1 /*
2 * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #ifndef CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
27 #define CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
28
29 #include "asm/assembler.inline.hpp"
30 #include "code/aotCodeCache.hpp"
31 #include "code/vmreg.hpp"
32 #include "metaprogramming/enableIf.hpp"
33 #include "oops/compressedOops.hpp"
34 #include "oops/compressedKlass.hpp"
35 #include "runtime/vm_version.hpp"
36 #include "utilities/globalDefinitions.hpp"
37 #include "utilities/powerOfTwo.hpp"
38
39 class OopMap;
40
41 // MacroAssembler extends Assembler by frequently used macros.
42 //
43 // Instructions for which a 'better' code sequence exists depending
44 // on arguments should also go in here.
45
46 class MacroAssembler: public Assembler {
47 friend class LIR_Assembler;
48
49 public:
50 using Assembler::mov;
51 using Assembler::movi;
52
53 protected:
54
55 // Support for VM calls
56 //
57 // This is the base routine called by the different versions of call_VM_leaf. The interpreter
58 // may customize this version by overriding it for its purposes (e.g., to save/restore
59 // additional registers when doing a VM call).
60 virtual void call_VM_leaf_base(
61 address entry_point, // the entry point
62 int number_of_arguments, // the number of arguments to pop after the call
63 Label *retaddr = nullptr
64 );
65
66 virtual void call_VM_leaf_base(
67 address entry_point, // the entry point
68 int number_of_arguments, // the number of arguments to pop after the call
69 Label &retaddr) {
70 call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
71 }
72
73 // This is the base routine called by the different versions of call_VM. The interpreter
74 // may customize this version by overriding it for its purposes (e.g., to save/restore
75 // additional registers when doing a VM call).
76 //
77 // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
78 // returns the register which contains the thread upon return. If a thread register has been
79 // specified, the return value will correspond to that register. If no last_java_sp is specified
80 // (noreg) than rsp will be used instead.
81 virtual void call_VM_base( // returns the register containing the thread upon return
82 Register oop_result, // where an oop-result ends up if any; use noreg otherwise
83 Register java_thread, // the thread if computed before ; use noreg otherwise
84 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise
85 Label* return_pc, // to set up last_Java_frame; use nullptr otherwise
86 address entry_point, // the entry point
87 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call
88 bool check_exceptions // whether to check for pending exceptions after return
89 );
90
91 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
92
93 enum KlassDecodeMode {
94 KlassDecodeNone,
95 KlassDecodeZero,
96 KlassDecodeXor,
97 KlassDecodeMovk
98 };
99
100 // Calculate decoding mode based on given parameters, used for checking then ultimately setting.
101 static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range);
102
103 private:
104 static KlassDecodeMode _klass_decode_mode;
105
106 // Returns above setting with asserts
107 static KlassDecodeMode klass_decode_mode();
108
109 public:
110 // Checks the decode mode and returns false if not compatible with preferred decoding mode.
111 static bool check_klass_decode_mode(address base, int shift, const size_t range);
112
113 // Sets the decode mode and returns false if cannot be set.
114 static bool set_klass_decode_mode(address base, int shift, const size_t range);
115
116 public:
117 MacroAssembler(CodeBuffer* code) : Assembler(code) {}
118
119 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
120 // The implementation is only non-empty for the InterpreterMacroAssembler,
121 // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
122 virtual void check_and_handle_popframe(Register java_thread);
123 virtual void check_and_handle_earlyret(Register java_thread);
124
125 void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp = rscratch1);
126 void rt_call(address dest, Register tmp = rscratch1);
127
128 // Load Effective Address
129 void lea(Register r, const Address &a) {
130 InstructionMark im(this);
131 a.lea(this, r);
132 }
133
134 // Whether materializing the given address for a LDR/STR requires an
135 // additional lea instruction.
136 static bool legitimize_address_requires_lea(const Address &a, int size) {
137 return a.getMode() == Address::base_plus_offset &&
138 !Address::offset_ok_for_immed(a.offset(), exact_log2(size));
139 }
140
141 /* Sometimes we get misaligned loads and stores, usually from Unsafe
142 accesses, and these can exceed the offset range. */
143 Address legitimize_address(const Address &a, int size, Register scratch) {
144 if (legitimize_address_requires_lea(a, size)) {
145 block_comment("legitimize_address {");
146 lea(scratch, a);
147 block_comment("} legitimize_address");
148 return Address(scratch);
149 }
150 return a;
151 }
152
153 void addmw(Address a, Register incr, Register scratch) {
154 ldrw(scratch, a);
155 addw(scratch, scratch, incr);
156 strw(scratch, a);
157 }
158
159 // Add constant to memory word
160 void addmw(Address a, int imm, Register scratch) {
161 ldrw(scratch, a);
162 if (imm > 0)
163 addw(scratch, scratch, (unsigned)imm);
164 else
165 subw(scratch, scratch, (unsigned)-imm);
166 strw(scratch, a);
167 }
168
169 void bind(Label& L) {
170 Assembler::bind(L);
171 code()->clear_last_insn();
172 code()->set_last_label(pc());
173 }
174
175 void membar(Membar_mask_bits order_constraint);
176
177 using Assembler::ldr;
178 using Assembler::str;
179 using Assembler::ldrw;
180 using Assembler::strw;
181
182 void ldr(Register Rx, const Address &adr);
183 void ldrw(Register Rw, const Address &adr);
184 void str(Register Rx, const Address &adr);
185 void strw(Register Rx, const Address &adr);
186
187 // Frame creation and destruction shared between JITs.
188 void build_frame(int framesize);
189 void remove_frame(int framesize);
190
191 virtual void _call_Unimplemented(address call_site) {
192 mov(rscratch2, call_site);
193 }
194
195 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
196 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
197 // https://reviews.llvm.org/D3311
198
199 #ifdef _WIN64
200 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
201 #else
202 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
203 #endif
204
205 // aliases defined in AARCH64 spec
206
207 template<class T>
208 inline void cmpw(Register Rd, T imm) { subsw(zr, Rd, imm); }
209
210 inline void cmp(Register Rd, unsigned char imm8) { subs(zr, Rd, imm8); }
211 inline void cmp(Register Rd, unsigned imm) = delete;
212
213 template<class T>
214 inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
215
216 inline void cmn(Register Rd, unsigned char imm8) { adds(zr, Rd, imm8); }
217 inline void cmn(Register Rd, unsigned imm) = delete;
218
219 void cset(Register Rd, Assembler::Condition cond) {
220 csinc(Rd, zr, zr, ~cond);
221 }
222 void csetw(Register Rd, Assembler::Condition cond) {
223 csincw(Rd, zr, zr, ~cond);
224 }
225
226 void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
227 csneg(Rd, Rn, Rn, ~cond);
228 }
229 void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
230 csnegw(Rd, Rn, Rn, ~cond);
231 }
232
233 inline void movw(Register Rd, Register Rn) {
234 if (Rd == sp || Rn == sp) {
235 Assembler::addw(Rd, Rn, 0U);
236 } else {
237 orrw(Rd, zr, Rn);
238 }
239 }
240 inline void mov(Register Rd, Register Rn) {
241 assert(Rd != r31_sp && Rn != r31_sp, "should be");
242 if (Rd == Rn) {
243 } else if (Rd == sp || Rn == sp) {
244 Assembler::add(Rd, Rn, 0U);
245 } else {
246 orr(Rd, zr, Rn);
247 }
248 }
249
250 inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
251 inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
252
253 inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
254 inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
255
256 inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
257 inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
258
259 inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
260 bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
261 }
262 inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
263 bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
264 }
265
266 inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
267 bfmw(Rd, Rn, lsb, (lsb + width - 1));
268 }
269 inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
270 bfm(Rd, Rn, lsb , (lsb + width - 1));
271 }
272
273 inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
274 sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
275 }
276 inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
277 sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
278 }
279
280 inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
281 sbfmw(Rd, Rn, lsb, (lsb + width - 1));
282 }
283 inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
284 sbfm(Rd, Rn, lsb , (lsb + width - 1));
285 }
286
287 inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
288 ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
289 }
290 inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
291 ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
292 }
293
294 inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
295 ubfmw(Rd, Rn, lsb, (lsb + width - 1));
296 }
297 inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
298 ubfm(Rd, Rn, lsb , (lsb + width - 1));
299 }
300
301 inline void asrw(Register Rd, Register Rn, unsigned imm) {
302 sbfmw(Rd, Rn, imm, 31);
303 }
304
305 inline void asr(Register Rd, Register Rn, unsigned imm) {
306 sbfm(Rd, Rn, imm, 63);
307 }
308
309 inline void lslw(Register Rd, Register Rn, unsigned imm) {
310 ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
311 }
312
313 inline void lsl(Register Rd, Register Rn, unsigned imm) {
314 ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
315 }
316
317 inline void lsrw(Register Rd, Register Rn, unsigned imm) {
318 ubfmw(Rd, Rn, imm, 31);
319 }
320
321 inline void lsr(Register Rd, Register Rn, unsigned imm) {
322 ubfm(Rd, Rn, imm, 63);
323 }
324
325 inline void rorw(Register Rd, Register Rn, unsigned imm) {
326 extrw(Rd, Rn, Rn, imm);
327 }
328
329 inline void ror(Register Rd, Register Rn, unsigned imm) {
330 extr(Rd, Rn, Rn, imm);
331 }
332
333 inline void rolw(Register Rd, Register Rn, unsigned imm) {
334 extrw(Rd, Rn, Rn, (32 - imm));
335 }
336
337 inline void rol(Register Rd, Register Rn, unsigned imm) {
338 extr(Rd, Rn, Rn, (64 - imm));
339 }
340
341 using Assembler::rax1;
342 using Assembler::eor3;
343
344 inline void rax1(Register Rd, Register Rn, Register Rm) {
345 eor(Rd, Rn, Rm, ROR, 63); // Rd = Rn ^ rol(Rm, 1)
346 }
347
348 inline void eor3(Register Rd, Register Rn, Register Rm, Register Rk) {
349 assert(Rd != Rn, "Use tmp register");
350 eor(Rd, Rm, Rk);
351 eor(Rd, Rd, Rn);
352 }
353
354 inline void sxtbw(Register Rd, Register Rn) {
355 sbfmw(Rd, Rn, 0, 7);
356 }
357 inline void sxthw(Register Rd, Register Rn) {
358 sbfmw(Rd, Rn, 0, 15);
359 }
360 inline void sxtb(Register Rd, Register Rn) {
361 sbfm(Rd, Rn, 0, 7);
362 }
363 inline void sxth(Register Rd, Register Rn) {
364 sbfm(Rd, Rn, 0, 15);
365 }
366 inline void sxtw(Register Rd, Register Rn) {
367 sbfm(Rd, Rn, 0, 31);
368 }
369
370 inline void uxtbw(Register Rd, Register Rn) {
371 ubfmw(Rd, Rn, 0, 7);
372 }
373 inline void uxthw(Register Rd, Register Rn) {
374 ubfmw(Rd, Rn, 0, 15);
375 }
376 inline void uxtb(Register Rd, Register Rn) {
377 ubfm(Rd, Rn, 0, 7);
378 }
379 inline void uxth(Register Rd, Register Rn) {
380 ubfm(Rd, Rn, 0, 15);
381 }
382 inline void uxtw(Register Rd, Register Rn) {
383 ubfm(Rd, Rn, 0, 31);
384 }
385
386 inline void cmnw(Register Rn, Register Rm) {
387 addsw(zr, Rn, Rm);
388 }
389 inline void cmn(Register Rn, Register Rm) {
390 adds(zr, Rn, Rm);
391 }
392
393 inline void cmpw(Register Rn, Register Rm) {
394 subsw(zr, Rn, Rm);
395 }
396 inline void cmp(Register Rn, Register Rm) {
397 subs(zr, Rn, Rm);
398 }
399
400 inline void negw(Register Rd, Register Rn) {
401 subw(Rd, zr, Rn);
402 }
403
404 inline void neg(Register Rd, Register Rn) {
405 sub(Rd, zr, Rn);
406 }
407
408 inline void negsw(Register Rd, Register Rn) {
409 subsw(Rd, zr, Rn);
410 }
411
412 inline void negs(Register Rd, Register Rn) {
413 subs(Rd, zr, Rn);
414 }
415
416 inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
417 addsw(zr, Rn, Rm, kind, shift);
418 }
419 inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
420 adds(zr, Rn, Rm, kind, shift);
421 }
422
423 inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
424 subsw(zr, Rn, Rm, kind, shift);
425 }
426 inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
427 subs(zr, Rn, Rm, kind, shift);
428 }
429
430 inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
431 subw(Rd, zr, Rn, kind, shift);
432 }
433
434 inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
435 sub(Rd, zr, Rn, kind, shift);
436 }
437
438 inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
439 subsw(Rd, zr, Rn, kind, shift);
440 }
441
442 inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
443 subs(Rd, zr, Rn, kind, shift);
444 }
445
446 inline void mnegw(Register Rd, Register Rn, Register Rm) {
447 msubw(Rd, Rn, Rm, zr);
448 }
449 inline void mneg(Register Rd, Register Rn, Register Rm) {
450 msub(Rd, Rn, Rm, zr);
451 }
452
453 inline void mulw(Register Rd, Register Rn, Register Rm) {
454 maddw(Rd, Rn, Rm, zr);
455 }
456 inline void mul(Register Rd, Register Rn, Register Rm) {
457 madd(Rd, Rn, Rm, zr);
458 }
459
460 inline void smnegl(Register Rd, Register Rn, Register Rm) {
461 smsubl(Rd, Rn, Rm, zr);
462 }
463 inline void smull(Register Rd, Register Rn, Register Rm) {
464 smaddl(Rd, Rn, Rm, zr);
465 }
466
467 inline void umnegl(Register Rd, Register Rn, Register Rm) {
468 umsubl(Rd, Rn, Rm, zr);
469 }
470 inline void umull(Register Rd, Register Rn, Register Rm) {
471 umaddl(Rd, Rn, Rm, zr);
472 }
473
474 #define WRAP(INSN) \
475 void INSN(Register Rd, Register Rn, Register Rm, Register Ra) { \
476 if (VM_Version::supports_a53mac() && Ra != zr) \
477 nop(); \
478 Assembler::INSN(Rd, Rn, Rm, Ra); \
479 }
480
481 WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
482 WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
483 #undef WRAP
484
485
486 // macro assembly operations needed for aarch64
487
488 public:
489
490 enum FpPushPopMode {
491 PushPopFull,
492 PushPopSVE,
493 PushPopNeon,
494 PushPopFp
495 };
496
497 // first two private routines for loading 32 bit or 64 bit constants
498 private:
499
500 void mov_immediate64(Register dst, uint64_t imm64);
501 void mov_immediate32(Register dst, uint32_t imm32);
502
503 void mov(Register dst, Address a);
504
505 public:
506
507 int push(RegSet regset, Register stack);
508 int pop(RegSet regset, Register stack);
509
510 int push_fp(FloatRegSet regset, Register stack, FpPushPopMode mode = PushPopFull);
511 int pop_fp(FloatRegSet regset, Register stack, FpPushPopMode mode = PushPopFull);
512
513 static RegSet call_clobbered_gp_registers();
514
515 int push_p(PRegSet regset, Register stack);
516 int pop_p(PRegSet regset, Register stack);
517
518 // Push and pop everything that might be clobbered by a native
519 // runtime call except rscratch1 and rscratch2. (They are always
520 // scratch, so we don't have to protect them.) Only save the lower
521 // 64 bits of each vector register. Additional registers can be excluded
522 // in a passed RegSet.
523 void push_call_clobbered_registers_except(RegSet exclude);
524 void pop_call_clobbered_registers_except(RegSet exclude);
525
526 void push_call_clobbered_registers() {
527 push_call_clobbered_registers_except(RegSet());
528 }
529 void pop_call_clobbered_registers() {
530 pop_call_clobbered_registers_except(RegSet());
531 }
532
533
534 // now mov instructions for loading absolute addresses and 32 or
535 // 64 bit integers
536
537 inline void mov(Register dst, address addr) { mov_immediate64(dst, (uint64_t)addr); }
538
539 template<typename T, ENABLE_IF(std::is_integral<T>::value)>
540 inline void mov(Register dst, T o) { mov_immediate64(dst, (uint64_t)o); }
541
542 inline void movw(Register dst, uint32_t imm32) { mov_immediate32(dst, imm32); }
543
544 void mov(Register dst, RegisterOrConstant src) {
545 if (src.is_register())
546 mov(dst, src.as_register());
547 else
548 mov(dst, src.as_constant());
549 }
550
551 void movptr(Register r, uintptr_t imm64);
552
553 void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
554
555 void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
556 orr(Vd, T, Vn, Vn);
557 }
558
559 void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
560 fcvtsh(tmp, src);
561 smov(dst, tmp, H, 0);
562 }
563
564 void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
565 mov(tmp, H, 0, src);
566 fcvths(dst, tmp);
567 }
568
569 // Generalized Test Bit And Branch, including a "far" variety which
570 // spans more than 32KiB.
571 void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
572 assert(cond == EQ || cond == NE, "must be");
573
574 if (isfar)
575 cond = ~cond;
576
577 void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
578 if (cond == Assembler::EQ)
579 branch = &Assembler::tbz;
580 else
581 branch = &Assembler::tbnz;
582
583 if (isfar) {
584 Label L;
585 (this->*branch)(Rt, bitpos, L);
586 b(dest);
587 bind(L);
588 } else {
589 (this->*branch)(Rt, bitpos, dest);
590 }
591 }
592
593 // macro instructions for accessing and updating floating point
594 // status register
595 //
596 // FPSR : op1 == 011
597 // CRn == 0100
598 // CRm == 0100
599 // op2 == 001
600
601 inline void get_fpsr(Register reg)
602 {
603 mrs(0b11, 0b0100, 0b0100, 0b001, reg);
604 }
605
606 inline void set_fpsr(Register reg)
607 {
608 msr(0b011, 0b0100, 0b0100, 0b001, reg);
609 }
610
611 inline void clear_fpsr()
612 {
613 msr(0b011, 0b0100, 0b0100, 0b001, zr);
614 }
615
616 // FPCR : op1 == 011
617 // CRn == 0100
618 // CRm == 0100
619 // op2 == 000
620
621 inline void get_fpcr(Register reg) {
622 mrs(0b11, 0b0100, 0b0100, 0b000, reg);
623 }
624
625 inline void set_fpcr(Register reg) {
626 msr(0b011, 0b0100, 0b0100, 0b000, reg);
627 }
628
629 // DCZID_EL0: op1 == 011
630 // CRn == 0000
631 // CRm == 0000
632 // op2 == 111
633 inline void get_dczid_el0(Register reg)
634 {
635 mrs(0b011, 0b0000, 0b0000, 0b111, reg);
636 }
637
638 // CTR_EL0: op1 == 011
639 // CRn == 0000
640 // CRm == 0000
641 // op2 == 001
642 inline void get_ctr_el0(Register reg)
643 {
644 mrs(0b011, 0b0000, 0b0000, 0b001, reg);
645 }
646
647 inline void get_nzcv(Register reg) {
648 mrs(0b011, 0b0100, 0b0010, 0b000, reg);
649 }
650
651 inline void set_nzcv(Register reg) {
652 msr(0b011, 0b0100, 0b0010, 0b000, reg);
653 }
654
655 // CNTVCTSS_EL0: op1 == 011
656 // CRn == 1110
657 // CRm == 0000
658 // op2 == 110
659 inline void get_cntvctss_el0(Register reg) {
660 mrs(0b011, 0b1110, 0b0000, 0b110, reg);
661 }
662
663 // idiv variant which deals with MINLONG as dividend and -1 as divisor
664 int corrected_idivl(Register result, Register ra, Register rb,
665 bool want_remainder, Register tmp = rscratch1);
666 int corrected_idivq(Register result, Register ra, Register rb,
667 bool want_remainder, Register tmp = rscratch1);
668
669 // Support for null-checks
670 //
671 // Generates code that causes a null OS exception if the content of reg is null.
672 // If the accessed location is M[reg + offset] and the offset is known, provide the
673 // offset. No explicit code generation is needed if the offset is within a certain
674 // range (0 <= offset <= page_size).
675
676 virtual void null_check(Register reg, int offset = -1);
677 static bool needs_explicit_null_check(intptr_t offset);
678 static bool uses_implicit_null_check(void* address);
679
680 static address target_addr_for_insn(address insn_addr);
681
682 // Required platform-specific helpers for Label::patch_instructions.
683 // They _shadow_ the declarations in AbstractAssembler, which are undefined.
684 static int pd_patch_instruction_size(address branch, address target);
685 static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
686 pd_patch_instruction_size(branch, target);
687 }
688 static address pd_call_destination(address branch) {
689 return target_addr_for_insn(branch);
690 }
691 #ifndef PRODUCT
692 static void pd_print_patched_instruction(address branch);
693 #endif
694
695 static int patch_oop(address insn_addr, address o);
696 static int patch_narrow_klass(address insn_addr, narrowKlass n);
697
698 // Return whether code is emitted to a scratch blob.
699 virtual bool in_scratch_emit_size() {
700 return false;
701 }
702 address emit_trampoline_stub(int insts_call_instruction_offset, address target);
703 static int max_trampoline_stub_size();
704 void emit_static_call_stub();
705 static int static_call_stub_size();
706
707 // The following 4 methods return the offset of the appropriate move instruction
708
709 // Support for fast byte/short loading with zero extension (depending on particular CPU)
710 int load_unsigned_byte(Register dst, Address src);
711 int load_unsigned_short(Register dst, Address src);
712
713 // Support for fast byte/short loading with sign extension (depending on particular CPU)
714 int load_signed_byte(Register dst, Address src);
715 int load_signed_short(Register dst, Address src);
716
717 int load_signed_byte32(Register dst, Address src);
718 int load_signed_short32(Register dst, Address src);
719
720 // Support for sign-extension (hi:lo = extend_sign(lo))
721 void extend_sign(Register hi, Register lo);
722
723 // Clean up a subword typed value to the representation in compliance with JVMS ยง2.3
724 void narrow_subword_type(Register reg, BasicType bt);
725
726 // Load and store values by size and signed-ness
727 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
728 void store_sized_value(Address dst, Register src, size_t size_in_bytes);
729
730 // Support for inc/dec with optimal instruction selection depending on value
731
732 // x86_64 aliases an unqualified register/address increment and
733 // decrement to call incrementq and decrementq but also supports
734 // explicitly sized calls to incrementq/decrementq or
735 // incrementl/decrementl
736
737 // for aarch64 the proper convention would be to use
738 // increment/decrement for 64 bit operations and
739 // incrementw/decrementw for 32 bit operations. so when porting
740 // x86_64 code we can leave calls to increment/decrement as is,
741 // replace incrementq/decrementq with increment/decrement and
742 // replace incrementl/decrementl with incrementw/decrementw.
743
744 // n.b. increment/decrement calls with an Address destination will
745 // need to use a scratch register to load the value to be
746 // incremented. increment/decrement calls which add or subtract a
747 // constant value greater than 2^12 will need to use a 2nd scratch
748 // register to hold the constant. so, a register increment/decrement
749 // may trash rscratch2 and an address increment/decrement trash
750 // rscratch and rscratch2
751
752 void decrementw(Address dst, int value = 1);
753 void decrementw(Register reg, int value = 1);
754
755 void decrement(Register reg, int value = 1);
756 void decrement(Address dst, int value = 1);
757
758 void incrementw(Address dst, int value = 1);
759 void incrementw(Register reg, int value = 1);
760
761 void increment(Register reg, int value = 1);
762 void increment(Address dst, int value = 1);
763
764
765 // Alignment
766 void align(int modulus);
767 void align(int modulus, int target);
768
769 // nop
770 void post_call_nop();
771
772 // Stack frame creation/removal
773 void enter(bool strip_ret_addr = false);
774 void leave();
775
776 // ROP Protection
777 void protect_return_address();
778 void protect_return_address(Register return_reg);
779 void authenticate_return_address();
780 void authenticate_return_address(Register return_reg);
781 void strip_return_address();
782 void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
783
784 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
785 // The pointer will be loaded into the thread register.
786 void get_thread(Register thread);
787
788 // support for argument shuffling
789 void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
790 void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
791 void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
792 void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
793 void object_move(
794 OopMap* map,
795 int oop_handle_offset,
796 int framesize_in_slots,
797 VMRegPair src,
798 VMRegPair dst,
799 bool is_receiver,
800 int* receiver_offset);
801
802
803 // Support for VM calls
804 //
805 // It is imperative that all calls into the VM are handled via the call_VM macros.
806 // They make sure that the stack linkage is setup correctly. call_VM's correspond
807 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
808
809
810 void call_VM(Register oop_result,
811 address entry_point,
812 bool check_exceptions = true);
813 void call_VM(Register oop_result,
814 address entry_point,
815 Register arg_1,
816 bool check_exceptions = true);
817 void call_VM(Register oop_result,
818 address entry_point,
819 Register arg_1, Register arg_2,
820 bool check_exceptions = true);
821 void call_VM(Register oop_result,
822 address entry_point,
823 Register arg_1, Register arg_2, Register arg_3,
824 bool check_exceptions = true);
825
826 // Overloadings with last_Java_sp
827 void call_VM(Register oop_result,
828 Register last_java_sp,
829 address entry_point,
830 int number_of_arguments = 0,
831 bool check_exceptions = true);
832 void call_VM(Register oop_result,
833 Register last_java_sp,
834 address entry_point,
835 Register arg_1, bool
836 check_exceptions = true);
837 void call_VM(Register oop_result,
838 Register last_java_sp,
839 address entry_point,
840 Register arg_1, Register arg_2,
841 bool check_exceptions = true);
842 void call_VM(Register oop_result,
843 Register last_java_sp,
844 address entry_point,
845 Register arg_1, Register arg_2, Register arg_3,
846 bool check_exceptions = true);
847
848 void get_vm_result_oop(Register oop_result, Register thread);
849 void get_vm_result_metadata(Register metadata_result, Register thread);
850
851 // These always tightly bind to MacroAssembler::call_VM_base
852 // bypassing the virtual implementation
853 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
854 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
855 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
856 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
857 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true);
858
859 void call_VM_leaf(address entry_point,
860 int number_of_arguments = 0);
861 void call_VM_leaf(address entry_point,
862 Register arg_1);
863 void call_VM_leaf(address entry_point,
864 Register arg_1, Register arg_2);
865 void call_VM_leaf(address entry_point,
866 Register arg_1, Register arg_2, Register arg_3);
867
868 // These always tightly bind to MacroAssembler::call_VM_leaf_base
869 // bypassing the virtual implementation
870 void super_call_VM_leaf(address entry_point);
871 void super_call_VM_leaf(address entry_point, Register arg_1);
872 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
873 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
874 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
875
876 // last Java Frame (fills frame anchor)
877 void set_last_Java_frame(Register last_java_sp,
878 Register last_java_fp,
879 address last_java_pc,
880 Register scratch);
881
882 void set_last_Java_frame(Register last_java_sp,
883 Register last_java_fp,
884 Label &last_java_pc,
885 Register scratch);
886
887 void set_last_Java_frame(Register last_java_sp,
888 Register last_java_fp,
889 Register last_java_pc,
890 Register scratch);
891
892 void reset_last_Java_frame(Register thread);
893
894 // thread in the default location (rthread)
895 void reset_last_Java_frame(bool clear_fp);
896
897 void resolve_jobject(Register value, Register tmp1, Register tmp2);
898 void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
899
900 // C 'boolean' to Java boolean: x == 0 ? 0 : 1
901 void c2bool(Register x);
902
903 void load_method_holder_cld(Register rresult, Register rmethod);
904 void load_method_holder(Register holder, Register method);
905
906 // oop manipulations
907 void load_narrow_klass_compact(Register dst, Register src);
908 void load_klass(Register dst, Register src);
909 void store_klass(Register dst, Register src);
910 void cmp_klass(Register obj, Register klass, Register tmp);
911 void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
912
913 void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
914 void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
915 void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
916
917 void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
918 Register tmp1, Register tmp2);
919
920 void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
921 Register tmp1, Register tmp2, Register tmp3);
922
923 void load_heap_oop(Register dst, Address src, Register tmp1,
924 Register tmp2, DecoratorSet decorators = 0);
925
926 void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
927 Register tmp2, DecoratorSet decorators = 0);
928 void store_heap_oop(Address dst, Register val, Register tmp1,
929 Register tmp2, Register tmp3, DecoratorSet decorators = 0);
930
931 // currently unimplemented
932 // Used for storing null. All other oop constants should be
933 // stored using routines that take a jobject.
934 void store_heap_oop_null(Address dst);
935
936 void store_klass_gap(Register dst, Register src);
937
938 // This dummy is to prevent a call to store_heap_oop from
939 // converting a zero (like null) into a Register by giving
940 // the compiler two choices it can't resolve
941
942 void store_heap_oop(Address dst, void* dummy);
943
944 void encode_heap_oop(Register d, Register s);
945 void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
946 void decode_heap_oop(Register d, Register s);
947 void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
948 void encode_heap_oop_not_null(Register r);
949 void decode_heap_oop_not_null(Register r);
950 void encode_heap_oop_not_null(Register dst, Register src);
951 void decode_heap_oop_not_null(Register dst, Register src);
952
953 void set_narrow_oop(Register dst, jobject obj);
954
955 void decode_klass_not_null_for_aot(Register dst, Register src);
956 void encode_klass_not_null_for_aot(Register dst, Register src);
957 void encode_klass_not_null(Register r);
958 void decode_klass_not_null(Register r);
959 void encode_klass_not_null(Register dst, Register src);
960 void decode_klass_not_null(Register dst, Register src);
961
962 void set_narrow_klass(Register dst, Klass* k);
963
964 // if heap base register is used - reinit it with the correct value
965 void reinit_heapbase();
966
967 DEBUG_ONLY(void verify_heapbase(const char* msg);)
968
969 void push_CPU_state(bool save_vectors = false, bool use_sve = false,
970 int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
971 void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
972 int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
973
974 void push_cont_fastpath(Register java_thread = rthread);
975 void pop_cont_fastpath(Register java_thread = rthread);
976
977 // Round up to a power of two
978 void round_to(Register reg, int modulus);
979
980 // java.lang.Math::round intrinsics
981 void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
982 void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
983
984 // allocation
985 void tlab_allocate(
986 Register obj, // result: pointer to object after successful allocation
987 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
988 int con_size_in_bytes, // object size in bytes if known at compile time
989 Register t1, // temp register
990 Register t2, // temp register
991 Label& slow_case // continuation point if fast allocation fails
992 );
993 void verify_tlab();
994
995 // interface method calling
996 void lookup_interface_method(Register recv_klass,
997 Register intf_klass,
998 RegisterOrConstant itable_index,
999 Register method_result,
1000 Register scan_temp,
1001 Label& no_such_interface,
1002 bool return_method = true);
1003
1004 void lookup_interface_method_stub(Register recv_klass,
1005 Register holder_klass,
1006 Register resolved_klass,
1007 Register method_result,
1008 Register temp_reg,
1009 Register temp_reg2,
1010 int itable_index,
1011 Label& L_no_such_interface);
1012
1013 // virtual method calling
1014 // n.b. x86 allows RegisterOrConstant for vtable_index
1015 void lookup_virtual_method(Register recv_klass,
1016 RegisterOrConstant vtable_index,
1017 Register method_result);
1018
1019 // Test sub_klass against super_klass, with fast and slow paths.
1020
1021 // The fast path produces a tri-state answer: yes / no / maybe-slow.
1022 // One of the three labels can be null, meaning take the fall-through.
1023 // If super_check_offset is -1, the value is loaded up from super_klass.
1024 // No registers are killed, except temp_reg.
1025 void check_klass_subtype_fast_path(Register sub_klass,
1026 Register super_klass,
1027 Register temp_reg,
1028 Label* L_success,
1029 Label* L_failure,
1030 Label* L_slow_path,
1031 Register super_check_offset = noreg);
1032
1033 // The rest of the type check; must be wired to a corresponding fast path.
1034 // It does not repeat the fast path logic, so don't use it standalone.
1035 // The temp_reg and temp2_reg can be noreg, if no temps are available.
1036 // Updates the sub's secondary super cache as necessary.
1037 // If set_cond_codes, condition codes will be Z on success, NZ on failure.
1038 void check_klass_subtype_slow_path(Register sub_klass,
1039 Register super_klass,
1040 Register temp_reg,
1041 Register temp2_reg,
1042 Label* L_success,
1043 Label* L_failure,
1044 bool set_cond_codes = false);
1045
1046 void check_klass_subtype_slow_path_linear(Register sub_klass,
1047 Register super_klass,
1048 Register temp_reg,
1049 Register temp2_reg,
1050 Label* L_success,
1051 Label* L_failure,
1052 bool set_cond_codes = false);
1053
1054 void check_klass_subtype_slow_path_table(Register sub_klass,
1055 Register super_klass,
1056 Register temp_reg,
1057 Register temp2_reg,
1058 Register temp3_reg,
1059 Register result_reg,
1060 FloatRegister vtemp_reg,
1061 Label* L_success,
1062 Label* L_failure,
1063 bool set_cond_codes = false);
1064
1065 // If r is valid, return r.
1066 // If r is invalid, remove a register r2 from available_regs, add r2
1067 // to regs_to_push, then return r2.
1068 Register allocate_if_noreg(const Register r,
1069 RegSetIterator<Register> &available_regs,
1070 RegSet ®s_to_push);
1071
1072 // Secondary subtype checking
1073 void lookup_secondary_supers_table_var(Register sub_klass,
1074 Register r_super_klass,
1075 Register temp1,
1076 Register temp2,
1077 Register temp3,
1078 FloatRegister vtemp,
1079 Register result,
1080 Label *L_success);
1081
1082
1083 // As above, but with a constant super_klass.
1084 // The result is in Register result, not the condition codes.
1085 bool lookup_secondary_supers_table_const(Register r_sub_klass,
1086 Register r_super_klass,
1087 Register temp1,
1088 Register temp2,
1089 Register temp3,
1090 FloatRegister vtemp,
1091 Register result,
1092 u1 super_klass_slot,
1093 bool stub_is_near = false);
1094
1095 void verify_secondary_supers_table(Register r_sub_klass,
1096 Register r_super_klass,
1097 Register temp1,
1098 Register temp2,
1099 Register result);
1100
1101 void lookup_secondary_supers_table_slow_path(Register r_super_klass,
1102 Register r_array_base,
1103 Register r_array_index,
1104 Register r_bitmap,
1105 Register temp1,
1106 Register result,
1107 bool is_stub = true);
1108
1109 // Simplified, combined version, good for typical uses.
1110 // Falls through on failure.
1111 void check_klass_subtype(Register sub_klass,
1112 Register super_klass,
1113 Register temp_reg,
1114 Label& L_success);
1115
1116 void clinit_barrier(Register klass,
1117 Register thread,
1118 Label* L_fast_path = nullptr,
1119 Label* L_slow_path = nullptr);
1120
1121 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1122
1123 void profile_receiver_type(Register recv, Register mdp, int mdp_offset);
1124
1125 void verify_sve_vector_length(Register tmp = rscratch1);
1126 void reinitialize_ptrue() {
1127 if (UseSVE > 0) {
1128 sve_ptrue(ptrue, B);
1129 }
1130 }
1131 void verify_ptrue();
1132
1133 // Debugging
1134
1135 // only if +VerifyOops
1136 void _verify_oop(Register reg, const char* s, const char* file, int line);
1137 void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1138
1139 void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1140 if (VerifyOops) {
1141 _verify_oop(reg, s, file, line);
1142 }
1143 }
1144 void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1145 if (VerifyOops) {
1146 _verify_oop_addr(reg, s, file, line);
1147 }
1148 }
1149
1150 // TODO: verify method and klass metadata (compare against vptr?)
1151 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1152 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1153
1154 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1155 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1156 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1157 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1158 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1159
1160 // Restore cpu control state after JNI call
1161 void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1162
1163 // prints msg, dumps registers and stops execution
1164 void stop(const char* msg);
1165
1166 static void debug64(char* msg, int64_t pc, int64_t regs[]);
1167
1168 void untested() { stop("untested"); }
1169
1170 void unimplemented(const char* what = "");
1171
1172 void should_not_reach_here() { stop("should not reach here"); }
1173
1174 void _assert_asm(Condition cc, const char* msg);
1175 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1176 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1177
1178 // Stack overflow checking
1179 void bang_stack_with_offset(int offset) {
1180 // stack grows down, caller passes positive offset
1181 assert(offset > 0, "must bang with negative offset");
1182 sub(rscratch2, sp, offset);
1183 str(zr, Address(rscratch2));
1184 }
1185
1186 // Writes to stack successive pages until offset reached to check for
1187 // stack overflow + shadow pages. Also, clobbers tmp
1188 void bang_stack_size(Register size, Register tmp);
1189
1190 // Check for reserved stack access in method being exited (for JIT)
1191 void reserved_stack_check();
1192
1193 // Arithmetics
1194
1195 // Clobber: rscratch1, rscratch2
1196 void addptr(const Address &dst, int32_t src);
1197
1198 // Clobber: rscratch1
1199 void cmpptr(Register src1, Address src2);
1200
1201 void cmpoop(Register obj1, Register obj2);
1202
1203 void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1204 void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1205 void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1206 void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1207
1208 void atomic_xchg(Register prev, Register newv, Register addr);
1209 void atomic_xchgw(Register prev, Register newv, Register addr);
1210 void atomic_xchgl(Register prev, Register newv, Register addr);
1211 void atomic_xchglw(Register prev, Register newv, Register addr);
1212 void atomic_xchgal(Register prev, Register newv, Register addr);
1213 void atomic_xchgalw(Register prev, Register newv, Register addr);
1214
1215 void orptr(Address adr, RegisterOrConstant src) {
1216 ldr(rscratch1, adr);
1217 if (src.is_register())
1218 orr(rscratch1, rscratch1, src.as_register());
1219 else
1220 orr(rscratch1, rscratch1, src.as_constant());
1221 str(rscratch1, adr);
1222 }
1223
1224 // A generic CAS; success or failure is in the EQ flag.
1225 // Clobbers rscratch1
1226 void cmpxchg(Register addr, Register expected, Register new_val,
1227 enum operand_size size,
1228 bool acquire, bool release, bool weak,
1229 Register result);
1230
1231 #ifdef ASSERT
1232 // Template short-hand support to clean-up after a failed call to trampoline
1233 // call generation (see trampoline_call() below), when a set of Labels must
1234 // be reset (before returning).
1235 template<typename Label, typename... More>
1236 void reset_labels(Label &lbl, More&... more) {
1237 lbl.reset(); reset_labels(more...);
1238 }
1239 template<typename Label>
1240 void reset_labels(Label &lbl) {
1241 lbl.reset();
1242 }
1243 #endif
1244
1245 private:
1246 void compare_eq(Register rn, Register rm, enum operand_size size);
1247
1248 public:
1249 // AArch64 OpenJDK uses four different types of calls:
1250 // - direct call: bl pc_relative_offset
1251 // This is the shortest and the fastest, but the offset has the range:
1252 // +/-128MB for the release build, +/-2MB for the debug build.
1253 //
1254 // - far call: adrp reg, pc_relative_offset; add; bl reg
1255 // This is longer than a direct call. The offset has
1256 // the range +/-4GB. As the code cache size is limited to 4GB,
1257 // far calls can reach anywhere in the code cache. If a jump is
1258 // needed rather than a call, a far jump 'b reg' can be used instead.
1259 // All instructions are embedded at a call site.
1260 //
1261 // - trampoline call:
1262 // This is only available in C1/C2-generated code (nmethod). It is a combination
1263 // of a direct call, which is used if the destination of a call is in range,
1264 // and a register-indirect call. It has the advantages of reaching anywhere in
1265 // the AArch64 address space and being patchable at runtime when the generated
1266 // code is being executed by other threads.
1267 //
1268 // [Main code section]
1269 // bl trampoline
1270 // [Stub code section]
1271 // trampoline:
1272 // ldr reg, pc + 8
1273 // br reg
1274 // <64-bit destination address>
1275 //
1276 // If the destination is in range when the generated code is moved to the code
1277 // cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1278 // is not used.
1279 // The optimization does not remove the trampoline from the stub section.
1280 // This is necessary because the trampoline may well be redirected later when
1281 // code is patched, and the new destination may not be reachable by a simple BR
1282 // instruction.
1283 //
1284 // - indirect call: move reg, address; blr reg
1285 // This too can reach anywhere in the address space, but it cannot be
1286 // patched while code is running, so it must only be modified at a safepoint.
1287 // This form of call is most suitable for targets at fixed addresses, which
1288 // will never be patched.
1289 //
1290 // The patching we do conforms to the "Concurrent modification and
1291 // execution of instructions" section of the Arm Architectural
1292 // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1293 // or SVC instructions to be modified while another thread is
1294 // executing them.
1295 //
1296 // To patch a trampoline call when the BL can't reach, we first modify
1297 // the 64-bit destination address in the trampoline, then modify the
1298 // BL to point to the trampoline, then flush the instruction cache to
1299 // broadcast the change to all executing threads. See
1300 // NativeCall::set_destination_mt_safe for the details.
1301 //
1302 // There is a benign race in that the other thread might observe the
1303 // modified BL before it observes the modified 64-bit destination
1304 // address. That does not matter because the destination method has been
1305 // invalidated, so there will be a trap at its start.
1306 // For this to work, the destination address in the trampoline is
1307 // always updated, even if we're not using the trampoline.
1308
1309 // Emit a direct call if the entry address will always be in range,
1310 // otherwise a trampoline call.
1311 // Supported entry.rspec():
1312 // - relocInfo::runtime_call_type
1313 // - relocInfo::opt_virtual_call_type
1314 // - relocInfo::static_call_type
1315 // - relocInfo::virtual_call_type
1316 //
1317 // Return: the call PC or null if CodeCache is full.
1318 // Clobbers: rscratch1
1319 address trampoline_call(Address entry);
1320
1321 static bool far_branches() {
1322 return ReservedCodeCacheSize > branch_range;
1323 }
1324
1325 // Check if branches to the non nmethod section require a far jump
1326 static bool codestub_branch_needs_far_jump() {
1327 if (AOTCodeCache::is_on_for_dump()) {
1328 // To calculate far_codestub_branch_size correctly.
1329 return true;
1330 }
1331 return CodeCache::max_distance_to_non_nmethod() > branch_range;
1332 }
1333
1334 // Emit a direct call/jump if the entry address will always be in range,
1335 // otherwise a far call/jump.
1336 // The address must be inside the code cache.
1337 // Supported entry.rspec():
1338 // - relocInfo::external_word_type
1339 // - relocInfo::runtime_call_type
1340 // - relocInfo::none
1341 // In the case of a far call/jump, the entry address is put in the tmp register.
1342 // The tmp register is invalidated.
1343 //
1344 // Far_jump returns the amount of the emitted code.
1345 void far_call(Address entry, Register tmp = rscratch1);
1346 int far_jump(Address entry, Register tmp = rscratch1);
1347
1348 static int far_codestub_branch_size() {
1349 if (codestub_branch_needs_far_jump()) {
1350 return 3 * 4; // adrp, add, br
1351 } else {
1352 return 4;
1353 }
1354 }
1355
1356 // Emit the CompiledIC call idiom
1357 address ic_call(address entry, jint method_index = 0);
1358 static int ic_check_size();
1359 int ic_check(int end_alignment);
1360
1361 public:
1362
1363 // Data
1364
1365 void mov_metadata(Register dst, Metadata* obj);
1366 Address allocate_metadata_address(Metadata* obj);
1367 Address constant_oop_address(jobject obj);
1368
1369 void movoop(Register dst, jobject obj);
1370
1371 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1372 void kernel_crc32(Register crc, Register buf, Register len,
1373 Register table0, Register table1, Register table2, Register table3,
1374 Register tmp, Register tmp2, Register tmp3);
1375 // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1376 void kernel_crc32c(Register crc, Register buf, Register len,
1377 Register table0, Register table1, Register table2, Register table3,
1378 Register tmp, Register tmp2, Register tmp3);
1379
1380 // Stack push and pop individual 64 bit registers
1381 void push(Register src);
1382 void pop(Register dst);
1383
1384 void repne_scan(Register addr, Register value, Register count,
1385 Register scratch);
1386 void repne_scanw(Register addr, Register value, Register count,
1387 Register scratch);
1388
1389 typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1390 typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1391
1392 // If a constant does not fit in an immediate field, generate some
1393 // number of MOV instructions and then perform the operation
1394 void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1395 add_sub_imm_insn insn1,
1396 add_sub_reg_insn insn2, bool is32);
1397 // Separate vsn which sets the flags
1398 void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1399 add_sub_imm_insn insn1,
1400 add_sub_reg_insn insn2, bool is32);
1401
1402 #define WRAP(INSN, is32) \
1403 void INSN(Register Rd, Register Rn, uint64_t imm) { \
1404 wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1405 } \
1406 \
1407 void INSN(Register Rd, Register Rn, Register Rm, \
1408 enum shift_kind kind, unsigned shift = 0) { \
1409 Assembler::INSN(Rd, Rn, Rm, kind, shift); \
1410 } \
1411 \
1412 void INSN(Register Rd, Register Rn, Register Rm) { \
1413 Assembler::INSN(Rd, Rn, Rm); \
1414 } \
1415 \
1416 void INSN(Register Rd, Register Rn, Register Rm, \
1417 ext::operation option, int amount = 0) { \
1418 Assembler::INSN(Rd, Rn, Rm, option, amount); \
1419 }
1420
1421 WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1422
1423 #undef WRAP
1424 #define WRAP(INSN, is32) \
1425 void INSN(Register Rd, Register Rn, uint64_t imm) { \
1426 wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1427 } \
1428 \
1429 void INSN(Register Rd, Register Rn, Register Rm, \
1430 enum shift_kind kind, unsigned shift = 0) { \
1431 Assembler::INSN(Rd, Rn, Rm, kind, shift); \
1432 } \
1433 \
1434 void INSN(Register Rd, Register Rn, Register Rm) { \
1435 Assembler::INSN(Rd, Rn, Rm); \
1436 } \
1437 \
1438 void INSN(Register Rd, Register Rn, Register Rm, \
1439 ext::operation option, int amount = 0) { \
1440 Assembler::INSN(Rd, Rn, Rm, option, amount); \
1441 }
1442
1443 WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1444
1445 void add(Register Rd, Register Rn, RegisterOrConstant increment);
1446 void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1447 void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1448 void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1449
1450 void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1451
1452 void tableswitch(Register index, jint lowbound, jint highbound,
1453 Label &jumptable, Label &jumptable_end, int stride = 1) {
1454 adr(rscratch1, jumptable);
1455 subsw(rscratch2, index, lowbound);
1456 subsw(zr, rscratch2, highbound - lowbound);
1457 br(Assembler::HS, jumptable_end);
1458 add(rscratch1, rscratch1, rscratch2,
1459 ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1460 br(rscratch1);
1461 }
1462
1463 // Form an address from base + offset in Rd. Rd may or may not
1464 // actually be used: you must use the Address that is returned. It
1465 // is up to you to ensure that the shift provided matches the size
1466 // of your data.
1467 Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1468
1469 // Return true iff an address is within the 48-bit AArch64 address
1470 // space.
1471 bool is_valid_AArch64_address(address a) {
1472 return ((uint64_t)a >> 48) == 0;
1473 }
1474
1475 // Load the base of the cardtable byte map into reg.
1476 void load_byte_map_base(Register reg);
1477
1478 // Load a constant address in the AOT Runtime Constants area
1479 void load_aotrc_address(Register reg, address a);
1480
1481 // Prolog generator routines to support switch between x86 code and
1482 // generated ARM code
1483
1484 // routine to generate an x86 prolog for a stub function which
1485 // bootstraps into the generated ARM code which directly follows the
1486 // stub
1487 //
1488
1489 public:
1490
1491 address read_polling_page(Register r, relocInfo::relocType rtype);
1492 void get_polling_page(Register dest, relocInfo::relocType rtype);
1493
1494 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1495 void update_byte_crc32(Register crc, Register val, Register table);
1496 void update_word_crc32(Register crc, Register v, Register tmp,
1497 Register table0, Register table1, Register table2, Register table3,
1498 bool upper = false);
1499
1500 address count_positives(Register ary1, Register len, Register result);
1501
1502 address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1503 Register tmp1, Register tmp2, Register tmp3, int elem_size);
1504
1505 // Ensure that the inline code and the stub use the same registers.
1506 #define ARRAYS_HASHCODE_REGISTERS \
1507 do { \
1508 assert(result == r0 && \
1509 ary == r1 && \
1510 cnt == r2 && \
1511 vdata0 == v3 && \
1512 vdata1 == v2 && \
1513 vdata2 == v1 && \
1514 vdata3 == v0 && \
1515 vmul0 == v4 && \
1516 vmul1 == v5 && \
1517 vmul2 == v6 && \
1518 vmul3 == v7 && \
1519 vpow == v12 && \
1520 vpowm == v13, "registers must match aarch64.ad"); \
1521 } while (0)
1522
1523 void string_equals(Register a1, Register a2, Register result, Register cnt1);
1524
1525 void fill_words(Register base, Register cnt, Register value);
1526 address zero_words(Register base, uint64_t cnt);
1527 address zero_words(Register ptr, Register cnt);
1528 void zero_dcache_blocks(Register base, Register cnt);
1529
1530 static const int zero_words_block_size;
1531
1532 address byte_array_inflate(Register src, Register dst, Register len,
1533 FloatRegister vtmp1, FloatRegister vtmp2,
1534 FloatRegister vtmp3, Register tmp4);
1535
1536 void char_array_compress(Register src, Register dst, Register len,
1537 Register res,
1538 FloatRegister vtmp0, FloatRegister vtmp1,
1539 FloatRegister vtmp2, FloatRegister vtmp3,
1540 FloatRegister vtmp4, FloatRegister vtmp5);
1541
1542 void encode_iso_array(Register src, Register dst,
1543 Register len, Register res, bool ascii,
1544 FloatRegister vtmp0, FloatRegister vtmp1,
1545 FloatRegister vtmp2, FloatRegister vtmp3,
1546 FloatRegister vtmp4, FloatRegister vtmp5);
1547
1548 void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1549 address pio2, address dsin_coef, address dcos_coef);
1550 private:
1551 // begin trigonometric functions support block
1552 void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1553 void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1554 void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1555 void generate_kernel_cos(FloatRegister x, address dcos_coef);
1556 // end trigonometric functions support block
1557 void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1558 Register src1, Register src2);
1559 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1560 add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1561 }
1562 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1563 Register y, Register y_idx, Register z,
1564 Register carry, Register product,
1565 Register idx, Register kdx);
1566 void multiply_128_x_128_loop(Register y, Register z,
1567 Register carry, Register carry2,
1568 Register idx, Register jdx,
1569 Register yz_idx1, Register yz_idx2,
1570 Register tmp, Register tmp3, Register tmp4,
1571 Register tmp7, Register product_hi);
1572 void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1573 Register len, Register tmp0, Register tmp1, Register tmp2,
1574 Register tmp3);
1575 void kernel_crc32_using_crc32(Register crc, Register buf,
1576 Register len, Register tmp0, Register tmp1, Register tmp2,
1577 Register tmp3);
1578 void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1579 Register len, Register tmp0, Register tmp1, Register tmp2,
1580 Register tmp3);
1581 void kernel_crc32c_using_crc32c(Register crc, Register buf,
1582 Register len, Register tmp0, Register tmp1, Register tmp2,
1583 Register tmp3);
1584 void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1585 Register len, Register tmp0, Register tmp1, Register tmp2,
1586 size_t table_offset);
1587
1588 void ghash_modmul (FloatRegister result,
1589 FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1590 FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1591 FloatRegister t1, FloatRegister t2, FloatRegister t3);
1592 void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1593 public:
1594 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1595 Register tmp0, Register tmp1, Register tmp2, Register tmp3,
1596 Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1597 void mul_add(Register out, Register in, Register offs, Register len, Register k);
1598 void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1599 FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1600 FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1601 void ghash_multiply_wide(int index,
1602 FloatRegister result_lo, FloatRegister result_hi,
1603 FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1604 FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1605 void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1606 FloatRegister p, FloatRegister z, FloatRegister t1);
1607 void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1608 FloatRegister p, FloatRegister z, FloatRegister t1);
1609 void ghash_processBlocks_wide(Label& p, Register state, Register subkeyH,
1610 Register data, Register blocks, int unrolls);
1611
1612
1613 void aesenc_loadkeys(Register key, Register keylen);
1614 void aesecb_encrypt(Register from, Register to, Register keylen,
1615 FloatRegister data = v0, int unrolls = 1);
1616 void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1617 void aes_round(FloatRegister input, FloatRegister subkey);
1618
1619 // ChaCha20 functions support block
1620 void cc20_qr_add4(FloatRegister (&addFirst)[4],
1621 FloatRegister (&addSecond)[4]);
1622 void cc20_qr_xor4(FloatRegister (&firstElem)[4],
1623 FloatRegister (&secondElem)[4], FloatRegister (&result)[4]);
1624 void cc20_qr_lrot4(FloatRegister (&sourceReg)[4],
1625 FloatRegister (&destReg)[4], int bits, FloatRegister table);
1626 void cc20_set_qr_registers(FloatRegister (&vectorSet)[4],
1627 const FloatRegister (&stateVectors)[16], int idx1, int idx2,
1628 int idx3, int idx4);
1629
1630 // Place an ISB after code may have been modified due to a safepoint.
1631 void safepoint_isb();
1632
1633 private:
1634 // Return the effective address r + (r1 << ext) + offset.
1635 // Uses rscratch2.
1636 Address offsetted_address(Register r, Register r1, Address::extend ext,
1637 int offset, int size);
1638
1639 private:
1640 // Returns an address on the stack which is reachable with a ldr/str of size
1641 // Uses rscratch2 if the address is not directly reachable
1642 Address spill_address(int size, int offset, Register tmp=rscratch2);
1643 Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1644
1645 bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1646
1647 // Check whether two loads/stores can be merged into ldp/stp.
1648 bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1649
1650 // Merge current load/store with previous load/store into ldp/stp.
1651 void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1652
1653 // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1654 bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1655
1656 public:
1657 void spill(Register Rx, bool is64, int offset) {
1658 if (is64) {
1659 str(Rx, spill_address(8, offset));
1660 } else {
1661 strw(Rx, spill_address(4, offset));
1662 }
1663 }
1664 void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1665 str(Vx, T, spill_address(1 << (int)T, offset));
1666 }
1667
1668 void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1669 sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1670 }
1671 void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1672 sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1673 }
1674
1675 void unspill(Register Rx, bool is64, int offset) {
1676 if (is64) {
1677 ldr(Rx, spill_address(8, offset));
1678 } else {
1679 ldrw(Rx, spill_address(4, offset));
1680 }
1681 }
1682 void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1683 ldr(Vx, T, spill_address(1 << (int)T, offset));
1684 }
1685
1686 void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1687 sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1688 }
1689 void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1690 sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1691 }
1692
1693 void spill_copy128(int src_offset, int dst_offset,
1694 Register tmp1=rscratch1, Register tmp2=rscratch2) {
1695 if (src_offset < 512 && (src_offset & 7) == 0 &&
1696 dst_offset < 512 && (dst_offset & 7) == 0) {
1697 ldp(tmp1, tmp2, Address(sp, src_offset));
1698 stp(tmp1, tmp2, Address(sp, dst_offset));
1699 } else {
1700 unspill(tmp1, true, src_offset);
1701 spill(tmp1, true, dst_offset);
1702 unspill(tmp1, true, src_offset+8);
1703 spill(tmp1, true, dst_offset+8);
1704 }
1705 }
1706 void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1707 int sve_vec_reg_size_in_bytes) {
1708 assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1709 for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1710 spill_copy128(src_offset, dst_offset);
1711 src_offset += 16;
1712 dst_offset += 16;
1713 }
1714 }
1715 void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1716 int sve_predicate_reg_size_in_bytes) {
1717 sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1718 sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1719 reinitialize_ptrue();
1720 }
1721 void cache_wb(Address line);
1722 void cache_wbsync(bool is_pre);
1723
1724 // Code for java.lang.Thread::onSpinWait() intrinsic.
1725 void spin_wait();
1726 void spin_wait_wfet(int delay_ns);
1727
1728 void fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow);
1729 void fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1730
1731 private:
1732 // Check the current thread doesn't need a cross modify fence.
1733 void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1734
1735 };
1736
1737 #ifdef ASSERT
1738 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1739 #endif
1740
1741 struct tableswitch {
1742 Register _reg;
1743 int _insn_index; jint _first_key; jint _last_key;
1744 Label _after;
1745 Label _branches;
1746 };
1747
1748 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP