1 /*
2 * Copyright (c) 2008, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #ifndef CPU_ARM_VM_MACROASSEMBLER_ARM_HPP
26 #define CPU_ARM_VM_MACROASSEMBLER_ARM_HPP
27
28 #include "code/relocInfo.hpp"
29 #include "code/relocInfo_ext.hpp"
30
31 class BiasedLockingCounters;
32
33 // Introduced AddressLiteral and its subclasses to ease portability from
34 // x86 and avoid relocation issues
35 class AddressLiteral {
36 RelocationHolder _rspec;
37 // Typically we use AddressLiterals we want to use their rval
38 // However in some situations we want the lval (effect address) of the item.
39 // We provide a special factory for making those lvals.
40 bool _is_lval;
41
42 address _target;
43
44 private:
45 static relocInfo::relocType reloc_for_target(address target) {
46 // Used for ExternalAddress or when the type is not specified
47 // Sometimes ExternalAddress is used for values which aren't
48 // exactly addresses, like the card table base.
49 // external_word_type can't be used for values in the first page
50 // so just skip the reloc in that case.
51 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none;
52 }
53
54 void set_rspec(relocInfo::relocType rtype);
55
56 protected:
57 // creation
58 AddressLiteral()
59 : _is_lval(false),
60 _target(NULL)
61 {}
62
63 public:
64
65 AddressLiteral(address target, relocInfo::relocType rtype) {
66 _is_lval = false;
67 _target = target;
68 set_rspec(rtype);
69 }
70
71 AddressLiteral(address target, RelocationHolder const& rspec)
72 : _rspec(rspec),
73 _is_lval(false),
74 _target(target)
75 {}
76
77 AddressLiteral(address target) {
78 _is_lval = false;
79 _target = target;
80 set_rspec(reloc_for_target(target));
81 }
82
83 AddressLiteral addr() {
84 AddressLiteral ret = *this;
85 ret._is_lval = true;
86 return ret;
87 }
88
89 private:
90
91 address target() { return _target; }
92 bool is_lval() { return _is_lval; }
93
94 relocInfo::relocType reloc() const { return _rspec.type(); }
95 const RelocationHolder& rspec() const { return _rspec; }
96
97 friend class Assembler;
98 friend class MacroAssembler;
99 friend class Address;
100 friend class LIR_Assembler;
101 friend class InlinedAddress;
102 };
103
104 class ExternalAddress: public AddressLiteral {
105
106 public:
107
108 ExternalAddress(address target) : AddressLiteral(target) {}
109
110 };
111
112 class InternalAddress: public AddressLiteral {
113
114 public:
115
116 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {}
117
118 };
119
120 // Inlined constants, for use with ldr_literal / bind_literal
121 // Note: InlinedInteger not supported (use move_slow(Register,int[,cond]))
122 class InlinedLiteral: StackObj {
123 public:
124 Label label; // need to be public for direct access with &
125 InlinedLiteral() {
126 }
127 };
128
129 class InlinedMetadata: public InlinedLiteral {
130 private:
131 Metadata *_data;
132
133 public:
134 InlinedMetadata(Metadata *data): InlinedLiteral() {
135 _data = data;
136 }
137 Metadata *data() { return _data; }
138 };
139
140 // Currently unused
141 // class InlinedOop: public InlinedLiteral {
142 // private:
143 // jobject _jobject;
144 //
145 // public:
146 // InlinedOop(jobject target): InlinedLiteral() {
147 // _jobject = target;
148 // }
149 // jobject jobject() { return _jobject; }
150 // };
151
152 class InlinedAddress: public InlinedLiteral {
153 private:
154 AddressLiteral _literal;
155
156 public:
157
158 InlinedAddress(jobject object): InlinedLiteral(), _literal((address)object, relocInfo::oop_type) {
159 ShouldNotReachHere(); // use mov_oop (or implement InlinedOop)
160 }
161
162 InlinedAddress(Metadata *data): InlinedLiteral(), _literal((address)data, relocInfo::metadata_type) {
163 ShouldNotReachHere(); // use InlinedMetadata or mov_metadata
164 }
165
166 InlinedAddress(address target, const RelocationHolder &rspec): InlinedLiteral(), _literal(target, rspec) {
167 assert(rspec.type() != relocInfo::oop_type, "Do not use InlinedAddress for oops");
168 assert(rspec.type() != relocInfo::metadata_type, "Do not use InlinedAddress for metadatas");
169 }
170
171 InlinedAddress(address target, relocInfo::relocType rtype): InlinedLiteral(), _literal(target, rtype) {
172 assert(rtype != relocInfo::oop_type, "Do not use InlinedAddress for oops");
173 assert(rtype != relocInfo::metadata_type, "Do not use InlinedAddress for metadatas");
174 }
175
176 // Note: default is relocInfo::none for InlinedAddress
177 InlinedAddress(address target): InlinedLiteral(), _literal(target, relocInfo::none) {
178 }
179
180 address target() { return _literal.target(); }
181
182 const RelocationHolder& rspec() const { return _literal.rspec(); }
183 };
184
185 class InlinedString: public InlinedLiteral {
186 private:
187 const char* _msg;
188
189 public:
190 InlinedString(const char* msg): InlinedLiteral() {
191 _msg = msg;
192 }
193 const char* msg() { return _msg; }
194 };
195
196 class MacroAssembler: public Assembler {
197 protected:
198
199 // Support for VM calls
200 //
201
202 // This is the base routine called by the different versions of call_VM_leaf.
203 void call_VM_leaf_helper(address entry_point, int number_of_arguments);
204
205 // This is the base routine called by the different versions of call_VM. The interpreter
206 // may customize this version by overriding it for its purposes (e.g., to save/restore
207 // additional registers when doing a VM call).
208 virtual void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions);
209 public:
210
211 MacroAssembler(CodeBuffer* code) : Assembler(code) {}
212
213 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
214 // The implementation is only non-empty for the InterpreterMacroAssembler,
215 // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
216 virtual void check_and_handle_popframe() {}
217 virtual void check_and_handle_earlyret() {}
218
219 // By default, we do not need relocation information for non
220 // patchable absolute addresses. However, when needed by some
221 // extensions, ignore_non_patchable_relocations can be modified,
222 // returning false to preserve all relocation information.
223 inline bool ignore_non_patchable_relocations() { return true; }
224
225 // Initially added to the Assembler interface as a pure virtual:
226 // RegisterConstant delayed_value(..)
227 // for:
228 // 6812678 macro assembler needs delayed binding of a few constants (for 6655638)
229 // this was subsequently modified to its present name and return type
230 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, Register tmp, int offset);
231
232 #ifdef AARCH64
233 # define NOT_IMPLEMENTED() unimplemented("NYI at " __FILE__ ":" XSTR(__LINE__))
234 # define NOT_TESTED() warn("Not tested at " __FILE__ ":" XSTR(__LINE__))
235 #endif
236
237 void align(int modulus);
238
239 // Support for VM calls
240 //
241 // It is imperative that all calls into the VM are handled via the call_VM methods.
242 // They make sure that the stack linkage is setup correctly. call_VM's correspond
243 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
244
245 void call_VM(Register oop_result, address entry_point, bool check_exceptions = true);
246 void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
247 void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
248 void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
249
250 // The following methods are required by templateTable.cpp,
251 // but not used on ARM.
252 void call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
253 void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
254 void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
255 void 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);
256
257 // Note: The super_call_VM calls are not used on ARM
258
259 // Raw call, without saving/restoring registers, exception handling, etc.
260 // Mainly used from various stubs.
261 // Note: if 'save_R9_if_scratched' is true, call_VM may on some
262 // platforms save values on the stack. Set it to false (and handle
263 // R9 in the callers) if the top of the stack must not be modified
264 // by call_VM.
265 void call_VM(address entry_point, bool save_R9_if_scratched);
266
267 void call_VM_leaf(address entry_point);
268 void call_VM_leaf(address entry_point, Register arg_1);
269 void call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
270 void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
271 void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
272
273 void get_vm_result(Register oop_result, Register tmp);
274 void get_vm_result_2(Register metadata_result, Register tmp);
275
276 // Always sets/resets sp, which default to SP if (last_sp == noreg)
277 // Optionally sets/resets fp (use noreg to avoid setting it)
278 // Always sets/resets pc on AArch64; optionally sets/resets pc on 32-bit ARM depending on save_last_java_pc flag
279 // Note: when saving PC, set_last_Java_frame returns PC's offset in the code section
280 // (for oop_maps offset computation)
281 int set_last_Java_frame(Register last_sp, Register last_fp, bool save_last_java_pc, Register tmp);
282 void reset_last_Java_frame(Register tmp);
283 // status set in set_last_Java_frame for reset_last_Java_frame
284 bool _fp_saved;
285 bool _pc_saved;
286
287 #ifdef PRODUCT
288 #define BLOCK_COMMENT(str) /* nothing */
289 #define STOP(error) __ stop(error)
290 #else
291 #define BLOCK_COMMENT(str) __ block_comment(str)
292 #define STOP(error) __ block_comment(error); __ stop(error)
293 #endif
294
295 void lookup_virtual_method(Register recv_klass,
296 Register vtable_index,
297 Register method_result);
298
299 // Test sub_klass against super_klass, with fast and slow paths.
300
301 // The fast path produces a tri-state answer: yes / no / maybe-slow.
302 // One of the three labels can be NULL, meaning take the fall-through.
303 // No registers are killed, except temp_regs.
304 void check_klass_subtype_fast_path(Register sub_klass,
305 Register super_klass,
306 Register temp_reg,
307 Register temp_reg2,
308 Label* L_success,
309 Label* L_failure,
310 Label* L_slow_path);
311
312 // The rest of the type check; must be wired to a corresponding fast path.
313 // It does not repeat the fast path logic, so don't use it standalone.
314 // temp_reg3 can be noreg, if no temps are available.
315 // Updates the sub's secondary super cache as necessary.
316 // If set_cond_codes:
317 // - condition codes will be Z on success, NZ on failure.
318 // - temp_reg will be 0 on success, non-0 on failure
319 void check_klass_subtype_slow_path(Register sub_klass,
320 Register super_klass,
321 Register temp_reg,
322 Register temp_reg2,
323 Register temp_reg3, // auto assigned if noreg
324 Label* L_success,
325 Label* L_failure,
326 bool set_cond_codes = false);
327
328 // Simplified, combined version, good for typical uses.
329 // temp_reg3 can be noreg, if no temps are available. It is used only on slow path.
330 // Falls through on failure.
331 void check_klass_subtype(Register sub_klass,
332 Register super_klass,
333 Register temp_reg,
334 Register temp_reg2,
335 Register temp_reg3, // auto assigned on slow path if noreg
336 Label& L_success);
337
338 // Returns address of receiver parameter, using tmp as base register. tmp and params_count can be the same.
339 Address receiver_argument_address(Register params_base, Register params_count, Register tmp);
340
341 void _verify_oop(Register reg, const char* s, const char* file, int line);
342 void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
343
344 // TODO: verify method and klass metadata (compare against vptr?)
345 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
346 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line) {}
347
348 #define verify_oop(reg) _verify_oop(reg, "broken oop " #reg, __FILE__, __LINE__)
349 #define verify_oop_addr(addr) _verify_oop_addr(addr, "broken oop ", __FILE__, __LINE__)
350 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
351 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
352
353 void null_check(Register reg, Register tmp, int offset = -1);
354 inline void null_check(Register reg) { null_check(reg, noreg, -1); } // for C1 lir_null_check
355
356 // Puts address of allocated object into register `obj` and end of allocated object into register `obj_end`.
357 void eden_allocate(Register obj, Register obj_end, Register tmp1, Register tmp2,
358 RegisterOrConstant size_expression, Label& slow_case);
359 void tlab_allocate(Register obj, Register obj_end, Register tmp1,
360 RegisterOrConstant size_expression, Label& slow_case);
361
362 void zero_memory(Register start, Register end, Register tmp);
363
364 static bool needs_explicit_null_check(intptr_t offset);
365
366 void arm_stack_overflow_check(int frame_size_in_bytes, Register tmp);
367 void arm_stack_overflow_check(Register Rsize, Register tmp);
368
369 void bang_stack_with_offset(int offset) {
370 ShouldNotReachHere();
371 }
372
373 // Biased locking support
374 // lock_reg and obj_reg must be loaded up with the appropriate values.
375 // swap_reg must be supplied.
376 // tmp_reg must be supplied.
377 // Optional slow case is for implementations (interpreter and C1) which branch to
378 // slow case directly. If slow_case is NULL, then leaves condition
379 // codes set (for C2's Fast_Lock node) and jumps to done label.
380 // Falls through for the fast locking attempt.
381 // Returns offset of first potentially-faulting instruction for null
382 // check info (currently consumed only by C1). If
383 // swap_reg_contains_mark is true then returns -1 as it is assumed
384 // the calling code has already passed any potential faults.
385 // Notes:
386 // - swap_reg and tmp_reg are scratched
387 // - Rtemp was (implicitly) scratched and can now be specified as the tmp2
388 int biased_locking_enter(Register obj_reg, Register swap_reg, Register tmp_reg,
389 bool swap_reg_contains_mark,
390 Register tmp2,
391 Label& done, Label& slow_case,
392 BiasedLockingCounters* counters = NULL);
393 void biased_locking_exit(Register obj_reg, Register temp_reg, Label& done);
394
395 // Building block for CAS cases of biased locking: makes CAS and records statistics.
396 // Optional slow_case label is used to transfer control if CAS fails. Otherwise leaves condition codes set.
397 void biased_locking_enter_with_cas(Register obj_reg, Register old_mark_reg, Register new_mark_reg,
398 Register tmp, Label& slow_case, int* counter_addr);
399
400 void resolve_jobject(Register value, Register tmp1, Register tmp2);
401
402 #ifndef AARCH64
403 void nop() {
404 mov(R0, R0);
405 }
406
407 void push(Register rd, AsmCondition cond = al) {
408 assert(rd != SP, "unpredictable instruction");
409 str(rd, Address(SP, -wordSize, pre_indexed), cond);
410 }
411
412 void push(RegisterSet reg_set, AsmCondition cond = al) {
413 assert(!reg_set.contains(SP), "unpredictable instruction");
414 stmdb(SP, reg_set, writeback, cond);
415 }
416
417 void pop(Register rd, AsmCondition cond = al) {
418 assert(rd != SP, "unpredictable instruction");
419 ldr(rd, Address(SP, wordSize, post_indexed), cond);
420 }
421
422 void pop(RegisterSet reg_set, AsmCondition cond = al) {
423 assert(!reg_set.contains(SP), "unpredictable instruction");
424 ldmia(SP, reg_set, writeback, cond);
425 }
426
427 void fpushd(FloatRegister fd, AsmCondition cond = al) {
428 fstmdbd(SP, FloatRegisterSet(fd), writeback, cond);
429 }
430
431 void fpushs(FloatRegister fd, AsmCondition cond = al) {
432 fstmdbs(SP, FloatRegisterSet(fd), writeback, cond);
433 }
434
435 void fpopd(FloatRegister fd, AsmCondition cond = al) {
436 fldmiad(SP, FloatRegisterSet(fd), writeback, cond);
437 }
438
439 void fpops(FloatRegister fd, AsmCondition cond = al) {
440 fldmias(SP, FloatRegisterSet(fd), writeback, cond);
441 }
442 #endif // !AARCH64
443
444 // Order access primitives
445 enum Membar_mask_bits {
446 StoreStore = 1 << 3,
447 LoadStore = 1 << 2,
448 StoreLoad = 1 << 1,
449 LoadLoad = 1 << 0
450 };
451
452 #ifdef AARCH64
453 // tmp register is not used on AArch64, this parameter is provided solely for better compatibility with 32-bit ARM
454 void membar(Membar_mask_bits order_constraint, Register tmp = noreg);
455 #else
456 void membar(Membar_mask_bits mask,
457 Register tmp,
458 bool preserve_flags = true,
459 Register load_tgt = noreg);
460 #endif
461
462 void breakpoint(AsmCondition cond = al);
463 void stop(const char* msg);
464 // prints msg and continues
465 void warn(const char* msg);
466 void unimplemented(const char* what = "");
467 void should_not_reach_here() { stop("should not reach here"); }
468 static void debug(const char* msg, const intx* registers);
469
470 // Create a walkable frame to help tracking down who called this code.
471 // Returns the frame size in words.
472 int should_not_call_this() {
473 raw_push(FP, LR);
474 should_not_reach_here();
475 flush();
476 return 2; // frame_size_in_words (FP+LR)
477 }
478
479 int save_all_registers();
480 void restore_all_registers();
481 int save_caller_save_registers();
482 void restore_caller_save_registers();
483
484 void add_rc(Register dst, Register arg1, RegisterOrConstant arg2);
485
486 // add_slow and mov_slow are used to manipulate offsets larger than 1024,
487 // these functions are not expected to handle all possible constants,
488 // only those that can really occur during compilation
489 void add_slow(Register rd, Register rn, int c);
490 void sub_slow(Register rd, Register rn, int c);
491
492 #ifdef AARCH64
493 static int mov_slow_helper(Register rd, intptr_t c, MacroAssembler* masm /* optional */);
494 #endif
495
496 void mov_slow(Register rd, intptr_t c NOT_AARCH64_ARG(AsmCondition cond = al));
497 void mov_slow(Register rd, const char *string);
498 void mov_slow(Register rd, address addr);
499
500 void patchable_mov_oop(Register rd, jobject o, int oop_index) {
501 mov_oop(rd, o, oop_index AARCH64_ONLY_ARG(true));
502 }
503 void mov_oop(Register rd, jobject o, int index = 0
504 AARCH64_ONLY_ARG(bool patchable = false)
505 NOT_AARCH64_ARG(AsmCondition cond = al));
506
507
508 void patchable_mov_metadata(Register rd, Metadata* o, int index) {
509 mov_metadata(rd, o, index AARCH64_ONLY_ARG(true));
510 }
511 void mov_metadata(Register rd, Metadata* o, int index = 0 AARCH64_ONLY_ARG(bool patchable = false));
512
513 void mov_float(FloatRegister fd, jfloat c NOT_AARCH64_ARG(AsmCondition cond = al));
514 void mov_double(FloatRegister fd, jdouble c NOT_AARCH64_ARG(AsmCondition cond = al));
515
516 #ifdef AARCH64
517 int mov_pc_to(Register rd) {
518 Label L;
519 adr(rd, L);
520 bind(L);
521 return offset();
522 }
523 #endif
524
525 // Note: this variant of mov_address assumes the address moves with
526 // the code. Do *not* implement it with non-relocated instructions,
527 // unless PC-relative.
528 #ifdef AARCH64
529 void mov_relative_address(Register rd, address addr) {
530 adr(rd, addr);
531 }
532 #else
533 void mov_relative_address(Register rd, address addr, AsmCondition cond = al) {
534 int offset = addr - pc() - 8;
535 assert((offset & 3) == 0, "bad alignment");
536 if (offset >= 0) {
537 assert(AsmOperand::is_rotated_imm(offset), "addr too far");
538 add(rd, PC, offset, cond);
539 } else {
540 assert(AsmOperand::is_rotated_imm(-offset), "addr too far");
541 sub(rd, PC, -offset, cond);
542 }
543 }
544 #endif // AARCH64
545
546 // Runtime address that may vary from one execution to another. The
547 // symbolic_reference describes what the address is, allowing
548 // the address to be resolved in a different execution context.
549 // Warning: do not implement as a PC relative address.
550 void mov_address(Register rd, address addr, symbolic_Relocation::symbolic_reference t) {
551 mov_address(rd, addr, RelocationHolder::none);
552 }
553
554 // rspec can be RelocationHolder::none (for ignored symbolic_Relocation).
555 // In that case, the address is absolute and the generated code need
556 // not be relocable.
557 void mov_address(Register rd, address addr, RelocationHolder const& rspec) {
558 assert(rspec.type() != relocInfo::runtime_call_type, "do not use mov_address for runtime calls");
559 assert(rspec.type() != relocInfo::static_call_type, "do not use mov_address for relocable calls");
560 if (rspec.type() == relocInfo::none) {
561 // absolute address, relocation not needed
562 mov_slow(rd, (intptr_t)addr);
563 return;
564 }
565 #ifndef AARCH64
566 if (VM_Version::supports_movw()) {
567 relocate(rspec);
568 int c = (int)addr;
569 movw(rd, c & 0xffff);
570 if ((unsigned int)c >> 16) {
571 movt(rd, (unsigned int)c >> 16);
572 }
573 return;
574 }
575 #endif
576 Label skip_literal;
577 InlinedAddress addr_literal(addr, rspec);
578 ldr_literal(rd, addr_literal);
579 b(skip_literal);
580 bind_literal(addr_literal);
581 // AARCH64 WARNING: because of alignment padding, extra padding
582 // may be required to get a consistent size for C2, or rules must
583 // overestimate size see MachEpilogNode::size
584 bind(skip_literal);
585 }
586
587 // Note: Do not define mov_address for a Label
588 //
589 // Load from addresses potentially within the code are now handled
590 // InlinedLiteral subclasses (to allow more flexibility on how the
591 // ldr_literal is performed).
592
593 void ldr_literal(Register rd, InlinedAddress& L) {
594 assert(L.rspec().type() != relocInfo::runtime_call_type, "avoid ldr_literal for calls");
595 assert(L.rspec().type() != relocInfo::static_call_type, "avoid ldr_literal for calls");
596 relocate(L.rspec());
597 #ifdef AARCH64
598 ldr(rd, target(L.label));
599 #else
600 ldr(rd, Address(PC, target(L.label) - pc() - 8));
601 #endif
602 }
603
604 void ldr_literal(Register rd, InlinedString& L) {
605 const char* msg = L.msg();
606 if (code()->consts()->contains((address)msg)) {
607 // string address moves with the code
608 #ifdef AARCH64
609 ldr(rd, (address)msg);
610 #else
611 ldr(rd, Address(PC, ((address)msg) - pc() - 8));
612 #endif
613 return;
614 }
615 // Warning: use external strings with care. They are not relocated
616 // if the code moves. If needed, use code_string to move them
617 // to the consts section.
618 #ifdef AARCH64
619 ldr(rd, target(L.label));
620 #else
621 ldr(rd, Address(PC, target(L.label) - pc() - 8));
622 #endif
623 }
624
625 void ldr_literal(Register rd, InlinedMetadata& L) {
626 // relocation done in the bind_literal for metadatas
627 #ifdef AARCH64
628 ldr(rd, target(L.label));
629 #else
630 ldr(rd, Address(PC, target(L.label) - pc() - 8));
631 #endif
632 }
633
634 void bind_literal(InlinedAddress& L) {
635 AARCH64_ONLY(align(wordSize));
636 bind(L.label);
637 assert(L.rspec().type() != relocInfo::metadata_type, "Must use InlinedMetadata");
638 // We currently do not use oop 'bound' literals.
639 // If the code evolves and the following assert is triggered,
640 // we need to implement InlinedOop (see InlinedMetadata).
641 assert(L.rspec().type() != relocInfo::oop_type, "Inlined oops not supported");
642 // Note: relocation is handled by relocate calls in ldr_literal
643 AbstractAssembler::emit_address((address)L.target());
644 }
645
646 void bind_literal(InlinedString& L) {
647 const char* msg = L.msg();
648 if (code()->consts()->contains((address)msg)) {
649 // The Label should not be used; avoid binding it
650 // to detect errors.
651 return;
652 }
653 AARCH64_ONLY(align(wordSize));
654 bind(L.label);
655 AbstractAssembler::emit_address((address)L.msg());
656 }
657
658 void bind_literal(InlinedMetadata& L) {
659 AARCH64_ONLY(align(wordSize));
660 bind(L.label);
661 relocate(metadata_Relocation::spec_for_immediate());
662 AbstractAssembler::emit_address((address)L.data());
663 }
664
665 void resolve_oop_handle(Register result);
666 void load_mirror(Register mirror, Register method, Register tmp);
667
668 // Porting layer between 32-bit ARM and AArch64
669
670 #define COMMON_INSTR_1(common_mnemonic, aarch64_mnemonic, arm32_mnemonic, arg_type) \
671 void common_mnemonic(arg_type arg) { \
672 AARCH64_ONLY(aarch64_mnemonic) NOT_AARCH64(arm32_mnemonic) (arg); \
673 }
674
675 #define COMMON_INSTR_2(common_mnemonic, aarch64_mnemonic, arm32_mnemonic, arg1_type, arg2_type) \
676 void common_mnemonic(arg1_type arg1, arg2_type arg2) { \
677 AARCH64_ONLY(aarch64_mnemonic) NOT_AARCH64(arm32_mnemonic) (arg1, arg2); \
678 }
679
680 #define COMMON_INSTR_3(common_mnemonic, aarch64_mnemonic, arm32_mnemonic, arg1_type, arg2_type, arg3_type) \
681 void common_mnemonic(arg1_type arg1, arg2_type arg2, arg3_type arg3) { \
682 AARCH64_ONLY(aarch64_mnemonic) NOT_AARCH64(arm32_mnemonic) (arg1, arg2, arg3); \
683 }
684
685 COMMON_INSTR_1(jump, br, bx, Register)
686 COMMON_INSTR_1(call, blr, blx, Register)
687
688 COMMON_INSTR_2(cbz_32, cbz_w, cbz, Register, Label&)
689 COMMON_INSTR_2(cbnz_32, cbnz_w, cbnz, Register, Label&)
690
691 COMMON_INSTR_2(ldr_u32, ldr_w, ldr, Register, Address)
692 COMMON_INSTR_2(ldr_s32, ldrsw, ldr, Register, Address)
693 COMMON_INSTR_2(str_32, str_w, str, Register, Address)
694
695 COMMON_INSTR_2(mvn_32, mvn_w, mvn, Register, Register)
696 COMMON_INSTR_2(cmp_32, cmp_w, cmp, Register, Register)
697 COMMON_INSTR_2(neg_32, neg_w, neg, Register, Register)
698 COMMON_INSTR_2(clz_32, clz_w, clz, Register, Register)
699 COMMON_INSTR_2(rbit_32, rbit_w, rbit, Register, Register)
700
701 COMMON_INSTR_2(cmp_32, cmp_w, cmp, Register, int)
702 COMMON_INSTR_2(cmn_32, cmn_w, cmn, Register, int)
703
704 COMMON_INSTR_3(add_32, add_w, add, Register, Register, Register)
705 COMMON_INSTR_3(sub_32, sub_w, sub, Register, Register, Register)
706 COMMON_INSTR_3(subs_32, subs_w, subs, Register, Register, Register)
707 COMMON_INSTR_3(mul_32, mul_w, mul, Register, Register, Register)
708 COMMON_INSTR_3(and_32, andr_w, andr, Register, Register, Register)
709 COMMON_INSTR_3(orr_32, orr_w, orr, Register, Register, Register)
710 COMMON_INSTR_3(eor_32, eor_w, eor, Register, Register, Register)
711
712 COMMON_INSTR_3(add_32, add_w, add, Register, Register, AsmOperand)
713 COMMON_INSTR_3(sub_32, sub_w, sub, Register, Register, AsmOperand)
714 COMMON_INSTR_3(orr_32, orr_w, orr, Register, Register, AsmOperand)
715 COMMON_INSTR_3(eor_32, eor_w, eor, Register, Register, AsmOperand)
716 COMMON_INSTR_3(and_32, andr_w, andr, Register, Register, AsmOperand)
717
718
719 COMMON_INSTR_3(add_32, add_w, add, Register, Register, int)
720 COMMON_INSTR_3(adds_32, adds_w, adds, Register, Register, int)
721 COMMON_INSTR_3(sub_32, sub_w, sub, Register, Register, int)
722 COMMON_INSTR_3(subs_32, subs_w, subs, Register, Register, int)
723
724 COMMON_INSTR_2(tst_32, tst_w, tst, Register, unsigned int)
725 COMMON_INSTR_2(tst_32, tst_w, tst, Register, AsmOperand)
726
727 COMMON_INSTR_3(and_32, andr_w, andr, Register, Register, uint)
728 COMMON_INSTR_3(orr_32, orr_w, orr, Register, Register, uint)
729 COMMON_INSTR_3(eor_32, eor_w, eor, Register, Register, uint)
730
731 COMMON_INSTR_1(cmp_zero_float, fcmp0_s, fcmpzs, FloatRegister)
732 COMMON_INSTR_1(cmp_zero_double, fcmp0_d, fcmpzd, FloatRegister)
733
734 COMMON_INSTR_2(ldr_float, ldr_s, flds, FloatRegister, Address)
735 COMMON_INSTR_2(str_float, str_s, fsts, FloatRegister, Address)
736 COMMON_INSTR_2(mov_float, fmov_s, fcpys, FloatRegister, FloatRegister)
737 COMMON_INSTR_2(neg_float, fneg_s, fnegs, FloatRegister, FloatRegister)
738 COMMON_INSTR_2(abs_float, fabs_s, fabss, FloatRegister, FloatRegister)
739 COMMON_INSTR_2(sqrt_float, fsqrt_s, fsqrts, FloatRegister, FloatRegister)
740 COMMON_INSTR_2(cmp_float, fcmp_s, fcmps, FloatRegister, FloatRegister)
741
742 COMMON_INSTR_3(add_float, fadd_s, fadds, FloatRegister, FloatRegister, FloatRegister)
743 COMMON_INSTR_3(sub_float, fsub_s, fsubs, FloatRegister, FloatRegister, FloatRegister)
744 COMMON_INSTR_3(mul_float, fmul_s, fmuls, FloatRegister, FloatRegister, FloatRegister)
745 COMMON_INSTR_3(div_float, fdiv_s, fdivs, FloatRegister, FloatRegister, FloatRegister)
746
747 COMMON_INSTR_2(ldr_double, ldr_d, fldd, FloatRegister, Address)
748 COMMON_INSTR_2(str_double, str_d, fstd, FloatRegister, Address)
749 COMMON_INSTR_2(mov_double, fmov_d, fcpyd, FloatRegister, FloatRegister)
750 COMMON_INSTR_2(neg_double, fneg_d, fnegd, FloatRegister, FloatRegister)
751 COMMON_INSTR_2(cmp_double, fcmp_d, fcmpd, FloatRegister, FloatRegister)
752 COMMON_INSTR_2(abs_double, fabs_d, fabsd, FloatRegister, FloatRegister)
753 COMMON_INSTR_2(sqrt_double, fsqrt_d, fsqrtd, FloatRegister, FloatRegister)
754
755 COMMON_INSTR_3(add_double, fadd_d, faddd, FloatRegister, FloatRegister, FloatRegister)
756 COMMON_INSTR_3(sub_double, fsub_d, fsubd, FloatRegister, FloatRegister, FloatRegister)
757 COMMON_INSTR_3(mul_double, fmul_d, fmuld, FloatRegister, FloatRegister, FloatRegister)
758 COMMON_INSTR_3(div_double, fdiv_d, fdivd, FloatRegister, FloatRegister, FloatRegister)
759
760 COMMON_INSTR_2(convert_f2d, fcvt_ds, fcvtds, FloatRegister, FloatRegister)
761 COMMON_INSTR_2(convert_d2f, fcvt_sd, fcvtsd, FloatRegister, FloatRegister)
762
763 COMMON_INSTR_2(mov_fpr2gpr_float, fmov_ws, fmrs, Register, FloatRegister)
764
765 #undef COMMON_INSTR_1
766 #undef COMMON_INSTR_2
767 #undef COMMON_INSTR_3
768
769
770 #ifdef AARCH64
771
772 void mov(Register dst, Register src, AsmCondition cond) {
773 if (cond == al) {
774 mov(dst, src);
775 } else {
776 csel(dst, src, dst, cond);
777 }
778 }
779
780 // Propagate other overloaded "mov" methods from Assembler.
781 void mov(Register dst, Register src) { Assembler::mov(dst, src); }
782 void mov(Register rd, int imm) { Assembler::mov(rd, imm); }
783
784 void mov(Register dst, int imm, AsmCondition cond) {
785 assert(imm == 0 || imm == 1, "");
786 if (imm == 0) {
787 mov(dst, ZR, cond);
788 } else if (imm == 1) {
789 csinc(dst, dst, ZR, inverse(cond));
790 } else if (imm == -1) {
791 csinv(dst, dst, ZR, inverse(cond));
792 } else {
793 fatal("illegal mov(R%d,%d,cond)", dst->encoding(), imm);
794 }
795 }
796
797 void movs(Register dst, Register src) { adds(dst, src, 0); }
798
799 #else // AARCH64
800
801 void tbz(Register rt, int bit, Label& L) {
802 assert(0 <= bit && bit < BitsPerWord, "bit number is out of range");
803 tst(rt, 1 << bit);
804 b(L, eq);
805 }
806
807 void tbnz(Register rt, int bit, Label& L) {
808 assert(0 <= bit && bit < BitsPerWord, "bit number is out of range");
809 tst(rt, 1 << bit);
810 b(L, ne);
811 }
812
813 void cbz(Register rt, Label& L) {
814 cmp(rt, 0);
815 b(L, eq);
816 }
817
818 void cbz(Register rt, address target) {
819 cmp(rt, 0);
820 b(target, eq);
821 }
822
823 void cbnz(Register rt, Label& L) {
824 cmp(rt, 0);
825 b(L, ne);
826 }
827
828 void ret(Register dst = LR) {
829 bx(dst);
830 }
831
832 #endif // AARCH64
833
834 Register zero_register(Register tmp) {
835 #ifdef AARCH64
836 return ZR;
837 #else
838 mov(tmp, 0);
839 return tmp;
840 #endif
841 }
842
843 void logical_shift_left(Register dst, Register src, int shift) {
844 #ifdef AARCH64
845 _lsl(dst, src, shift);
846 #else
847 mov(dst, AsmOperand(src, lsl, shift));
848 #endif
849 }
850
851 void logical_shift_left_32(Register dst, Register src, int shift) {
852 #ifdef AARCH64
853 _lsl_w(dst, src, shift);
854 #else
855 mov(dst, AsmOperand(src, lsl, shift));
856 #endif
857 }
858
859 void logical_shift_right(Register dst, Register src, int shift) {
860 #ifdef AARCH64
861 _lsr(dst, src, shift);
862 #else
863 mov(dst, AsmOperand(src, lsr, shift));
864 #endif
865 }
866
867 void arith_shift_right(Register dst, Register src, int shift) {
868 #ifdef AARCH64
869 _asr(dst, src, shift);
870 #else
871 mov(dst, AsmOperand(src, asr, shift));
872 #endif
873 }
874
875 void asr_32(Register dst, Register src, int shift) {
876 #ifdef AARCH64
877 _asr_w(dst, src, shift);
878 #else
879 mov(dst, AsmOperand(src, asr, shift));
880 #endif
881 }
882
883 // If <cond> holds, compares r1 and r2. Otherwise, flags are set so that <cond> does not hold.
884 void cond_cmp(Register r1, Register r2, AsmCondition cond) {
885 #ifdef AARCH64
886 ccmp(r1, r2, flags_for_condition(inverse(cond)), cond);
887 #else
888 cmp(r1, r2, cond);
889 #endif
890 }
891
892 // If <cond> holds, compares r and imm. Otherwise, flags are set so that <cond> does not hold.
893 void cond_cmp(Register r, int imm, AsmCondition cond) {
894 #ifdef AARCH64
895 ccmp(r, imm, flags_for_condition(inverse(cond)), cond);
896 #else
897 cmp(r, imm, cond);
898 #endif
899 }
900
901 void align_reg(Register dst, Register src, int align) {
902 assert (is_power_of_2(align), "should be");
903 #ifdef AARCH64
904 andr(dst, src, ~(uintx)(align-1));
905 #else
906 bic(dst, src, align-1);
907 #endif
908 }
909
910 void prefetch_read(Address addr) {
911 #ifdef AARCH64
912 prfm(pldl1keep, addr);
913 #else
914 pld(addr);
915 #endif
916 }
917
918 void raw_push(Register r1, Register r2) {
919 #ifdef AARCH64
920 stp(r1, r2, Address(SP, -2*wordSize, pre_indexed));
921 #else
922 assert(r1->encoding() < r2->encoding(), "should be ordered");
923 push(RegisterSet(r1) | RegisterSet(r2));
924 #endif
925 }
926
927 void raw_pop(Register r1, Register r2) {
928 #ifdef AARCH64
929 ldp(r1, r2, Address(SP, 2*wordSize, post_indexed));
930 #else
931 assert(r1->encoding() < r2->encoding(), "should be ordered");
932 pop(RegisterSet(r1) | RegisterSet(r2));
933 #endif
934 }
935
936 void raw_push(Register r1, Register r2, Register r3) {
937 #ifdef AARCH64
938 raw_push(r1, r2);
939 raw_push(r3, ZR);
940 #else
941 assert(r1->encoding() < r2->encoding() && r2->encoding() < r3->encoding(), "should be ordered");
942 push(RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3));
943 #endif
944 }
945
946 void raw_pop(Register r1, Register r2, Register r3) {
947 #ifdef AARCH64
948 raw_pop(r3, ZR);
949 raw_pop(r1, r2);
950 #else
951 assert(r1->encoding() < r2->encoding() && r2->encoding() < r3->encoding(), "should be ordered");
952 pop(RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3));
953 #endif
954 }
955
956 // Restores registers r1 and r2 previously saved by raw_push(r1, r2, ret_addr) and returns by ret_addr. Clobbers LR.
957 void raw_pop_and_ret(Register r1, Register r2) {
958 #ifdef AARCH64
959 raw_pop(r1, r2, LR);
960 ret();
961 #else
962 raw_pop(r1, r2, PC);
963 #endif
964 }
965
966 void indirect_jump(Address addr, Register scratch) {
967 #ifdef AARCH64
968 ldr(scratch, addr);
969 br(scratch);
970 #else
971 ldr(PC, addr);
972 #endif
973 }
974
975 void indirect_jump(InlinedAddress& literal, Register scratch) {
976 #ifdef AARCH64
977 ldr_literal(scratch, literal);
978 br(scratch);
979 #else
980 ldr_literal(PC, literal);
981 #endif
982 }
983
984 #ifndef AARCH64
985 void neg(Register dst, Register src) {
986 rsb(dst, src, 0);
987 }
988 #endif
989
990 void branch_if_negative_32(Register r, Label& L) {
991 // Note about branch_if_negative_32() / branch_if_any_negative_32() implementation for AArch64:
992 // tbnz is not used instead of tst & b.mi because destination may be out of tbnz range (+-32KB)
993 // since these methods are used in LIR_Assembler::emit_arraycopy() to jump to stub entry.
994 tst_32(r, r);
995 b(L, mi);
996 }
997
998 void branch_if_any_negative_32(Register r1, Register r2, Register tmp, Label& L) {
999 #ifdef AARCH64
1000 orr_32(tmp, r1, r2);
1001 tst_32(tmp, tmp);
1002 #else
1003 orrs(tmp, r1, r2);
1004 #endif
1005 b(L, mi);
1006 }
1007
1008 void branch_if_any_negative_32(Register r1, Register r2, Register r3, Register tmp, Label& L) {
1009 orr_32(tmp, r1, r2);
1010 #ifdef AARCH64
1011 orr_32(tmp, tmp, r3);
1012 tst_32(tmp, tmp);
1013 #else
1014 orrs(tmp, tmp, r3);
1015 #endif
1016 b(L, mi);
1017 }
1018
1019 void add_ptr_scaled_int32(Register dst, Register r1, Register r2, int shift) {
1020 #ifdef AARCH64
1021 add(dst, r1, r2, ex_sxtw, shift);
1022 #else
1023 add(dst, r1, AsmOperand(r2, lsl, shift));
1024 #endif
1025 }
1026
1027 void sub_ptr_scaled_int32(Register dst, Register r1, Register r2, int shift) {
1028 #ifdef AARCH64
1029 sub(dst, r1, r2, ex_sxtw, shift);
1030 #else
1031 sub(dst, r1, AsmOperand(r2, lsl, shift));
1032 #endif
1033 }
1034
1035
1036 // klass oop manipulations if compressed
1037
1038 #ifdef AARCH64
1039 void load_klass(Register dst_klass, Register src_oop);
1040 #else
1041 void load_klass(Register dst_klass, Register src_oop, AsmCondition cond = al);
1042 #endif // AARCH64
1043
1044 void store_klass(Register src_klass, Register dst_oop);
1045
1046 #ifdef AARCH64
1047 void store_klass_gap(Register dst);
1048 #endif // AARCH64
1049
1050 // oop manipulations
1051
1052 void load_heap_oop(Register dst, Address src, Register tmp1 = noreg, Register tmp2 = noreg, Register tmp3 = noreg, DecoratorSet decorators = 0);
1053 void store_heap_oop(Address obj, Register new_val, Register tmp1 = noreg, Register tmp2 = noreg, Register tmp3 = noreg, DecoratorSet decorators = 0);
1054 void store_heap_oop_null(Address obj, Register new_val, Register tmp1 = noreg, Register tmp2 = noreg, Register tmp3 = noreg, DecoratorSet decorators = 0);
1055
1056 void access_load_at(BasicType type, DecoratorSet decorators, Address src, Register dst, Register tmp1, Register tmp2, Register tmp3);
1057 void access_store_at(BasicType type, DecoratorSet decorators, Address obj, Register new_val, Register tmp1, Register tmp2, Register tmp3, bool is_null);
1058
1059 #ifdef AARCH64
1060 void encode_heap_oop(Register dst, Register src);
1061 void encode_heap_oop(Register r) {
1062 encode_heap_oop(r, r);
1063 }
1064 void decode_heap_oop(Register dst, Register src);
1065 void decode_heap_oop(Register r) {
1066 decode_heap_oop(r, r);
1067 }
1068
1069 #ifdef COMPILER2
1070 void encode_heap_oop_not_null(Register dst, Register src);
1071 void decode_heap_oop_not_null(Register dst, Register src);
1072
1073 void set_narrow_klass(Register dst, Klass* k);
1074 void set_narrow_oop(Register dst, jobject obj);
1075 #endif
1076
1077 void encode_klass_not_null(Register r);
1078 void encode_klass_not_null(Register dst, Register src);
1079 void decode_klass_not_null(Register r);
1080 void decode_klass_not_null(Register dst, Register src);
1081
1082 void reinit_heapbase();
1083
1084 #ifdef ASSERT
1085 void verify_heapbase(const char* msg);
1086 #endif // ASSERT
1087
1088 static int instr_count_for_mov_slow(intptr_t c);
1089 static int instr_count_for_mov_slow(address addr);
1090 static int instr_count_for_decode_klass_not_null();
1091 #endif // AARCH64
1092
1093 void ldr_global_ptr(Register reg, address address_of_global);
1094 void ldr_global_s32(Register reg, address address_of_global);
1095 void ldrb_global(Register reg, address address_of_global);
1096
1097 // address_placeholder_instruction is invalid instruction and is used
1098 // as placeholder in code for address of label
1099 enum { address_placeholder_instruction = 0xFFFFFFFF };
1100
1101 void emit_address(Label& L) {
1102 assert(!L.is_bound(), "otherwise address will not be patched");
1103 target(L); // creates relocation which will be patched later
1104
1105 assert ((offset() & (wordSize-1)) == 0, "should be aligned by word size");
1106
1107 #ifdef AARCH64
1108 emit_int32(address_placeholder_instruction);
1109 emit_int32(address_placeholder_instruction);
1110 #else
1111 AbstractAssembler::emit_address((address)address_placeholder_instruction);
1112 #endif
1113 }
1114
1115 void b(address target, AsmCondition cond = al) {
1116 Assembler::b(target, cond); \
1117 }
1118 void b(Label& L, AsmCondition cond = al) {
1119 // internal jumps
1120 Assembler::b(target(L), cond);
1121 }
1122
1123 void bl(address target NOT_AARCH64_ARG(AsmCondition cond = al)) {
1124 Assembler::bl(target NOT_AARCH64_ARG(cond));
1125 }
1126 void bl(Label& L NOT_AARCH64_ARG(AsmCondition cond = al)) {
1127 // internal calls
1128 Assembler::bl(target(L) NOT_AARCH64_ARG(cond));
1129 }
1130
1131 #ifndef AARCH64
1132 void adr(Register dest, Label& L, AsmCondition cond = al) {
1133 int delta = target(L) - pc() - 8;
1134 if (delta >= 0) {
1135 add(dest, PC, delta, cond);
1136 } else {
1137 sub(dest, PC, -delta, cond);
1138 }
1139 }
1140 #endif // !AARCH64
1141
1142 // Variable-length jump and calls. We now distinguish only the
1143 // patchable case from the other cases. Patchable must be
1144 // distinguised from relocable. Relocable means the generated code
1145 // containing the jump/call may move. Patchable means that the
1146 // targeted address may be changed later.
1147
1148 // Non patchable versions.
1149 // - used only for relocInfo::runtime_call_type and relocInfo::none
1150 // - may use relative or absolute format (do not use relocInfo::none
1151 // if the generated code may move)
1152 // - the implementation takes into account switch to THUMB mode if the
1153 // destination is a THUMB address
1154 // - the implementation supports far targets
1155 //
1156 // To reduce regression risk, scratch still defaults to noreg on
1157 // arm32. This results in patchable instructions. However, if
1158 // patching really matters, the call sites should be modified and
1159 // use patchable_call or patchable_jump. If patching is not required
1160 // and if a register can be cloberred, it should be explicitly
1161 // specified to allow future optimizations.
1162 void jump(address target,
1163 relocInfo::relocType rtype = relocInfo::runtime_call_type,
1164 Register scratch = AARCH64_ONLY(Rtemp) NOT_AARCH64(noreg)
1165 #ifndef AARCH64
1166 , AsmCondition cond = al
1167 #endif
1168 );
1169
1170 void call(address target,
1171 RelocationHolder rspec
1172 NOT_AARCH64_ARG(AsmCondition cond = al));
1173
1174 void call(address target,
1175 relocInfo::relocType rtype = relocInfo::runtime_call_type
1176 NOT_AARCH64_ARG(AsmCondition cond = al)) {
1177 call(target, Relocation::spec_simple(rtype) NOT_AARCH64_ARG(cond));
1178 }
1179
1180 void jump(AddressLiteral dest) {
1181 jump(dest.target(), dest.reloc());
1182 }
1183 #ifndef AARCH64
1184 void jump(address dest, relocInfo::relocType rtype, AsmCondition cond) {
1185 jump(dest, rtype, Rtemp, cond);
1186 }
1187 #endif
1188
1189 void call(AddressLiteral dest) {
1190 call(dest.target(), dest.reloc());
1191 }
1192
1193 // Patchable version:
1194 // - set_destination can be used to atomically change the target
1195 //
1196 // The targets for patchable_jump and patchable_call must be in the
1197 // code cache.
1198 // [ including possible extensions of the code cache, like AOT code ]
1199 //
1200 // To reduce regression risk, scratch still defaults to noreg on
1201 // arm32. If a register can be cloberred, it should be explicitly
1202 // specified to allow future optimizations.
1203 void patchable_jump(address target,
1204 relocInfo::relocType rtype = relocInfo::runtime_call_type,
1205 Register scratch = AARCH64_ONLY(Rtemp) NOT_AARCH64(noreg)
1206 #ifndef AARCH64
1207 , AsmCondition cond = al
1208 #endif
1209 );
1210
1211 // patchable_call may scratch Rtemp
1212 int patchable_call(address target,
1213 RelocationHolder const& rspec,
1214 bool c2 = false);
1215
1216 int patchable_call(address target,
1217 relocInfo::relocType rtype,
1218 bool c2 = false) {
1219 return patchable_call(target, Relocation::spec_simple(rtype), c2);
1220 }
1221
1222 #if defined(AARCH64) && defined(COMPILER2)
1223 static int call_size(address target, bool far, bool patchable);
1224 #endif
1225
1226 #ifdef AARCH64
1227 static bool page_reachable_from_cache(address target);
1228 #endif
1229 static bool _reachable_from_cache(address target);
1230 static bool _cache_fully_reachable();
1231 bool cache_fully_reachable();
1232 bool reachable_from_cache(address target);
1233
1234 void zero_extend(Register rd, Register rn, int bits);
1235 void sign_extend(Register rd, Register rn, int bits);
1236
1237 inline void zap_high_non_significant_bits(Register r) {
1238 #ifdef AARCH64
1239 if(ZapHighNonSignificantBits) {
1240 movk(r, 0xBAAD, 48);
1241 movk(r, 0xF00D, 32);
1242 }
1243 #endif
1244 }
1245
1246 #ifndef AARCH64
1247 void cmpoop(Register obj1, Register obj2);
1248
1249 void long_move(Register rd_lo, Register rd_hi,
1250 Register rn_lo, Register rn_hi,
1251 AsmCondition cond = al);
1252 void long_shift(Register rd_lo, Register rd_hi,
1253 Register rn_lo, Register rn_hi,
1254 AsmShift shift, Register count);
1255 void long_shift(Register rd_lo, Register rd_hi,
1256 Register rn_lo, Register rn_hi,
1257 AsmShift shift, int count);
1258
1259 void atomic_cas(Register tmpreg1, Register tmpreg2, Register oldval, Register newval, Register base, int offset);
1260 void atomic_cas_bool(Register oldval, Register newval, Register base, int offset, Register tmpreg);
1261 void atomic_cas64(Register temp_lo, Register temp_hi, Register temp_result, Register oldval_lo, Register oldval_hi, Register newval_lo, Register newval_hi, Register base, int offset);
1262 #endif // !AARCH64
1263
1264 void cas_for_lock_acquire(Register oldval, Register newval, Register base, Register tmp, Label &slow_case, bool allow_fallthrough_on_failure = false, bool one_shot = false);
1265 void cas_for_lock_release(Register oldval, Register newval, Register base, Register tmp, Label &slow_case, bool allow_fallthrough_on_failure = false, bool one_shot = false);
1266
1267 #ifndef PRODUCT
1268 // Preserves flags and all registers.
1269 // On SMP the updated value might not be visible to external observers without a sychronization barrier
1270 void cond_atomic_inc32(AsmCondition cond, int* counter_addr);
1271 #endif // !PRODUCT
1272
1273 // unconditional non-atomic increment
1274 void inc_counter(address counter_addr, Register tmpreg1, Register tmpreg2);
1275 void inc_counter(int* counter_addr, Register tmpreg1, Register tmpreg2) {
1276 inc_counter((address) counter_addr, tmpreg1, tmpreg2);
1277 }
1278
1279 void pd_patch_instruction(address branch, address target, const char* file, int line);
1280
1281 // Loading and storing values by size and signed-ness;
1282 // size must not exceed wordSize (i.e. 8-byte values are not supported on 32-bit ARM);
1283 // each of these calls generates exactly one load or store instruction,
1284 // so src can be pre- or post-indexed address.
1285 #ifdef AARCH64
1286 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
1287 void store_sized_value(Register src, Address dst, size_t size_in_bytes);
1288 #else
1289 // 32-bit ARM variants also support conditional execution
1290 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, AsmCondition cond = al);
1291 void store_sized_value(Register src, Address dst, size_t size_in_bytes, AsmCondition cond = al);
1292 #endif
1293
1294 void lookup_interface_method(Register recv_klass,
1295 Register intf_klass,
1296 RegisterOrConstant itable_index,
1297 Register method_result,
1298 Register temp_reg1,
1299 Register temp_reg2,
1300 Label& L_no_such_interface);
1301
1302 // Compare char[] arrays aligned to 4 bytes.
1303 void char_arrays_equals(Register ary1, Register ary2,
1304 Register limit, Register result,
1305 Register chr1, Register chr2, Label& Ldone);
1306
1307
1308 void floating_cmp(Register dst);
1309
1310 // improved x86 portability (minimizing source code changes)
1311
1312 void ldr_literal(Register rd, AddressLiteral addr) {
1313 relocate(addr.rspec());
1314 #ifdef AARCH64
1315 ldr(rd, addr.target());
1316 #else
1317 ldr(rd, Address(PC, addr.target() - pc() - 8));
1318 #endif
1319 }
1320
1321 void lea(Register Rd, AddressLiteral addr) {
1322 // Never dereferenced, as on x86 (lval status ignored)
1323 mov_address(Rd, addr.target(), addr.rspec());
1324 }
1325
1326 void restore_default_fp_mode();
1327
1328 #ifdef COMPILER2
1329 #ifdef AARCH64
1330 // Code used by cmpFastLock and cmpFastUnlock mach instructions in .ad file.
1331 void fast_lock(Register obj, Register box, Register scratch, Register scratch2, Register scratch3);
1332 void fast_unlock(Register obj, Register box, Register scratch, Register scratch2, Register scratch3);
1333 #else
1334 void fast_lock(Register obj, Register box, Register scratch, Register scratch2);
1335 void fast_unlock(Register obj, Register box, Register scratch, Register scratch2);
1336 #endif
1337 #endif
1338
1339 #ifdef AARCH64
1340
1341 #define F(mnemonic) \
1342 void mnemonic(Register rt, address target) { \
1343 Assembler::mnemonic(rt, target); \
1344 } \
1345 void mnemonic(Register rt, Label& L) { \
1346 Assembler::mnemonic(rt, target(L)); \
1347 }
1348
1349 F(cbz_w);
1350 F(cbnz_w);
1351 F(cbz);
1352 F(cbnz);
1353
1354 #undef F
1355
1356 #define F(mnemonic) \
1357 void mnemonic(Register rt, int bit, address target) { \
1358 Assembler::mnemonic(rt, bit, target); \
1359 } \
1360 void mnemonic(Register rt, int bit, Label& L) { \
1361 Assembler::mnemonic(rt, bit, target(L)); \
1362 }
1363
1364 F(tbz);
1365 F(tbnz);
1366 #undef F
1367
1368 #endif // AARCH64
1369
1370 };
1371
1372
1373 // The purpose of this class is to build several code fragments of the same size
1374 // in order to allow fast table branch.
1375
1376 class FixedSizeCodeBlock {
1377 public:
1378 FixedSizeCodeBlock(MacroAssembler* masm, int size_in_instrs, bool enabled);
1379 ~FixedSizeCodeBlock();
1380
1381 private:
1382 MacroAssembler* _masm;
1383 address _start;
1384 int _size_in_instrs;
1385 bool _enabled;
1386 };
1387
1388
1389 #endif // CPU_ARM_VM_MACROASSEMBLER_ARM_HPP
1390