:SI 3 "register_operand" "r"))))] | |
4616 | "" | |
5051c121 | 4617 | "<logical>\\t%w0, %w3, %w1, ror #(32 - %2)" |
57b77d46 RE |
4618 | [(set_attr "type" "logic_shift_imm")] |
4619 | ) | |
4620 | ||
43e9d192 | 4621 | (define_insn "one_cmpl<mode>2" |
fe82d1f2 AL |
4622 | [(set (match_operand:GPI 0 "register_operand" "=r,w") |
4623 | (not:GPI (match_operand:GPI 1 "register_operand" "r,w")))] | |
43e9d192 | 4624 | "" |
fe82d1f2 AL |
4625 | "@ |
4626 | mvn\\t%<w>0, %<w>1 | |
4627 | mvn\\t%0.8b, %1.8b" | |
4628 | [(set_attr "type" "logic_reg,neon_logic") | |
488461d8 | 4629 | (set_attr "arch" "*,simd")] |
99e6af15 | 4630 | ) |
43e9d192 | 4631 | |
1bd3a8af AC |
4632 | (define_insn "*one_cmpl_zero_extend" |
4633 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4634 | (zero_extend:DI | |
4635 | (not:SI (match_operand:SI 1 "register_operand" "r"))))] | |
4636 | "" | |
4637 | "mvn\\t%w0, %w1" | |
4638 | [(set_attr "type" "logic_reg")] | |
4639 | ) | |
4640 | ||
43e9d192 IB |
4641 | (define_insn "*one_cmpl_<optab><mode>2" |
4642 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4643 | (not:GPI (SHIFT:GPI (match_operand:GPI 1 "register_operand" "r") | |
4644 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))))] | |
4645 | "" | |
4646 | "mvn\\t%<w>0, %<w>1, <shift> %2" | |
99e6af15 JG |
4647 | [(set_attr "type" "logic_shift_imm")] |
4648 | ) | |
43e9d192 | 4649 | |
84be6032 AL |
4650 | ;; Binary logical operators negating one operand, i.e. (a & !b), (a | !b). |
4651 | ||
4652 | (define_insn "*<NLOGICAL:optab>_one_cmpl<mode>3" | |
4653 | [(set (match_operand:GPI 0 "register_operand" "=r,w") | |
4654 | (NLOGICAL:GPI (not:GPI (match_operand:GPI 1 "register_operand" "r,w")) | |
4655 | (match_operand:GPI 2 "register_operand" "r,w")))] | |
43e9d192 | 4656 | "" |
84be6032 AL |
4657 | "@ |
4658 | <NLOGICAL:nlogical>\\t%<w>0, %<w>2, %<w>1 | |
4659 | <NLOGICAL:nlogical>\\t%0.<Vbtype>, %2.<Vbtype>, %1.<Vbtype>" | |
4660 | [(set_attr "type" "logic_reg,neon_logic") | |
488461d8 | 4661 | (set_attr "arch" "*,simd")] |
84be6032 AL |
4662 | ) |
4663 | ||
8409e468 KT |
4664 | (define_insn "*<NLOGICAL:optab>_one_cmplsidi3_ze" |
4665 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4666 | (zero_extend:DI | |
4667 | (NLOGICAL:SI (not:SI (match_operand:SI 1 "register_operand" "r")) | |
4668 | (match_operand:SI 2 "register_operand" "r"))))] | |
4669 | "" | |
4670 | "<NLOGICAL:nlogical>\\t%w0, %w2, %w1" | |
4671 | [(set_attr "type" "logic_reg")] | |
4672 | ) | |
4673 | ||
4674 | (define_insn "*xor_one_cmplsidi3_ze" | |
4675 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4676 | (zero_extend:DI | |
4677 | (not:SI (xor:SI (match_operand:SI 1 "register_operand" "r") | |
4678 | (match_operand:SI 2 "register_operand" "r")))))] | |
4679 | "" | |
4680 | "eon\\t%w0, %w1, %w2" | |
4681 | [(set_attr "type" "logic_reg")] | |
4682 | ) | |
4683 | ||
84be6032 AL |
4684 | ;; (xor (not a) b) is simplify_rtx-ed down to (not (xor a b)). |
4685 | ;; eon does not operate on SIMD registers so the vector variant must be split. | |
4686 | (define_insn_and_split "*xor_one_cmpl<mode>3" | |
4687 | [(set (match_operand:GPI 0 "register_operand" "=r,w") | |
4688 | (not:GPI (xor:GPI (match_operand:GPI 1 "register_operand" "r,?w") | |
4689 | (match_operand:GPI 2 "register_operand" "r,w"))))] | |
4690 | "" | |
eeaccc07 AL |
4691 | "@ |
4692 | eon\\t%<w>0, %<w>1, %<w>2 | |
4693 | #" | |
4694 | "reload_completed && FP_REGNUM_P (REGNO (operands[0]))" ;; For SIMD registers. | |
84be6032 AL |
4695 | [(set (match_operand:GPI 0 "register_operand" "=w") |
4696 | (xor:GPI (match_operand:GPI 1 "register_operand" "w") | |
4697 | (match_operand:GPI 2 "register_operand" "w"))) | |
4698 | (set (match_dup 0) (not:GPI (match_dup 0)))] | |
4699 | "" | |
4700 | [(set_attr "type" "logic_reg,multiple") | |
488461d8 | 4701 | (set_attr "arch" "*,simd")] |
99e6af15 | 4702 | ) |
43e9d192 | 4703 | |
add3c965 IB |
4704 | (define_insn "*and_one_cmpl<mode>3_compare0" |
4705 | [(set (reg:CC_NZ CC_REGNUM) | |
4706 | (compare:CC_NZ | |
4707 | (and:GPI (not:GPI | |
4708 | (match_operand:GPI 1 "register_operand" "r")) | |
4709 | (match_operand:GPI 2 "register_operand" "r")) | |
4710 | (const_int 0))) | |
4711 | (set (match_operand:GPI 0 "register_operand" "=r") | |
4712 | (and:GPI (not:GPI (match_dup 1)) (match_dup 2)))] | |
4713 | "" | |
4714 | "bics\\t%<w>0, %<w>2, %<w>1" | |
99e6af15 JG |
4715 | [(set_attr "type" "logics_reg")] |
4716 | ) | |
add3c965 IB |
4717 | |
4718 | ;; zero_extend version of above | |
4719 | (define_insn "*and_one_cmplsi3_compare0_uxtw" | |
4720 | [(set (reg:CC_NZ CC_REGNUM) | |
4721 | (compare:CC_NZ | |
4722 | (and:SI (not:SI | |
4723 | (match_operand:SI 1 "register_operand" "r")) | |
4724 | (match_operand:SI 2 "register_operand" "r")) | |
4725 | (const_int 0))) | |
4726 | (set (match_operand:DI 0 "register_operand" "=r") | |
4727 | (zero_extend:DI (and:SI (not:SI (match_dup 1)) (match_dup 2))))] | |
4728 | "" | |
4729 | "bics\\t%w0, %w2, %w1" | |
99e6af15 JG |
4730 | [(set_attr "type" "logics_reg")] |
4731 | ) | |
add3c965 | 4732 | |
cd8fc5ee AV |
4733 | (define_insn "*and_one_cmpl<mode>3_compare0_no_reuse" |
4734 | [(set (reg:CC_NZ CC_REGNUM) | |
4735 | (compare:CC_NZ | |
4736 | (and:GPI (not:GPI | |
4737 | (match_operand:GPI 0 "register_operand" "r")) | |
4738 | (match_operand:GPI 1 "register_operand" "r")) | |
4739 | (const_int 0)))] | |
4740 | "" | |
4741 | "bics\\t<w>zr, %<w>1, %<w>0" | |
4742 | [(set_attr "type" "logics_reg")] | |
4743 | ) | |
4744 | ||
cc05c2c3 | 4745 | (define_insn "<LOGICAL:optab>_one_cmpl_<SHIFT:optab><mode>3" |
43e9d192 IB |
4746 | [(set (match_operand:GPI 0 "register_operand" "=r") |
4747 | (LOGICAL:GPI (not:GPI | |
4748 | (SHIFT:GPI | |
4749 | (match_operand:GPI 1 "register_operand" "r") | |
4750 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))) | |
4751 | (match_operand:GPI 3 "register_operand" "r")))] | |
4752 | "" | |
4753 | "<LOGICAL:nlogical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2" | |
dba606bf | 4754 | [(set_attr "type" "logic_shift_imm")] |
99e6af15 | 4755 | ) |
43e9d192 | 4756 | |
6365da9e KT |
4757 | (define_insn "*eor_one_cmpl_<SHIFT:optab><mode>3_alt" |
4758 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4759 | (not:GPI (xor:GPI | |
4760 | (SHIFT:GPI | |
4761 | (match_operand:GPI 1 "register_operand" "r") | |
4762 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n")) | |
4763 | (match_operand:GPI 3 "register_operand" "r"))))] | |
4764 | "" | |
4765 | "eon\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2" | |
4766 | [(set_attr "type" "logic_shift_imm")] | |
4767 | ) | |
4768 | ||
4769 | ;; Zero-extend version of the above. | |
4770 | (define_insn "*eor_one_cmpl_<SHIFT:optab>sidi3_alt_ze" | |
4771 | [(set (match_operand:DI 0 "register_operand" "=r") | |
4772 | (zero_extend:DI | |
4773 | (not:SI (xor:SI | |
4774 | (SHIFT:SI | |
4775 | (match_operand:SI 1 "register_operand" "r") | |
4776 | (match_operand:QI 2 "aarch64_shift_imm_si" "n")) | |
4777 | (match_operand:SI 3 "register_operand" "r")))))] | |
4778 | "" | |
4779 | "eon\\t%w0, %w3, %w1, <SHIFT:shift> %2" | |
4780 | [(set_attr "type" "logic_shift_imm")] | |
4781 | ) | |
4782 | ||
add3c965 IB |
4783 | (define_insn "*and_one_cmpl_<SHIFT:optab><mode>3_compare0" |
4784 | [(set (reg:CC_NZ CC_REGNUM) | |
4785 | (compare:CC_NZ | |
4786 | (and:GPI (not:GPI | |
4787 | (SHIFT:GPI | |
4788 | (match_operand:GPI 1 "register_operand" "r") | |
4789 | (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))) | |
4790 | (match_operand:GPI 3 "register_operand" "r")) | |
4791 | (const_int 0))) | |
4792 | (set (match_operand:GPI 0 "register_operand" "=r") | |
4793 | (and:GPI (not:GPI | |
4794 | (SHIFT:GPI | |
4795 | (match_dup 1) (match_dup 2))) (match_dup 3)))] | |
4796 | "" | |
4797 | "bics\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2" | |
99e6af15 JG |
4798 | [(set_attr "type" "logics_shift_imm")] |
4799 | ) | |
add3c965 IB |
4800 | |
4801 | ;; zero_extend version of above | |
4802 | (define_insn "*and_one_cmpl_<SHIFT:optab>si3_compare0_uxtw" | |
4803 | [(set (reg:CC_NZ CC_REGNUM) | |
4804 | (compare:CC_NZ | |
4805 | (and:SI (not:SI | |
4806 | (SHIFT:SI | |
4807 | (match_operand:SI 1 "register_operand" "r") | |
4808 | (match_operand:QI 2 "aarch64_shift_imm_si" "n"))) | |
4809 | (match_operand:SI 3 "register_operand" "r")) | |
4810 | (const_int 0))) | |
4811 | (set (match_operand:DI 0 "register_operand" "=r") | |
4812 | (zero_extend:DI (and:SI | |
4813 | (not:SI | |
4814 | (SHIFT:SI (match_dup 1) (match_dup 2))) (match_dup 3))))] | |
4815 | "" | |
4816 | "bics\\t%w0, %w3, %w1, <SHIFT:shift> %2" | |
99e6af15 JG |
4817 | [(set_attr "type" "logics_shift_imm")] |
4818 | ) | |
add3c965 | 4819 | |
cd8fc5ee AV |
4820 | (define_insn "*and_one_cmpl_<SHIFT:optab><mode>3_compare0_no_reuse" |
4821 | [(set (reg:CC_NZ CC_REGNUM) | |
4822 | (compare:CC_NZ | |
4823 | (and:GPI (not:GPI | |
4824 | (SHIFT:GPI | |
4825 | (match_operand:GPI 0 "register_operand" "r") | |
4826 | (match_operand:QI 1 "aarch64_shift_imm_<mode>" "n"))) | |
4827 | (match_operand:GPI 2 "register_operand" "r")) | |
4828 | (const_int 0)))] | |
4829 | "" | |
4830 | "bics\\t<w>zr, %<w>2, %<w>0, <SHIFT:shift> %1" | |
4831 | [(set_attr "type" "logics_shift_imm")] | |
4832 | ) | |
4833 | ||
43e9d192 IB |
4834 | (define_insn "clz<mode>2" |
4835 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4836 | (clz:GPI (match_operand:GPI 1 "register_operand" "r")))] | |
4837 | "" | |
4838 | "clz\\t%<w>0, %<w>1" | |
99e6af15 JG |
4839 | [(set_attr "type" "clz")] |
4840 | ) | |
43e9d192 IB |
4841 | |
4842 | (define_expand "ffs<mode>2" | |
4843 | [(match_operand:GPI 0 "register_operand") | |
4844 | (match_operand:GPI 1 "register_operand")] | |
4845 | "" | |
4846 | { | |
4847 | rtx ccreg = aarch64_gen_compare_reg (EQ, operands[1], const0_rtx); | |
4848 | rtx x = gen_rtx_NE (VOIDmode, ccreg, const0_rtx); | |
4849 | ||
4850 | emit_insn (gen_rbit<mode>2 (operands[0], operands[1])); | |
4851 | emit_insn (gen_clz<mode>2 (operands[0], operands[0])); | |
cd5660ab | 4852 | emit_insn (gen_csinc3<mode>_insn (operands[0], x, operands[0], const0_rtx)); |
43e9d192 IB |
4853 | DONE; |
4854 | } | |
4855 | ) | |
4856 | ||
9d5019a1 AP |
4857 | ;; Pop count be done via the "CNT" instruction in AdvSIMD. |
4858 | ;; | |
4859 | ;; MOV v.1d, x0 | |
4860 | ;; CNT v1.8b, v.8b | |
4861 | ;; ADDV b2, v1.8b | |
4862 | ;; MOV w0, v2.b[0] | |
4863 | ||
4864 | (define_expand "popcount<mode>2" | |
4865 | [(match_operand:GPI 0 "register_operand") | |
4866 | (match_operand:GPI 1 "register_operand")] | |
4867 | "TARGET_SIMD" | |
4868 | { | |
4869 | rtx v = gen_reg_rtx (V8QImode); | |
4870 | rtx v1 = gen_reg_rtx (V8QImode); | |
9d5019a1 AP |
4871 | rtx in = operands[1]; |
4872 | rtx out = operands[0]; | |
4873 | if(<MODE>mode == SImode) | |
4874 | { | |
4875 | rtx tmp; | |
4876 | tmp = gen_reg_rtx (DImode); | |
4877 | /* If we have SImode, zero extend to DImode, pop count does | |
4878 | not change if we have extra zeros. */ | |
4879 | emit_insn (gen_zero_extendsidi2 (tmp, in)); | |
4880 | in = tmp; | |
4881 | } | |
4882 | emit_move_insn (v, gen_lowpart (V8QImode, in)); | |
4883 | emit_insn (gen_popcountv8qi2 (v1, v)); | |
9921bbf9 | 4884 | emit_insn (gen_aarch64_zero_extend<mode>_reduc_plus_v8qi (out, v1)); |
9d5019a1 AP |
4885 | DONE; |
4886 | }) | |
4887 | ||
43e9d192 IB |
4888 | (define_insn "clrsb<mode>2" |
4889 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
dfc98d99 | 4890 | (clrsb:GPI (match_operand:GPI 1 "register_operand" "r")))] |
43e9d192 IB |
4891 | "" |
4892 | "cls\\t%<w>0, %<w>1" | |
99e6af15 JG |
4893 | [(set_attr "type" "clz")] |
4894 | ) | |
43e9d192 IB |
4895 | |
4896 | (define_insn "rbit<mode>2" | |
4897 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4898 | (unspec:GPI [(match_operand:GPI 1 "register_operand" "r")] UNSPEC_RBIT))] | |
4899 | "" | |
4900 | "rbit\\t%<w>0, %<w>1" | |
99e6af15 JG |
4901 | [(set_attr "type" "rbit")] |
4902 | ) | |
43e9d192 | 4903 | |
e8536e2b KT |
4904 | ;; Split after reload into RBIT + CLZ. Since RBIT is represented as an UNSPEC |
4905 | ;; it is unlikely to fold with any other operation, so keep this as a CTZ | |
4906 | ;; expression and split after reload to enable scheduling them apart if | |
4907 | ;; needed. | |
4908 | ||
4909 | (define_insn_and_split "ctz<mode>2" | |
4910 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
4911 | (ctz:GPI (match_operand:GPI 1 "register_operand" "r")))] | |
43e9d192 | 4912 | "" |
e8536e2b KT |
4913 | "#" |
4914 | "reload_completed" | |
4915 | [(const_int 0)] | |
4916 | " | |
4917 | emit_insn (gen_rbit<mode>2 (operands[0], operands[1])); | |
4918 | emit_insn (gen_clz<mode>2 (operands[0], operands[0])); | |
4919 | DONE; | |
4920 | ") | |
43e9d192 | 4921 | |
2b8568fe KT |
4922 | (define_insn "*and<mode>_compare0" |
4923 | [(set (reg:CC_NZ CC_REGNUM) | |
4924 | (compare:CC_NZ | |
4925 | (match_operand:SHORT 0 "register_operand" "r") | |
4926 | (const_int 0)))] | |
4927 | "" | |
4928 | "tst\\t%<w>0, <short_mask>" | |
4929 | [(set_attr "type" "alus_imm")] | |
4930 | ) | |
4931 | ||
e7ba492a | 4932 | (define_insn "*ands<GPI:mode>_compare0" |
b06335f9 KT |
4933 | [(set (reg:CC_NZ CC_REGNUM) |
4934 | (compare:CC_NZ | |
4935 | (zero_extend:GPI (match_operand:SHORT 1 "register_operand" "r")) | |
4936 | (const_int 0))) | |
4937 | (set (match_operand:GPI 0 "register_operand" "=r") | |
4938 | (zero_extend:GPI (match_dup 1)))] | |
4939 | "" | |
4940 | "ands\\t%<GPI:w>0, %<GPI:w>1, <short_mask>" | |
4941 | [(set_attr "type" "alus_imm")] | |
4942 | ) | |
4943 | ||
43e9d192 | 4944 | (define_insn "*and<mode>3nr_compare0" |
d803a491 IB |
4945 | [(set (reg:CC_NZ CC_REGNUM) |
4946 | (compare:CC_NZ | |
43e9d192 IB |
4947 | (and:GPI (match_operand:GPI 0 "register_operand" "%r,r") |
4948 | (match_operand:GPI 1 "aarch64_logical_operand" "r,<lconst>")) | |
4949 | (const_int 0)))] | |
4950 | "" | |
954224d7 WD |
4951 | "@ |
4952 | tst\\t%<w>0, %<w>1 | |
4953 | tst\\t%<w>0, %1" | |
1d70b920 | 4954 | [(set_attr "type" "logics_reg,logics_imm")] |
99e6af15 | 4955 | ) |
43e9d192 | 4956 | |
2c2789d5 KT |
4957 | (define_split |
4958 | [(set (reg:CC_NZ CC_REGNUM) | |
4959 | (compare:CC_NZ | |
4960 | (and:GPI (match_operand:GPI 0 "register_operand") | |
4961 | (match_operand:GPI 1 "aarch64_mov_imm_operand")) | |
4962 | (const_int 0))) | |
4963 | (clobber (match_operand:SI 2 "register_operand"))] | |
4964 | "" | |
4965 | [(set (match_dup 2) (match_dup 1)) | |
4966 | (set (reg:CC_NZ CC_REGNUM) | |
4967 | (compare:CC_NZ | |
4968 | (and:GPI (match_dup 0) | |
4969 | (match_dup 2)) | |
4970 | (const_int 0)))] | |
4971 | ) | |
4972 | ||
7325d85a KT |
4973 | (define_insn "*and<mode>3nr_compare0_zextract" |
4974 | [(set (reg:CC_NZ CC_REGNUM) | |
4975 | (compare:CC_NZ | |
4976 | (zero_extract:GPI (match_operand:GPI 0 "register_operand" "r") | |
4977 | (match_operand:GPI 1 "const_int_operand" "n") | |
4978 | (match_operand:GPI 2 "const_int_operand" "n")) | |
4979 | (const_int 0)))] | |
4980 | "INTVAL (operands[1]) > 0 | |
4981 | && ((INTVAL (operands[1]) + INTVAL (operands[2])) | |
4982 | <= GET_MODE_BITSIZE (<MODE>mode)) | |
4983 | && aarch64_bitmask_imm ( | |
4984 | UINTVAL (aarch64_mask_from_zextract_ops (operands[1], | |
4985 | operands[2])), | |
4986 | <MODE>mode)" | |
4987 | { | |
4988 | operands[1] | |
4989 | = aarch64_mask_from_zextract_ops (operands[1], operands[2]); | |
4990 | return "tst\\t%<w>0, %1"; | |
4991 | } | |
4992 | [(set_attr "type" "logics_shift_imm")] | |
4993 | ) | |
4994 | ||
43e9d192 | 4995 | (define_insn "*and_<SHIFT:optab><mode>3nr_compare0" |
d803a491 IB |
4996 | [(set (reg:CC_NZ CC_REGNUM) |
4997 | (compare:CC_NZ | |
43e9d192 IB |
4998 | (and:GPI (SHIFT:GPI |
4999 | (match_operand:GPI 0 "register_operand" "r") | |
5000 | (match_operand:QI 1 "aarch64_shift_imm_<mode>" "n")) | |
5001 | (match_operand:GPI 2 "register_operand" "r")) | |
5002 | (const_int 0)))] | |
5003 | "" | |
5004 | "tst\\t%<w>2, %<w>0, <SHIFT:shift> %1" | |
99e6af15 JG |
5005 | [(set_attr "type" "logics_shift_imm")] |
5006 | ) | |
43e9d192 | 5007 | |
2c2789d5 KT |
5008 | (define_split |
5009 | [(set (reg:CC_NZ CC_REGNUM) | |
5010 | (compare:CC_NZ | |
5011 | (and:GPI (SHIFT:GPI | |
5012 | (match_operand:GPI 0 "register_operand") | |
5013 | (match_operand:QI 1 "aarch64_shift_imm_<mode>")) | |
5014 | (match_operand:GPI 2 "aarch64_mov_imm_operand")) | |
5015 | (const_int 0))) | |
5016 | (clobber (match_operand:SI 3 "register_operand"))] | |
5017 | "" | |
5018 | [(set (match_dup 3) (match_dup 2)) | |
5019 | (set (reg:CC_NZ CC_REGNUM) | |
5020 | (compare:CC_NZ | |
5021 | (and:GPI (SHIFT:GPI | |
5022 | (match_dup 0) | |
5023 | (match_dup 1)) | |
5024 | (match_dup 3)) | |
5025 | (const_int 0)))] | |
5026 | ) | |
5027 | ||
43e9d192 IB |
5028 | ;; ------------------------------------------------------------------- |
5029 | ;; Shifts | |
5030 | ;; ------------------------------------------------------------------- | |
5031 | ||
5032 | (define_expand "<optab><mode>3" | |
5033 | [(set (match_operand:GPI 0 "register_operand") | |
5034 | (ASHIFT:GPI (match_operand:GPI 1 "register_operand") | |
f32c3adb | 5035 | (match_operand:QI 2 "aarch64_reg_or_imm")))] |
43e9d192 IB |
5036 | "" |
5037 | { | |
5038 | if (CONST_INT_P (operands[2])) | |
5039 | { | |
5040 | operands[2] = GEN_INT (INTVAL (operands[2]) | |
5041 | & (GET_MODE_BITSIZE (<MODE>mode) - 1)); | |
5042 | ||
5043 | if (operands[2] == const0_rtx) | |
5044 | { | |
5045 | emit_insn (gen_mov<mode> (operands[0], operands[1])); | |
5046 | DONE; | |
5047 | } | |
5048 | } | |
5049 | } | |
5050 | ) | |
5051 | ||
5052 | (define_expand "ashl<mode>3" | |
5053 | [(set (match_operand:SHORT 0 "register_operand") | |
5054 | (ashift:SHORT (match_operand:SHORT 1 "register_operand") | |
4bc72f35 | 5055 | (match_operand:QI 2 "const_int_operand")))] |
43e9d192 IB |
5056 | "" |
5057 | { | |
4bc72f35 | 5058 | operands[2] = GEN_INT (INTVAL (operands[2]) & GET_MODE_MASK (<MODE>mode)); |
43e9d192 | 5059 | |
4bc72f35 KT |
5060 | if (operands[2] == const0_rtx) |
5061 | { | |
5062 | emit_insn (gen_mov<mode> (operands[0], operands[1])); | |
5063 | DONE; | |
43e9d192 IB |
5064 | } |
5065 | } | |
5066 | ) | |
5067 | ||
5068 | (define_expand "rotr<mode>3" | |
5069 | [(set (match_operand:GPI 0 "register_operand") | |
5070 | (rotatert:GPI (match_operand:GPI 1 "register_operand") | |
f32c3adb | 5071 | (match_operand:QI 2 "aarch64_reg_or_imm")))] |
43e9d192 IB |
5072 | "" |
5073 | { | |
5074 | if (CONST_INT_P (operands[2])) | |
5075 | { | |
5076 | operands[2] = GEN_INT (INTVAL (operands[2]) | |
5077 | & (GET_MODE_BITSIZE (<MODE>mode) - 1)); | |
5078 | ||
5079 | if (operands[2] == const0_rtx) | |
5080 | { | |
5081 | emit_insn (gen_mov<mode> (operands[0], operands[1])); | |
5082 | DONE; | |
5083 | } | |
5084 | } | |
5085 | } | |
5086 | ) | |
5087 | ||
5088 | (define_expand "rotl<mode>3" | |
5089 | [(set (match_operand:GPI 0 "register_operand") | |
5090 | (rotatert:GPI (match_operand:GPI 1 "register_operand") | |
f32c3adb | 5091 | (match_operand:QI 2 "aarch64_reg_or_imm")))] |
43e9d192 IB |
5092 | "" |
5093 | { | |
5094 | /* (SZ - cnt) % SZ == -cnt % SZ */ | |
5095 | if (CONST_INT_P (operands[2])) | |
5096 | { | |
618ae596 | 5097 | operands[2] = GEN_INT ((-UINTVAL (operands[2])) |
43e9d192 IB |
5098 | & (GET_MODE_BITSIZE (<MODE>mode) - 1)); |
5099 | if (operands[2] == const0_rtx) | |
5100 | { | |
5101 | emit_insn (gen_mov<mode> (operands[0], operands[1])); | |
5102 | DONE; | |
5103 | } | |
5104 | } | |
5105 | else | |
5106 | operands[2] = expand_simple_unop (QImode, NEG, operands[2], | |
5107 | NULL_RTX, 1); | |
5108 | } | |
5109 | ) | |
5110 | ||
7813b280 KT |
5111 | ;; When the LSL, LSR, ASR, ROR instructions operate on all register arguments |
5112 | ;; they truncate the shift/rotate amount by the size of the registers they | |
5113 | ;; operate on: 32 for W-regs, 64 for X-regs. This allows us to optimise away | |
5114 | ;; such redundant masking instructions. GCC can do that automatically when | |
5115 | ;; SHIFT_COUNT_TRUNCATED is true, but we can't enable it for TARGET_SIMD | |
5116 | ;; because some of the SISD shift alternatives don't perform this truncations. | |
5117 | ;; So this pattern exists to catch such cases. | |
5118 | ||
5119 | (define_insn "*aarch64_<optab>_reg_<mode>3_mask1" | |
5120 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5121 | (SHIFT:GPI | |
5122 | (match_operand:GPI 1 "register_operand" "r") | |
5123 | (match_operator 4 "subreg_lowpart_operator" | |
5124 | [(and:GPI (match_operand:GPI 2 "register_operand" "r") | |
5125 | (match_operand 3 "const_int_operand" "n"))])))] | |
5126 | "(~INTVAL (operands[3]) & (GET_MODE_BITSIZE (<MODE>mode) - 1)) == 0" | |
5127 | "<shift>\t%<w>0, %<w>1, %<w>2" | |
5128 | [(set_attr "type" "shift_reg")] | |
5129 | ) | |
5130 | ||
cdeba3e0 | 5131 | (define_insn_and_split "*aarch64_<optab>_reg_<mode>3_neg_mask2" |
7813b280 KT |
5132 | [(set (match_operand:GPI 0 "register_operand" "=&r") |
5133 | (SHIFT:GPI | |
5134 | (match_operand:GPI 1 "register_operand" "r") | |
5135 | (match_operator 4 "subreg_lowpart_operator" | |
5136 | [(neg:SI (and:SI (match_operand:SI 2 "register_operand" "r") | |
5137 | (match_operand 3 "const_int_operand" "n")))])))] | |
5138 | "((~INTVAL (operands[3]) & (GET_MODE_BITSIZE (<MODE>mode) - 1)) == 0)" | |
5139 | "#" | |
5140 | "&& true" | |
5141 | [(const_int 0)] | |
5142 | { | |
5143 | rtx tmp = (can_create_pseudo_p () ? gen_reg_rtx (SImode) | |
cdeba3e0 | 5144 | : lowpart_subreg (SImode, operands[0], <MODE>mode)); |
7813b280 KT |
5145 | emit_insn (gen_negsi2 (tmp, operands[2])); |
5146 | ||
5147 | rtx and_op = gen_rtx_AND (SImode, tmp, operands[3]); | |
5148 | rtx subreg_tmp = gen_rtx_SUBREG (GET_MODE (operands[4]), and_op, | |
5149 | SUBREG_BYTE (operands[4])); | |
5150 | emit_insn (gen_<optab><mode>3 (operands[0], operands[1], subreg_tmp)); | |
5151 | DONE; | |
5152 | } | |
5153 | ) | |
5154 | ||
cdeba3e0 | 5155 | (define_insn_and_split "*aarch64_ashl_reg_<mode>3_minus_mask" |
7813b280 KT |
5156 | [(set (match_operand:GPI 0 "register_operand" "=&r") |
5157 | (ashift:GPI | |
5158 | (match_operand:GPI 1 "register_operand" "r") | |
5159 | (minus:QI (match_operand 2 "const_int_operand" "n") | |
5160 | (match_operator 5 "subreg_lowpart_operator" | |
5161 | [(and:SI (match_operand:SI 3 "register_operand" "r") | |
5162 | (match_operand 4 "const_int_operand" "n"))]))))] | |
5163 | "((~INTVAL (operands[4]) & (GET_MODE_BITSIZE (<MODE>mode) - 1)) == 0) | |
5164 | && INTVAL (operands[2]) == GET_MODE_BITSIZE (<MODE>mode)" | |
5165 | "#" | |
5166 | "&& true" | |
5167 | [(const_int 0)] | |
5168 | { | |
5169 | rtx tmp = (can_create_pseudo_p () ? gen_reg_rtx (SImode) | |
5170 | : operands[0]); | |
5171 | ||
5172 | emit_insn (gen_negsi2 (tmp, operands[3])); | |
5173 | ||
5174 | rtx and_op = gen_rtx_AND (SImode, tmp, operands[4]); | |
5175 | rtx subreg_tmp = gen_rtx_SUBREG (GET_MODE (operands[5]), and_op, | |
5176 | SUBREG_BYTE (operands[5])); | |
5177 | ||
5178 | emit_insn (gen_ashl<mode>3 (operands[0], operands[1], subreg_tmp)); | |
5179 | DONE; | |
5180 | } | |
5181 | ) | |
5182 | ||
5183 | (define_insn "*aarch64_<optab>_reg_di3_mask2" | |
5184 | [(set (match_operand:DI 0 "register_operand" "=r") | |
5185 | (SHIFT:DI | |
5186 | (match_operand:DI 1 "register_operand" "r") | |
5187 | (match_operator 4 "subreg_lowpart_operator" | |
5188 | [(and:SI (match_operand:SI 2 "register_operand" "r") | |
cdeba3e0 JJ |
5189 | (match_operand 3 "const_int_operand" "n"))])))] |
5190 | "((~INTVAL (operands[3]) & (GET_MODE_BITSIZE (DImode) - 1)) == 0)" | |
7813b280 KT |
5191 | { |
5192 | rtx xop[3]; | |
5193 | xop[0] = operands[0]; | |
5194 | xop[1] = operands[1]; | |
5195 | xop[2] = gen_lowpart (GET_MODE (operands[4]), operands[2]); | |
5196 | output_asm_insn ("<shift>\t%x0, %x1, %x2", xop); | |
5197 | return ""; | |
5198 | } | |
5199 | [(set_attr "type" "shift_reg")] | |
5200 | ) | |
5201 | ||
cdeba3e0 | 5202 | (define_insn_and_split "*aarch64_<optab>_reg_minus<mode>3" |
59abe903 MC |
5203 | [(set (match_operand:GPI 0 "register_operand" "=&r") |
5204 | (ASHIFT:GPI | |
5205 | (match_operand:GPI 1 "register_operand" "r") | |
5206 | (minus:QI (match_operand 2 "const_int_operand" "n") | |
5207 | (match_operand:QI 3 "register_operand" "r"))))] | |
5208 | "INTVAL (operands[2]) == GET_MODE_BITSIZE (<MODE>mode)" | |
5209 | "#" | |
5210 | "&& true" | |
5211 | [(const_int 0)] | |
5212 | { | |
5213 | rtx subreg_tmp = gen_lowpart (SImode, operands[3]); | |
5214 | ||
5215 | rtx tmp = (can_create_pseudo_p () ? gen_reg_rtx (SImode) | |
5216 | : gen_lowpart (SImode, operands[0])); | |
5217 | ||
5218 | emit_insn (gen_negsi2 (tmp, subreg_tmp)); | |
5219 | ||
5220 | rtx and_op = gen_rtx_AND (SImode, tmp, | |
5221 | GEN_INT (GET_MODE_BITSIZE (<MODE>mode) - 1)); | |
5222 | ||
5223 | rtx subreg_tmp2 = gen_lowpart_SUBREG (QImode, and_op); | |
5224 | ||
5225 | emit_insn (gen_<optab><mode>3 (operands[0], operands[1], subreg_tmp2)); | |
5226 | DONE; | |
5227 | } | |
5228 | [(set_attr "length" "8")] | |
5229 | ) | |
5230 | ||
2e100703 VP |
5231 | ;; Logical left shift using SISD or Integer instruction |
5232 | (define_insn "*aarch64_ashl_sisd_or_int_<mode>3" | |
3362d764 WD |
5233 | [(set (match_operand:GPI 0 "register_operand" "=r,r,w,w") |
5234 | (ashift:GPI | |
5235 | (match_operand:GPI 1 "register_operand" "r,r,w,w") | |
5236 | (match_operand:QI 2 "aarch64_reg_or_shift_imm_<mode>" "Us<cmode>,r,Us<cmode>,w")))] | |
2e100703 VP |
5237 | "" |
5238 | "@ | |
3362d764 | 5239 | lsl\t%<w>0, %<w>1, %2 |
d99dacc9 | 5240 | lsl\t%<w>0, %<w>1, %<w>2 |
2e100703 | 5241 | shl\t%<rtn>0<vas>, %<rtn>1<vas>, %2 |
d99dacc9 | 5242 | ushl\t%<rtn>0<vas>, %<rtn>1<vas>, %<rtn>2<vas>" |
488461d8 MM |
5243 | [(set_attr "type" "bfx,shift_reg,neon_shift_imm<q>, neon_shift_reg<q>") |
5244 | (set_attr "arch" "*,*,simd,simd")] | |
2e100703 VP |
5245 | ) |
5246 | ||
5247 | ;; Logical right shift using SISD or Integer instruction | |
5248 | (define_insn "*aarch64_lshr_sisd_or_int_<mode>3" | |
3362d764 WD |
5249 | [(set (match_operand:GPI 0 "register_operand" "=r,r,w,&w,&w") |
5250 | (lshiftrt:GPI | |
5251 | (match_operand:GPI 1 "register_operand" "r,r,w,w,w") | |
0603375c KT |
5252 | (match_operand:QI 2 "aarch64_reg_or_shift_imm_<mode>" |
5253 | "Us<cmode>,r,Us<cmode_simd>,w,0")))] | |
2e100703 VP |
5254 | "" |
5255 | "@ | |
3362d764 | 5256 | lsr\t%<w>0, %<w>1, %2 |
ad7b853a | 5257 | lsr\t%<w>0, %<w>1, %<w>2 |
2e100703 VP |
5258 | ushr\t%<rtn>0<vas>, %<rtn>1<vas>, %2 |
5259 | # | |
ad7b853a | 5260 | #" |
488461d8 MM |
5261 | [(set_attr "type" "bfx,shift_reg,neon_shift_imm<q>,neon_shift_reg<q>,neon_shift_reg<q>") |
5262 | (set_attr "arch" "*,*,simd,simd,simd")] | |
2e100703 VP |
5263 | ) |
5264 | ||
5265 | (define_split | |
5266 | [(set (match_operand:DI 0 "aarch64_simd_register") | |
5267 | (lshiftrt:DI | |
5268 | (match_operand:DI 1 "aarch64_simd_register") | |
5269 | (match_operand:QI 2 "aarch64_simd_register")))] | |
5270 | "TARGET_SIMD && reload_completed" | |
498b13e2 | 5271 | [(set (match_dup 3) |
2e100703 VP |
5272 | (unspec:QI [(match_dup 2)] UNSPEC_SISD_NEG)) |
5273 | (set (match_dup 0) | |
498b13e2 KT |
5274 | (unspec:DI [(match_dup 1) (match_dup 3)] UNSPEC_SISD_USHL))] |
5275 | { | |
5276 | operands[3] = gen_lowpart (QImode, operands[0]); | |
5277 | } | |
2e100703 VP |
5278 | ) |
5279 | ||
5280 | (define_split | |
5281 | [(set (match_operand:SI 0 "aarch64_simd_register") | |
5282 | (lshiftrt:SI | |
5283 | (match_operand:SI 1 "aarch64_simd_register") | |
5284 | (match_operand:QI 2 "aarch64_simd_register")))] | |
5285 | "TARGET_SIMD && reload_completed" | |
498b13e2 | 5286 | [(set (match_dup 3) |
2e100703 VP |
5287 | (unspec:QI [(match_dup 2)] UNSPEC_SISD_NEG)) |
5288 | (set (match_dup 0) | |
498b13e2 KT |
5289 | (unspec:SI [(match_dup 1) (match_dup 3)] UNSPEC_USHL_2S))] |
5290 | { | |
5291 | operands[3] = gen_lowpart (QImode, operands[0]); | |
5292 | } | |
2e100703 VP |
5293 | ) |
5294 | ||
5295 | ;; Arithmetic right shift using SISD or Integer instruction | |
5296 | (define_insn "*aarch64_ashr_sisd_or_int_<mode>3" | |
3362d764 | 5297 | [(set (match_operand:GPI 0 "register_operand" "=r,r,w,&w,&w") |
0603375c KT |
5298 | (ashiftrt:GPI |
5299 | (match_operand:GPI 1 "register_operand" "r,r,w,w,w") | |
5300 | (match_operand:QI 2 "aarch64_reg_or_shift_imm_di" | |
5301 | "Us<cmode>,r,Us<cmode_simd>,w,0")))] | |
2e100703 VP |
5302 | "" |
5303 | "@ | |
3362d764 | 5304 | asr\t%<w>0, %<w>1, %2 |
ad7b853a | 5305 | asr\t%<w>0, %<w>1, %<w>2 |
2e100703 VP |
5306 | sshr\t%<rtn>0<vas>, %<rtn>1<vas>, %2 |
5307 | # | |
ad7b853a | 5308 | #" |
488461d8 MM |
5309 | [(set_attr "type" "bfx,shift_reg,neon_shift_imm<q>,neon_shift_reg<q>,neon_shift_reg<q>") |
5310 | (set_attr "arch" "*,*,simd,simd,simd")] | |
2e100703 VP |
5311 | ) |
5312 | ||
5313 | (define_split | |
5314 | [(set (match_operand:DI 0 "aarch64_simd_register") | |
5315 | (ashiftrt:DI | |
5316 | (match_operand:DI 1 "aarch64_simd_register") | |
5317 | (match_operand:QI 2 "aarch64_simd_register")))] | |
5318 | "TARGET_SIMD && reload_completed" | |
21a7b5e1 | 5319 | [(set (match_dup 3) |
2e100703 VP |
5320 | (unspec:QI [(match_dup 2)] UNSPEC_SISD_NEG)) |
5321 | (set (match_dup 0) | |
21a7b5e1 MS |
5322 | (unspec:DI [(match_dup 1) (match_dup 3)] UNSPEC_SISD_SSHL))] |
5323 | { | |
5324 | operands[3] = gen_lowpart (QImode, operands[0]); | |
5325 | } | |
2e100703 VP |
5326 | ) |
5327 | ||
5328 | (define_split | |
5329 | [(set (match_operand:SI 0 "aarch64_simd_register") | |
5330 | (ashiftrt:SI | |
5331 | (match_operand:SI 1 "aarch64_simd_register") | |
5332 | (match_operand:QI 2 "aarch64_simd_register")))] | |
5333 | "TARGET_SIMD && reload_completed" | |
21a7b5e1 | 5334 | [(set (match_dup 3) |
2e100703 VP |
5335 | (unspec:QI [(match_dup 2)] UNSPEC_SISD_NEG)) |
5336 | (set (match_dup 0) | |
21a7b5e1 MS |
5337 | (unspec:SI [(match_dup 1) (match_dup 3)] UNSPEC_SSHL_2S))] |
5338 | { | |
5339 | operands[3] = gen_lowpart (QImode, operands[0]); | |
5340 | } | |
2e100703 VP |
5341 | ) |
5342 | ||
5343 | (define_insn "*aarch64_sisd_ushl" | |
5344 | [(set (match_operand:DI 0 "register_operand" "=w") | |
5345 | (unspec:DI [(match_operand:DI 1 "register_operand" "w") | |
5346 | (match_operand:QI 2 "register_operand" "w")] | |
5347 | UNSPEC_SISD_USHL))] | |
5348 | "TARGET_SIMD" | |
5349 | "ushl\t%d0, %d1, %d2" | |
488461d8 | 5350 | [(set_attr "type" "neon_shift_reg")] |
2e100703 VP |
5351 | ) |
5352 | ||
5353 | (define_insn "*aarch64_ushl_2s" | |
5354 | [(set (match_operand:SI 0 "register_operand" "=w") | |
5355 | (unspec:SI [(match_operand:SI 1 "register_operand" "w") | |
5356 | (match_operand:QI 2 "register_operand" "w")] | |
5357 | UNSPEC_USHL_2S))] | |
5358 | "TARGET_SIMD" | |
5359 | "ushl\t%0.2s, %1.2s, %2.2s" | |
488461d8 | 5360 | [(set_attr "type" "neon_shift_reg")] |
2e100703 VP |
5361 | ) |
5362 | ||
5363 | (define_insn "*aarch64_sisd_sshl" | |
5364 | [(set (match_operand:DI 0 "register_operand" "=w") | |
5365 | (unspec:DI [(match_operand:DI 1 "register_operand" "w") | |
5366 | (match_operand:QI 2 "register_operand" "w")] | |
5367 | UNSPEC_SISD_SSHL))] | |
5368 | "TARGET_SIMD" | |
5369 | "sshl\t%d0, %d1, %d2" | |
488461d8 | 5370 | [(set_attr "type" "neon_shift_reg")] |
2e100703 VP |
5371 | ) |
5372 | ||
5373 | (define_insn "*aarch64_sshl_2s" | |
5374 | [(set (match_operand:SI 0 "register_operand" "=w") | |
5375 | (unspec:SI [(match_operand:SI 1 "register_operand" "w") | |
5376 | (match_operand:QI 2 "register_operand" "w")] | |
5377 | UNSPEC_SSHL_2S))] | |
5378 | "TARGET_SIMD" | |
5379 | "sshl\t%0.2s, %1.2s, %2.2s" | |
488461d8 | 5380 | [(set_attr "type" "neon_shift_reg")] |
2e100703 VP |
5381 | ) |
5382 | ||
5383 | (define_insn "*aarch64_sisd_neg_qi" | |
5384 | [(set (match_operand:QI 0 "register_operand" "=w") | |
5385 | (unspec:QI [(match_operand:QI 1 "register_operand" "w")] | |
5386 | UNSPEC_SISD_NEG))] | |
5387 | "TARGET_SIMD" | |
5388 | "neg\t%d0, %d1" | |
488461d8 | 5389 | [(set_attr "type" "neon_neg")] |
2e100703 VP |
5390 | ) |
5391 | ||
5392 | ;; Rotate right | |
5393 | (define_insn "*ror<mode>3_insn" | |
b81f1ee3 JG |
5394 | [(set (match_operand:GPI 0 "register_operand" "=r,r") |
5395 | (rotatert:GPI | |
5396 | (match_operand:GPI 1 "register_operand" "r,r") | |
3362d764 | 5397 | (match_operand:QI 2 "aarch64_reg_or_shift_imm_<mode>" "Us<cmode>,r")))] |
43e9d192 | 5398 | "" |
3362d764 WD |
5399 | "@ |
5400 | ror\\t%<w>0, %<w>1, %2 | |
5401 | ror\\t%<w>0, %<w>1, %<w>2" | |
5402 | [(set_attr "type" "rotate_imm,shift_reg")] | |
43e9d192 IB |
5403 | ) |
5404 | ||
d0b6bb1b IB |
5405 | ;; zero_extend version of above |
5406 | (define_insn "*<optab>si3_insn_uxtw" | |
3362d764 | 5407 | [(set (match_operand:DI 0 "register_operand" "=r,r") |
d0b6bb1b | 5408 | (zero_extend:DI (SHIFT:SI |
3362d764 WD |
5409 | (match_operand:SI 1 "register_operand" "r,r") |
5410 | (match_operand:QI 2 "aarch64_reg_or_shift_imm_si" "Uss,r"))))] | |
d0b6bb1b | 5411 | "" |
3362d764 WD |
5412 | "@ |
5413 | <shift>\\t%w0, %w1, %2 | |
5414 | <shift>\\t%w0, %w1, %w2" | |
94f7a25e | 5415 | [(set_attr "type" "bfx,shift_reg")] |
d0b6bb1b IB |
5416 | ) |
5417 | ||
43e9d192 IB |
5418 | (define_insn "*<optab><mode>3_insn" |
5419 | [(set (match_operand:SHORT 0 "register_operand" "=r") | |
5420 | (ASHIFT:SHORT (match_operand:SHORT 1 "register_operand" "r") | |
5421 | (match_operand 2 "const_int_operand" "n")))] | |
5422 | "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<MODE>mode)" | |
5423 | { | |
5424 | operands[3] = GEN_INT (<sizen> - UINTVAL (operands[2])); | |
5425 | return "<bfshift>\t%w0, %w1, %2, %3"; | |
5426 | } | |
94f7a25e | 5427 | [(set_attr "type" "bfx")] |
43e9d192 IB |
5428 | ) |
5429 | ||
bd83ff2c IB |
5430 | (define_insn "*extr<mode>5_insn" |
5431 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5432 | (ior:GPI (ashift:GPI (match_operand:GPI 1 "register_operand" "r") | |
5433 | (match_operand 3 "const_int_operand" "n")) | |
5434 | (lshiftrt:GPI (match_operand:GPI 2 "register_operand" "r") | |
5435 | (match_operand 4 "const_int_operand" "n"))))] | |
5436 | "UINTVAL (operands[3]) < GET_MODE_BITSIZE (<MODE>mode) && | |
5437 | (UINTVAL (operands[3]) + UINTVAL (operands[4]) == GET_MODE_BITSIZE (<MODE>mode))" | |
5438 | "extr\\t%<w>0, %<w>1, %<w>2, %4" | |
3362d764 | 5439 | [(set_attr "type" "rotate_imm")] |
bd83ff2c IB |
5440 | ) |
5441 | ||
fb0cb7fa KT |
5442 | ;; There are no canonicalisation rules for ashift and lshiftrt inside an ior |
5443 | ;; so we have to match both orderings. | |
5444 | (define_insn "*extr<mode>5_insn_alt" | |
5445 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5446 | (ior:GPI (lshiftrt:GPI (match_operand:GPI 2 "register_operand" "r") | |
5447 | (match_operand 4 "const_int_operand" "n")) | |
5448 | (ashift:GPI (match_operand:GPI 1 "register_operand" "r") | |
5449 | (match_operand 3 "const_int_operand" "n"))))] | |
5450 | "UINTVAL (operands[3]) < GET_MODE_BITSIZE (<MODE>mode) | |
5451 | && (UINTVAL (operands[3]) + UINTVAL (operands[4]) | |
5452 | == GET_MODE_BITSIZE (<MODE>mode))" | |
5453 | "extr\\t%<w>0, %<w>1, %<w>2, %4" | |
3362d764 | 5454 | [(set_attr "type" "rotate_imm")] |
fb0cb7fa KT |
5455 | ) |
5456 | ||
bd83ff2c IB |
5457 | ;; zero_extend version of the above |
5458 | (define_insn "*extrsi5_insn_uxtw" | |
5459 | [(set (match_operand:DI 0 "register_operand" "=r") | |
5460 | (zero_extend:DI | |
5461 | (ior:SI (ashift:SI (match_operand:SI 1 "register_operand" "r") | |
5462 | (match_operand 3 "const_int_operand" "n")) | |
5463 | (lshiftrt:SI (match_operand:SI 2 "register_operand" "r") | |
5464 | (match_operand 4 "const_int_operand" "n")))))] | |
5465 | "UINTVAL (operands[3]) < 32 && | |
5466 | (UINTVAL (operands[3]) + UINTVAL (operands[4]) == 32)" | |
5467 | "extr\\t%w0, %w1, %w2, %4" | |
3362d764 | 5468 | [(set_attr "type" "rotate_imm")] |
bd83ff2c IB |
5469 | ) |
5470 | ||
fb0cb7fa KT |
5471 | (define_insn "*extrsi5_insn_uxtw_alt" |
5472 | [(set (match_operand:DI 0 "register_operand" "=r") | |
5473 | (zero_extend:DI | |
5474 | (ior:SI (lshiftrt:SI (match_operand:SI 2 "register_operand" "r") | |
5475 | (match_operand 4 "const_int_operand" "n")) | |
5476 | (ashift:SI (match_operand:SI 1 "register_operand" "r") | |
5477 | (match_operand 3 "const_int_operand" "n")))))] | |
5478 | "UINTVAL (operands[3]) < 32 && | |
5479 | (UINTVAL (operands[3]) + UINTVAL (operands[4]) == 32)" | |
5480 | "extr\\t%w0, %w1, %w2, %4" | |
3362d764 | 5481 | [(set_attr "type" "rotate_imm")] |
fb0cb7fa KT |
5482 | ) |
5483 | ||
47f42744 JJ |
5484 | (define_insn "*extrsi5_insn_di" |
5485 | [(set (match_operand:SI 0 "register_operand" "=r") | |
5486 | (ior:SI (ashift:SI (match_operand:SI 1 "register_operand" "r") | |
5487 | (match_operand 3 "const_int_operand" "n")) | |
5488 | (match_operator:SI 6 "subreg_lowpart_operator" | |
5489 | [(zero_extract:DI | |
5490 | (match_operand:DI 2 "register_operand" "r") | |
5491 | (match_operand 5 "const_int_operand" "n") | |
5492 | (match_operand 4 "const_int_operand" "n"))])))] | |
5493 | "UINTVAL (operands[3]) < 32 | |
5494 | && UINTVAL (operands[3]) + UINTVAL (operands[4]) == 32 | |
5495 | && INTVAL (operands[3]) == INTVAL (operands[5])" | |
5496 | "extr\\t%w0, %w1, %w2, %4" | |
5497 | [(set_attr "type" "rotate_imm")] | |
5498 | ) | |
5499 | ||
5977a10d IB |
5500 | (define_insn "*ror<mode>3_insn" |
5501 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5502 | (rotate:GPI (match_operand:GPI 1 "register_operand" "r") | |
5503 | (match_operand 2 "const_int_operand" "n")))] | |
5504 | "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<MODE>mode)" | |
5505 | { | |
5506 | operands[3] = GEN_INT (<sizen> - UINTVAL (operands[2])); | |
5507 | return "ror\\t%<w>0, %<w>1, %3"; | |
5508 | } | |
b81f1ee3 | 5509 | [(set_attr "type" "rotate_imm")] |
5977a10d IB |
5510 | ) |
5511 | ||
5512 | ;; zero_extend version of the above | |
5513 | (define_insn "*rorsi3_insn_uxtw" | |
5514 | [(set (match_operand:DI 0 "register_operand" "=r") | |
5515 | (zero_extend:DI | |
5516 | (rotate:SI (match_operand:SI 1 "register_operand" "r") | |
5517 | (match_operand 2 "const_int_operand" "n"))))] | |
5518 | "UINTVAL (operands[2]) < 32" | |
5519 | { | |
5520 | operands[3] = GEN_INT (32 - UINTVAL (operands[2])); | |
5521 | return "ror\\t%w0, %w1, %3"; | |
5522 | } | |
b81f1ee3 | 5523 | [(set_attr "type" "rotate_imm")] |
5977a10d IB |
5524 | ) |
5525 | ||
43e9d192 IB |
5526 | (define_insn "*<ANY_EXTEND:optab><GPI:mode>_ashl<SHORT:mode>" |
5527 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5528 | (ANY_EXTEND:GPI | |
5529 | (ashift:SHORT (match_operand:SHORT 1 "register_operand" "r") | |
5530 | (match_operand 2 "const_int_operand" "n"))))] | |
5531 | "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<SHORT:MODE>mode)" | |
5532 | { | |
5533 | operands[3] = GEN_INT (<SHORT:sizen> - UINTVAL (operands[2])); | |
5534 | return "<su>bfiz\t%<GPI:w>0, %<GPI:w>1, %2, %3"; | |
5535 | } | |
94f7a25e | 5536 | [(set_attr "type" "bfx")] |
43e9d192 IB |
5537 | ) |
5538 | ||
5539 | (define_insn "*zero_extend<GPI:mode>_lshr<SHORT:mode>" | |
5540 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5541 | (zero_extend:GPI | |
5542 | (lshiftrt:SHORT (match_operand:SHORT 1 "register_operand" "r") | |
5543 | (match_operand 2 "const_int_operand" "n"))))] | |
5544 | "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<SHORT:MODE>mode)" | |
5545 | { | |
5546 | operands[3] = GEN_INT (<SHORT:sizen> - UINTVAL (operands[2])); | |
5547 | return "ubfx\t%<GPI:w>0, %<GPI:w>1, %2, %3"; | |
5548 | } | |
94f7a25e | 5549 | [(set_attr "type" "bfx")] |
43e9d192 IB |
5550 | ) |
5551 | ||
5552 | (define_insn "*extend<GPI:mode>_ashr<SHORT:mode>" | |
5553 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5554 | (sign_extend:GPI | |
5555 | (ashiftrt:SHORT (match_operand:SHORT 1 "register_operand" "r") | |
5556 | (match_operand 2 "const_int_operand" "n"))))] | |
5557 | "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<SHORT:MODE>mode)" | |
5558 | { | |
5559 | operands[3] = GEN_INT (<SHORT:sizen> - UINTVAL (operands[2])); | |
5560 | return "sbfx\\t%<GPI:w>0, %<GPI:w>1, %2, %3"; | |
5561 | } | |
94f7a25e | 5562 | [(set_attr "type" "bfx")] |
43e9d192 IB |
5563 | ) |
5564 | ||
5565 | ;; ------------------------------------------------------------------- | |
5566 | ;; Bitfields | |
5567 | ;; ------------------------------------------------------------------- | |
5568 | ||
5569 | (define_expand "<optab>" | |
1bbffb87 | 5570 | [(set (match_operand:DI 0 "register_operand") |
e27bfda2 KT |
5571 | (ANY_EXTRACT:DI (match_operand:DI 1 "register_operand") |
5572 | (match_operand 2 | |
5573 | "aarch64_simd_shift_imm_offset_di") | |
5574 | (match_operand 3 "aarch64_simd_shift_imm_di")))] | |
43e9d192 | 5575 | "" |
e27bfda2 KT |
5576 | { |
5577 | if (!IN_RANGE (INTVAL (operands[2]) + INTVAL (operands[3]), | |
5578 | 1, GET_MODE_BITSIZE (DImode) - 1)) | |
5579 | FAIL; | |
5580 | } | |
43e9d192 IB |
5581 | ) |
5582 | ||
e27bfda2 | 5583 | |
43e9d192 IB |
5584 | (define_insn "*<optab><mode>" |
5585 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5586 | (ANY_EXTRACT:GPI (match_operand:GPI 1 "register_operand" "r") | |
e27bfda2 KT |
5587 | (match_operand 2 |
5588 | "aarch64_simd_shift_imm_offset_<mode>" "n") | |
5589 | (match_operand 3 | |
5590 | "aarch64_simd_shift_imm_<mode>" "n")))] | |
5591 | "IN_RANGE (INTVAL (operands[2]) + INTVAL (operands[3]), | |
5592 | 1, GET_MODE_BITSIZE (<MODE>mode) - 1)" | |
43e9d192 | 5593 | "<su>bfx\\t%<w>0, %<w>1, %3, %2" |
94f7a25e | 5594 | [(set_attr "type" "bfx")] |
65f9e789 IB |
5595 | ) |
5596 | ||
bcb036c5 KT |
5597 | ;; When the bit position and width add up to 32 we can use a W-reg LSR |
5598 | ;; instruction taking advantage of the implicit zero-extension of the X-reg. | |
5599 | (define_split | |
5600 | [(set (match_operand:DI 0 "register_operand") | |
5601 | (zero_extract:DI (match_operand:DI 1 "register_operand") | |
5602 | (match_operand 2 | |
5603 | "aarch64_simd_shift_imm_offset_di") | |
5604 | (match_operand 3 | |
5605 | "aarch64_simd_shift_imm_di")))] | |
5606 | "IN_RANGE (INTVAL (operands[2]) + INTVAL (operands[3]), 1, | |
5607 | GET_MODE_BITSIZE (DImode) - 1) | |
5608 | && (INTVAL (operands[2]) + INTVAL (operands[3])) | |
5609 | == GET_MODE_BITSIZE (SImode)" | |
5610 | [(set (match_dup 0) | |
5611 | (zero_extend:DI (lshiftrt:SI (match_dup 4) (match_dup 3))))] | |
5612 | { | |
5613 | operands[4] = gen_lowpart (SImode, operands[1]); | |
5614 | } | |
5615 | ) | |
5616 | ||
65f9e789 IB |
5617 | ;; Bitfield Insert (insv) |
5618 | (define_expand "insv<mode>" | |
5619 | [(set (zero_extract:GPI (match_operand:GPI 0 "register_operand") | |
5620 | (match_operand 1 "const_int_operand") | |
5621 | (match_operand 2 "const_int_operand")) | |
5622 | (match_operand:GPI 3 "general_operand"))] | |
5623 | "" | |
5624 | { | |
5625 | unsigned HOST_WIDE_INT width = UINTVAL (operands[1]); | |
5626 | unsigned HOST_WIDE_INT pos = UINTVAL (operands[2]); | |
5627 | rtx value = operands[3]; | |
5628 | ||
5629 | if (width == 0 || (pos + width) > GET_MODE_BITSIZE (<MODE>mode)) | |
5630 | FAIL; | |
5631 | ||
5632 | if (CONST_INT_P (value)) | |
5633 | { | |
5634 | unsigned HOST_WIDE_INT mask = ((unsigned HOST_WIDE_INT)1 << width) - 1; | |
5635 | ||
5636 | /* Prefer AND/OR for inserting all zeros or all ones. */ | |
5637 | if ((UINTVAL (value) & mask) == 0 | |
5638 | || (UINTVAL (value) & mask) == mask) | |
5639 | FAIL; | |
5640 | ||
5641 | /* 16-bit aligned 16-bit wide insert is handled by insv_imm. */ | |
5642 | if (width == 16 && (pos % 16) == 0) | |
5643 | DONE; | |
5644 | } | |
5645 | operands[3] = force_reg (<MODE>mode, value); | |
5646 | }) | |
5647 | ||
5648 | (define_insn "*insv_reg<mode>" | |
5649 | [(set (zero_extract:GPI (match_operand:GPI 0 "register_operand" "+r") | |
5650 | (match_operand 1 "const_int_operand" "n") | |
5651 | (match_operand 2 "const_int_operand" "n")) | |
5652 | (match_operand:GPI 3 "register_operand" "r"))] | |
5653 | "!(UINTVAL (operands[1]) == 0 | |
5654 | || (UINTVAL (operands[2]) + UINTVAL (operands[1]) | |
5655 | > GET_MODE_BITSIZE (<MODE>mode)))" | |
5656 | "bfi\\t%<w>0, %<w>3, %2, %1" | |
99e6af15 | 5657 | [(set_attr "type" "bfm")] |
2879bb2b IB |
5658 | ) |
5659 | ||
e4f019d3 RL |
5660 | (define_insn "*aarch64_bfi<GPI:mode><ALLX:mode>4" |
5661 | [(set (zero_extract:GPI (match_operand:GPI 0 "register_operand" "+r") | |
5662 | (match_operand 1 "const_int_operand" "n") | |
5663 | (match_operand 2 "const_int_operand" "n")) | |
5664 | (zero_extend:GPI (match_operand:ALLX 3 "register_operand" "r")))] | |
5665 | "UINTVAL (operands[1]) <= <ALLX:sizen>" | |
5666 | "bfi\\t%<GPI:w>0, %<GPI:w>3, %2, %1" | |
5667 | [(set_attr "type" "bfm")] | |
5668 | ) | |
5669 | ||
6a0d3939 SE |
5670 | ;; Match a bfi instruction where the shift of OP3 means that we are |
5671 | ;; actually copying the least significant bits of OP3 into OP0 by way | |
5672 | ;; of the AND masks and the IOR instruction. A similar instruction | |
5673 | ;; with the two parts of the IOR swapped around was never triggered | |
5674 | ;; in a bootstrap build and test of GCC so it was not included. | |
5675 | ||
5676 | (define_insn "*aarch64_bfi<GPI:mode>5_shift" | |
5677 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5678 | (ior:GPI (and:GPI (match_operand:GPI 1 "register_operand" "0") | |
5679 | (match_operand:GPI 2 "const_int_operand" "n")) | |
5680 | (and:GPI (ashift:GPI | |
5681 | (match_operand:GPI 3 "register_operand" "r") | |
5682 | (match_operand:GPI 4 "aarch64_simd_shift_imm_<mode>" "n")) | |
5683 | (match_operand:GPI 5 "const_int_operand" "n"))))] | |
5684 | "aarch64_masks_and_shift_for_bfi_p (<MODE>mode, UINTVAL (operands[2]), | |
5685 | UINTVAL (operands[4]), | |
5686 | UINTVAL(operands[5]))" | |
5687 | "bfi\t%<GPI:w>0, %<GPI:w>3, %4, %P5" | |
5688 | [(set_attr "type" "bfm")] | |
5689 | ) | |
5690 | ||
5691 | (define_insn "*aarch64_bfi<GPI:mode>5_shift_alt" | |
5692 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5693 | (ior:GPI (and:GPI (ashift:GPI | |
5694 | (match_operand:GPI 1 "register_operand" "r") | |
5695 | (match_operand:GPI 2 "aarch64_simd_shift_imm_<mode>" "n")) | |
5696 | (match_operand:GPI 3 "const_int_operand" "n")) | |
5697 | (and:GPI (match_operand:GPI 4 "register_operand" "0") | |
5698 | (match_operand:GPI 5 "const_int_operand" "n"))))] | |
5699 | "aarch64_masks_and_shift_for_bfi_p (<MODE>mode, UINTVAL (operands[5]), | |
5700 | UINTVAL (operands[2]), | |
5701 | UINTVAL(operands[3]))" | |
5702 | "bfi\t%<GPI:w>0, %<GPI:w>1, %2, %P3" | |
5703 | [(set_attr "type" "bfm")] | |
5704 | ) | |
5705 | ||
5706 | ;; Like *aarch64_bfi<GPI:mode>5_shift but with no and of the ashift because | |
5707 | ;; the shift is large enough to remove the need for an AND instruction. | |
5708 | ||
5709 | (define_insn "*aarch64_bfi<GPI:mode>4_noand" | |
5710 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5711 | (ior:GPI (and:GPI (match_operand:GPI 1 "register_operand" "0") | |
5712 | (match_operand:GPI 2 "const_int_operand" "n")) | |
5713 | (ashift:GPI | |
5714 | (match_operand:GPI 3 "register_operand" "r") | |
5715 | (match_operand:GPI 4 "aarch64_simd_shift_imm_<mode>" "n"))))] | |
5716 | "aarch64_masks_and_shift_for_bfi_p (<MODE>mode, UINTVAL (operands[2]), | |
5717 | UINTVAL (operands[4]), | |
5718 | HOST_WIDE_INT_M1U << UINTVAL (operands[4]) )" | |
5719 | { | |
5720 | operands[5] = GEN_INT (GET_MODE_BITSIZE (<MODE>mode) - UINTVAL (operands[4])); | |
5721 | return "bfi\t%<GPI:w>0, %<GPI:w>3, %4, %5"; | |
5722 | } | |
5723 | [(set_attr "type" "bfm")] | |
5724 | ) | |
5725 | ||
5726 | (define_insn "*aarch64_bfi<GPI:mode>4_noand_alt" | |
5727 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5728 | (ior:GPI (ashift:GPI | |
5729 | (match_operand:GPI 1 "register_operand" "r") | |
5730 | (match_operand:GPI 2 "aarch64_simd_shift_imm_<mode>" "n")) | |
5731 | (and:GPI (match_operand:GPI 3 "register_operand" "0") | |
5732 | (match_operand:GPI 4 "const_int_operand" "n"))))] | |
5733 | "aarch64_masks_and_shift_for_bfi_p (<MODE>mode, UINTVAL (operands[4]), | |
5734 | UINTVAL (operands[2]), | |
5735 | HOST_WIDE_INT_M1U << UINTVAL (operands[2]) )" | |
5736 | { | |
5737 | operands[5] = GEN_INT (GET_MODE_BITSIZE (<MODE>mode) - UINTVAL (operands[2])); | |
5738 | return "bfi\t%<GPI:w>0, %<GPI:w>1, %2, %5"; | |
5739 | } | |
5740 | [(set_attr "type" "bfm")] | |
5741 | ) | |
5742 | ||
5743 | ;; Like *aarch64_bfi<GPI:mode>5_shift but with no shifting, we are just | |
f0a8f6bc SE |
5744 | ;; copying the least significant bits of OP3 to OP0. We need two versions |
5745 | ;; of the instruction to handle different checks on the constant values. | |
6a0d3939 SE |
5746 | |
5747 | (define_insn "*aarch64_bfi<GPI:mode>4_noshift" | |
5748 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5749 | (ior:GPI (and:GPI (match_operand:GPI 1 "register_operand" "0") | |
5750 | (match_operand:GPI 2 "const_int_operand" "n")) | |
5751 | (and:GPI (match_operand:GPI 3 "register_operand" "r") | |
5752 | (match_operand:GPI 4 "const_int_operand" "n"))))] | |
5753 | "aarch64_masks_and_shift_for_bfi_p (<MODE>mode, UINTVAL (operands[2]), 0, | |
5754 | UINTVAL (operands[4]))" | |
5755 | "bfi\t%<GPI:w>0, %<GPI:w>3, 0, %P4" | |
5756 | [(set_attr "type" "bfm")] | |
5757 | ) | |
5758 | ||
f0a8f6bc SE |
5759 | (define_insn "*aarch64_bfi<GPI:mode>4_noshift_alt" |
5760 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5761 | (ior:GPI (and:GPI (match_operand:GPI 3 "register_operand" "r") | |
5762 | (match_operand:GPI 4 "const_int_operand" "n")) | |
5763 | (and:GPI (match_operand:GPI 1 "register_operand" "0") | |
5764 | (match_operand:GPI 2 "const_int_operand" "n"))))] | |
5765 | "aarch64_masks_and_shift_for_bfi_p (<MODE>mode, UINTVAL (operands[2]), 0, | |
5766 | UINTVAL (operands[4]))" | |
5767 | "bfi\t%<GPI:w>0, %<GPI:w>3, 0, %P4" | |
5768 | [(set_attr "type" "bfm")] | |
5769 | ) | |
5770 | ||
f6ba5d03 JJ |
5771 | (define_insn "*aarch64_bfxil<mode>_extr" |
5772 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5773 | (ior:GPI (and:GPI (match_operand:GPI 1 "register_operand" "0") | |
5774 | (match_operand:GPI 2 "const_int_operand" "n")) | |
5775 | (zero_extract:GPI | |
5776 | (match_operand:GPI 3 "register_operand" "r") | |
5777 | (match_operand:GPI 4 "aarch64_simd_shift_imm_<mode>" "n") | |
5778 | (match_operand:GPI 5 "aarch64_simd_shift_imm_<mode>" "n"))))] | |
5779 | "UINTVAL (operands[2]) == HOST_WIDE_INT_M1U << INTVAL (operands[4]) | |
5780 | && INTVAL (operands[4]) | |
5781 | && (UINTVAL (operands[4]) + UINTVAL (operands[5]) | |
5782 | <= GET_MODE_BITSIZE (<MODE>mode))" | |
5783 | "bfxil\t%<GPI:w>0, %<GPI:w>3, %5, %4" | |
5784 | [(set_attr "type" "bfm")] | |
5785 | ) | |
5786 | ||
5787 | (define_insn "*aarch64_bfxilsi_extrdi" | |
5788 | [(set (match_operand:SI 0 "register_operand" "=r") | |
5789 | (ior:SI (and:SI (match_operand:SI 1 "register_operand" "0") | |
5790 | (match_operand:SI 2 "const_int_operand" "n")) | |
5791 | (match_operator:SI 6 "subreg_lowpart_operator" | |
5792 | [(zero_extract:DI | |
5793 | (match_operand:DI 3 "register_operand" "r") | |
5794 | (match_operand:SI 4 "aarch64_simd_shift_imm_si" "n") | |
5795 | (match_operand:SI 5 "aarch64_simd_shift_imm_si" "n"))])))] | |
5796 | "UINTVAL (operands[2]) == HOST_WIDE_INT_M1U << INTVAL (operands[4]) | |
5797 | && INTVAL (operands[4]) | |
5798 | && UINTVAL (operands[4]) + UINTVAL (operands[5]) <= 32" | |
5799 | "bfxil\t%w0, %w3, %5, %4" | |
5800 | [(set_attr "type" "bfm")] | |
5801 | ) | |
5802 | ||
2879bb2b IB |
5803 | (define_insn "*extr_insv_lower_reg<mode>" |
5804 | [(set (zero_extract:GPI (match_operand:GPI 0 "register_operand" "+r") | |
5805 | (match_operand 1 "const_int_operand" "n") | |
5806 | (const_int 0)) | |
98fa5ef3 | 5807 | (zero_extract:GPI (match_operand:GPI 2 "register_operand" "r") |
2879bb2b IB |
5808 | (match_dup 1) |
5809 | (match_operand 3 "const_int_operand" "n")))] | |
5810 | "!(UINTVAL (operands[1]) == 0 | |
5811 | || (UINTVAL (operands[3]) + UINTVAL (operands[1]) | |
5812 | > GET_MODE_BITSIZE (<MODE>mode)))" | |
5813 | "bfxil\\t%<w>0, %<w>2, %3, %1" | |
99e6af15 | 5814 | [(set_attr "type" "bfm")] |
43e9d192 IB |
5815 | ) |
5816 | ||
5817 | (define_insn "*<optab><ALLX:mode>_shft_<GPI:mode>" | |
5818 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5819 | (ashift:GPI (ANY_EXTEND:GPI | |
5820 | (match_operand:ALLX 1 "register_operand" "r")) | |
5821 | (match_operand 2 "const_int_operand" "n")))] | |
5822 | "UINTVAL (operands[2]) < <GPI:sizen>" | |
5823 | { | |
5824 | operands[3] = (<ALLX:sizen> <= (<GPI:sizen> - UINTVAL (operands[2]))) | |
5825 | ? GEN_INT (<ALLX:sizen>) | |
5826 | : GEN_INT (<GPI:sizen> - UINTVAL (operands[2])); | |
5827 | return "<su>bfiz\t%<GPI:w>0, %<GPI:w>1, %2, %3"; | |
5828 | } | |
94f7a25e | 5829 | [(set_attr "type" "bfx")] |
43e9d192 IB |
5830 | ) |
5831 | ||
5832 | ;; XXX We should match (any_extend (ashift)) here, like (and (ashift)) below | |
5833 | ||
5834 | (define_insn "*andim_ashift<mode>_bfiz" | |
5835 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5836 | (and:GPI (ashift:GPI (match_operand:GPI 1 "register_operand" "r") | |
5837 | (match_operand 2 "const_int_operand" "n")) | |
5838 | (match_operand 3 "const_int_operand" "n")))] | |
8c83f71d | 5839 | "aarch64_mask_and_shift_for_ubfiz_p (<MODE>mode, operands[3], operands[2])" |
43e9d192 | 5840 | "ubfiz\\t%<w>0, %<w>1, %2, %P3" |
94f7a25e | 5841 | [(set_attr "type" "bfx")] |
43e9d192 IB |
5842 | ) |
5843 | ||
58c2ad42 LM |
5844 | ;; Match sbfiz pattern in a shift left + shift right operation. |
5845 | ||
5846 | (define_insn "*ashift<mode>_extv_bfiz" | |
5847 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5848 | (ashift:GPI (sign_extract:GPI (match_operand:GPI 1 "register_operand" "r") | |
5849 | (match_operand 2 "aarch64_simd_shift_imm_offset_<mode>" "n") | |
5850 | (const_int 0)) | |
5851 | (match_operand 3 "aarch64_simd_shift_imm_<mode>" "n")))] | |
5852 | "IN_RANGE (INTVAL (operands[2]) + INTVAL (operands[3]), | |
5853 | 1, GET_MODE_BITSIZE (<MODE>mode) - 1)" | |
5854 | "sbfiz\\t%<w>0, %<w>1, %3, %2" | |
5855 | [(set_attr "type" "bfx")] | |
5856 | ) | |
5857 | ||
b65a1eb3 RS |
5858 | (define_insn "*ashiftsi_extvdi_bfiz" |
5859 | [(set (match_operand:SI 0 "register_operand" "=r") | |
5860 | (ashift:SI | |
5861 | (match_operator:SI 4 "subreg_lowpart_operator" | |
5862 | [(sign_extract:DI | |
5863 | (match_operand:DI 1 "register_operand" "r") | |
5864 | (match_operand 2 "aarch64_simd_shift_imm_offset_si") | |
5865 | (const_int 0))]) | |
5866 | (match_operand 3 "aarch64_simd_shift_imm_si")))] | |
5867 | "IN_RANGE (INTVAL (operands[2]) + INTVAL (operands[3]), | |
5868 | 1, GET_MODE_BITSIZE (SImode) - 1)" | |
5869 | "sbfiz\\t%w0, %w1, %3, %2" | |
5870 | [(set_attr "type" "bfx")] | |
5871 | ) | |
5872 | ||
680153bd KT |
5873 | ;; When the bit position and width of the equivalent extraction add up to 32 |
5874 | ;; we can use a W-reg LSL instruction taking advantage of the implicit | |
5875 | ;; zero-extension of the X-reg. | |
5876 | (define_split | |
5877 | [(set (match_operand:DI 0 "register_operand") | |
5878 | (and:DI (ashift:DI (match_operand:DI 1 "register_operand") | |
5879 | (match_operand 2 "const_int_operand")) | |
5880 | (match_operand 3 "const_int_operand")))] | |
5881 | "aarch64_mask_and_shift_for_ubfiz_p (DImode, operands[3], operands[2]) | |
5882 | && (INTVAL (operands[2]) + popcount_hwi (INTVAL (operands[3]))) | |
5883 | == GET_MODE_BITSIZE (SImode)" | |
5884 | [(set (match_dup 0) | |
5885 | (zero_extend:DI (ashift:SI (match_dup 4) (match_dup 2))))] | |
5886 | { | |
5887 | operands[4] = gen_lowpart (SImode, operands[1]); | |
5888 | } | |
5889 | ) | |
5890 | ||
43e9d192 IB |
5891 | (define_insn "bswap<mode>2" |
5892 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5893 | (bswap:GPI (match_operand:GPI 1 "register_operand" "r")))] | |
5894 | "" | |
5895 | "rev\\t%<w>0, %<w>1" | |
99e6af15 | 5896 | [(set_attr "type" "rev")] |
43e9d192 IB |
5897 | ) |
5898 | ||
c16c63bb IB |
5899 | (define_insn "bswaphi2" |
5900 | [(set (match_operand:HI 0 "register_operand" "=r") | |
5901 | (bswap:HI (match_operand:HI 1 "register_operand" "r")))] | |
5902 | "" | |
5903 | "rev16\\t%w0, %w1" | |
99e6af15 | 5904 | [(set_attr "type" "rev")] |
c16c63bb IB |
5905 | ) |
5906 | ||
231c52ae ST |
5907 | (define_insn "*aarch64_bfxil<mode>" |
5908 | [(set (match_operand:GPI 0 "register_operand" "=r,r") | |
5909 | (ior:GPI (and:GPI (match_operand:GPI 1 "register_operand" "r,0") | |
5910 | (match_operand:GPI 3 "const_int_operand" "n, Ulc")) | |
5911 | (and:GPI (match_operand:GPI 2 "register_operand" "0,r") | |
5912 | (match_operand:GPI 4 "const_int_operand" "Ulc, n"))))] | |
5913 | "(INTVAL (operands[3]) == ~INTVAL (operands[4])) | |
5914 | && (aarch64_high_bits_all_ones_p (INTVAL (operands[3])) | |
5915 | || aarch64_high_bits_all_ones_p (INTVAL (operands[4])))" | |
5916 | { | |
5917 | switch (which_alternative) | |
5918 | { | |
5919 | case 0: | |
5920 | operands[3] = GEN_INT (ctz_hwi (~INTVAL (operands[3]))); | |
5921 | return "bfxil\\t%<w>0, %<w>1, 0, %3"; | |
5922 | case 1: | |
5923 | operands[3] = GEN_INT (ctz_hwi (~INTVAL (operands[4]))); | |
5924 | return "bfxil\\t%<w>0, %<w>2, 0, %3"; | |
5925 | default: | |
5926 | gcc_unreachable (); | |
5927 | } | |
5928 | } | |
5929 | [(set_attr "type" "bfm")] | |
5930 | ) | |
5931 | ||
2dc09f66 ST |
5932 | ; Zero-extended version of above (aarch64_bfxil) |
5933 | (define_insn "*aarch64_bfxilsi_uxtw" | |
5934 | [(set (match_operand:DI 0 "register_operand" "=r,r") | |
5935 | (zero_extend:DI (ior:SI (and:SI (match_operand:SI 1 "register_operand" | |
5936 | "r,0") | |
5937 | (match_operand:SI 3 "const_int_operand" "n, Ulc")) | |
5938 | (and:SI (match_operand:SI 2 "register_operand" "0,r") | |
5939 | (match_operand:SI 4 "const_int_operand" "Ulc, n")))))] | |
5940 | "(INTVAL (operands[3]) == ~INTVAL (operands[4])) | |
5941 | && (aarch64_high_bits_all_ones_p (INTVAL (operands[3])) | |
5942 | || aarch64_high_bits_all_ones_p (INTVAL (operands[4])))" | |
5943 | { | |
5944 | switch (which_alternative) | |
5945 | { | |
5946 | case 0: | |
5947 | operands[3] = GEN_INT (ctz_hwi (~INTVAL (operands[3]))); | |
55163419 | 5948 | return "bfxil\\t%w0, %w1, 0, %3"; |
2dc09f66 ST |
5949 | case 1: |
5950 | operands[3] = GEN_INT (ctz_hwi (~INTVAL (operands[4]))); | |
55163419 | 5951 | return "bfxil\\t%w0, %w2, 0, %3"; |
2dc09f66 ST |
5952 | default: |
5953 | gcc_unreachable (); | |
5954 | } | |
5955 | } | |
5956 | [(set_attr "type" "bfm")] | |
5957 | ) | |
5958 | ||
f7d5cf8d KT |
5959 | ;; There are no canonicalisation rules for the position of the lshiftrt, ashift |
5960 | ;; operations within an IOR/AND RTX, therefore we have two patterns matching | |
5961 | ;; each valid permutation. | |
5962 | ||
5963 | (define_insn "rev16<mode>2" | |
5964 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5965 | (ior:GPI (and:GPI (ashift:GPI (match_operand:GPI 1 "register_operand" "r") | |
5966 | (const_int 8)) | |
5967 | (match_operand:GPI 3 "const_int_operand" "n")) | |
5968 | (and:GPI (lshiftrt:GPI (match_dup 1) | |
5969 | (const_int 8)) | |
5970 | (match_operand:GPI 2 "const_int_operand" "n"))))] | |
5971 | "aarch_rev16_shleft_mask_imm_p (operands[3], <MODE>mode) | |
5972 | && aarch_rev16_shright_mask_imm_p (operands[2], <MODE>mode)" | |
5973 | "rev16\\t%<w>0, %<w>1" | |
5974 | [(set_attr "type" "rev")] | |
5975 | ) | |
5976 | ||
5977 | (define_insn "rev16<mode>2_alt" | |
5978 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
5979 | (ior:GPI (and:GPI (lshiftrt:GPI (match_operand:GPI 1 "register_operand" "r") | |
5980 | (const_int 8)) | |
5981 | (match_operand:GPI 2 "const_int_operand" "n")) | |
5982 | (and:GPI (ashift:GPI (match_dup 1) | |
5983 | (const_int 8)) | |
5984 | (match_operand:GPI 3 "const_int_operand" "n"))))] | |
5985 | "aarch_rev16_shleft_mask_imm_p (operands[3], <MODE>mode) | |
5986 | && aarch_rev16_shright_mask_imm_p (operands[2], <MODE>mode)" | |
5987 | "rev16\\t%<w>0, %<w>1" | |
5988 | [(set_attr "type" "rev")] | |
5989 | ) | |
5990 | ||
d0b6bb1b IB |
5991 | ;; zero_extend version of above |
5992 | (define_insn "*bswapsi2_uxtw" | |
5993 | [(set (match_operand:DI 0 "register_operand" "=r") | |
5994 | (zero_extend:DI (bswap:SI (match_operand:SI 1 "register_operand" "r"))))] | |
5995 | "" | |
5996 | "rev\\t%w0, %w1" | |
99e6af15 | 5997 | [(set_attr "type" "rev")] |
d0b6bb1b IB |
5998 | ) |
5999 | ||
43e9d192 IB |
6000 | ;; ------------------------------------------------------------------- |
6001 | ;; Floating-point intrinsics | |
6002 | ;; ------------------------------------------------------------------- | |
6003 | ||
42fc9a7f | 6004 | ;; frint floating-point round to integral standard patterns. |
16ce822e | 6005 | ;; Expands to btrunc, ceil, floor, nearbyint, rint, round, roundeven. |
43e9d192 | 6006 | |
42fc9a7f | 6007 | (define_insn "<frint_pattern><mode>2" |
d7f33f07 JW |
6008 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6009 | (unspec:GPF_F16 [(match_operand:GPF_F16 1 "register_operand" "w")] | |
42fc9a7f | 6010 | FRINT))] |
43e9d192 | 6011 | "TARGET_FLOAT" |
42fc9a7f | 6012 | "frint<frint_suffix>\\t%<s>0, %<s>1" |
d7f33f07 | 6013 | [(set_attr "type" "f_rint<stype>")] |
43e9d192 IB |
6014 | ) |
6015 | ||
42fc9a7f JG |
6016 | ;; frcvt floating-point round to integer and convert standard patterns. |
6017 | ;; Expands to lbtrunc, lceil, lfloor, lround. | |
d7f33f07 | 6018 | (define_insn "l<fcvt_pattern><su_optab><GPF_F16:mode><GPI:mode>2" |
43e9d192 | 6019 | [(set (match_operand:GPI 0 "register_operand" "=r") |
d7f33f07 JW |
6020 | (FIXUORS:GPI |
6021 | (unspec:GPF_F16 [(match_operand:GPF_F16 1 "register_operand" "w")] | |
6022 | FCVT)))] | |
43e9d192 | 6023 | "TARGET_FLOAT" |
d7f33f07 | 6024 | "fcvt<frint_suffix><su>\\t%<GPI:w>0, %<GPF_F16:s>1" |
99e6af15 | 6025 | [(set_attr "type" "f_cvtf2i")] |
43e9d192 IB |
6026 | ) |
6027 | ||
39252973 KT |
6028 | (define_insn "*aarch64_fcvt<su_optab><GPF:mode><GPI:mode>2_mult" |
6029 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
6030 | (FIXUORS:GPI | |
6031 | (mult:GPF | |
6032 | (match_operand:GPF 1 "register_operand" "w") | |
6033 | (match_operand:GPF 2 "aarch64_fp_pow2" "F"))))] | |
6034 | "TARGET_FLOAT | |
6035 | && IN_RANGE (aarch64_fpconst_pow_of_2 (operands[2]), 1, | |
6036 | GET_MODE_BITSIZE (<GPI:MODE>mode))" | |
6037 | { | |
6038 | int fbits = aarch64_fpconst_pow_of_2 (operands[2]); | |
6039 | char buf[64]; | |
6040 | snprintf (buf, 64, "fcvtz<su>\\t%%<GPI:w>0, %%<GPF:s>1, #%d", fbits); | |
6041 | output_asm_insn (buf, operands); | |
6042 | return ""; | |
6043 | } | |
6044 | [(set_attr "type" "f_cvtf2i")] | |
6045 | ) | |
6046 | ||
d6e6e8b6 WD |
6047 | ;; fma - expand fma into patterns with the accumulator operand first since |
6048 | ;; reusing the accumulator results in better register allocation. | |
6049 | ;; The register allocator considers copy preferences in operand order, | |
6050 | ;; so this prefers fmadd s0, s1, s2, s0 over fmadd s1, s1, s2, s0. | |
6051 | ||
6052 | (define_expand "fma<mode>4" | |
6053 | [(set (match_operand:GPF_F16 0 "register_operand") | |
6054 | (fma:GPF_F16 (match_operand:GPF_F16 1 "register_operand") | |
6055 | (match_operand:GPF_F16 2 "register_operand") | |
6056 | (match_operand:GPF_F16 3 "register_operand")))] | |
6057 | "TARGET_FLOAT" | |
6058 | ) | |
43e9d192 | 6059 | |
d6e6e8b6 | 6060 | (define_insn "*aarch64_fma<mode>4" |
9a594ad6 | 6061 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
d6e6e8b6 WD |
6062 | (fma:GPF_F16 (match_operand:GPF_F16 2 "register_operand" "w") |
6063 | (match_operand:GPF_F16 3 "register_operand" "w") | |
6064 | (match_operand:GPF_F16 1 "register_operand" "w")))] | |
43e9d192 | 6065 | "TARGET_FLOAT" |
d6e6e8b6 | 6066 | "fmadd\\t%<s>0, %<s>2, %<s>3, %<s>1" |
9a594ad6 | 6067 | [(set_attr "type" "fmac<stype>")] |
43e9d192 IB |
6068 | ) |
6069 | ||
d6e6e8b6 WD |
6070 | (define_expand "fnma<mode>4" |
6071 | [(set (match_operand:GPF_F16 0 "register_operand") | |
6072 | (fma:GPF_F16 | |
6073 | (neg:GPF_F16 (match_operand:GPF_F16 1 "register_operand")) | |
6074 | (match_operand:GPF_F16 2 "register_operand") | |
6075 | (match_operand:GPF_F16 3 "register_operand")))] | |
6076 | "TARGET_FLOAT" | |
6077 | ) | |
6078 | ||
6079 | (define_insn "*aarch64_fnma<mode>4" | |
9a594ad6 JW |
6080 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6081 | (fma:GPF_F16 | |
d6e6e8b6 WD |
6082 | (neg:GPF_F16 (match_operand:GPF_F16 2 "register_operand" "w")) |
6083 | (match_operand:GPF_F16 3 "register_operand" "w") | |
6084 | (match_operand:GPF_F16 1 "register_operand" "w")))] | |
43e9d192 | 6085 | "TARGET_FLOAT" |
d6e6e8b6 | 6086 | "fmsub\\t%<s>0, %<s>2, %<s>3, %<s>1" |
9a594ad6 | 6087 | [(set_attr "type" "fmac<stype>")] |
43e9d192 IB |
6088 | ) |
6089 | ||
d6e6e8b6 WD |
6090 | |
6091 | (define_expand "fms<mode>4" | |
6092 | [(set (match_operand:GPF 0 "register_operand") | |
6093 | (fma:GPF (match_operand:GPF 1 "register_operand") | |
6094 | (match_operand:GPF 2 "register_operand") | |
6095 | (neg:GPF (match_operand:GPF 3 "register_operand"))))] | |
6096 | "TARGET_FLOAT" | |
6097 | ) | |
6098 | ||
6099 | (define_insn "*aarch64_fms<mode>4" | |
43e9d192 | 6100 | [(set (match_operand:GPF 0 "register_operand" "=w") |
d6e6e8b6 WD |
6101 | (fma:GPF (match_operand:GPF 2 "register_operand" "w") |
6102 | (match_operand:GPF 3 "register_operand" "w") | |
6103 | (neg:GPF (match_operand:GPF 1 "register_operand" "w"))))] | |
43e9d192 | 6104 | "TARGET_FLOAT" |
d6e6e8b6 | 6105 | "fnmsub\\t%<s>0, %<s>2, %<s>3, %<s>1" |
99e6af15 | 6106 | [(set_attr "type" "fmac<s>")] |
43e9d192 IB |
6107 | ) |
6108 | ||
d6e6e8b6 WD |
6109 | (define_expand "fnms<mode>4" |
6110 | [(set (match_operand:GPF 0 "register_operand") | |
6111 | (fma:GPF (neg:GPF (match_operand:GPF 1 "register_operand")) | |
6112 | (match_operand:GPF 2 "register_operand") | |
6113 | (neg:GPF (match_operand:GPF 3 "register_operand"))))] | |
6114 | "TARGET_FLOAT" | |
6115 | ) | |
6116 | ||
6117 | (define_insn "*aarch64_fnms<mode>4" | |
43e9d192 | 6118 | [(set (match_operand:GPF 0 "register_operand" "=w") |
d6e6e8b6 WD |
6119 | (fma:GPF (neg:GPF (match_operand:GPF 2 "register_operand" "w")) |
6120 | (match_operand:GPF 3 "register_operand" "w") | |
6121 | (neg:GPF (match_operand:GPF 1 "register_operand" "w"))))] | |
43e9d192 | 6122 | "TARGET_FLOAT" |
d6e6e8b6 | 6123 | "fnmadd\\t%<s>0, %<s>2, %<s>3, %<s>1" |
99e6af15 | 6124 | [(set_attr "type" "fmac<s>")] |
43e9d192 IB |
6125 | ) |
6126 | ||
6127 | ;; If signed zeros are ignored, -(a * b + c) = -a * b - c. | |
d6e6e8b6 | 6128 | (define_insn "*aarch64_fnmadd<mode>4" |
43e9d192 | 6129 | [(set (match_operand:GPF 0 "register_operand" "=w") |
d6e6e8b6 WD |
6130 | (neg:GPF (fma:GPF (match_operand:GPF 2 "register_operand" "w") |
6131 | (match_operand:GPF 3 "register_operand" "w") | |
6132 | (match_operand:GPF 1 "register_operand" "w"))))] | |
43e9d192 | 6133 | "!HONOR_SIGNED_ZEROS (<MODE>mode) && TARGET_FLOAT" |
d6e6e8b6 | 6134 | "fnmadd\\t%<s>0, %<s>2, %<s>3, %<s>1" |
99e6af15 | 6135 | [(set_attr "type" "fmac<s>")] |
43e9d192 IB |
6136 | ) |
6137 | ||
6138 | ;; ------------------------------------------------------------------- | |
6139 | ;; Floating-point conversions | |
6140 | ;; ------------------------------------------------------------------- | |
6141 | ||
6142 | (define_insn "extendsfdf2" | |
6143 | [(set (match_operand:DF 0 "register_operand" "=w") | |
6144 | (float_extend:DF (match_operand:SF 1 "register_operand" "w")))] | |
6145 | "TARGET_FLOAT" | |
6146 | "fcvt\\t%d0, %s1" | |
99e6af15 | 6147 | [(set_attr "type" "f_cvt")] |
43e9d192 IB |
6148 | ) |
6149 | ||
c2ec330c AL |
6150 | (define_insn "extendhfsf2" |
6151 | [(set (match_operand:SF 0 "register_operand" "=w") | |
6152 | (float_extend:SF (match_operand:HF 1 "register_operand" "w")))] | |
6153 | "TARGET_FLOAT" | |
6154 | "fcvt\\t%s0, %h1" | |
6155 | [(set_attr "type" "f_cvt")] | |
6156 | ) | |
6157 | ||
6158 | (define_insn "extendhfdf2" | |
6159 | [(set (match_operand:DF 0 "register_operand" "=w") | |
6160 | (float_extend:DF (match_operand:HF 1 "register_operand" "w")))] | |
6161 | "TARGET_FLOAT" | |
6162 | "fcvt\\t%d0, %h1" | |
6163 | [(set_attr "type" "f_cvt")] | |
6164 | ) | |
6165 | ||
43e9d192 IB |
6166 | (define_insn "truncdfsf2" |
6167 | [(set (match_operand:SF 0 "register_operand" "=w") | |
6168 | (float_truncate:SF (match_operand:DF 1 "register_operand" "w")))] | |
6169 | "TARGET_FLOAT" | |
6170 | "fcvt\\t%s0, %d1" | |
99e6af15 | 6171 | [(set_attr "type" "f_cvt")] |
43e9d192 IB |
6172 | ) |
6173 | ||
c2ec330c AL |
6174 | (define_insn "truncsfhf2" |
6175 | [(set (match_operand:HF 0 "register_operand" "=w") | |
6176 | (float_truncate:HF (match_operand:SF 1 "register_operand" "w")))] | |
6177 | "TARGET_FLOAT" | |
6178 | "fcvt\\t%h0, %s1" | |
6179 | [(set_attr "type" "f_cvt")] | |
6180 | ) | |
6181 | ||
6182 | (define_insn "truncdfhf2" | |
6183 | [(set (match_operand:HF 0 "register_operand" "=w") | |
6184 | (float_truncate:HF (match_operand:DF 1 "register_operand" "w")))] | |
6185 | "TARGET_FLOAT" | |
6186 | "fcvt\\t%h0, %d1" | |
6187 | [(set_attr "type" "f_cvt")] | |
6188 | ) | |
6189 | ||
22be0d08 MC |
6190 | ;; Convert SF -> SI or DF -> DI while preferring w = w register constraints |
6191 | ;; and making r = w more expensive | |
6192 | ||
6193 | (define_insn "<optab>_trunc<fcvt_target><GPI:mode>2" | |
43d0a8ee | 6194 | [(set (match_operand:GPI 0 "register_operand" "=w,?r") |
22be0d08 MC |
6195 | (FIXUORS:GPI (match_operand:<FCVT_TARGET> 1 "register_operand" "w,w")))] |
6196 | "TARGET_FLOAT" | |
6197 | "@ | |
43d0a8ee WD |
6198 | fcvtz<su>\t%<s>0, %<s>1 |
6199 | fcvtz<su>\t%<w>0, %<s>1" | |
cc2fda13 MS |
6200 | [(set_attr "type" "neon_fp_to_int_s,f_cvtf2i") |
6201 | (set_attr "arch" "simd,fp")] | |
22be0d08 MC |
6202 | ) |
6203 | ||
6204 | ;; Convert HF -> SI or DI | |
6205 | ||
6206 | (define_insn "<optab>_trunchf<GPI:mode>2" | |
6207 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
6208 | (FIXUORS:GPI (match_operand:HF 1 "register_operand" "w")))] | |
6209 | "TARGET_FP_F16INST" | |
6210 | "fcvtz<su>\t%<w>0, %h1" | |
6211 | [(set_attr "type" "f_cvtf2i")] | |
6212 | ) | |
6213 | ||
6214 | ;; Convert DF -> SI or SF -> DI which can only be accomplished with | |
6215 | ;; input in a fp register and output in a integer register | |
6216 | ||
6217 | (define_insn "<optab>_trunc<fcvt_change_mode><GPI:mode>2" | |
43e9d192 | 6218 | [(set (match_operand:GPI 0 "register_operand" "=r") |
22be0d08 | 6219 | (FIXUORS:GPI (match_operand:<FCVT_CHANGE_MODE> 1 "register_operand" "w")))] |
43e9d192 | 6220 | "TARGET_FLOAT" |
22be0d08 | 6221 | "fcvtz<su>\t%<w>0, %<fpw>1" |
99e6af15 | 6222 | [(set_attr "type" "f_cvtf2i")] |
43e9d192 IB |
6223 | ) |
6224 | ||
eeb59c16 MC |
6225 | (define_insn "*fix_to_zero_extend<mode>di2" |
6226 | [(set (match_operand:DI 0 "register_operand" "=r") | |
6227 | (zero_extend:DI | |
6228 | (unsigned_fix:SI | |
6229 | (match_operand:GPF 1 "register_operand" "w"))))] | |
6230 | "TARGET_FLOAT" | |
6231 | "fcvtzu\t%w0, %<s>1" | |
6232 | [(set_attr "type" "f_cvtf2i")] | |
6233 | ) | |
6234 | ||
188d0079 JH |
6235 | ;; Equal width integer to fp and multiply combine. |
6236 | (define_insn "*aarch64_<su_optab>cvtf<fcvt_target><GPF:mode>2_mult" | |
6237 | [(set (match_operand:GPF 0 "register_operand" "=w,w") | |
6238 | (mult:GPF (FLOATUORS:GPF | |
6239 | (match_operand:<FCVT_TARGET> 1 "register_operand" "w,?r")) | |
6240 | (match_operand:GPF 2 "aarch64_fp_pow2_recip" "Dt,Dt")))] | |
6241 | "TARGET_FLOAT" | |
6242 | { | |
6243 | operands[2] = GEN_INT (aarch64_fpconst_pow2_recip (operands[2])); | |
6244 | switch (which_alternative) | |
6245 | { | |
6246 | case 0: | |
6247 | return "<su_optab>cvtf\t%<GPF:s>0, %<s>1, #%2"; | |
6248 | case 1: | |
6249 | return "<su_optab>cvtf\t%<GPF:s>0, %<w1>1, #%2"; | |
6250 | default: | |
6251 | gcc_unreachable (); | |
6252 | } | |
6253 | } | |
6254 | [(set_attr "type" "neon_int_to_fp_<Vetype>,f_cvti2f") | |
6255 | (set_attr "arch" "simd,fp")] | |
6256 | ) | |
6257 | ||
6258 | ;; Unequal width integer to fp and multiply combine. | |
6259 | (define_insn "*aarch64_<su_optab>cvtf<fcvt_iesize><GPF:mode>2_mult" | |
6260 | [(set (match_operand:GPF 0 "register_operand" "=w") | |
6261 | (mult:GPF (FLOATUORS:GPF | |
6262 | (match_operand:<FCVT_IESIZE> 1 "register_operand" "r")) | |
6263 | (match_operand:GPF 2 "aarch64_fp_pow2_recip" "Dt")))] | |
6264 | "TARGET_FLOAT" | |
6265 | { | |
6266 | operands[2] = GEN_INT (aarch64_fpconst_pow2_recip (operands[2])); | |
6267 | return "<su_optab>cvtf\t%<GPF:s>0, %<w2>1, #%2"; | |
6268 | } | |
6269 | [(set_attr "type" "f_cvti2f")] | |
6270 | ) | |
6271 | ||
6272 | ;; Equal width integer to fp conversion. | |
0d35c5c2 VP |
6273 | (define_insn "<optab><fcvt_target><GPF:mode>2" |
6274 | [(set (match_operand:GPF 0 "register_operand" "=w,w") | |
43d0a8ee | 6275 | (FLOATUORS:GPF (match_operand:<FCVT_TARGET> 1 "register_operand" "w,?r")))] |
31ca7cba | 6276 | "TARGET_FLOAT" |
0d35c5c2 VP |
6277 | "@ |
6278 | <su_optab>cvtf\t%<GPF:s>0, %<s>1 | |
6279 | <su_optab>cvtf\t%<GPF:s>0, %<w1>1" | |
488461d8 MM |
6280 | [(set_attr "type" "neon_int_to_fp_<Vetype>,f_cvti2f") |
6281 | (set_attr "arch" "simd,fp")] | |
43e9d192 IB |
6282 | ) |
6283 | ||
188d0079 | 6284 | ;; Unequal width integer to fp conversions. |
0d35c5c2 | 6285 | (define_insn "<optab><fcvt_iesize><GPF:mode>2" |
43e9d192 | 6286 | [(set (match_operand:GPF 0 "register_operand" "=w") |
0d35c5c2 | 6287 | (FLOATUORS:GPF (match_operand:<FCVT_IESIZE> 1 "register_operand" "r")))] |
43e9d192 | 6288 | "TARGET_FLOAT" |
0d35c5c2 VP |
6289 | "<su_optab>cvtf\t%<GPF:s>0, %<w2>1" |
6290 | [(set_attr "type" "f_cvti2f")] | |
43e9d192 IB |
6291 | ) |
6292 | ||
bf1e3646 JG |
6293 | ;; If we do not have ARMv8.2-A 16-bit floating point extensions, the |
6294 | ;; midend will arrange for an SImode conversion to HFmode to first go | |
6295 | ;; through DFmode, then to HFmode. But first it will try converting | |
6296 | ;; to DImode then down, which would match our DImode pattern below and | |
6297 | ;; give very poor code-generation. So, we must provide our own emulation | |
6298 | ;; of the mid-end logic. | |
6299 | ||
6300 | (define_insn "aarch64_fp16_<optab><mode>hf2" | |
d7f33f07 JW |
6301 | [(set (match_operand:HF 0 "register_operand" "=w") |
6302 | (FLOATUORS:HF (match_operand:GPI 1 "register_operand" "r")))] | |
6303 | "TARGET_FP_F16INST" | |
6304 | "<su_optab>cvtf\t%h0, %<w>1" | |
6305 | [(set_attr "type" "f_cvti2f")] | |
6306 | ) | |
6307 | ||
bf1e3646 JG |
6308 | (define_expand "<optab>sihf2" |
6309 | [(set (match_operand:HF 0 "register_operand") | |
6310 | (FLOATUORS:HF (match_operand:SI 1 "register_operand")))] | |
6311 | "TARGET_FLOAT" | |
6312 | { | |
6313 | if (TARGET_FP_F16INST) | |
6314 | emit_insn (gen_aarch64_fp16_<optab>sihf2 (operands[0], operands[1])); | |
6315 | else | |
6316 | { | |
6317 | rtx convert_target = gen_reg_rtx (DFmode); | |
6318 | emit_insn (gen_<optab>sidf2 (convert_target, operands[1])); | |
6319 | emit_insn (gen_truncdfhf2 (operands[0], convert_target)); | |
6320 | } | |
6321 | DONE; | |
6322 | } | |
6323 | ) | |
6324 | ||
6325 | ;; For DImode there is no wide enough floating-point mode that we | |
6326 | ;; can convert through natively (TFmode would work, but requires a library | |
6327 | ;; call). However, we know that any value >= 65504 will be rounded | |
6328 | ;; to infinity on conversion. This is well within the range of SImode, so | |
6329 | ;; we can: | |
6330 | ;; Saturate to SImode. | |
6331 | ;; Convert from that to DFmode | |
6332 | ;; Convert from that to HFmode (phew!). | |
6333 | ;; Note that the saturation to SImode requires the SIMD extensions. If | |
6334 | ;; we ever need to provide this pattern where the SIMD extensions are not | |
6335 | ;; available, we would need a different approach. | |
6336 | ||
6337 | (define_expand "<optab>dihf2" | |
6338 | [(set (match_operand:HF 0 "register_operand") | |
6339 | (FLOATUORS:HF (match_operand:DI 1 "register_operand")))] | |
6340 | "TARGET_FLOAT && (TARGET_FP_F16INST || TARGET_SIMD)" | |
6341 | { | |
6342 | if (TARGET_FP_F16INST) | |
6343 | emit_insn (gen_aarch64_fp16_<optab>dihf2 (operands[0], operands[1])); | |
6344 | else | |
6345 | { | |
6346 | rtx sat_target = gen_reg_rtx (SImode); | |
6347 | emit_insn (gen_aarch64_<su_optab>qmovndi (sat_target, operands[1])); | |
6348 | emit_insn (gen_<optab>sihf2 (operands[0], sat_target)); | |
6349 | } | |
6350 | ||
6351 | DONE; | |
6352 | } | |
6353 | ) | |
6354 | ||
3f598afe JW |
6355 | ;; Convert between fixed-point and floating-point (scalar modes) |
6356 | ||
6357 | (define_insn "<FCVT_F2FIXED:fcvt_fixed_insn><GPF:mode>3" | |
6358 | [(set (match_operand:<GPF:FCVT_TARGET> 0 "register_operand" "=r, w") | |
6359 | (unspec:<GPF:FCVT_TARGET> [(match_operand:GPF 1 "register_operand" "w, w") | |
6360 | (match_operand:SI 2 "immediate_operand" "i, i")] | |
6361 | FCVT_F2FIXED))] | |
6362 | "" | |
6363 | "@ | |
1f0e9e34 JG |
6364 | <FCVT_F2FIXED:fcvt_fixed_insn>\t%<GPF:w1>0, %<GPF:s>1, #%2 |
6365 | <FCVT_F2FIXED:fcvt_fixed_insn>\t%<GPF:s>0, %<GPF:s>1, #%2" | |
3f598afe | 6366 | [(set_attr "type" "f_cvtf2i, neon_fp_to_int_<GPF:Vetype>") |
488461d8 | 6367 | (set_attr "arch" "fp,simd")] |
3f598afe JW |
6368 | ) |
6369 | ||
6370 | (define_insn "<FCVT_FIXED2F:fcvt_fixed_insn><GPI:mode>3" | |
6371 | [(set (match_operand:<GPI:FCVT_TARGET> 0 "register_operand" "=w, w") | |
6372 | (unspec:<GPI:FCVT_TARGET> [(match_operand:GPI 1 "register_operand" "r, w") | |
6373 | (match_operand:SI 2 "immediate_operand" "i, i")] | |
6374 | FCVT_FIXED2F))] | |
6375 | "" | |
6376 | "@ | |
1f0e9e34 JG |
6377 | <FCVT_FIXED2F:fcvt_fixed_insn>\t%<GPI:v>0, %<GPI:w>1, #%2 |
6378 | <FCVT_FIXED2F:fcvt_fixed_insn>\t%<GPI:v>0, %<GPI:v>1, #%2" | |
3f598afe | 6379 | [(set_attr "type" "f_cvti2f, neon_int_to_fp_<GPI:Vetype>") |
488461d8 | 6380 | (set_attr "arch" "fp,simd")] |
3f598afe JW |
6381 | ) |
6382 | ||
68ad28c3 JW |
6383 | (define_insn "<FCVT_F2FIXED:fcvt_fixed_insn>hf<mode>3" |
6384 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
6385 | (unspec:GPI [(match_operand:HF 1 "register_operand" "w") | |
6386 | (match_operand:SI 2 "immediate_operand" "i")] | |
6387 | FCVT_F2FIXED))] | |
6388 | "TARGET_FP_F16INST" | |
6389 | "<FCVT_F2FIXED:fcvt_fixed_insn>\t%<GPI:w>0, %h1, #%2" | |
6390 | [(set_attr "type" "f_cvtf2i")] | |
6391 | ) | |
6392 | ||
6393 | (define_insn "<FCVT_FIXED2F:fcvt_fixed_insn><mode>hf3" | |
6394 | [(set (match_operand:HF 0 "register_operand" "=w") | |
6395 | (unspec:HF [(match_operand:GPI 1 "register_operand" "r") | |
6396 | (match_operand:SI 2 "immediate_operand" "i")] | |
6397 | FCVT_FIXED2F))] | |
6398 | "TARGET_FP_F16INST" | |
6399 | "<FCVT_FIXED2F:fcvt_fixed_insn>\t%h0, %<GPI:w>1, #%2" | |
6400 | [(set_attr "type" "f_cvti2f")] | |
6401 | ) | |
6402 | ||
6403 | (define_insn "<FCVT_F2FIXED:fcvt_fixed_insn>hf3" | |
6404 | [(set (match_operand:HI 0 "register_operand" "=w") | |
6405 | (unspec:HI [(match_operand:HF 1 "register_operand" "w") | |
6406 | (match_operand:SI 2 "immediate_operand" "i")] | |
6407 | FCVT_F2FIXED))] | |
6408 | "TARGET_SIMD" | |
6409 | "<FCVT_F2FIXED:fcvt_fixed_insn>\t%h0, %h1, #%2" | |
6410 | [(set_attr "type" "neon_fp_to_int_s")] | |
6411 | ) | |
6412 | ||
6413 | (define_insn "<FCVT_FIXED2F:fcvt_fixed_insn>hi3" | |
6414 | [(set (match_operand:HF 0 "register_operand" "=w") | |
6415 | (unspec:HF [(match_operand:HI 1 "register_operand" "w") | |
6416 | (match_operand:SI 2 "immediate_operand" "i")] | |
6417 | FCVT_FIXED2F))] | |
6418 | "TARGET_SIMD" | |
6419 | "<FCVT_FIXED2F:fcvt_fixed_insn>\t%h0, %h1, #%2" | |
6420 | [(set_attr "type" "neon_int_to_fp_s")] | |
6421 | ) | |
6422 | ||
43e9d192 IB |
6423 | ;; ------------------------------------------------------------------- |
6424 | ;; Floating-point arithmetic | |
6425 | ;; ------------------------------------------------------------------- | |
6426 | ||
6427 | (define_insn "add<mode>3" | |
68ad28c3 JW |
6428 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6429 | (plus:GPF_F16 | |
6430 | (match_operand:GPF_F16 1 "register_operand" "w") | |
6431 | (match_operand:GPF_F16 2 "register_operand" "w")))] | |
43e9d192 IB |
6432 | "TARGET_FLOAT" |
6433 | "fadd\\t%<s>0, %<s>1, %<s>2" | |
68ad28c3 | 6434 | [(set_attr "type" "fadd<stype>")] |
43e9d192 IB |
6435 | ) |
6436 | ||
6437 | (define_insn "sub<mode>3" | |
68ad28c3 JW |
6438 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6439 | (minus:GPF_F16 | |
6440 | (match_operand:GPF_F16 1 "register_operand" "w") | |
6441 | (match_operand:GPF_F16 2 "register_operand" "w")))] | |
43e9d192 IB |
6442 | "TARGET_FLOAT" |
6443 | "fsub\\t%<s>0, %<s>1, %<s>2" | |
68ad28c3 | 6444 | [(set_attr "type" "fadd<stype>")] |
43e9d192 IB |
6445 | ) |
6446 | ||
6447 | (define_insn "mul<mode>3" | |
68ad28c3 JW |
6448 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6449 | (mult:GPF_F16 | |
6450 | (match_operand:GPF_F16 1 "register_operand" "w") | |
6451 | (match_operand:GPF_F16 2 "register_operand" "w")))] | |
43e9d192 IB |
6452 | "TARGET_FLOAT" |
6453 | "fmul\\t%<s>0, %<s>1, %<s>2" | |
68ad28c3 | 6454 | [(set_attr "type" "fmul<stype>")] |
43e9d192 IB |
6455 | ) |
6456 | ||
6457 | (define_insn "*fnmul<mode>3" | |
6458 | [(set (match_operand:GPF 0 "register_operand" "=w") | |
6459 | (mult:GPF | |
6460 | (neg:GPF (match_operand:GPF 1 "register_operand" "w")) | |
6461 | (match_operand:GPF 2 "register_operand" "w")))] | |
2b3fe2e9 SN |
6462 | "TARGET_FLOAT && !flag_rounding_math" |
6463 | "fnmul\\t%<s>0, %<s>1, %<s>2" | |
6464 | [(set_attr "type" "fmul<s>")] | |
6465 | ) | |
6466 | ||
6467 | (define_insn "*fnmul<mode>3" | |
6468 | [(set (match_operand:GPF 0 "register_operand" "=w") | |
6469 | (neg:GPF (mult:GPF | |
6470 | (match_operand:GPF 1 "register_operand" "w") | |
6471 | (match_operand:GPF 2 "register_operand" "w"))))] | |
43e9d192 IB |
6472 | "TARGET_FLOAT" |
6473 | "fnmul\\t%<s>0, %<s>1, %<s>2" | |
99e6af15 | 6474 | [(set_attr "type" "fmul<s>")] |
43e9d192 IB |
6475 | ) |
6476 | ||
79a2bc2d | 6477 | (define_expand "div<mode>3" |
68ad28c3 JW |
6478 | [(set (match_operand:GPF_F16 0 "register_operand") |
6479 | (div:GPF_F16 (match_operand:GPF_F16 1 "general_operand") | |
6480 | (match_operand:GPF_F16 2 "register_operand")))] | |
1be49a38 | 6481 | "TARGET_FLOAT" |
79a2bc2d EM |
6482 | { |
6483 | if (aarch64_emit_approx_div (operands[0], operands[1], operands[2])) | |
6484 | DONE; | |
6485 | ||
6486 | operands[1] = force_reg (<MODE>mode, operands[1]); | |
6487 | }) | |
6488 | ||
6489 | (define_insn "*div<mode>3" | |
68ad28c3 JW |
6490 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6491 | (div:GPF_F16 (match_operand:GPF_F16 1 "register_operand" "w") | |
6492 | (match_operand:GPF_F16 2 "register_operand" "w")))] | |
43e9d192 IB |
6493 | "TARGET_FLOAT" |
6494 | "fdiv\\t%<s>0, %<s>1, %<s>2" | |
68ad28c3 | 6495 | [(set_attr "type" "fdiv<stype>")] |
43e9d192 IB |
6496 | ) |
6497 | ||
6498 | (define_insn "neg<mode>2" | |
d7f33f07 JW |
6499 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6500 | (neg:GPF_F16 (match_operand:GPF_F16 1 "register_operand" "w")))] | |
43e9d192 IB |
6501 | "TARGET_FLOAT" |
6502 | "fneg\\t%<s>0, %<s>1" | |
d7f33f07 | 6503 | [(set_attr "type" "ffarith<stype>")] |
43e9d192 IB |
6504 | ) |
6505 | ||
98daafa0 | 6506 | (define_expand "sqrt<mode>2" |
1bbffb87 DZ |
6507 | [(set (match_operand:GPF_F16 0 "register_operand") |
6508 | (sqrt:GPF_F16 (match_operand:GPF_F16 1 "register_operand")))] | |
98daafa0 EM |
6509 | "TARGET_FLOAT" |
6510 | { | |
6511 | if (aarch64_emit_approx_sqrt (operands[0], operands[1], false)) | |
6512 | DONE; | |
6513 | }) | |
6514 | ||
6515 | (define_insn "*sqrt<mode>2" | |
d7f33f07 JW |
6516 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6517 | (sqrt:GPF_F16 (match_operand:GPF_F16 1 "register_operand" "w")))] | |
43e9d192 IB |
6518 | "TARGET_FLOAT" |
6519 | "fsqrt\\t%<s>0, %<s>1" | |
d7f33f07 | 6520 | [(set_attr "type" "fsqrt<stype>")] |
43e9d192 IB |
6521 | ) |
6522 | ||
6523 | (define_insn "abs<mode>2" | |
d7f33f07 JW |
6524 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6525 | (abs:GPF_F16 (match_operand:GPF_F16 1 "register_operand" "w")))] | |
43e9d192 IB |
6526 | "TARGET_FLOAT" |
6527 | "fabs\\t%<s>0, %<s>1" | |
d7f33f07 | 6528 | [(set_attr "type" "ffarith<stype>")] |
43e9d192 IB |
6529 | ) |
6530 | ||
6531 | ;; Given that smax/smin do not specify the result when either input is NaN, | |
6532 | ;; we could use either FMAXNM or FMAX for smax, and either FMINNM or FMIN | |
6533 | ;; for smin. | |
6534 | ||
6535 | (define_insn "smax<mode>3" | |
6536 | [(set (match_operand:GPF 0 "register_operand" "=w") | |
6537 | (smax:GPF (match_operand:GPF 1 "register_operand" "w") | |
6538 | (match_operand:GPF 2 "register_operand" "w")))] | |
6539 | "TARGET_FLOAT" | |
6540 | "fmaxnm\\t%<s>0, %<s>1, %<s>2" | |
99e6af15 | 6541 | [(set_attr "type" "f_minmax<s>")] |
43e9d192 IB |
6542 | ) |
6543 | ||
6544 | (define_insn "smin<mode>3" | |
6545 | [(set (match_operand:GPF 0 "register_operand" "=w") | |
6546 | (smin:GPF (match_operand:GPF 1 "register_operand" "w") | |
6547 | (match_operand:GPF 2 "register_operand" "w")))] | |
6548 | "TARGET_FLOAT" | |
6549 | "fminnm\\t%<s>0, %<s>1, %<s>2" | |
99e6af15 | 6550 | [(set_attr "type" "f_minmax<s>")] |
43e9d192 IB |
6551 | ) |
6552 | ||
1efafef3 TC |
6553 | ;; Scalar forms for fmax, fmin, fmaxnm, fminnm. |
6554 | ;; fmaxnm and fminnm are used for the fmax<mode>3 standard pattern names, | |
6555 | ;; which implement the IEEE fmax ()/fmin () functions. | |
6d331688 | 6556 | (define_insn "<fmaxmin><mode>3" |
68ad28c3 JW |
6557 | [(set (match_operand:GPF_F16 0 "register_operand" "=w") |
6558 | (unspec:GPF_F16 [(match_operand:GPF_F16 1 "register_operand" "w") | |
6559 | (match_operand:GPF_F16 2 "register_operand" "w")] | |
1efafef3 | 6560 | FMAXMIN_UNS))] |
202d0c11 | 6561 | "TARGET_FLOAT" |
1efafef3 | 6562 | "<maxmin_uns_op>\\t%<s>0, %<s>1, %<s>2" |
68ad28c3 | 6563 | [(set_attr "type" "f_minmax<stype>")] |
202d0c11 DS |
6564 | ) |
6565 | ||
f66e6423 TC |
6566 | (define_expand "lrint<GPF:mode><GPI:mode>2" |
6567 | [(match_operand:GPI 0 "register_operand") | |
6568 | (match_operand:GPF 1 "register_operand")] | |
1013465f | 6569 | "TARGET_FLOAT |
fb802d91 | 6570 | && ((GET_MODE_BITSIZE (<GPF:MODE>mode) <= LONG_TYPE_SIZE) |
1013465f | 6571 | || !flag_trapping_math || flag_fp_int_builtin_inexact)" |
f66e6423 TC |
6572 | { |
6573 | rtx cvt = gen_reg_rtx (<GPF:MODE>mode); | |
6574 | emit_insn (gen_rint<GPF:mode>2 (cvt, operands[1])); | |
6575 | emit_insn (gen_lbtrunc<GPF:mode><GPI:mode>2 (operands[0], cvt)); | |
6576 | DONE; | |
6577 | } | |
6578 | ) | |
6579 | ||
fb0f04fe JG |
6580 | ;; For copysign (x, y), we want to generate: |
6581 | ;; | |
6582 | ;; LDR d2, #(1 << 63) | |
6583 | ;; BSL v2.8b, [y], [x] | |
6584 | ;; | |
315fdae8 RE |
6585 | ;; or another, equivalent, sequence using one of BSL/BIT/BIF. Because |
6586 | ;; we expect these operations to nearly always operate on | |
6587 | ;; floating-point values, we do not want the operation to be | |
6588 | ;; simplified into a bit-field insert operation that operates on the | |
6589 | ;; integer side, since typically that would involve three inter-bank | |
6590 | ;; register copies. As we do not expect copysign to be followed by | |
6591 | ;; other logical operations on the result, it seems preferable to keep | |
6592 | ;; this as an unspec operation, rather than exposing the underlying | |
6593 | ;; logic to the compiler. | |
6594 | ||
6595 | (define_expand "copysign<GPF:mode>3" | |
6596 | [(match_operand:GPF 0 "register_operand") | |
6597 | (match_operand:GPF 1 "register_operand") | |
6598 | (match_operand:GPF 2 "register_operand")] | |
fb0f04fe JG |
6599 | "TARGET_FLOAT && TARGET_SIMD" |
6600 | { | |
315fdae8 RE |
6601 | rtx bitmask = gen_reg_rtx (<V_INT_EQUIV>mode); |
6602 | emit_move_insn (bitmask, GEN_INT (HOST_WIDE_INT_M1U | |
6603 | << (GET_MODE_BITSIZE (<MODE>mode) - 1))); | |
6604 | emit_insn (gen_copysign<mode>3_insn (operands[0], operands[1], operands[2], | |
6605 | bitmask)); | |
fb0f04fe JG |
6606 | DONE; |
6607 | } | |
6608 | ) | |
6609 | ||
315fdae8 RE |
6610 | (define_insn "copysign<GPF:mode>3_insn" |
6611 | [(set (match_operand:GPF 0 "register_operand" "=w,w,w,r") | |
6612 | (unspec:GPF [(match_operand:GPF 1 "register_operand" "w,0,w,r") | |
6613 | (match_operand:GPF 2 "register_operand" "w,w,0,0") | |
6614 | (match_operand:<V_INT_EQUIV> 3 "register_operand" "0,w,w,X")] | |
6615 | UNSPEC_COPYSIGN))] | |
fb0f04fe | 6616 | "TARGET_FLOAT && TARGET_SIMD" |
315fdae8 RE |
6617 | "@ |
6618 | bsl\\t%0.<Vbtype>, %2.<Vbtype>, %1.<Vbtype> | |
6619 | bit\\t%0.<Vbtype>, %2.<Vbtype>, %3.<Vbtype> | |
6620 | bif\\t%0.<Vbtype>, %1.<Vbtype>, %3.<Vbtype> | |
6621 | bfxil\\t%<w1>0, %<w1>1, #0, <sizem1>" | |
6622 | [(set_attr "type" "neon_bsl<q>,neon_bsl<q>,neon_bsl<q>,bfm")] | |
fb0f04fe JG |
6623 | ) |
6624 | ||
315fdae8 | 6625 | |
4261463d TC |
6626 | ;; For xorsign (x, y), we want to generate: |
6627 | ;; | |
6628 | ;; LDR d2, #1<<63 | |
6629 | ;; AND v3.8B, v1.8B, v2.8B | |
6630 | ;; EOR v0.8B, v0.8B, v3.8B | |
6631 | ;; | |
6632 | ||
6633 | (define_expand "xorsign<mode>3" | |
6634 | [(match_operand:GPF 0 "register_operand") | |
6635 | (match_operand:GPF 1 "register_operand") | |
6636 | (match_operand:GPF 2 "register_operand")] | |
6637 | "TARGET_FLOAT && TARGET_SIMD" | |
6638 | { | |
6639 | ||
5f565314 | 6640 | machine_mode imode = <V_INT_EQUIV>mode; |
4261463d TC |
6641 | rtx mask = gen_reg_rtx (imode); |
6642 | rtx op1x = gen_reg_rtx (imode); | |
6643 | rtx op2x = gen_reg_rtx (imode); | |
6644 | ||
6645 | int bits = GET_MODE_BITSIZE (<MODE>mode) - 1; | |
6646 | emit_move_insn (mask, GEN_INT (trunc_int_for_mode (HOST_WIDE_INT_M1U << bits, | |
6647 | imode))); | |
6648 | ||
5f565314 RS |
6649 | emit_insn (gen_and<v_int_equiv>3 (op2x, mask, |
6650 | lowpart_subreg (imode, operands[2], | |
6651 | <MODE>mode))); | |
6652 | emit_insn (gen_xor<v_int_equiv>3 (op1x, | |
6653 | lowpart_subreg (imode, operands[1], | |
6654 | <MODE>mode), | |
6655 | op2x)); | |
4261463d TC |
6656 | emit_move_insn (operands[0], |
6657 | lowpart_subreg (<MODE>mode, op1x, imode)); | |
6658 | DONE; | |
6659 | } | |
6660 | ) | |
6661 | ||
43e9d192 IB |
6662 | ;; ------------------------------------------------------------------- |
6663 | ;; Reload support | |
6664 | ;; ------------------------------------------------------------------- | |
b4f50fd4 RR |
6665 | ;; Reload Scalar Floating point modes from constant pool. |
6666 | ;; The AArch64 port doesn't have __int128 constant move support. | |
1bbffb87 | 6667 | ;; The patterns need constraints due to TARGET_SECONDARY_RELOAD hook. |
0016d8d9 | 6668 | (define_expand "@aarch64_reload_movcp<GPF_TF:mode><P:mode>" |
b4f50fd4 RR |
6669 | [(set (match_operand:GPF_TF 0 "register_operand" "=w") |
6670 | (mem:GPF_TF (match_operand 1 "aarch64_constant_pool_symref" "S"))) | |
6671 | (clobber (match_operand:P 2 "register_operand" "=&r"))] | |
197d1c09 | 6672 | "TARGET_FLOAT" |
b4f50fd4 RR |
6673 | { |
6674 | aarch64_expand_mov_immediate (operands[2], XEXP (operands[1], 0)); | |
6675 | emit_move_insn (operands[0], gen_rtx_MEM (<GPF_TF:MODE>mode, operands[2])); | |
6676 | DONE; | |
6677 | } | |
6678 | ) | |
6679 | ||
6680 | ;; Reload Vector modes from constant pool. | |
0016d8d9 | 6681 | (define_expand "@aarch64_reload_movcp<VALL:mode><P:mode>" |
b4f50fd4 RR |
6682 | [(set (match_operand:VALL 0 "register_operand" "=w") |
6683 | (mem:VALL (match_operand 1 "aarch64_constant_pool_symref" "S"))) | |
6684 | (clobber (match_operand:P 2 "register_operand" "=&r"))] | |
197d1c09 | 6685 | "TARGET_FLOAT" |
b4f50fd4 RR |
6686 | { |
6687 | aarch64_expand_mov_immediate (operands[2], XEXP (operands[1], 0)); | |
6688 | emit_move_insn (operands[0], gen_rtx_MEM (<VALL:MODE>mode, operands[2])); | |
6689 | DONE; | |
6690 | } | |
6691 | ) | |
43e9d192 | 6692 | |
0016d8d9 | 6693 | (define_expand "@aarch64_reload_mov<mode>" |
43e9d192 IB |
6694 | [(set (match_operand:TX 0 "register_operand" "=w") |
6695 | (match_operand:TX 1 "register_operand" "w")) | |
6696 | (clobber (match_operand:DI 2 "register_operand" "=&r")) | |
6697 | ] | |
a636bb40 | 6698 | "TARGET_FLOAT" |
43e9d192 IB |
6699 | { |
6700 | rtx op0 = simplify_gen_subreg (TImode, operands[0], <MODE>mode, 0); | |
6701 | rtx op1 = simplify_gen_subreg (TImode, operands[1], <MODE>mode, 0); | |
6702 | gen_aarch64_movtilow_tilow (op0, op1); | |
6703 | gen_aarch64_movdi_tihigh (operands[2], op1); | |
6704 | gen_aarch64_movtihigh_di (op0, operands[2]); | |
6705 | DONE; | |
6706 | } | |
6707 | ) | |
6708 | ||
6709 | ;; The following secondary reload helpers patterns are invoked | |
6710 | ;; after or during reload as we don't want these patterns to start | |
6711 | ;; kicking in during the combiner. | |
2a9704d0 | 6712 | |
0016d8d9 | 6713 | (define_insn "@aarch64_movdi_<mode>low" |
43e9d192 | 6714 | [(set (match_operand:DI 0 "register_operand" "=r") |
4c81f958 JG |
6715 | (zero_extract:DI (match_operand:TX 1 "register_operand" "w") |
6716 | (const_int 64) (const_int 0)))] | |
a636bb40 | 6717 | "TARGET_FLOAT && (reload_completed || reload_in_progress)" |
43e9d192 | 6718 | "fmov\\t%x0, %d1" |
c1dfcec7 | 6719 | [(set_attr "type" "f_mrc") |
43e9d192 IB |
6720 | (set_attr "length" "4") |
6721 | ]) | |
6722 | ||
0016d8d9 | 6723 | (define_insn "@aarch64_movdi_<mode>high" |
43e9d192 | 6724 | [(set (match_operand:DI 0 "register_operand" "=r") |
4c81f958 JG |
6725 | (zero_extract:DI (match_operand:TX 1 "register_operand" "w") |
6726 | (const_int 64) (const_int 64)))] | |
a636bb40 | 6727 | "TARGET_FLOAT && (reload_completed || reload_in_progress)" |
43e9d192 | 6728 | "fmov\\t%x0, %1.d[1]" |
c1dfcec7 | 6729 | [(set_attr "type" "f_mrc") |
43e9d192 IB |
6730 | (set_attr "length" "4") |
6731 | ]) | |
6732 | ||
0016d8d9 | 6733 | (define_insn "@aarch64_mov<mode>high_di" |
12dc6974 | 6734 | [(set (zero_extract:TX (match_operand:TX 0 "register_operand" "+w") |
43e9d192 | 6735 | (const_int 64) (const_int 64)) |
12dc6974 | 6736 | (zero_extend:TX (match_operand:DI 1 "register_operand" "r")))] |
a636bb40 | 6737 | "TARGET_FLOAT && (reload_completed || reload_in_progress)" |
43e9d192 | 6738 | "fmov\\t%0.d[1], %x1" |
c1dfcec7 | 6739 | [(set_attr "type" "f_mcr") |
43e9d192 IB |
6740 | (set_attr "length" "4") |
6741 | ]) | |
6742 | ||
0016d8d9 | 6743 | (define_insn "@aarch64_mov<mode>low_di" |
12dc6974 SN |
6744 | [(set (match_operand:TX 0 "register_operand" "=w") |
6745 | (zero_extend:TX (match_operand:DI 1 "register_operand" "r")))] | |
a636bb40 | 6746 | "TARGET_FLOAT && (reload_completed || reload_in_progress)" |
43e9d192 | 6747 | "fmov\\t%d0, %x1" |
c1dfcec7 | 6748 | [(set_attr "type" "f_mcr") |
43e9d192 IB |
6749 | (set_attr "length" "4") |
6750 | ]) | |
6751 | ||
6752 | (define_insn "aarch64_movtilow_tilow" | |
6753 | [(set (match_operand:TI 0 "register_operand" "=w") | |
a636bb40 | 6754 | (zero_extend:TI |
43e9d192 | 6755 | (truncate:DI (match_operand:TI 1 "register_operand" "w"))))] |
a636bb40 | 6756 | "TARGET_FLOAT && (reload_completed || reload_in_progress)" |
43e9d192 | 6757 | "fmov\\t%d0, %d1" |
d7256bb8 | 6758 | [(set_attr "type" "fmov") |
43e9d192 IB |
6759 | (set_attr "length" "4") |
6760 | ]) | |
6761 | ||
6762 | ;; There is a deliberate reason why the parameters of high and lo_sum's | |
6763 | ;; don't have modes for ADRP and ADD instructions. This is to allow high | |
6764 | ;; and lo_sum's to be used with the labels defining the jump tables in | |
6765 | ;; rodata section. | |
6766 | ||
28514dda | 6767 | (define_expand "add_losym" |
1bbffb87 DZ |
6768 | [(set (match_operand 0 "register_operand") |
6769 | (lo_sum (match_operand 1 "register_operand") | |
6770 | (match_operand 2 "aarch64_valid_symref")))] | |
43e9d192 | 6771 | "" |
28514dda | 6772 | { |
ef4bddc2 | 6773 | machine_mode mode = GET_MODE (operands[0]); |
28514dda YZ |
6774 | |
6775 | emit_insn ((mode == DImode | |
6776 | ? gen_add_losym_di | |
6777 | : gen_add_losym_si) (operands[0], | |
6778 | operands[1], | |
6779 | operands[2])); | |
6780 | DONE; | |
6781 | }) | |
6782 | ||
6783 | (define_insn "add_losym_<mode>" | |
6784 | [(set (match_operand:P 0 "register_operand" "=r") | |
6785 | (lo_sum:P (match_operand:P 1 "register_operand" "r") | |
6786 | (match_operand 2 "aarch64_valid_symref" "S")))] | |
6787 | "" | |
d6591257 | 6788 | "add\\t%<w>0, %<w>1, :lo12:%c2" |
d8e11262 | 6789 | [(set_attr "type" "alu_imm")] |
28514dda | 6790 | ) |
43e9d192 | 6791 | |
1b1e81f8 JW |
6792 | (define_insn "ldr_got_small_28k_<mode>" |
6793 | [(set (match_operand:PTR 0 "register_operand" "=r") | |
6794 | (unspec:PTR [(mem:PTR (lo_sum:PTR | |
6795 | (match_operand:PTR 1 "register_operand" "r") | |
6796 | (match_operand:PTR 2 "aarch64_valid_symref" "S")))] | |
6797 | UNSPEC_GOTSMALLPIC28K))] | |
6798 | "" | |
d6591257 | 6799 | "ldr\\t%<w>0, [%1, #:<got_modifier>:%c2]" |
db46a2e6 | 6800 | [(set_attr "type" "load_<ldst_sz>")] |
1b1e81f8 JW |
6801 | ) |
6802 | ||
6803 | (define_insn "ldr_got_small_28k_sidi" | |
6804 | [(set (match_operand:DI 0 "register_operand" "=r") | |
6805 | (zero_extend:DI | |
6806 | (unspec:SI [(mem:SI (lo_sum:DI | |
6807 | (match_operand:DI 1 "register_operand" "r") | |
6808 | (match_operand:DI 2 "aarch64_valid_symref" "S")))] | |
6809 | UNSPEC_GOTSMALLPIC28K)))] | |
6810 | "TARGET_ILP32" | |
d6591257 | 6811 | "ldr\\t%w0, [%1, #:gotpage_lo14:%c2]" |
89b2133e | 6812 | [(set_attr "type" "load_4")] |
1b1e81f8 JW |
6813 | ) |
6814 | ||
d91480de D |
6815 | (define_insn "@ldr_got_tiny_<mode>" |
6816 | [(set (match_operand:PTR 0 "register_operand" "=r") | |
6817 | (unspec:PTR [(match_operand:PTR 1 "aarch64_valid_symref" "S")] | |
6818 | UNSPEC_GOTTINYPIC))] | |
87dd8ab0 | 6819 | "" |
d91480de D |
6820 | "ldr\t%<w>0, %L1" |
6821 | [(set_attr "type" "load_<ldst_sz>")] | |
6822 | ) | |
6823 | ||
6824 | (define_insn "ldr_got_tiny_sidi" | |
6825 | [(set (match_operand:DI 0 "register_operand" "=r") | |
6826 | (zero_extend:DI | |
6827 | (unspec:SI [(match_operand:DI 1 "aarch64_valid_symref" "S")] | |
6828 | UNSPEC_GOTTINYPIC)))] | |
6829 | "TARGET_ILP32" | |
6830 | "ldr\t%w0, %L1" | |
6831 | [(set_attr "type" "load_4")] | |
87dd8ab0 MS |
6832 | ) |
6833 | ||
43e9d192 IB |
6834 | (define_insn "aarch64_load_tp_hard" |
6835 | [(set (match_operand:DI 0 "register_operand" "=r") | |
6836 | (unspec:DI [(const_int 0)] UNSPEC_TLS))] | |
6837 | "" | |
6838 | "mrs\\t%0, tpidr_el0" | |
99e6af15 | 6839 | [(set_attr "type" "mrs")] |
43e9d192 IB |
6840 | ) |
6841 | ||
6842 | ;; The TLS ABI specifically requires that the compiler does not schedule | |
6843 | ;; instructions in the TLS stubs, in order to enable linker relaxation. | |
6844 | ;; Therefore we treat the stubs as an atomic sequence. | |
23b88fda | 6845 | (define_expand "tlsgd_small_<mode>" |
87ca615a | 6846 | [(parallel [(set (match_operand:PTR 0 "register_operand") |
43e9d192 | 6847 | (call (mem:DI (match_dup 2)) (const_int 1))) |
08cc4d92 | 6848 | (unspec:DI [(const_int 0)] UNSPEC_CALLEE_ABI) |
87ca615a | 6849 | (unspec:DI [(match_operand 1 "aarch64_valid_symref")] UNSPEC_GOTSMALLTLS) |
43e9d192 IB |
6850 | (clobber (reg:DI LR_REGNUM))])] |
6851 | "" | |
6852 | { | |
6853 | operands[2] = aarch64_tls_get_addr (); | |
6854 | }) | |
6855 | ||
23b88fda | 6856 | (define_insn "*tlsgd_small_<mode>" |
87ca615a | 6857 | [(set (match_operand:PTR 0 "register_operand" "") |
43e9d192 | 6858 | (call (mem:DI (match_operand:DI 2 "" "")) (const_int 1))) |
08cc4d92 | 6859 | (unspec:DI [(const_int 0)] UNSPEC_CALLEE_ABI) |
87ca615a | 6860 | (unspec:DI [(match_operand 1 "aarch64_valid_symref" "S")] UNSPEC_GOTSMALLTLS) |
43e9d192 IB |
6861 | (clobber (reg:DI LR_REGNUM)) |
6862 | ] | |
6863 | "" | |
6864 | "adrp\\tx0, %A1\;add\\tx0, x0, %L1\;bl\\t%2\;nop" | |
c1dfcec7 | 6865 | [(set_attr "type" "call") |
43e9d192 IB |
6866 | (set_attr "length" "16")]) |
6867 | ||
621ad2de AP |
6868 | (define_insn "tlsie_small_<mode>" |
6869 | [(set (match_operand:PTR 0 "register_operand" "=r") | |
6870 | (unspec:PTR [(match_operand 1 "aarch64_tls_ie_symref" "S")] | |
43e9d192 IB |
6871 | UNSPEC_GOTSMALLTLS))] |
6872 | "" | |
621ad2de | 6873 | "adrp\\t%0, %A1\;ldr\\t%<w>0, [%0, #%L1]" |
89b2133e | 6874 | [(set_attr "type" "load_4") |
43e9d192 IB |
6875 | (set_attr "length" "8")] |
6876 | ) | |
6877 | ||
621ad2de | 6878 | (define_insn "tlsie_small_sidi" |
43e9d192 | 6879 | [(set (match_operand:DI 0 "register_operand" "=r") |
621ad2de AP |
6880 | (zero_extend:DI |
6881 | (unspec:SI [(match_operand 1 "aarch64_tls_ie_symref" "S")] | |
6882 | UNSPEC_GOTSMALLTLS)))] | |
6883 | "" | |
6884 | "adrp\\t%0, %A1\;ldr\\t%w0, [%0, #%L1]" | |
89b2133e | 6885 | [(set_attr "type" "load_4") |
621ad2de AP |
6886 | (set_attr "length" "8")] |
6887 | ) | |
6888 | ||
5ae7caad JW |
6889 | (define_insn "tlsie_tiny_<mode>" |
6890 | [(set (match_operand:PTR 0 "register_operand" "=&r") | |
6891 | (unspec:PTR [(match_operand 1 "aarch64_tls_ie_symref" "S") | |
6892 | (match_operand:PTR 2 "register_operand" "r")] | |
6893 | UNSPEC_GOTTINYTLS))] | |
6894 | "" | |
6895 | "ldr\\t%<w>0, %L1\;add\\t%<w>0, %<w>0, %<w>2" | |
6896 | [(set_attr "type" "multiple") | |
6897 | (set_attr "length" "8")] | |
6898 | ) | |
6899 | ||
6900 | (define_insn "tlsie_tiny_sidi" | |
6901 | [(set (match_operand:DI 0 "register_operand" "=&r") | |
6902 | (zero_extend:DI | |
6903 | (unspec:SI [(match_operand 1 "aarch64_tls_ie_symref" "S") | |
6904 | (match_operand:DI 2 "register_operand" "r") | |
6905 | ] | |
6906 | UNSPEC_GOTTINYTLS)))] | |
6907 | "" | |
5b295a81 | 6908 | "ldr\\t%w0, %L1\;add\\t%w0, %w0, %w2" |
5ae7caad JW |
6909 | [(set_attr "type" "multiple") |
6910 | (set_attr "length" "8")] | |
6911 | ) | |
6912 | ||
cbf5629e JW |
6913 | (define_insn "tlsle12_<mode>" |
6914 | [(set (match_operand:P 0 "register_operand" "=r") | |
6915 | (unspec:P [(match_operand:P 1 "register_operand" "r") | |
6916 | (match_operand 2 "aarch64_tls_le_symref" "S")] | |
6917 | UNSPEC_TLSLE12))] | |
621ad2de | 6918 | "" |
cbf5629e JW |
6919 | "add\\t%<w>0, %<w>1, #%L2"; |
6920 | [(set_attr "type" "alu_sreg") | |
6921 | (set_attr "length" "4")] | |
6922 | ) | |
621ad2de | 6923 | |
cbf5629e | 6924 | (define_insn "tlsle24_<mode>" |
621ad2de | 6925 | [(set (match_operand:P 0 "register_operand" "=r") |
cbf5629e JW |
6926 | (unspec:P [(match_operand:P 1 "register_operand" "r") |
6927 | (match_operand 2 "aarch64_tls_le_symref" "S")] | |
6928 | UNSPEC_TLSLE24))] | |
43e9d192 | 6929 | "" |
0699caae | 6930 | "add\\t%<w>0, %<w>1, #%G2, lsl #12\;add\\t%<w>0, %<w>0, #%L2" |
cbf5629e | 6931 | [(set_attr "type" "multiple") |
43e9d192 IB |
6932 | (set_attr "length" "8")] |
6933 | ) | |
6934 | ||
cbf5629e JW |
6935 | (define_insn "tlsle32_<mode>" |
6936 | [(set (match_operand:P 0 "register_operand" "=r") | |
6937 | (unspec:P [(match_operand 1 "aarch64_tls_le_symref" "S")] | |
6938 | UNSPEC_TLSLE32))] | |
6939 | "" | |
6940 | "movz\\t%<w>0, #:tprel_g1:%1\;movk\\t%<w>0, #:tprel_g0_nc:%1" | |
6941 | [(set_attr "type" "multiple") | |
6942 | (set_attr "length" "8")] | |
6943 | ) | |
6944 | ||
6945 | (define_insn "tlsle48_<mode>" | |
6946 | [(set (match_operand:P 0 "register_operand" "=r") | |
6947 | (unspec:P [(match_operand 1 "aarch64_tls_le_symref" "S")] | |
6948 | UNSPEC_TLSLE48))] | |
6949 | "" | |
6950 | "movz\\t%<w>0, #:tprel_g2:%1\;movk\\t%<w>0, #:tprel_g1_nc:%1\;movk\\t%<w>0, #:tprel_g0_nc:%1" | |
6951 | [(set_attr "type" "multiple") | |
6952 | (set_attr "length" "12")] | |
6953 | ) | |
6954 | ||
80c13ac5 RS |
6955 | (define_expand "tlsdesc_small_<mode>" |
6956 | [(unspec:PTR [(match_operand 0 "aarch64_valid_symref")] UNSPEC_TLSDESC)] | |
6957 | "TARGET_TLS_DESC" | |
6958 | { | |
6959 | if (TARGET_SVE) | |
bb6ce448 RS |
6960 | { |
6961 | rtx abi = gen_int_mode (aarch64_tlsdesc_abi_id (), DImode); | |
6962 | rtx_insn *call | |
6963 | = emit_call_insn (gen_tlsdesc_small_sve_<mode> (operands[0], abi)); | |
6964 | RTL_CONST_CALL_P (call) = 1; | |
6965 | } | |
80c13ac5 RS |
6966 | else |
6967 | emit_insn (gen_tlsdesc_small_advsimd_<mode> (operands[0])); | |
6968 | DONE; | |
6969 | } | |
6970 | ) | |
6971 | ||
6972 | ;; tlsdesc calls preserve all core and Advanced SIMD registers except | |
6973 | ;; R0 and LR. | |
6974 | (define_insn "tlsdesc_small_advsimd_<mode>" | |
621ad2de AP |
6975 | [(set (reg:PTR R0_REGNUM) |
6976 | (unspec:PTR [(match_operand 0 "aarch64_valid_symref" "S")] | |
80c13ac5 | 6977 | UNSPEC_TLSDESC)) |
43e9d192 | 6978 | (clobber (reg:DI LR_REGNUM)) |
0bc11714 | 6979 | (clobber (reg:CC CC_REGNUM)) |
0e510d18 ST |
6980 | (clobber (match_scratch:DI 1 "=r")) |
6981 | (use (reg:DI FP_REGNUM))] | |
80c13ac5 RS |
6982 | "TARGET_TLS_DESC && !TARGET_SVE" |
6983 | "adrp\\tx0, %A0\;ldr\\t%<w>1, [x0, #%L0]\;add\\t<w>0, <w>0, %L0\;.tlsdesccall\\t%0\;blr\\t%1" | |
6984 | [(set_attr "type" "call") | |
6985 | (set_attr "length" "16")]) | |
6986 | ||
bb6ce448 RS |
6987 | ;; For SVE, model tlsdesc calls as normal calls, with the callee ABI |
6988 | ;; describing the extra call-preserved guarantees. This would work | |
6989 | ;; for non-SVE too, but avoiding a call is probably better if we can. | |
80c13ac5 RS |
6990 | (define_insn "tlsdesc_small_sve_<mode>" |
6991 | [(set (reg:PTR R0_REGNUM) | |
bb6ce448 RS |
6992 | (call (mem:DI (unspec:PTR |
6993 | [(match_operand 0 "aarch64_valid_symref")] | |
6994 | UNSPEC_TLSDESC)) | |
6995 | (const_int 0))) | |
6996 | (unspec:DI [(match_operand:DI 1 "const_int_operand")] UNSPEC_CALLEE_ABI) | |
ba2913f6 | 6997 | (clobber (reg:DI LR_REGNUM))] |
80c13ac5 | 6998 | "TARGET_TLS_DESC && TARGET_SVE" |
ba2913f6 | 6999 | "adrp\\tx0, %A0\;ldr\\t<w>30, [x0, #%L0]\;add\\t<w>0, <w>0, %L0\;.tlsdesccall\\t%0\;blr\\tx30" |
c1dfcec7 | 7000 | [(set_attr "type" "call") |
43e9d192 IB |
7001 | (set_attr "length" "16")]) |
7002 | ||
7003 | (define_insn "stack_tie" | |
7004 | [(set (mem:BLK (scratch)) | |
7005 | (unspec:BLK [(match_operand:DI 0 "register_operand" "rk") | |
7006 | (match_operand:DI 1 "register_operand" "rk")] | |
7007 | UNSPEC_PRLG_STK))] | |
7008 | "" | |
7009 | "" | |
7010 | [(set_attr "length" "0")] | |
7011 | ) | |
7012 | ||
e1d5d19e KT |
7013 | (define_insn "aarch64_fjcvtzs" |
7014 | [(set (match_operand:SI 0 "register_operand" "=r") | |
7015 | (unspec:SI [(match_operand:DF 1 "register_operand" "w")] | |
d2b86e14 AC |
7016 | UNSPEC_FJCVTZS)) |
7017 | (clobber (reg:CC CC_REGNUM))] | |
e1d5d19e KT |
7018 | "TARGET_JSCVT" |
7019 | "fjcvtzs\\t%w0, %d1" | |
7020 | [(set_attr "type" "f_cvtf2i")] | |
7021 | ) | |
7022 | ||
db58fd89 JW |
7023 | ;; Pointer authentication patterns are always provided. In architecture |
7024 | ;; revisions prior to ARMv8.3-A these HINT instructions operate as NOPs. | |
7025 | ;; This lets the user write portable software which authenticates pointers | |
7026 | ;; when run on something which implements ARMv8.3-A, and which runs | |
7027 | ;; correctly, but does not authenticate pointers, where ARMv8.3-A is not | |
7028 | ;; implemented. | |
7029 | ||
7030 | ;; Signing/Authenticating R30 using SP as the salt. | |
7031 | ||
7032 | (define_insn "<pauth_mnem_prefix>sp" | |
7033 | [(set (reg:DI R30_REGNUM) | |
7034 | (unspec:DI [(reg:DI R30_REGNUM) (reg:DI SP_REGNUM)] PAUTH_LR_SP))] | |
7035 | "" | |
8fc16d72 | 7036 | "hint\t<pauth_hint_num> // <pauth_mnem_prefix>sp"; |
db58fd89 JW |
7037 | ) |
7038 | ||
7039 | ;; Signing/Authenticating X17 using X16 as the salt. | |
7040 | ||
7041 | (define_insn "<pauth_mnem_prefix>1716" | |
7042 | [(set (reg:DI R17_REGNUM) | |
7043 | (unspec:DI [(reg:DI R17_REGNUM) (reg:DI R16_REGNUM)] PAUTH_17_16))] | |
7044 | "" | |
8fc16d72 | 7045 | "hint\t<pauth_hint_num> // <pauth_mnem_prefix>1716"; |
db58fd89 JW |
7046 | ) |
7047 | ||
7048 | ;; Stripping the signature in R30. | |
7049 | ||
7050 | (define_insn "xpaclri" | |
7051 | [(set (reg:DI R30_REGNUM) (unspec:DI [(reg:DI R30_REGNUM)] UNSPEC_XPACLRI))] | |
7052 | "" | |
7053 | "hint\t7 // xpaclri" | |
7054 | ) | |
7055 | ||
a3eb8a52 EB |
7056 | ;; UNSPEC_VOLATILE is considered to use and clobber all hard registers and |
7057 | ;; all of memory. This blocks insns from being moved across this point. | |
7058 | ||
7059 | (define_insn "blockage" | |
7060 | [(unspec_volatile [(const_int 0)] UNSPECV_BLOCKAGE)] | |
7061 | "" | |
7062 | "" | |
7063 | [(set_attr "length" "0") | |
7064 | (set_attr "type" "block")] | |
7065 | ) | |
7066 | ||
5f5c5e0f | 7067 | (define_insn "probe_stack_range" |
cd1bef27 | 7068 | [(set (match_operand:DI 0 "register_operand" "=rk") |
5f5c5e0f EB |
7069 | (unspec_volatile:DI [(match_operand:DI 1 "register_operand" "0") |
7070 | (match_operand:DI 2 "register_operand" "r")] | |
7071 | UNSPECV_PROBE_STACK_RANGE))] | |
a3eb8a52 EB |
7072 | "" |
7073 | { | |
7074 | return aarch64_output_probe_stack_range (operands[0], operands[2]); | |
7075 | } | |
7076 | [(set_attr "length" "32")] | |
7077 | ) | |
7078 | ||
eb471ba3 TC |
7079 | ;; This instruction is used to generate the stack clash stack adjustment and |
7080 | ;; probing loop. We can't change the control flow during prologue and epilogue | |
7081 | ;; code generation. So we must emit a volatile unspec and expand it later on. | |
7082 | ||
7083 | (define_insn "@probe_sve_stack_clash_<mode>" | |
7084 | [(set (match_operand:P 0 "register_operand" "=rk") | |
7085 | (unspec_volatile:P [(match_operand:P 1 "register_operand" "0") | |
7086 | (match_operand:P 2 "register_operand" "r") | |
7087 | (match_operand:P 3 "const_int_operand" "n") | |
7088 | (match_operand:P 4 "aarch64_plus_immediate" "L")] | |
7089 | UNSPECV_PROBE_STACK_RANGE))] | |
7090 | "TARGET_SVE" | |
7091 | { | |
7092 | return aarch64_output_probe_sve_stack_clash (operands[0], operands[2], | |
7093 | operands[3], operands[4]); | |
7094 | } | |
7095 | [(set_attr "length" "28")] | |
7096 | ) | |
7097 | ||
a5bc806c AP |
7098 | ;; Named pattern for expanding thread pointer reference. |
7099 | (define_expand "get_thread_pointerdi" | |
1bbffb87 | 7100 | [(match_operand:DI 0 "register_operand")] |
a5bc806c AP |
7101 | "" |
7102 | { | |
7103 | rtx tmp = aarch64_load_tp (operands[0]); | |
7104 | if (tmp != operands[0]) | |
7105 | emit_move_insn (operands[0], tmp); | |
7106 | DONE; | |
7107 | }) | |
7108 | ||
74b27d8e RS |
7109 | ;; Defined for -mstack-protector-guard=sysreg, which goes through this |
7110 | ;; pattern rather than stack_protect_combined_set. Our implementation | |
7111 | ;; of the latter can handle both. | |
36e17020 VK |
7112 | (define_expand "stack_protect_set" |
7113 | [(match_operand 0 "memory_operand") | |
74b27d8e | 7114 | (match_operand 1 "")] |
36e17020 VK |
7115 | "" |
7116 | { | |
74b27d8e RS |
7117 | emit_insn (gen_stack_protect_combined_set (operands[0], operands[1])); |
7118 | DONE; | |
7119 | }) | |
36e17020 | 7120 | |
74b27d8e RS |
7121 | (define_expand "stack_protect_combined_set" |
7122 | [(match_operand 0 "memory_operand") | |
7123 | (match_operand 1 "")] | |
7124 | "" | |
7125 | { | |
7126 | machine_mode mode = GET_MODE (operands[0]); | |
7127 | operands[1] = aarch64_stack_protect_canary_mem (mode, operands[1], | |
7128 | AARCH64_SALT_SSP_SET); | |
36e17020 VK |
7129 | emit_insn ((mode == DImode |
7130 | ? gen_stack_protect_set_di | |
7131 | : gen_stack_protect_set_si) (operands[0], operands[1])); | |
7132 | DONE; | |
7133 | }) | |
7134 | ||
74b27d8e | 7135 | ;; Operand 1 is either AARCH64_SALT_SSP_SET or AARCH64_SALT_SSP_TEST. |
cd0b2d36 RR |
7136 | (define_insn "reg_stack_protect_address_<mode>" |
7137 | [(set (match_operand:PTR 0 "register_operand" "=r") | |
74b27d8e RS |
7138 | (unspec:PTR [(match_operand 1 "const_int_operand")] |
7139 | UNSPEC_SSP_SYSREG))] | |
cd0b2d36 RR |
7140 | "aarch64_stack_protector_guard != SSP_GLOBAL" |
7141 | { | |
7142 | char buf[150]; | |
7143 | snprintf (buf, 150, "mrs\\t%%<w>0, %s", | |
7144 | aarch64_stack_protector_guard_reg_str); | |
7145 | output_asm_insn (buf, operands); | |
7146 | return ""; | |
7147 | } | |
7148 | [(set_attr "type" "mrs")]) | |
7149 | ||
a7e73b41 RE |
7150 | ;; DO NOT SPLIT THIS PATTERN. It is important for security reasons that the |
7151 | ;; canary value does not live beyond the life of this sequence. | |
36e17020 VK |
7152 | (define_insn "stack_protect_set_<mode>" |
7153 | [(set (match_operand:PTR 0 "memory_operand" "=m") | |
7154 | (unspec:PTR [(match_operand:PTR 1 "memory_operand" "m")] | |
7155 | UNSPEC_SP_SET)) | |
7156 | (set (match_scratch:PTR 2 "=&r") (const_int 0))] | |
7157 | "" | |
a7e73b41 | 7158 | "ldr\\t%<w>2, %1\;str\\t%<w>2, %0\;mov\t%<w>2, 0" |
36e17020 VK |
7159 | [(set_attr "length" "12") |
7160 | (set_attr "type" "multiple")]) | |
7161 | ||
74b27d8e RS |
7162 | ;; Defined for -mstack-protector-guard=sysreg, which goes through this |
7163 | ;; pattern rather than stack_protect_combined_test. Our implementation | |
7164 | ;; of the latter can handle both. | |
36e17020 VK |
7165 | (define_expand "stack_protect_test" |
7166 | [(match_operand 0 "memory_operand") | |
74b27d8e | 7167 | (match_operand 1 "") |
36e17020 VK |
7168 | (match_operand 2)] |
7169 | "" | |
7170 | { | |
74b27d8e RS |
7171 | emit_insn (gen_stack_protect_combined_test (operands[0], operands[1], |
7172 | operands[2])); | |
7173 | DONE; | |
7174 | }) | |
9c6ab05f | 7175 | |
74b27d8e RS |
7176 | (define_expand "stack_protect_combined_test" |
7177 | [(match_operand 0 "memory_operand") | |
7178 | (match_operand 1 "") | |
7179 | (match_operand 2)] | |
7180 | "" | |
7181 | { | |
7182 | machine_mode mode = GET_MODE (operands[0]); | |
7183 | operands[1] = aarch64_stack_protect_canary_mem (mode, operands[1], | |
7184 | AARCH64_SALT_SSP_TEST); | |
36e17020 | 7185 | emit_insn ((mode == DImode |
fe1a2642 RS |
7186 | ? gen_stack_protect_test_di |
7187 | : gen_stack_protect_test_si) (operands[0], operands[1])); | |
7188 | ||
7189 | rtx cc_reg = gen_rtx_REG (CCmode, CC_REGNUM); | |
7190 | emit_jump_insn (gen_condjump (gen_rtx_EQ (VOIDmode, cc_reg, const0_rtx), | |
7191 | cc_reg, operands[2])); | |
36e17020 VK |
7192 | DONE; |
7193 | }) | |
7194 | ||
fe1a2642 RS |
7195 | ;; DO NOT SPLIT THIS PATTERN. It is important for security reasons that the |
7196 | ;; canary value does not live beyond the end of this sequence. | |
36e17020 | 7197 | (define_insn "stack_protect_test_<mode>" |
fe1a2642 RS |
7198 | [(set (reg:CC CC_REGNUM) |
7199 | (unspec:CC [(match_operand:PTR 0 "memory_operand" "m") | |
7200 | (match_operand:PTR 1 "memory_operand" "m")] | |
7201 | UNSPEC_SP_TEST)) | |
7202 | (clobber (match_scratch:PTR 2 "=&r")) | |
b4e9251d | 7203 | (clobber (match_scratch:PTR 3 "=&r"))] |
36e17020 | 7204 | "" |
fe1a2642 RS |
7205 | "ldr\t%<w>2, %0\;ldr\t%<w>3, %1\;subs\t%<w>2, %<w>2, %<w>3\;mov\t%3, 0" |
7206 | [(set_attr "length" "16") | |
36e17020 VK |
7207 | (set_attr "type" "multiple")]) |
7208 | ||
0d7e5fa6 AC |
7209 | ;; Write into the Floating-point Status or Control Register. |
7210 | (define_insn "@aarch64_set_<fpscr_name><GPI:mode>" | |
7211 | [(unspec_volatile [(match_operand:GPI 0 "register_operand" "r")] SET_FPSCR)] | |
aa87aced | 7212 | "" |
0d7e5fa6 | 7213 | "msr\\t<fpscr_name>, %0" |
aa87aced KV |
7214 | [(set_attr "type" "mrs")]) |
7215 | ||
0d7e5fa6 AC |
7216 | ;; Read into the Floating-point Status or Control Register. |
7217 | (define_insn "@aarch64_get_<fpscr_name><GPI:mode>" | |
7218 | [(set (match_operand:GPI 0 "register_operand" "=r") | |
7219 | (unspec_volatile:GPI [(const_int 0)] GET_FPSCR))] | |
aa87aced | 7220 | "" |
0d7e5fa6 | 7221 | "mrs\\t%0, <fpscr_name>" |
aa87aced KV |
7222 | [(set_attr "type" "mrs")]) |
7223 | ||
594bdd53 FY |
7224 | ;; Define the subtract-one-and-jump insns so loop.c |
7225 | ;; knows what to generate. | |
7226 | (define_expand "doloop_end" | |
7227 | [(use (match_operand 0 "" "")) ; loop pseudo | |
7228 | (use (match_operand 1 "" ""))] ; label | |
7229 | "optimize > 0 && flag_modulo_sched" | |
7230 | { | |
7231 | rtx s0; | |
7232 | rtx bcomp; | |
7233 | rtx loc_ref; | |
7234 | rtx cc_reg; | |
7235 | rtx insn; | |
7236 | rtx cmp; | |
7237 | ||
7238 | /* Currently SMS relies on the do-loop pattern to recognize loops | |
7239 | where (1) the control part consists of all insns defining and/or | |
7240 | using a certain 'count' register and (2) the loop count can be | |
7241 | adjusted by modifying this register prior to the loop. | |
7242 | ??? The possible introduction of a new block to initialize the | |
7243 | new IV can potentially affect branch optimizations. */ | |
7244 | ||
7245 | if (GET_MODE (operands[0]) != DImode) | |
7246 | FAIL; | |
7247 | ||
7248 | s0 = operands [0]; | |
7249 | insn = emit_insn (gen_adddi3_compare0 (s0, s0, GEN_INT (-1))); | |
7250 | ||
7251 | cmp = XVECEXP (PATTERN (insn), 0, 0); | |
7252 | cc_reg = SET_DEST (cmp); | |
7253 | bcomp = gen_rtx_NE (VOIDmode, cc_reg, const0_rtx); | |
7254 | loc_ref = gen_rtx_LABEL_REF (VOIDmode, operands [1]); | |
f7df4a84 | 7255 | emit_jump_insn (gen_rtx_SET (pc_rtx, |
594bdd53 FY |
7256 | gen_rtx_IF_THEN_ELSE (VOIDmode, bcomp, |
7257 | loc_ref, pc_rtx))); | |
7258 | DONE; | |
7259 | }) | |
7260 | ||
3751345d RE |
7261 | ;; Track speculation through conditional branches. We assume that |
7262 | ;; SPECULATION_TRACKER_REGNUM is reserved for this purpose when necessary. | |
7263 | (define_insn "speculation_tracker" | |
7264 | [(set (reg:DI SPECULATION_TRACKER_REGNUM) | |
996bea89 | 7265 | (unspec:DI [(reg:DI SPECULATION_TRACKER_REGNUM) (match_operand 0)] |
3751345d RE |
7266 | UNSPEC_SPECULATION_TRACKER))] |
7267 | "" | |
7268 | { | |
7269 | operands[1] = gen_rtx_REG (DImode, SPECULATION_TRACKER_REGNUM); | |
7270 | output_asm_insn ("csel\\t%1, %1, xzr, %m0", operands); | |
7271 | return ""; | |
7272 | } | |
7273 | [(set_attr "type" "csel")] | |
7274 | ) | |
7275 | ||
2214053f RS |
7276 | ;; Like speculation_tracker, but track the inverse condition. |
7277 | (define_insn "speculation_tracker_rev" | |
7278 | [(set (reg:DI SPECULATION_TRACKER_REGNUM) | |
7279 | (unspec:DI [(reg:DI SPECULATION_TRACKER_REGNUM) (match_operand 0)] | |
7280 | UNSPEC_SPECULATION_TRACKER_REV))] | |
7281 | "" | |
7282 | { | |
7283 | operands[1] = gen_rtx_REG (DImode, SPECULATION_TRACKER_REGNUM); | |
7284 | output_asm_insn ("csel\\t%1, %1, xzr, %M0", operands); | |
7285 | return ""; | |
7286 | } | |
7287 | [(set_attr "type" "csel")] | |
7288 | ) | |
7289 | ||
b5f794b4 SD |
7290 | ;; BTI <target> instructions |
7291 | (define_insn "bti_noarg" | |
7292 | [(unspec_volatile [(const_int 0)] UNSPECV_BTI_NOARG)] | |
7293 | "" | |
7294 | "hint\t32 // bti" | |
7295 | [(set_attr "type" "no_insn")] | |
7296 | ) | |
7297 | ||
7298 | (define_insn "bti_c" | |
7299 | [(unspec_volatile [(const_int 0)] UNSPECV_BTI_C)] | |
7300 | "" | |
7301 | "hint\t34 // bti c" | |
7302 | [(set_attr "type" "no_insn")] | |
7303 | ) | |
7304 | ||
7305 | (define_insn "bti_j" | |
7306 | [(unspec_volatile [(const_int 0)] UNSPECV_BTI_J)] | |
7307 | "" | |
7308 | "hint\t36 // bti j" | |
7309 | [(set_attr "type" "no_insn")] | |
7310 | ) | |
7311 | ||
7312 | (define_insn "bti_jc" | |
7313 | [(unspec_volatile [(const_int 0)] UNSPECV_BTI_JC)] | |
7314 | "" | |
7315 | "hint\t38 // bti jc" | |
7316 | [(set_attr "type" "no_insn")] | |
7317 | ) | |
7318 | ||
def745b2 RE |
7319 | ;; Hard speculation barrier. |
7320 | (define_insn "speculation_barrier" | |
7321 | [(unspec_volatile [(const_int 0)] UNSPECV_SPECULATION_BARRIER)] | |
7322 | "" | |
7323 | "isb\;dsb\\tsy" | |
7324 | [(set_attr "length" "8") | |
3751345d RE |
7325 | (set_attr "type" "block") |
7326 | (set_attr "speculation_barrier" "true")] | |
def745b2 RE |
7327 | ) |
7328 | ||
c0111dc4 RE |
7329 | ;; Support for __builtin_speculation_safe_value when we have speculation |
7330 | ;; tracking enabled. Use the speculation tracker to decide whether to | |
7331 | ;; copy operand 1 to the target, or to copy the fail value (operand 2). | |
21cebf90 | 7332 | (define_expand "@despeculate_copy<ALLI_TI:mode>" |
1bbffb87 | 7333 | [(set (match_operand:ALLI_TI 0 "register_operand") |
c0111dc4 | 7334 | (unspec_volatile:ALLI_TI |
1bbffb87 DZ |
7335 | [(match_operand:ALLI_TI 1 "register_operand") |
7336 | (match_operand:ALLI_TI 2 "aarch64_reg_or_zero") | |
c0111dc4 RE |
7337 | (use (reg:DI SPECULATION_TRACKER_REGNUM)) |
7338 | (clobber (reg:CC CC_REGNUM))] UNSPECV_SPECULATION_BARRIER))] | |
7339 | "" | |
7340 | " | |
7341 | { | |
7342 | if (operands[2] == const0_rtx) | |
7343 | { | |
7344 | rtx tracker; | |
7345 | if (<MODE>mode == TImode) | |
7346 | tracker = gen_rtx_REG (DImode, SPECULATION_TRACKER_REGNUM); | |
7347 | else | |
7348 | tracker = gen_rtx_REG (<MODE>mode, SPECULATION_TRACKER_REGNUM); | |
7349 | ||
7350 | emit_insn (gen_despeculate_simple<mode> (operands[0], operands[1], | |
7351 | tracker)); | |
7352 | DONE; | |
7353 | } | |
7354 | } | |
7355 | " | |
7356 | ) | |
7357 | ||
7358 | ;; Patterns to match despeculate_copy<mode>. Note that "hint 0x14" is the | |
7359 | ;; encoding for CSDB, but will work in older versions of the assembler. | |
7360 | (define_insn "*despeculate_copy<ALLI:mode>_insn" | |
7361 | [(set (match_operand:ALLI 0 "register_operand" "=r") | |
7362 | (unspec_volatile:ALLI | |
7363 | [(match_operand:ALLI 1 "register_operand" "r") | |
7364 | (match_operand:ALLI 2 "aarch64_reg_or_zero" "rZ") | |
7365 | (use (reg:DI SPECULATION_TRACKER_REGNUM)) | |
7366 | (clobber (reg:CC CC_REGNUM))] UNSPECV_SPECULATION_BARRIER))] | |
7367 | "" | |
7368 | { | |
7369 | operands[3] = gen_rtx_REG (DImode, SPECULATION_TRACKER_REGNUM); | |
7370 | output_asm_insn ("cmp\\t%3, #0\;csel\\t%<w>0, %<w>1, %<w>2, ne\;hint\t0x14 // csdb", | |
7371 | operands); | |
7372 | return ""; | |
7373 | } | |
7374 | [(set_attr "length" "12") | |
7375 | (set_attr "type" "block") | |
7376 | (set_attr "speculation_barrier" "true")] | |
7377 | ) | |
7378 | ||
7379 | ;; Pattern to match despeculate_copyti | |
7380 | (define_insn "*despeculate_copyti_insn" | |
7381 | [(set (match_operand:TI 0 "register_operand" "=r") | |
7382 | (unspec_volatile:TI | |
7383 | [(match_operand:TI 1 "register_operand" "r") | |
7384 | (match_operand:TI 2 "aarch64_reg_or_zero" "rZ") | |
7385 | (use (reg:DI SPECULATION_TRACKER_REGNUM)) | |
7386 | (clobber (reg:CC CC_REGNUM))] UNSPECV_SPECULATION_BARRIER))] | |
7387 | "" | |
7388 | { | |
7389 | operands[3] = gen_rtx_REG (DImode, SPECULATION_TRACKER_REGNUM); | |
7390 | output_asm_insn | |
7391 | ("cmp\\t%3, #0\;csel\\t%0, %1, %2, ne\;csel\\t%H0, %H1, %H2, ne\;hint\t0x14 // csdb", | |
7392 | operands); | |
7393 | return ""; | |
7394 | } | |
7395 | [(set_attr "length" "16") | |
7396 | (set_attr "type" "block") | |
7397 | (set_attr "speculation_barrier" "true")] | |
7398 | ) | |
7399 | ||
7400 | (define_insn "despeculate_simple<ALLI:mode>" | |
7401 | [(set (match_operand:ALLI 0 "register_operand" "=r") | |
7402 | (unspec_volatile:ALLI | |
7403 | [(match_operand:ALLI 1 "register_operand" "r") | |
7404 | (use (match_operand:ALLI 2 "register_operand" ""))] | |
7405 | UNSPECV_SPECULATION_BARRIER))] | |
7406 | "" | |
7407 | "and\\t%<w>0, %<w>1, %<w>2\;hint\t0x14 // csdb" | |
7408 | [(set_attr "type" "block") | |
7409 | (set_attr "length" "8") | |
7410 | (set_attr "speculation_barrier" "true")] | |
7411 | ) | |
7412 | ||
7413 | (define_insn "despeculate_simpleti" | |
7414 | [(set (match_operand:TI 0 "register_operand" "=r") | |
7415 | (unspec_volatile:TI | |
7416 | [(match_operand:TI 1 "register_operand" "r") | |
7417 | (use (match_operand:DI 2 "register_operand" ""))] | |
7418 | UNSPECV_SPECULATION_BARRIER))] | |
7419 | "" | |
7420 | "and\\t%0, %1, %2\;and\\t%H0, %H1, %2\;hint\t0x14 // csdb" | |
7421 | [(set_attr "type" "block") | |
7422 | (set_attr "length" "12") | |
7423 | (set_attr "speculation_barrier" "true")] | |
7424 | ) | |
7425 | ||
10bd1d96 KT |
7426 | (define_insn "aarch64_<frintnzs_op><mode>" |
7427 | [(set (match_operand:VSFDF 0 "register_operand" "=w") | |
7428 | (unspec:VSFDF [(match_operand:VSFDF 1 "register_operand" "w")] | |
7429 | FRINTNZX))] | |
7430 | "TARGET_FRINT && TARGET_FLOAT | |
7431 | && !(VECTOR_MODE_P (<MODE>mode) && !TARGET_SIMD)" | |
7432 | "<frintnzs_op>\\t%<v>0<Vmtype>, %<v>1<Vmtype>" | |
7433 | [(set_attr "type" "f_rint<stype>")] | |
7434 | ) | |
7435 | ||
89626179 SD |
7436 | ;; Transactional Memory Extension (TME) instructions. |
7437 | ||
7438 | (define_insn "tstart" | |
7439 | [(set (match_operand:DI 0 "register_operand" "=r") | |
7440 | (unspec_volatile:DI [(const_int 0)] UNSPECV_TSTART)) | |
7441 | (clobber (mem:BLK (scratch)))] | |
7442 | "TARGET_TME" | |
7443 | "tstart\\t%0" | |
7444 | [(set_attr "type" "tme")] | |
7445 | ) | |
7446 | ||
7447 | (define_insn "ttest" | |
7448 | [(set (match_operand:DI 0 "register_operand" "=r") | |
7449 | (unspec_volatile:DI [(const_int 0)] UNSPEC_TTEST)) | |
7450 | (clobber (mem:BLK (scratch)))] | |
7451 | "TARGET_TME" | |
7452 | "ttest\\t%0" | |
7453 | [(set_attr "type" "tme")] | |
7454 | ) | |
7455 | ||
7456 | (define_insn "tcommit" | |
7457 | [(unspec_volatile:BLK [(const_int 0)] UNSPECV_TCOMMIT) | |
7458 | (clobber (mem:BLK (scratch)))] | |
7459 | "TARGET_TME" | |
7460 | "tcommit" | |
7461 | [(set_attr "type" "tme")] | |
7462 | ) | |
7463 | ||
7464 | (define_insn "tcancel" | |
7465 | [(unspec_volatile:BLK | |
7466 | [(match_operand 0 "const_int_operand" "n")] UNSPECV_TCANCEL) | |
7467 | (clobber (mem:BLK (scratch)))] | |
7468 | "TARGET_TME && (UINTVAL (operands[0]) <= 65535)" | |
7469 | "tcancel\\t#%0" | |
7470 | [(set_attr "type" "tme")] | |
7471 | ) | |
7472 | ||
c5dc215d KT |
7473 | (define_insn "aarch64_rndr" |
7474 | [(set (match_operand:DI 0 "register_operand" "=r") | |
7475 | (unspec_volatile:DI [(const_int 0)] UNSPEC_RNDR)) | |
7476 | (set (reg:CC_Z CC_REGNUM) | |
7477 | (unspec_volatile:CC_Z [(const_int 0)] UNSPEC_RNDR))] | |
7478 | "TARGET_RNG" | |
7479 | "mrs\t%0, RNDR" | |
7480 | [(set_attr "type" "mrs")] | |
7481 | ) | |
7482 | ||
7483 | (define_insn "aarch64_rndrrs" | |
7484 | [(set (match_operand:DI 0 "register_operand" "=r") | |
7485 | (unspec_volatile:DI [(const_int 0)] UNSPEC_RNDRRS)) | |
7486 | (set (reg:CC_Z CC_REGNUM) | |
7487 | (unspec_volatile:CC_Z [(const_int 0)] UNSPEC_RNDRRS))] | |
7488 | "TARGET_RNG" | |
7489 | "mrs\t%0, RNDRRS" | |
7490 | [(set_attr "type" "mrs")] | |
7491 | ) | |
7492 | ||
ef01e6bb DZ |
7493 | ;; Memory Tagging Extension (MTE) instructions. |
7494 | ||
7495 | (define_insn "irg" | |
7496 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
7497 | (ior:DI | |
7498 | (and:DI (match_operand:DI 1 "register_operand" "rk") | |
7499 | (const_int -1080863910568919041)) ;; 0xf0ff... | |
7500 | (ashift:DI (unspec:QI [(match_operand:DI 2 "register_operand" "r")] | |
7501 | UNSPEC_GEN_TAG_RND) | |
7502 | (const_int 56))))] | |
7503 | "TARGET_MEMTAG" | |
7504 | "irg\\t%0, %1, %2" | |
7505 | [(set_attr "type" "memtag")] | |
7506 | ) | |
7507 | ||
7508 | (define_insn "gmi" | |
7509 | [(set (match_operand:DI 0 "register_operand" "=r") | |
7510 | (ior:DI (ashift:DI | |
7511 | (const_int 1) | |
7512 | (and:QI (lshiftrt:DI | |
7513 | (match_operand:DI 1 "register_operand" "rk") | |
7514 | (const_int 56)) (const_int 15))) | |
7515 | (match_operand:DI 2 "register_operand" "r")))] | |
7516 | "TARGET_MEMTAG" | |
7517 | "gmi\\t%0, %1, %2" | |
7518 | [(set_attr "type" "memtag")] | |
7519 | ) | |
7520 | ||
7521 | (define_insn "addg" | |
7522 | [(set (match_operand:DI 0 "register_operand" "=rk") | |
7523 | (ior:DI | |
7524 | (and:DI (plus:DI (match_operand:DI 1 "register_operand" "rk") | |
7525 | (match_operand:DI 2 "aarch64_granule16_uimm6" "i")) | |
7526 | (const_int -1080863910568919041)) ;; 0xf0ff... | |
7527 | (ashift:DI | |
7528 | (unspec:QI | |
7529 | [(and:QI (lshiftrt:DI (match_dup 1) (const_int 56)) (const_int 15)) | |
7530 | (match_operand:QI 3 "aarch64_memtag_tag_offset" "i")] | |
7531 | UNSPEC_GEN_TAG) | |
7532 | (const_int 56))))] | |
7533 | "TARGET_MEMTAG" | |
7534 | "addg\\t%0, %1, #%2, #%3" | |
7535 | [(set_attr "type" "memtag")] | |
7536 | ) | |
7537 | ||
7538 | (define_insn "subp" | |
7539 | [(set (match_operand:DI 0 "register_operand" "=r") | |
7540 | (minus:DI | |
7541 | (and:DI (match_operand:DI 1 "register_operand" "rk") | |
7542 | (const_int 72057594037927935)) ;; 0x00ff... | |
7543 | (and:DI (match_operand:DI 2 "register_operand" "rk") | |
7544 | (const_int 72057594037927935))))] ;; 0x00ff... | |
7545 | "TARGET_MEMTAG" | |
7546 | "subp\\t%0, %1, %2" | |
7547 | [(set_attr "type" "memtag")] | |
7548 | ) | |
7549 | ||
7550 | ;; LDG will use the 16-byte aligned value of the address. | |
7551 | (define_insn "ldg" | |
7552 | [(set (match_operand:DI 0 "register_operand" "+r") | |
7553 | (ior:DI | |
7554 | (and:DI (match_dup 0) (const_int -1080863910568919041)) ;; 0xf0ff... | |
7555 | (ashift:DI | |
7556 | (mem:QI (unspec:DI | |
7557 | [(and:DI (plus:DI (match_operand:DI 1 "register_operand" "rk") | |
7558 | (match_operand:DI 2 "aarch64_granule16_simm9" "i")) | |
7559 | (const_int -16))] UNSPEC_TAG_SPACE)) | |
7560 | (const_int 56))))] | |
7561 | "TARGET_MEMTAG" | |
7562 | "ldg\\t%0, [%1, #%2]" | |
7563 | [(set_attr "type" "memtag")] | |
7564 | ) | |
7565 | ||
7566 | ;; STG doesn't align the address but aborts with alignment fault | |
7567 | ;; when the address is not 16-byte aligned. | |
7568 | (define_insn "stg" | |
7569 | [(set (mem:QI (unspec:DI | |
7570 | [(plus:DI (match_operand:DI 1 "register_operand" "rk") | |
7571 | (match_operand:DI 2 "aarch64_granule16_simm9" "i"))] | |
7572 | UNSPEC_TAG_SPACE)) | |
7573 | (and:QI (lshiftrt:DI (match_operand:DI 0 "register_operand" "rk") | |
7574 | (const_int 56)) (const_int 15)))] | |
7575 | "TARGET_MEMTAG" | |
7576 | "stg\\t%0, [%1, #%2]" | |
7577 | [(set_attr "type" "memtag")] | |
7578 | ) | |
7579 | ||
fdcddba8 PW |
7580 | ;; Load/Store 64-bit (LS64) instructions. |
7581 | (define_insn "ld64b" | |
7582 | [(set (match_operand:V8DI 0 "register_operand" "=r") | |
7583 | (unspec_volatile:V8DI | |
7584 | [(mem:V8DI (match_operand:DI 1 "register_operand" "r"))] | |
7585 | UNSPEC_LD64B) | |
7586 | )] | |
7587 | "TARGET_LS64" | |
7588 | "ld64b\\t%0, [%1]" | |
7589 | [(set_attr "type" "ls64")] | |
7590 | ) | |
7591 | ||
7592 | (define_insn "st64b" | |
7593 | [(set (mem:V8DI (match_operand:DI 0 "register_operand" "=r")) | |
7594 | (unspec_volatile:V8DI [(match_operand:V8DI 1 "register_operand" "r")] | |
7595 | UNSPEC_ST64B) | |
7596 | )] | |
7597 | "TARGET_LS64" | |
7598 | "st64b\\t%1, [%0]" | |
7599 | [(set_attr "type" "ls64")] | |
7600 | ) | |
7601 | ||
7602 | (define_insn "st64bv" | |
7603 | [(set (match_operand:DI 0 "register_operand" "=r") | |
7604 | (unspec_volatile:DI [(const_int 0)] UNSPEC_ST64BV_RET)) | |
7605 | (set (mem:V8DI (match_operand:DI 1 "register_operand" "r")) | |
7606 | (unspec_volatile:V8DI [(match_operand:V8DI 2 "register_operand" "r")] | |
7607 | UNSPEC_ST64BV) | |
7608 | )] | |
7609 | "TARGET_LS64" | |
7610 | "st64bv\\t%0, %2, [%1]" | |
7611 | [(set_attr "type" "ls64")] | |
7612 | ) | |
7613 | ||
7614 | (define_insn "st64bv0" | |
7615 | [(set (match_operand:DI 0 "register_operand" "=r") | |
7616 | (unspec_volatile:DI [(const_int 0)] UNSPEC_ST64BV0_RET)) | |
7617 | (set (mem:V8DI (match_operand:DI 1 "register_operand" "r")) | |
7618 | (unspec_volatile:V8DI [(match_operand:V8DI 2 "register_operand" "r")] | |
7619 | UNSPEC_ST64BV0) | |
7620 | )] | |
7621 | "TARGET_LS64" | |
7622 | "st64bv0\\t%0, %2, [%1]" | |
7623 | [(set_attr "type" "ls64")] | |
7624 | ) | |
7625 | ||
43e9d192 IB |
7626 | ;; AdvSIMD Stuff |
7627 | (include "aarch64-simd.md") | |
7628 | ||
0462169c SN |
7629 | ;; Atomic Operations |
7630 | (include "atomics.md") | |
350013bc BC |
7631 | |
7632 | ;; ldp/stp peephole patterns | |
7633 | (include "aarch64-ldpstp.md") | |
43cacb12 RS |
7634 | |
7635 | ;; SVE. | |
7636 | (include "aarch64-sve.md") | |
0617e23c AM |
7637 | |
7638 | ;; SVE2. | |
7639 | (include "aarch64-sve2.md") |