1 /*
2 * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2015, 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_VM_MACROASSEMBLER_AARCH64_HPP
27 #define CPU_AARCH64_VM_MACROASSEMBLER_AARCH64_HPP
28
29 #include "asm/assembler.hpp"
30
31 #ifdef ASSERT
32 #include "gc/shared/collectedHeap.hpp"
33 #endif
34
35 // MacroAssembler extends Assembler by frequently used macros.
36 //
37 // Instructions for which a 'better' code sequence exists depending
38 // on arguments should also go in here.
39
40 class MacroAssembler: public Assembler {
41 friend class LIR_Assembler;
42
43 public:
44 using Assembler::mov;
45 using Assembler::movi;
46
47 protected:
48
49 // Support for VM calls
50 //
51 // This is the base routine called by the different versions of call_VM_leaf. The interpreter
52 // may customize this version by overriding it for its purposes (e.g., to save/restore
53 // additional registers when doing a VM call).
54 virtual void call_VM_leaf_base(
55 address entry_point, // the entry point
56 int number_of_arguments, // the number of arguments to pop after the call
57 Label *retaddr = NULL
58 );
59
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) {
64 call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
65 }
66
67 // This is the base routine called by the different versions of call_VM. The interpreter
68 // may customize this version by overriding it for its purposes (e.g., to save/restore
69 // additional registers when doing a VM call).
70 //
71 // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
72 // returns the register which contains the thread upon return. If a thread register has been
73 // specified, the return value will correspond to that register. If no last_java_sp is specified
74 // (noreg) than rsp will be used instead.
75 virtual void call_VM_base( // returns the register containing the thread upon return
76 Register oop_result, // where an oop-result ends up if any; use noreg otherwise
77 Register java_thread, // the thread if computed before ; use noreg otherwise
78 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise
79 address entry_point, // the entry point
80 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call
81 bool check_exceptions // whether to check for pending exceptions after return
82 );
83
84 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
85 // The implementation is only non-empty for the InterpreterMacroAssembler,
86 // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
87 virtual void check_and_handle_popframe(Register java_thread);
88 virtual void check_and_handle_earlyret(Register java_thread);
89
90 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
91
92 // Maximum size of class area in Metaspace when compressed
93 uint64_t use_XOR_for_compressed_class_base;
94
95 public:
96 MacroAssembler(CodeBuffer* code) : Assembler(code) {
97 use_XOR_for_compressed_class_base
98 = (operand_valid_for_logical_immediate(false /*is32*/,
99 (uint64_t)Universe::narrow_klass_base())
100 && ((uint64_t)Universe::narrow_klass_base()
101 > (1u << log2_intptr(CompressedClassSpaceSize))));
102 }
103
104 // Biased locking support
105 // lock_reg and obj_reg must be loaded up with the appropriate values.
106 // swap_reg is killed.
107 // tmp_reg must be supplied and must not be rscratch1 or rscratch2
108 // Optional slow case is for implementations (interpreter and C1) which branch to
109 // slow case directly. Leaves condition codes set for C2's Fast_Lock node.
110 // Returns offset of first potentially-faulting instruction for null
111 // check info (currently consumed only by C1). If
112 // swap_reg_contains_mark is true then returns -1 as it is assumed
113 // the calling code has already passed any potential faults.
114 int biased_locking_enter(Register lock_reg, Register obj_reg,
115 Register swap_reg, Register tmp_reg,
116 bool swap_reg_contains_mark,
117 Label& done, Label* slow_case = NULL,
118 BiasedLockingCounters* counters = NULL);
119 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done);
120
121
122 // Helper functions for statistics gathering.
123 // Unconditional atomic increment.
124 void atomic_incw(Register counter_addr, Register tmp, Register tmp2);
125 void atomic_incw(Address counter_addr, Register tmp1, Register tmp2, Register tmp3) {
126 lea(tmp1, counter_addr);
127 atomic_incw(tmp1, tmp2, tmp3);
128 }
129 // Load Effective Address
130 void lea(Register r, const Address &a) {
131 InstructionMark im(this);
132 code_section()->relocate(inst_mark(), a.rspec());
133 a.lea(this, r);
134 }
135
136 void addmw(Address a, Register incr, Register scratch) {
137 ldrw(scratch, a);
138 addw(scratch, scratch, incr);
139 strw(scratch, a);
140 }
141
142 // Add constant to memory word
143 void addmw(Address a, int imm, Register scratch) {
144 ldrw(scratch, a);
145 if (imm > 0)
146 addw(scratch, scratch, (unsigned)imm);
147 else
148 subw(scratch, scratch, (unsigned)-imm);
149 strw(scratch, a);
150 }
151
152 void bind(Label& L) {
153 Assembler::bind(L);
154 code()->clear_last_membar();
155 }
156
157 void membar(Membar_mask_bits order_constraint);
158
159 // Frame creation and destruction shared between JITs.
160 void build_frame(int framesize);
161 void remove_frame(int framesize);
162
163 virtual void _call_Unimplemented(address call_site) {
164 mov(rscratch2, call_site);
165 haltsim();
166 }
167
168 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
169
170 virtual void notify(int type);
171
172 // aliases defined in AARCH64 spec
173
174 template<class T>
175 inline void cmpw(Register Rd, T imm) { subsw(zr, Rd, imm); }
176 inline void cmp(Register Rd, unsigned imm) { subs(zr, Rd, imm); }
177
178 inline void cmnw(Register Rd, unsigned imm) { addsw(zr, Rd, imm); }
179 inline void cmn(Register Rd, unsigned imm) { adds(zr, Rd, imm); }
180
181 void cset(Register Rd, Assembler::Condition cond) {
182 csinc(Rd, zr, zr, ~cond);
183 }
184 void csetw(Register Rd, Assembler::Condition cond) {
185 csincw(Rd, zr, zr, ~cond);
186 }
187
188 void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
189 csneg(Rd, Rn, Rn, ~cond);
190 }
191 void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
192 csnegw(Rd, Rn, Rn, ~cond);
193 }
194
195 inline void movw(Register Rd, Register Rn) {
196 if (Rd == sp || Rn == sp) {
197 addw(Rd, Rn, 0U);
198 } else {
199 orrw(Rd, zr, Rn);
200 }
201 }
202 inline void mov(Register Rd, Register Rn) {
203 assert(Rd != r31_sp && Rn != r31_sp, "should be");
204 if (Rd == Rn) {
205 } else if (Rd == sp || Rn == sp) {
206 add(Rd, Rn, 0U);
207 } else {
208 orr(Rd, zr, Rn);
209 }
210 }
211
212 inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
213 inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
214
215 inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
216 inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
217
218 inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
219 inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
220
221 inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
222 bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
223 }
224 inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
225 bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
226 }
227
228 inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
229 bfmw(Rd, Rn, lsb, (lsb + width - 1));
230 }
231 inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
232 bfm(Rd, Rn, lsb , (lsb + width - 1));
233 }
234
235 inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
236 sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
237 }
238 inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
239 sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
240 }
241
242 inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
243 sbfmw(Rd, Rn, lsb, (lsb + width - 1));
244 }
245 inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
246 sbfm(Rd, Rn, lsb , (lsb + width - 1));
247 }
248
249 inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
250 ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
251 }
252 inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
253 ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
254 }
255
256 inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
257 ubfmw(Rd, Rn, lsb, (lsb + width - 1));
258 }
259 inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
260 ubfm(Rd, Rn, lsb , (lsb + width - 1));
261 }
262
263 inline void asrw(Register Rd, Register Rn, unsigned imm) {
264 sbfmw(Rd, Rn, imm, 31);
265 }
266
267 inline void asr(Register Rd, Register Rn, unsigned imm) {
268 sbfm(Rd, Rn, imm, 63);
269 }
270
271 inline void lslw(Register Rd, Register Rn, unsigned imm) {
272 ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
273 }
274
275 inline void lsl(Register Rd, Register Rn, unsigned imm) {
276 ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
277 }
278
279 inline void lsrw(Register Rd, Register Rn, unsigned imm) {
280 ubfmw(Rd, Rn, imm, 31);
281 }
282
283 inline void lsr(Register Rd, Register Rn, unsigned imm) {
284 ubfm(Rd, Rn, imm, 63);
285 }
286
287 inline void rorw(Register Rd, Register Rn, unsigned imm) {
288 extrw(Rd, Rn, Rn, imm);
289 }
290
291 inline void ror(Register Rd, Register Rn, unsigned imm) {
292 extr(Rd, Rn, Rn, imm);
293 }
294
295 inline void sxtbw(Register Rd, Register Rn) {
296 sbfmw(Rd, Rn, 0, 7);
297 }
298 inline void sxthw(Register Rd, Register Rn) {
299 sbfmw(Rd, Rn, 0, 15);
300 }
301 inline void sxtb(Register Rd, Register Rn) {
302 sbfm(Rd, Rn, 0, 7);
303 }
304 inline void sxth(Register Rd, Register Rn) {
305 sbfm(Rd, Rn, 0, 15);
306 }
307 inline void sxtw(Register Rd, Register Rn) {
308 sbfm(Rd, Rn, 0, 31);
309 }
310
311 inline void uxtbw(Register Rd, Register Rn) {
312 ubfmw(Rd, Rn, 0, 7);
313 }
314 inline void uxthw(Register Rd, Register Rn) {
315 ubfmw(Rd, Rn, 0, 15);
316 }
317 inline void uxtb(Register Rd, Register Rn) {
318 ubfm(Rd, Rn, 0, 7);
319 }
320 inline void uxth(Register Rd, Register Rn) {
321 ubfm(Rd, Rn, 0, 15);
322 }
323 inline void uxtw(Register Rd, Register Rn) {
324 ubfm(Rd, Rn, 0, 31);
325 }
326
327 inline void cmnw(Register Rn, Register Rm) {
328 addsw(zr, Rn, Rm);
329 }
330 inline void cmn(Register Rn, Register Rm) {
331 adds(zr, Rn, Rm);
332 }
333
334 inline void cmpw(Register Rn, Register Rm) {
335 subsw(zr, Rn, Rm);
336 }
337 inline void cmp(Register Rn, Register Rm) {
338 subs(zr, Rn, Rm);
339 }
340
341 inline void negw(Register Rd, Register Rn) {
342 subw(Rd, zr, Rn);
343 }
344
345 inline void neg(Register Rd, Register Rn) {
346 sub(Rd, zr, Rn);
347 }
348
349 inline void negsw(Register Rd, Register Rn) {
350 subsw(Rd, zr, Rn);
351 }
352
353 inline void negs(Register Rd, Register Rn) {
354 subs(Rd, zr, Rn);
355 }
356
357 inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
358 addsw(zr, Rn, Rm, kind, shift);
359 }
360 inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
361 adds(zr, Rn, Rm, kind, shift);
362 }
363
364 inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
365 subsw(zr, Rn, Rm, kind, shift);
366 }
367 inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
368 subs(zr, Rn, Rm, kind, shift);
369 }
370
371 inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
372 subw(Rd, zr, Rn, kind, shift);
373 }
374
375 inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
376 sub(Rd, zr, Rn, kind, shift);
377 }
378
379 inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
380 subsw(Rd, zr, Rn, kind, shift);
381 }
382
383 inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
384 subs(Rd, zr, Rn, kind, shift);
385 }
386
387 inline void mnegw(Register Rd, Register Rn, Register Rm) {
388 msubw(Rd, Rn, Rm, zr);
389 }
390 inline void mneg(Register Rd, Register Rn, Register Rm) {
391 msub(Rd, Rn, Rm, zr);
392 }
393
394 inline void mulw(Register Rd, Register Rn, Register Rm) {
395 maddw(Rd, Rn, Rm, zr);
396 }
397 inline void mul(Register Rd, Register Rn, Register Rm) {
398 madd(Rd, Rn, Rm, zr);
399 }
400
401 inline void smnegl(Register Rd, Register Rn, Register Rm) {
402 smsubl(Rd, Rn, Rm, zr);
403 }
404 inline void smull(Register Rd, Register Rn, Register Rm) {
405 smaddl(Rd, Rn, Rm, zr);
406 }
407
408 inline void umnegl(Register Rd, Register Rn, Register Rm) {
409 umsubl(Rd, Rn, Rm, zr);
410 }
411 inline void umull(Register Rd, Register Rn, Register Rm) {
412 umaddl(Rd, Rn, Rm, zr);
413 }
414
415 #define WRAP(INSN) \
416 void INSN(Register Rd, Register Rn, Register Rm, Register Ra) { \
417 if ((VM_Version::features() & VM_Version::CPU_A53MAC) && Ra != zr) \
418 nop(); \
419 Assembler::INSN(Rd, Rn, Rm, Ra); \
420 }
421
422 WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
423 WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
424 #undef WRAP
425
426
427 // macro assembly operations needed for aarch64
428
429 // first two private routines for loading 32 bit or 64 bit constants
430 private:
431
432 void mov_immediate64(Register dst, u_int64_t imm64);
433 void mov_immediate32(Register dst, u_int32_t imm32);
434
435 int push(unsigned int bitset, Register stack);
436 int pop(unsigned int bitset, Register stack);
437
438 void mov(Register dst, Address a);
439
440 public:
441 void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); }
442 void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); }
443
444 // Push and pop everything that might be clobbered by a native
445 // runtime call except rscratch1 and rscratch2. (They are always
446 // scratch, so we don't have to protect them.) Only save the lower
447 // 64 bits of each vector register.
448 void push_call_clobbered_registers();
449 void pop_call_clobbered_registers();
450 void push_call_clobbered_fp_registers();
451 void pop_call_clobbered_fp_registers();
452
453 // now mov instructions for loading absolute addresses and 32 or
454 // 64 bit integers
455
456 inline void mov(Register dst, address addr) {
457 assert(Universe::heap() == NULL
458 || !Universe::heap()->is_in(addr), "use movptr for oop pointers");
459 mov_immediate64(dst, (uintptr_t)addr);
460 }
461
462 inline void mov(Register dst, u_int64_t imm64)
463 {
464 mov_immediate64(dst, imm64);
465 }
466
467 inline void movw(Register dst, u_int32_t imm32)
468 {
469 mov_immediate32(dst, imm32);
470 }
471
472 inline void mov(Register dst, long l)
473 {
474 mov(dst, (u_int64_t)l);
475 }
476
477 inline void mov(Register dst, int i)
478 {
479 mov(dst, (long)i);
480 }
481
482 void mov(Register dst, RegisterOrConstant src) {
483 if (src.is_register())
484 mov(dst, src.as_register());
485 else
486 mov(dst, src.as_constant());
487 }
488
489 void movptr(Register r, uintptr_t imm64);
490
491 void mov(FloatRegister Vd, SIMD_Arrangement T, u_int32_t imm32);
492
493 void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
494 orr(Vd, T, Vn, Vn);
495 }
496
497 public:
498
499 // Generalized Test Bit And Branch, including a "far" variety which
500 // spans more than 32KiB.
501 void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool far = false) {
502 assert(cond == EQ || cond == NE, "must be");
503
504 if (far)
505 cond = ~cond;
506
507 void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
508 if (cond == Assembler::EQ)
509 branch = &Assembler::tbz;
510 else
511 branch = &Assembler::tbnz;
512
513 if (far) {
514 Label L;
515 (this->*branch)(Rt, bitpos, L);
516 b(dest);
517 bind(L);
518 } else {
519 (this->*branch)(Rt, bitpos, dest);
520 }
521 }
522
523 // macro instructions for accessing and updating floating point
524 // status register
525 //
526 // FPSR : op1 == 011
527 // CRn == 0100
528 // CRm == 0100
529 // op2 == 001
530
531 inline void get_fpsr(Register reg)
532 {
533 mrs(0b11, 0b0100, 0b0100, 0b001, reg);
534 }
535
536 inline void set_fpsr(Register reg)
537 {
538 msr(0b011, 0b0100, 0b0100, 0b001, reg);
539 }
540
541 inline void clear_fpsr()
542 {
543 msr(0b011, 0b0100, 0b0100, 0b001, zr);
544 }
545
546 // Macro instructions for accessing and updating the condition flags
547 inline void get_nzcv(Register reg)
548 {
549 mrs(0b011, 0b0100, 0b0010, 0b000, reg);
550 }
551
552 inline void set_nzcv(Register reg)
553 {
554 msr(0b011, 0b0100, 0b0010, 0b000, reg);
555 }
556
557 // DCZID_EL0: op1 == 011
558 // CRn == 0000
559 // CRm == 0000
560 // op2 == 111
561 inline void get_dczid_el0(Register reg)
562 {
563 mrs(0b011, 0b0000, 0b0000, 0b111, reg);
564 }
565
566 // CTR_EL0: op1 == 011
567 // CRn == 0000
568 // CRm == 0000
569 // op2 == 001
570 inline void get_ctr_el0(Register reg)
571 {
572 mrs(0b011, 0b0000, 0b0000, 0b001, reg);
573 }
574
575 // idiv variant which deals with MINLONG as dividend and -1 as divisor
576 int corrected_idivl(Register result, Register ra, Register rb,
577 bool want_remainder, Register tmp = rscratch1);
578 int corrected_idivq(Register result, Register ra, Register rb,
579 bool want_remainder, Register tmp = rscratch1);
580
581 // Support for NULL-checks
582 //
583 // Generates code that causes a NULL OS exception if the content of reg is NULL.
584 // If the accessed location is M[reg + offset] and the offset is known, provide the
585 // offset. No explicit code generation is needed if the offset is within a certain
586 // range (0 <= offset <= page_size).
587
588 virtual void null_check(Register reg, int offset = -1);
589 static bool needs_explicit_null_check(intptr_t offset);
590
591 static address target_addr_for_insn(address insn_addr, unsigned insn);
592 static address target_addr_for_insn(address insn_addr) {
593 unsigned insn = *(unsigned*)insn_addr;
594 return target_addr_for_insn(insn_addr, insn);
595 }
596
597 // Required platform-specific helpers for Label::patch_instructions.
598 // They _shadow_ the declarations in AbstractAssembler, which are undefined.
599 static int pd_patch_instruction_size(address branch, address target);
600 static void pd_patch_instruction(address branch, address target) {
601 pd_patch_instruction_size(branch, target);
602 }
603 static address pd_call_destination(address branch) {
604 return target_addr_for_insn(branch);
605 }
606 #ifndef PRODUCT
607 static void pd_print_patched_instruction(address branch);
608 #endif
609
610 static int patch_oop(address insn_addr, address o);
611
612 address emit_trampoline_stub(int insts_call_instruction_offset, address target);
613
614 // The following 4 methods return the offset of the appropriate move instruction
615
616 // Support for fast byte/short loading with zero extension (depending on particular CPU)
617 int load_unsigned_byte(Register dst, Address src);
618 int load_unsigned_short(Register dst, Address src);
619
620 // Support for fast byte/short loading with sign extension (depending on particular CPU)
621 int load_signed_byte(Register dst, Address src);
622 int load_signed_short(Register dst, Address src);
623
624 int load_signed_byte32(Register dst, Address src);
625 int load_signed_short32(Register dst, Address src);
626
627 // Support for sign-extension (hi:lo = extend_sign(lo))
628 void extend_sign(Register hi, Register lo);
629
630 // Load and store values by size and signed-ness
631 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg);
632 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg);
633
634 // Support for inc/dec with optimal instruction selection depending on value
635
636 // x86_64 aliases an unqualified register/address increment and
637 // decrement to call incrementq and decrementq but also supports
638 // explicitly sized calls to incrementq/decrementq or
639 // incrementl/decrementl
640
641 // for aarch64 the proper convention would be to use
642 // increment/decrement for 64 bit operatons and
643 // incrementw/decrementw for 32 bit operations. so when porting
644 // x86_64 code we can leave calls to increment/decrement as is,
645 // replace incrementq/decrementq with increment/decrement and
646 // replace incrementl/decrementl with incrementw/decrementw.
647
648 // n.b. increment/decrement calls with an Address destination will
649 // need to use a scratch register to load the value to be
650 // incremented. increment/decrement calls which add or subtract a
651 // constant value greater than 2^12 will need to use a 2nd scratch
652 // register to hold the constant. so, a register increment/decrement
653 // may trash rscratch2 and an address increment/decrement trash
654 // rscratch and rscratch2
655
656 void decrementw(Address dst, int value = 1);
657 void decrementw(Register reg, int value = 1);
658
659 void decrement(Register reg, int value = 1);
660 void decrement(Address dst, int value = 1);
661
662 void incrementw(Address dst, int value = 1);
663 void incrementw(Register reg, int value = 1);
664
665 void increment(Register reg, int value = 1);
666 void increment(Address dst, int value = 1);
667
668
669 // Alignment
670 void align(int modulus);
671
672 // Stack frame creation/removal
673 void enter()
674 {
675 stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
676 mov(rfp, sp);
677 }
678 void leave()
679 {
680 mov(sp, rfp);
681 ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
682 }
683
684 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
685 // The pointer will be loaded into the thread register.
686 void get_thread(Register thread);
687
688
689 // Support for VM calls
690 //
691 // It is imperative that all calls into the VM are handled via the call_VM macros.
692 // They make sure that the stack linkage is setup correctly. call_VM's correspond
693 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
694
695
696 void call_VM(Register oop_result,
697 address entry_point,
698 bool check_exceptions = true);
699 void call_VM(Register oop_result,
700 address entry_point,
701 Register arg_1,
702 bool check_exceptions = true);
703 void call_VM(Register oop_result,
704 address entry_point,
705 Register arg_1, Register arg_2,
706 bool check_exceptions = true);
707 void call_VM(Register oop_result,
708 address entry_point,
709 Register arg_1, Register arg_2, Register arg_3,
710 bool check_exceptions = true);
711
712 // Overloadings with last_Java_sp
713 void call_VM(Register oop_result,
714 Register last_java_sp,
715 address entry_point,
716 int number_of_arguments = 0,
717 bool check_exceptions = true);
718 void call_VM(Register oop_result,
719 Register last_java_sp,
720 address entry_point,
721 Register arg_1, bool
722 check_exceptions = true);
723 void call_VM(Register oop_result,
724 Register last_java_sp,
725 address entry_point,
726 Register arg_1, Register arg_2,
727 bool check_exceptions = true);
728 void call_VM(Register oop_result,
729 Register last_java_sp,
730 address entry_point,
731 Register arg_1, Register arg_2, Register arg_3,
732 bool check_exceptions = true);
733
734 void get_vm_result (Register oop_result, Register thread);
735 void get_vm_result_2(Register metadata_result, Register thread);
736
737 // These always tightly bind to MacroAssembler::call_VM_base
738 // bypassing the virtual implementation
739 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
740 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
741 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
742 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);
743 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);
744
745 void call_VM_leaf(address entry_point,
746 int number_of_arguments = 0);
747 void call_VM_leaf(address entry_point,
748 Register arg_1);
749 void call_VM_leaf(address entry_point,
750 Register arg_1, Register arg_2);
751 void call_VM_leaf(address entry_point,
752 Register arg_1, Register arg_2, Register arg_3);
753
754 // These always tightly bind to MacroAssembler::call_VM_leaf_base
755 // bypassing the virtual implementation
756 void super_call_VM_leaf(address entry_point);
757 void super_call_VM_leaf(address entry_point, Register arg_1);
758 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
759 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
760 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
761
762 // last Java Frame (fills frame anchor)
763 void set_last_Java_frame(Register last_java_sp,
764 Register last_java_fp,
765 address last_java_pc,
766 Register scratch);
767
768 void set_last_Java_frame(Register last_java_sp,
769 Register last_java_fp,
770 Label &last_java_pc,
771 Register scratch);
772
773 void set_last_Java_frame(Register last_java_sp,
774 Register last_java_fp,
775 Register last_java_pc,
776 Register scratch);
777
778 void reset_last_Java_frame(Register thread);
779
780 // thread in the default location (rthread)
781 void reset_last_Java_frame(bool clear_fp);
782
783 // Stores
784 void store_check(Register obj); // store check for obj - register is destroyed afterwards
785 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed)
786
787 #if INCLUDE_ALL_GCS
788
789 void g1_write_barrier_pre(Register obj,
790 Register pre_val,
791 Register thread,
792 Register tmp,
793 bool tosca_live,
794 bool expand_call);
795
796 void g1_write_barrier_post(Register store_addr,
797 Register new_val,
798 Register thread,
799 Register tmp,
800 Register tmp2);
801
802 void shenandoah_write_barrier(Register dst);
803
804 #endif // INCLUDE_ALL_GCS
805
806 // oop manipulations
807 void load_klass(Register dst, Register src);
808 void store_klass(Register dst, Register src);
809 void cmp_klass(Register oop, Register trial_klass, Register tmp);
810
811 void load_mirror(Register dst, Register method);
812
813 void load_heap_oop(Register dst, Address src);
814
815 void load_heap_oop_not_null(Register dst, Address src);
816 void store_heap_oop(Address dst, Register src);
817
818 // currently unimplemented
819 // Used for storing NULL. All other oop constants should be
820 // stored using routines that take a jobject.
821 void store_heap_oop_null(Address dst);
822
823 void load_prototype_header(Register dst, Register src);
824
825 void store_klass_gap(Register dst, Register src);
826
827 // This dummy is to prevent a call to store_heap_oop from
828 // converting a zero (like NULL) into a Register by giving
829 // the compiler two choices it can't resolve
830
831 void store_heap_oop(Address dst, void* dummy);
832
833 void encode_heap_oop(Register d, Register s);
834 void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
835 void decode_heap_oop(Register d, Register s);
836 void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
837 void encode_heap_oop_not_null(Register r);
838 void decode_heap_oop_not_null(Register r);
839 void encode_heap_oop_not_null(Register dst, Register src);
840 void decode_heap_oop_not_null(Register dst, Register src);
841
842 void set_narrow_oop(Register dst, jobject obj);
843
844 void encode_klass_not_null(Register r);
845 void decode_klass_not_null(Register r);
846 void encode_klass_not_null(Register dst, Register src);
847 void decode_klass_not_null(Register dst, Register src);
848
849 void set_narrow_klass(Register dst, Klass* k);
850
851 // if heap base register is used - reinit it with the correct value
852 void reinit_heapbase();
853
854 DEBUG_ONLY(void verify_heapbase(const char* msg);)
855
856 void push_CPU_state(bool save_vectors = false);
857 void pop_CPU_state(bool restore_vectors = false) ;
858
859 // Round up to a power of two
860 void round_to(Register reg, int modulus);
861
862 // allocation
863 void eden_allocate(
864 Register obj, // result: pointer to object after successful allocation
865 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
866 int con_size_in_bytes, // object size in bytes if known at compile time
867 Register t1, // temp register
868 Label& slow_case // continuation point if fast allocation fails
869 );
870 void tlab_allocate(
871 Register obj, // result: pointer to object after successful allocation
872 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
873 int con_size_in_bytes, // object size in bytes if known at compile time
874 Register t1, // temp register
875 Register t2, // temp register
876 Label& slow_case // continuation point if fast allocation fails
877 );
878 Register tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); // returns TLS address
879 void verify_tlab();
880
881 void incr_allocated_bytes(Register thread,
882 Register var_size_in_bytes, int con_size_in_bytes,
883 Register t1 = noreg);
884
885 // interface method calling
886 void lookup_interface_method(Register recv_klass,
887 Register intf_klass,
888 RegisterOrConstant itable_index,
889 Register method_result,
890 Register scan_temp,
891 Label& no_such_interface);
892
893 // virtual method calling
894 // n.b. x86 allows RegisterOrConstant for vtable_index
895 void lookup_virtual_method(Register recv_klass,
896 RegisterOrConstant vtable_index,
897 Register method_result);
898
899 // Test sub_klass against super_klass, with fast and slow paths.
900
901 // The fast path produces a tri-state answer: yes / no / maybe-slow.
902 // One of the three labels can be NULL, meaning take the fall-through.
903 // If super_check_offset is -1, the value is loaded up from super_klass.
904 // No registers are killed, except temp_reg.
905 void check_klass_subtype_fast_path(Register sub_klass,
906 Register super_klass,
907 Register temp_reg,
908 Label* L_success,
909 Label* L_failure,
910 Label* L_slow_path,
911 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
912
913 // The rest of the type check; must be wired to a corresponding fast path.
914 // It does not repeat the fast path logic, so don't use it standalone.
915 // The temp_reg and temp2_reg can be noreg, if no temps are available.
916 // Updates the sub's secondary super cache as necessary.
917 // If set_cond_codes, condition codes will be Z on success, NZ on failure.
918 void check_klass_subtype_slow_path(Register sub_klass,
919 Register super_klass,
920 Register temp_reg,
921 Register temp2_reg,
922 Label* L_success,
923 Label* L_failure,
924 bool set_cond_codes = false);
925
926 // Simplified, combined version, good for typical uses.
927 // Falls through on failure.
928 void check_klass_subtype(Register sub_klass,
929 Register super_klass,
930 Register temp_reg,
931 Label& L_success);
932
933 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
934
935
936 // Debugging
937
938 // only if +VerifyOops
939 void verify_oop(Register reg, const char* s = "broken oop");
940 void verify_oop_addr(Address addr, const char * s = "broken oop addr");
941
942 // TODO: verify method and klass metadata (compare against vptr?)
943 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
944 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
945
946 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
947 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
948
949 // only if +VerifyFPU
950 void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
951
952 // prints msg, dumps registers and stops execution
953 void stop(const char* msg, Label *l = NULL);
954
955 // prints msg and continues
956 void warn(const char* msg);
957
958 static void debug64(char* msg, int64_t pc, int64_t regs[]);
959
960 void untested() { stop("untested"); }
961
962 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); }
963
964 void should_not_reach_here() { stop("should not reach here"); }
965
966 // Stack overflow checking
967 void bang_stack_with_offset(int offset) {
968 // stack grows down, caller passes positive offset
969 assert(offset > 0, "must bang with negative offset");
970 mov(rscratch2, -offset);
971 str(zr, Address(sp, rscratch2));
972 }
973
974 // Writes to stack successive pages until offset reached to check for
975 // stack overflow + shadow pages. Also, clobbers tmp
976 void bang_stack_size(Register size, Register tmp);
977
978 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr,
979 Register tmp,
980 int offset);
981
982 // Support for serializing memory accesses between threads
983 void serialize_memory(Register thread, Register tmp);
984
985 // Arithmetics
986
987 void addptr(const Address &dst, int32_t src);
988 void cmpptr(Register src1, Address src2);
989
990 // Various forms of CAS
991
992 void cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
993 Label &suceed, Label *fail);
994
995 void cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
996 Label &suceed, Label *fail);
997
998 void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
999 void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1000 void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1001 void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1002
1003 void atomic_xchg(Register prev, Register newv, Register addr);
1004 void atomic_xchgw(Register prev, Register newv, Register addr);
1005 void atomic_xchgal(Register prev, Register newv, Register addr);
1006 void atomic_xchgalw(Register prev, Register newv, Register addr);
1007
1008 void orptr(Address adr, RegisterOrConstant src) {
1009 ldr(rscratch2, adr);
1010 if (src.is_register())
1011 orr(rscratch2, rscratch2, src.as_register());
1012 else
1013 orr(rscratch2, rscratch2, src.as_constant());
1014 str(rscratch2, adr);
1015 }
1016
1017 // A generic CAS; success or failure is in the EQ flag.
1018 // Clobbers rscratch1
1019 void cmpxchg(Register addr, Register expected, Register new_val,
1020 enum operand_size size,
1021 bool acquire, bool release, bool weak,
1022 Register result);
1023
1024 void cmpxchg_oop_shenandoah(Register res, Register addr, Register expected, Register new_val,
1025 bool narrow,
1026 bool acquire, bool release,
1027 Register tmp1 = rscratch1, Register tmp2 = rscratch2);
1028 // Calls
1029
1030 address trampoline_call(Address entry, CodeBuffer *cbuf = NULL);
1031
1032 static bool far_branches() {
1033 return ReservedCodeCacheSize > branch_range;
1034 }
1035
1036 // Jumps that can reach anywhere in the code cache.
1037 // Trashes tmp.
1038 void far_call(Address entry, CodeBuffer *cbuf = NULL, Register tmp = rscratch1);
1039 void far_jump(Address entry, CodeBuffer *cbuf = NULL, Register tmp = rscratch1);
1040
1041 static int far_branch_size() {
1042 if (far_branches()) {
1043 return 3 * 4; // adrp, add, br
1044 } else {
1045 return 4;
1046 }
1047 }
1048
1049 // Emit the CompiledIC call idiom
1050 address ic_call(address entry, jint method_index = 0);
1051
1052 public:
1053
1054 // Data
1055
1056 void mov_metadata(Register dst, Metadata* obj);
1057 Address allocate_metadata_address(Metadata* obj);
1058 Address constant_oop_address(jobject obj);
1059
1060 void movoop(Register dst, jobject obj, bool immediate = false);
1061
1062 // CRC32 code for java.util.zip.CRC32::updateBytes() instrinsic.
1063 void kernel_crc32(Register crc, Register buf, Register len,
1064 Register table0, Register table1, Register table2, Register table3,
1065 Register tmp, Register tmp2, Register tmp3);
1066 // CRC32 code for java.util.zip.CRC32C::updateBytes() instrinsic.
1067 void kernel_crc32c(Register crc, Register buf, Register len,
1068 Register table0, Register table1, Register table2, Register table3,
1069 Register tmp, Register tmp2, Register tmp3);
1070
1071 // Stack push and pop individual 64 bit registers
1072 void push(Register src);
1073 void pop(Register dst);
1074
1075 // push all registers onto the stack
1076 void pusha();
1077 void popa();
1078
1079 void repne_scan(Register addr, Register value, Register count,
1080 Register scratch);
1081 void repne_scanw(Register addr, Register value, Register count,
1082 Register scratch);
1083
1084 typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1085 typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1086
1087 // If a constant does not fit in an immediate field, generate some
1088 // number of MOV instructions and then perform the operation
1089 void wrap_add_sub_imm_insn(Register Rd, Register Rn, unsigned imm,
1090 add_sub_imm_insn insn1,
1091 add_sub_reg_insn insn2);
1092 // Seperate vsn which sets the flags
1093 void wrap_adds_subs_imm_insn(Register Rd, Register Rn, unsigned imm,
1094 add_sub_imm_insn insn1,
1095 add_sub_reg_insn insn2);
1096
1097 #define WRAP(INSN) \
1098 void INSN(Register Rd, Register Rn, unsigned imm) { \
1099 wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN); \
1100 } \
1101 \
1102 void INSN(Register Rd, Register Rn, Register Rm, \
1103 enum shift_kind kind, unsigned shift = 0) { \
1104 Assembler::INSN(Rd, Rn, Rm, kind, shift); \
1105 } \
1106 \
1107 void INSN(Register Rd, Register Rn, Register Rm) { \
1108 Assembler::INSN(Rd, Rn, Rm); \
1109 } \
1110 \
1111 void INSN(Register Rd, Register Rn, Register Rm, \
1112 ext::operation option, int amount = 0) { \
1113 Assembler::INSN(Rd, Rn, Rm, option, amount); \
1114 }
1115
1116 WRAP(add) WRAP(addw) WRAP(sub) WRAP(subw)
1117
1118 #undef WRAP
1119 #define WRAP(INSN) \
1120 void INSN(Register Rd, Register Rn, unsigned imm) { \
1121 wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN); \
1122 } \
1123 \
1124 void INSN(Register Rd, Register Rn, Register Rm, \
1125 enum shift_kind kind, unsigned shift = 0) { \
1126 Assembler::INSN(Rd, Rn, Rm, kind, shift); \
1127 } \
1128 \
1129 void INSN(Register Rd, Register Rn, Register Rm) { \
1130 Assembler::INSN(Rd, Rn, Rm); \
1131 } \
1132 \
1133 void INSN(Register Rd, Register Rn, Register Rm, \
1134 ext::operation option, int amount = 0) { \
1135 Assembler::INSN(Rd, Rn, Rm, option, amount); \
1136 }
1137
1138 WRAP(adds) WRAP(addsw) WRAP(subs) WRAP(subsw)
1139
1140 void add(Register Rd, Register Rn, RegisterOrConstant increment);
1141 void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1142 void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1143 void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1144
1145 void adrp(Register reg1, const Address &dest, unsigned long &byte_offset);
1146
1147 void tableswitch(Register index, jint lowbound, jint highbound,
1148 Label &jumptable, Label &jumptable_end, int stride = 1) {
1149 adr(rscratch1, jumptable);
1150 subsw(rscratch2, index, lowbound);
1151 subsw(zr, rscratch2, highbound - lowbound);
1152 br(Assembler::HS, jumptable_end);
1153 add(rscratch1, rscratch1, rscratch2,
1154 ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1155 br(rscratch1);
1156 }
1157
1158 // Form an address from base + offset in Rd. Rd may or may not
1159 // actually be used: you must use the Address that is returned. It
1160 // is up to you to ensure that the shift provided matches the size
1161 // of your data.
1162 Address form_address(Register Rd, Register base, long byte_offset, int shift);
1163
1164 // Return true iff an address is within the 48-bit AArch64 address
1165 // space.
1166 bool is_valid_AArch64_address(address a) {
1167 return ((uint64_t)a >> 48) == 0;
1168 }
1169
1170 // Load the base of the cardtable byte map into reg.
1171 void load_byte_map_base(Register reg);
1172
1173 // Prolog generator routines to support switch between x86 code and
1174 // generated ARM code
1175
1176 // routine to generate an x86 prolog for a stub function which
1177 // bootstraps into the generated ARM code which directly follows the
1178 // stub
1179 //
1180
1181 public:
1182 // enum used for aarch64--x86 linkage to define return type of x86 function
1183 enum ret_type { ret_type_void, ret_type_integral, ret_type_float, ret_type_double};
1184
1185 #ifdef BUILTIN_SIM
1186 void c_stub_prolog(int gp_arg_count, int fp_arg_count, int ret_type, address *prolog_ptr = NULL);
1187 #else
1188 void c_stub_prolog(int gp_arg_count, int fp_arg_count, int ret_type) { }
1189 #endif
1190
1191 // special version of call_VM_leaf_base needed for aarch64 simulator
1192 // where we need to specify both the gp and fp arg counts and the
1193 // return type so that the linkage routine from aarch64 to x86 and
1194 // back knows which aarch64 registers to copy to x86 registers and
1195 // which x86 result register to copy back to an aarch64 register
1196
1197 void call_VM_leaf_base1(
1198 address entry_point, // the entry point
1199 int number_of_gp_arguments, // the number of gp reg arguments to pass
1200 int number_of_fp_arguments, // the number of fp reg arguments to pass
1201 ret_type type, // the return type for the call
1202 Label* retaddr = NULL
1203 );
1204
1205 void ldr_constant(Register dest, const Address &const_addr) {
1206 if (NearCpool) {
1207 ldr(dest, const_addr);
1208 } else {
1209 unsigned long offset;
1210 adrp(dest, InternalAddress(const_addr.target()), offset);
1211 ldr(dest, Address(dest, offset));
1212 }
1213 }
1214
1215 address read_polling_page(Register r, address page, relocInfo::relocType rtype);
1216 address read_polling_page(Register r, relocInfo::relocType rtype);
1217
1218 // CRC32 code for java.util.zip.CRC32::updateBytes() instrinsic.
1219 void update_byte_crc32(Register crc, Register val, Register table);
1220 void update_word_crc32(Register crc, Register v, Register tmp,
1221 Register table0, Register table1, Register table2, Register table3,
1222 bool upper = false);
1223
1224 void string_compare(Register str1, Register str2,
1225 Register cnt1, Register cnt2, Register result,
1226 Register tmp1,
1227 FloatRegister vtmp, FloatRegister vtmpZ, int ae);
1228
1229 void arrays_equals(Register a1, Register a2,
1230 Register result, Register cnt1,
1231 int elem_size, bool is_string);
1232
1233 void fill_words(Register base, Register cnt, Register value);
1234 void zero_words(Register base, u_int64_t cnt);
1235 void zero_words(Register base, Register cnt);
1236 void block_zero(Register base, Register cnt, bool is_large = false);
1237
1238 void byte_array_inflate(Register src, Register dst, Register len,
1239 FloatRegister vtmp1, FloatRegister vtmp2,
1240 FloatRegister vtmp3, Register tmp4);
1241
1242 void char_array_compress(Register src, Register dst, Register len,
1243 FloatRegister tmp1Reg, FloatRegister tmp2Reg,
1244 FloatRegister tmp3Reg, FloatRegister tmp4Reg,
1245 Register result);
1246
1247 void encode_iso_array(Register src, Register dst,
1248 Register len, Register result,
1249 FloatRegister Vtmp1, FloatRegister Vtmp2,
1250 FloatRegister Vtmp3, FloatRegister Vtmp4);
1251 void string_indexof(Register str1, Register str2,
1252 Register cnt1, Register cnt2,
1253 Register tmp1, Register tmp2,
1254 Register tmp3, Register tmp4,
1255 int int_cnt1, Register result, int ae);
1256
1257 void in_heap_check(Register r, Register tmp, Label &nope);
1258
1259 private:
1260 void shenandoah_cset_check(Register obj, Register tmp1, Register tmp2, Label& done);
1261
1262 public:
1263 void _shenandoah_store_addr_check(Register addr, const char* msg, const char* file, int line);
1264 void _shenandoah_store_addr_check(Address addr, const char* msg, const char* file, int line);
1265 #define shenandoah_store_addr_check(reg) _shenandoah_store_addr_check(reg, "oop not safe for writing", __FILE__, __LINE__)
1266
1267 void _shenandoah_store_check(Address addr, Register value, const char* msg, const char* file, int line);
1268 void _shenandoah_store_check(Register addr, Register value, const char* msg, const char* file, int line);
1269 #define shenandoah_store_check(addr, value) _shenandoah_store_check(addr, value, "oop not safe for writing", __FILE__, __LINE__)
1270
1271 private:
1272 void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1273 Register src1, Register src2);
1274 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1275 add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1276 }
1277 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1278 Register y, Register y_idx, Register z,
1279 Register carry, Register product,
1280 Register idx, Register kdx);
1281 void multiply_128_x_128_loop(Register y, Register z,
1282 Register carry, Register carry2,
1283 Register idx, Register jdx,
1284 Register yz_idx1, Register yz_idx2,
1285 Register tmp, Register tmp3, Register tmp4,
1286 Register tmp7, Register product_hi);
1287 public:
1288 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1289 Register zlen, Register tmp1, Register tmp2, Register tmp3,
1290 Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1291 // ISB may be needed because of a safepoint
1292 void maybe_isb() { isb(); }
1293
1294 private:
1295 // Return the effective address r + (r1 << ext) + offset.
1296 // Uses rscratch2.
1297 Address offsetted_address(Register r, Register r1, Address::extend ext,
1298 int offset, int size);
1299
1300 private:
1301 // Returns an address on the stack which is reachable with a ldr/str of size
1302 // Uses rscratch2 if the address is not directly reachable
1303 Address spill_address(int size, int offset, Register tmp=rscratch2);
1304
1305 public:
1306 void spill(Register Rx, bool is64, int offset) {
1307 if (is64) {
1308 str(Rx, spill_address(8, offset));
1309 } else {
1310 strw(Rx, spill_address(4, offset));
1311 }
1312 }
1313 void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1314 str(Vx, T, spill_address(1 << (int)T, offset));
1315 }
1316 void unspill(Register Rx, bool is64, int offset) {
1317 if (is64) {
1318 ldr(Rx, spill_address(8, offset));
1319 } else {
1320 ldrw(Rx, spill_address(4, offset));
1321 }
1322 }
1323 void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1324 ldr(Vx, T, spill_address(1 << (int)T, offset));
1325 }
1326 void spill_copy128(int src_offset, int dst_offset,
1327 Register tmp1=rscratch1, Register tmp2=rscratch2) {
1328 if (src_offset < 512 && (src_offset & 7) == 0 &&
1329 dst_offset < 512 && (dst_offset & 7) == 0) {
1330 ldp(tmp1, tmp2, Address(sp, src_offset));
1331 stp(tmp1, tmp2, Address(sp, dst_offset));
1332 } else {
1333 unspill(tmp1, true, src_offset);
1334 spill(tmp1, true, dst_offset);
1335 unspill(tmp1, true, src_offset+8);
1336 spill(tmp1, true, dst_offset+8);
1337 }
1338 }
1339 };
1340
1341 #ifdef ASSERT
1342 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1343 #endif
1344
1345 /**
1346 * class SkipIfEqual:
1347 *
1348 * Instantiating this class will result in assembly code being output that will
1349 * jump around any code emitted between the creation of the instance and it's
1350 * automatic destruction at the end of a scope block, depending on the value of
1351 * the flag passed to the constructor, which will be checked at run-time.
1352 */
1353 class SkipIfEqual {
1354 private:
1355 MacroAssembler* _masm;
1356 Label _label;
1357
1358 public:
1359 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value);
1360 ~SkipIfEqual();
1361 };
1362
1363 struct tableswitch {
1364 Register _reg;
1365 int _insn_index; jint _first_key; jint _last_key;
1366 Label _after;
1367 Label _branches;
1368 };
1369
1370 #endif // CPU_AARCH64_VM_MACROASSEMBLER_AARCH64_HPP