1 /*
2 * Copyright (c) 2003, 2011, 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 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "assembler_x86.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "nativeInst_x86.hpp"
30 #include "oops/instanceOop.hpp"
31 #include "oops/methodOop.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "oops/oop.inline.hpp"
34 #include "prims/methodHandles.hpp"
35 #include "runtime/frame.inline.hpp"
36 #include "runtime/handles.inline.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubCodeGenerator.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "utilities/top.hpp"
41 #ifdef TARGET_OS_FAMILY_linux
42 # include "thread_linux.inline.hpp"
43 #endif
44 #ifdef TARGET_OS_FAMILY_solaris
45 # include "thread_solaris.inline.hpp"
46 #endif
47 #ifdef TARGET_OS_FAMILY_windows
48 # include "thread_windows.inline.hpp"
49 #endif
50 #ifdef COMPILER2
51 #include "opto/runtime.hpp"
52 #endif
53
54 // Declaration and definition of StubGenerator (no .hpp file).
55 // For a more detailed description of the stub routine structure
56 // see the comment in stubRoutines.hpp
57
58 #define __ _masm->
59 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
60 #define a__ ((Assembler*)_masm)->
61
62 #ifdef PRODUCT
63 #define BLOCK_COMMENT(str) /* nothing */
64 #else
65 #define BLOCK_COMMENT(str) __ block_comment(str)
66 #endif
67
68 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
69 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
70
71 // Stub Code definitions
72
73 static address handle_unsafe_access() {
74 JavaThread* thread = JavaThread::current();
75 address pc = thread->saved_exception_pc();
76 // pc is the instruction which we must emulate
77 // doing a no-op is fine: return garbage from the load
78 // therefore, compute npc
79 address npc = Assembler::locate_next_instruction(pc);
80
81 // request an async exception
82 thread->set_pending_unsafe_access_error();
83
84 // return address of next instruction to execute
85 return npc;
86 }
87
88 class StubGenerator: public StubCodeGenerator {
89 private:
90
91 #ifdef PRODUCT
92 #define inc_counter_np(counter) (0)
93 #else
94 void inc_counter_np_(int& counter) {
95 __ incrementl(ExternalAddress((address)&counter));
96 }
97 #define inc_counter_np(counter) \
98 BLOCK_COMMENT("inc_counter " #counter); \
99 inc_counter_np_(counter);
100 #endif
101
102 // Call stubs are used to call Java from C
103 //
104 // Linux Arguments:
105 // c_rarg0: call wrapper address address
106 // c_rarg1: result address
107 // c_rarg2: result type BasicType
108 // c_rarg3: method methodOop
109 // c_rarg4: (interpreter) entry point address
110 // c_rarg5: parameters intptr_t*
111 // 16(rbp): parameter size (in words) int
112 // 24(rbp): thread Thread*
113 //
114 // [ return_from_Java ] <--- rsp
115 // [ argument word n ]
116 // ...
117 // -12 [ argument word 1 ]
118 // -11 [ saved r15 ] <--- rsp_after_call
119 // -10 [ saved r14 ]
120 // -9 [ saved r13 ]
121 // -8 [ saved r12 ]
122 // -7 [ saved rbx ]
123 // -6 [ call wrapper ]
124 // -5 [ result ]
125 // -4 [ result type ]
126 // -3 [ method ]
127 // -2 [ entry point ]
128 // -1 [ parameters ]
129 // 0 [ saved rbp ] <--- rbp
130 // 1 [ return address ]
131 // 2 [ parameter size ]
132 // 3 [ thread ]
133 //
134 // Windows Arguments:
135 // c_rarg0: call wrapper address address
136 // c_rarg1: result address
137 // c_rarg2: result type BasicType
138 // c_rarg3: method methodOop
139 // 48(rbp): (interpreter) entry point address
140 // 56(rbp): parameters intptr_t*
141 // 64(rbp): parameter size (in words) int
142 // 72(rbp): thread Thread*
143 //
144 // [ return_from_Java ] <--- rsp
145 // [ argument word n ]
146 // ...
147 // -28 [ argument word 1 ]
148 // -27 [ saved xmm15 ] <--- rsp_after_call
149 // [ saved xmm7-xmm14 ]
150 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
151 // -7 [ saved r15 ]
152 // -6 [ saved r14 ]
153 // -5 [ saved r13 ]
154 // -4 [ saved r12 ]
155 // -3 [ saved rdi ]
156 // -2 [ saved rsi ]
157 // -1 [ saved rbx ]
158 // 0 [ saved rbp ] <--- rbp
159 // 1 [ return address ]
160 // 2 [ call wrapper ]
161 // 3 [ result ]
162 // 4 [ result type ]
163 // 5 [ method ]
164 // 6 [ entry point ]
165 // 7 [ parameters ]
166 // 8 [ parameter size ]
167 // 9 [ thread ]
168 //
169 // Windows reserves the callers stack space for arguments 1-4.
170 // We spill c_rarg0-c_rarg3 to this space.
171
172 // Call stub stack layout word offsets from rbp
173 enum call_stub_layout {
174 #ifdef _WIN64
175 xmm_save_first = 6, // save from xmm6
176 xmm_save_last = 15, // to xmm15
177 xmm_save_base = -9,
178 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
179 r15_off = -7,
180 r14_off = -6,
181 r13_off = -5,
182 r12_off = -4,
183 rdi_off = -3,
184 rsi_off = -2,
185 rbx_off = -1,
186 rbp_off = 0,
187 retaddr_off = 1,
188 call_wrapper_off = 2,
189 result_off = 3,
190 result_type_off = 4,
191 method_off = 5,
192 entry_point_off = 6,
193 parameters_off = 7,
194 parameter_size_off = 8,
195 thread_off = 9
196 #else
197 rsp_after_call_off = -12,
198 mxcsr_off = rsp_after_call_off,
199 r15_off = -11,
200 r14_off = -10,
201 r13_off = -9,
202 r12_off = -8,
203 rbx_off = -7,
204 call_wrapper_off = -6,
205 result_off = -5,
206 result_type_off = -4,
207 method_off = -3,
208 entry_point_off = -2,
209 parameters_off = -1,
210 rbp_off = 0,
211 retaddr_off = 1,
212 parameter_size_off = 2,
213 thread_off = 3
214 #endif
215 };
216
217 #ifdef _WIN64
218 Address xmm_save(int reg) {
219 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
220 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
221 }
222 #endif
223
224 address generate_call_stub(address& return_address) {
225 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
226 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
227 "adjust this code");
228 StubCodeMark mark(this, "StubRoutines", "call_stub");
229 address start = __ pc();
230
231 // same as in generate_catch_exception()!
232 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
233
234 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
235 const Address result (rbp, result_off * wordSize);
236 const Address result_type (rbp, result_type_off * wordSize);
237 const Address method (rbp, method_off * wordSize);
238 const Address entry_point (rbp, entry_point_off * wordSize);
239 const Address parameters (rbp, parameters_off * wordSize);
240 const Address parameter_size(rbp, parameter_size_off * wordSize);
241
242 // same as in generate_catch_exception()!
243 const Address thread (rbp, thread_off * wordSize);
244
245 const Address r15_save(rbp, r15_off * wordSize);
246 const Address r14_save(rbp, r14_off * wordSize);
247 const Address r13_save(rbp, r13_off * wordSize);
248 const Address r12_save(rbp, r12_off * wordSize);
249 const Address rbx_save(rbp, rbx_off * wordSize);
250
251 // stub code
252 __ enter();
253 __ subptr(rsp, -rsp_after_call_off * wordSize);
254
255 // save register parameters
256 #ifndef _WIN64
257 __ movptr(parameters, c_rarg5); // parameters
258 __ movptr(entry_point, c_rarg4); // entry_point
259 #endif
260
261 __ movptr(method, c_rarg3); // method
262 __ movl(result_type, c_rarg2); // result type
263 __ movptr(result, c_rarg1); // result
264 __ movptr(call_wrapper, c_rarg0); // call wrapper
265
266 // save regs belonging to calling function
267 __ movptr(rbx_save, rbx);
268 __ movptr(r12_save, r12);
269 __ movptr(r13_save, r13);
270 __ movptr(r14_save, r14);
271 __ movptr(r15_save, r15);
272 #ifdef _WIN64
273 for (int i = 6; i <= 15; i++) {
274 __ movdqu(xmm_save(i), as_XMMRegister(i));
275 }
276
277 const Address rdi_save(rbp, rdi_off * wordSize);
278 const Address rsi_save(rbp, rsi_off * wordSize);
279
280 __ movptr(rsi_save, rsi);
281 __ movptr(rdi_save, rdi);
282 #else
283 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
284 {
285 Label skip_ldmx;
286 __ stmxcsr(mxcsr_save);
287 __ movl(rax, mxcsr_save);
288 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
289 ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
290 __ cmp32(rax, mxcsr_std);
291 __ jcc(Assembler::equal, skip_ldmx);
292 __ ldmxcsr(mxcsr_std);
293 __ bind(skip_ldmx);
294 }
295 #endif
296
297 // Load up thread register
298 __ movptr(r15_thread, thread);
299 __ reinit_heapbase();
300
301 #ifdef ASSERT
302 // make sure we have no pending exceptions
303 {
304 Label L;
305 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
306 __ jcc(Assembler::equal, L);
307 __ stop("StubRoutines::call_stub: entered with pending exception");
308 __ bind(L);
309 }
310 #endif
311
312 // pass parameters if any
313 BLOCK_COMMENT("pass parameters if any");
314 Label parameters_done;
315 __ movl(c_rarg3, parameter_size);
316 __ testl(c_rarg3, c_rarg3);
317 __ jcc(Assembler::zero, parameters_done);
318
319 Label loop;
320 __ movptr(c_rarg2, parameters); // parameter pointer
321 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
322 __ BIND(loop);
323 __ movptr(rax, Address(c_rarg2, 0));// get parameter
324 __ addptr(c_rarg2, wordSize); // advance to next parameter
325 __ decrementl(c_rarg1); // decrement counter
326 __ push(rax); // pass parameter
327 __ jcc(Assembler::notZero, loop);
328
329 // call Java function
330 __ BIND(parameters_done);
331 __ movptr(rbx, method); // get methodOop
332 __ movptr(c_rarg1, entry_point); // get entry_point
333 __ mov(r13, rsp); // set sender sp
334 BLOCK_COMMENT("call Java function");
335 __ call(c_rarg1);
336
337 BLOCK_COMMENT("call_stub_return_address:");
338 return_address = __ pc();
339
340 // store result depending on type (everything that is not
341 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
342 __ movptr(c_rarg0, result);
343 Label is_long, is_float, is_double, exit;
344 __ movl(c_rarg1, result_type);
345 __ cmpl(c_rarg1, T_OBJECT);
346 __ jcc(Assembler::equal, is_long);
347 __ cmpl(c_rarg1, T_LONG);
348 __ jcc(Assembler::equal, is_long);
349 __ cmpl(c_rarg1, T_FLOAT);
350 __ jcc(Assembler::equal, is_float);
351 __ cmpl(c_rarg1, T_DOUBLE);
352 __ jcc(Assembler::equal, is_double);
353
354 // handle T_INT case
355 __ movl(Address(c_rarg0, 0), rax);
356
357 __ BIND(exit);
358
359 // pop parameters
360 __ lea(rsp, rsp_after_call);
361
362 #ifdef ASSERT
363 // verify that threads correspond
364 {
365 Label L, S;
366 __ cmpptr(r15_thread, thread);
367 __ jcc(Assembler::notEqual, S);
368 __ get_thread(rbx);
369 __ cmpptr(r15_thread, rbx);
370 __ jcc(Assembler::equal, L);
371 __ bind(S);
372 __ jcc(Assembler::equal, L);
373 __ stop("StubRoutines::call_stub: threads must correspond");
374 __ bind(L);
375 }
376 #endif
377
378 // restore regs belonging to calling function
379 #ifdef _WIN64
380 for (int i = 15; i >= 6; i--) {
381 __ movdqu(as_XMMRegister(i), xmm_save(i));
382 }
383 #endif
384 __ movptr(r15, r15_save);
385 __ movptr(r14, r14_save);
386 __ movptr(r13, r13_save);
387 __ movptr(r12, r12_save);
388 __ movptr(rbx, rbx_save);
389
390 #ifdef _WIN64
391 __ movptr(rdi, rdi_save);
392 __ movptr(rsi, rsi_save);
393 #else
394 __ ldmxcsr(mxcsr_save);
395 #endif
396
397 // restore rsp
398 __ addptr(rsp, -rsp_after_call_off * wordSize);
399
400 // return
401 __ pop(rbp);
402 __ ret(0);
403
404 // handle return types different from T_INT
405 __ BIND(is_long);
406 __ movq(Address(c_rarg0, 0), rax);
407 __ jmp(exit);
408
409 __ BIND(is_float);
410 __ movflt(Address(c_rarg0, 0), xmm0);
411 __ jmp(exit);
412
413 __ BIND(is_double);
414 __ movdbl(Address(c_rarg0, 0), xmm0);
415 __ jmp(exit);
416
417 return start;
418 }
419
420 // Return point for a Java call if there's an exception thrown in
421 // Java code. The exception is caught and transformed into a
422 // pending exception stored in JavaThread that can be tested from
423 // within the VM.
424 //
425 // Note: Usually the parameters are removed by the callee. In case
426 // of an exception crossing an activation frame boundary, that is
427 // not the case if the callee is compiled code => need to setup the
428 // rsp.
429 //
430 // rax: exception oop
431
432 address generate_catch_exception() {
433 StubCodeMark mark(this, "StubRoutines", "catch_exception");
434 address start = __ pc();
435
436 // same as in generate_call_stub():
437 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
438 const Address thread (rbp, thread_off * wordSize);
439
440 #ifdef ASSERT
441 // verify that threads correspond
442 {
443 Label L, S;
444 __ cmpptr(r15_thread, thread);
445 __ jcc(Assembler::notEqual, S);
446 __ get_thread(rbx);
447 __ cmpptr(r15_thread, rbx);
448 __ jcc(Assembler::equal, L);
449 __ bind(S);
450 __ stop("StubRoutines::catch_exception: threads must correspond");
451 __ bind(L);
452 }
453 #endif
454
455 // set pending exception
456 __ verify_oop(rax);
457
458 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
459 __ lea(rscratch1, ExternalAddress((address)__FILE__));
460 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
461 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
462
463 // complete return to VM
464 assert(StubRoutines::_call_stub_return_address != NULL,
465 "_call_stub_return_address must have been generated before");
466 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
467
468 return start;
469 }
470
471 // Continuation point for runtime calls returning with a pending
472 // exception. The pending exception check happened in the runtime
473 // or native call stub. The pending exception in Thread is
474 // converted into a Java-level exception.
475 //
476 // Contract with Java-level exception handlers:
477 // rax: exception
478 // rdx: throwing pc
479 //
480 // NOTE: At entry of this stub, exception-pc must be on stack !!
481
482 address generate_forward_exception() {
483 StubCodeMark mark(this, "StubRoutines", "forward exception");
484 address start = __ pc();
485
486 // Upon entry, the sp points to the return address returning into
487 // Java (interpreted or compiled) code; i.e., the return address
488 // becomes the throwing pc.
489 //
490 // Arguments pushed before the runtime call are still on the stack
491 // but the exception handler will reset the stack pointer ->
492 // ignore them. A potential result in registers can be ignored as
493 // well.
494
495 #ifdef ASSERT
496 // make sure this code is only executed if there is a pending exception
497 {
498 Label L;
499 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
500 __ jcc(Assembler::notEqual, L);
501 __ stop("StubRoutines::forward exception: no pending exception (1)");
502 __ bind(L);
503 }
504 #endif
505
506 // compute exception handler into rbx
507 __ movptr(c_rarg0, Address(rsp, 0));
508 BLOCK_COMMENT("call exception_handler_for_return_address");
509 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
510 SharedRuntime::exception_handler_for_return_address),
511 r15_thread, c_rarg0);
512 __ mov(rbx, rax);
513
514 // setup rax & rdx, remove return address & clear pending exception
515 __ pop(rdx);
516 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
517 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
518
519 #ifdef ASSERT
520 // make sure exception is set
521 {
522 Label L;
523 __ testptr(rax, rax);
524 __ jcc(Assembler::notEqual, L);
525 __ stop("StubRoutines::forward exception: no pending exception (2)");
526 __ bind(L);
527 }
528 #endif
529
530 // continue at exception handler (return address removed)
531 // rax: exception
532 // rbx: exception handler
533 // rdx: throwing pc
534 __ verify_oop(rax);
535 __ jmp(rbx);
536
537 return start;
538 }
539
540 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
541 //
542 // Arguments :
543 // c_rarg0: exchange_value
544 // c_rarg0: dest
545 //
546 // Result:
547 // *dest <- ex, return (orig *dest)
548 address generate_atomic_xchg() {
549 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
550 address start = __ pc();
551
552 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
553 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
554 __ ret(0);
555
556 return start;
557 }
558
559 // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
560 //
561 // Arguments :
562 // c_rarg0: exchange_value
563 // c_rarg1: dest
564 //
565 // Result:
566 // *dest <- ex, return (orig *dest)
567 address generate_atomic_xchg_ptr() {
568 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
569 address start = __ pc();
570
571 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
572 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
573 __ ret(0);
574
575 return start;
576 }
577
578 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
579 // jint compare_value)
580 //
581 // Arguments :
582 // c_rarg0: exchange_value
583 // c_rarg1: dest
584 // c_rarg2: compare_value
585 //
586 // Result:
587 // if ( compare_value == *dest ) {
588 // *dest = exchange_value
589 // return compare_value;
590 // else
591 // return *dest;
592 address generate_atomic_cmpxchg() {
593 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
594 address start = __ pc();
595
596 __ movl(rax, c_rarg2);
597 if ( os::is_MP() ) __ lock();
598 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
599 __ ret(0);
600
601 return start;
602 }
603
604 // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
605 // volatile jlong* dest,
606 // jlong compare_value)
607 // Arguments :
608 // c_rarg0: exchange_value
609 // c_rarg1: dest
610 // c_rarg2: compare_value
611 //
612 // Result:
613 // if ( compare_value == *dest ) {
614 // *dest = exchange_value
615 // return compare_value;
616 // else
617 // return *dest;
618 address generate_atomic_cmpxchg_long() {
619 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
620 address start = __ pc();
621
622 __ movq(rax, c_rarg2);
623 if ( os::is_MP() ) __ lock();
624 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
625 __ ret(0);
626
627 return start;
628 }
629
630 // Support for jint atomic::add(jint add_value, volatile jint* dest)
631 //
632 // Arguments :
633 // c_rarg0: add_value
634 // c_rarg1: dest
635 //
636 // Result:
637 // *dest += add_value
638 // return *dest;
639 address generate_atomic_add() {
640 StubCodeMark mark(this, "StubRoutines", "atomic_add");
641 address start = __ pc();
642
643 __ movl(rax, c_rarg0);
644 if ( os::is_MP() ) __ lock();
645 __ xaddl(Address(c_rarg1, 0), c_rarg0);
646 __ addl(rax, c_rarg0);
647 __ ret(0);
648
649 return start;
650 }
651
652 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
653 //
654 // Arguments :
655 // c_rarg0: add_value
656 // c_rarg1: dest
657 //
658 // Result:
659 // *dest += add_value
660 // return *dest;
661 address generate_atomic_add_ptr() {
662 StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
663 address start = __ pc();
664
665 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
666 if ( os::is_MP() ) __ lock();
667 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
668 __ addptr(rax, c_rarg0);
669 __ ret(0);
670
671 return start;
672 }
673
674 // Support for intptr_t OrderAccess::fence()
675 //
676 // Arguments :
677 //
678 // Result:
679 address generate_orderaccess_fence() {
680 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
681 address start = __ pc();
682 __ membar(Assembler::StoreLoad);
683 __ ret(0);
684
685 return start;
686 }
687
688 // Support for intptr_t get_previous_fp()
689 //
690 // This routine is used to find the previous frame pointer for the
691 // caller (current_frame_guess). This is used as part of debugging
692 // ps() is seemingly lost trying to find frames.
693 // This code assumes that caller current_frame_guess) has a frame.
694 address generate_get_previous_fp() {
695 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
696 const Address old_fp(rbp, 0);
697 const Address older_fp(rax, 0);
698 address start = __ pc();
699
700 __ enter();
701 __ movptr(rax, old_fp); // callers fp
702 __ movptr(rax, older_fp); // the frame for ps()
703 __ pop(rbp);
704 __ ret(0);
705
706 return start;
707 }
708
709 //----------------------------------------------------------------------------------------------------
710 // Support for void verify_mxcsr()
711 //
712 // This routine is used with -Xcheck:jni to verify that native
713 // JNI code does not return to Java code without restoring the
714 // MXCSR register to our expected state.
715
716 address generate_verify_mxcsr() {
717 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
718 address start = __ pc();
719
720 const Address mxcsr_save(rsp, 0);
721
722 if (CheckJNICalls) {
723 Label ok_ret;
724 __ push(rax);
725 __ subptr(rsp, wordSize); // allocate a temp location
726 __ stmxcsr(mxcsr_save);
727 __ movl(rax, mxcsr_save);
728 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
729 __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
730 __ jcc(Assembler::equal, ok_ret);
731
732 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
733
734 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
735
736 __ bind(ok_ret);
737 __ addptr(rsp, wordSize);
738 __ pop(rax);
739 }
740
741 __ ret(0);
742
743 return start;
744 }
745
746 address generate_f2i_fixup() {
747 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
748 Address inout(rsp, 5 * wordSize); // return address + 4 saves
749
750 address start = __ pc();
751
752 Label L;
753
754 __ push(rax);
755 __ push(c_rarg3);
756 __ push(c_rarg2);
757 __ push(c_rarg1);
758
759 __ movl(rax, 0x7f800000);
760 __ xorl(c_rarg3, c_rarg3);
761 __ movl(c_rarg2, inout);
762 __ movl(c_rarg1, c_rarg2);
763 __ andl(c_rarg1, 0x7fffffff);
764 __ cmpl(rax, c_rarg1); // NaN? -> 0
765 __ jcc(Assembler::negative, L);
766 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
767 __ movl(c_rarg3, 0x80000000);
768 __ movl(rax, 0x7fffffff);
769 __ cmovl(Assembler::positive, c_rarg3, rax);
770
771 __ bind(L);
772 __ movptr(inout, c_rarg3);
773
774 __ pop(c_rarg1);
775 __ pop(c_rarg2);
776 __ pop(c_rarg3);
777 __ pop(rax);
778
779 __ ret(0);
780
781 return start;
782 }
783
784 address generate_f2l_fixup() {
785 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
786 Address inout(rsp, 5 * wordSize); // return address + 4 saves
787 address start = __ pc();
788
789 Label L;
790
791 __ push(rax);
792 __ push(c_rarg3);
793 __ push(c_rarg2);
794 __ push(c_rarg1);
795
796 __ movl(rax, 0x7f800000);
797 __ xorl(c_rarg3, c_rarg3);
798 __ movl(c_rarg2, inout);
799 __ movl(c_rarg1, c_rarg2);
800 __ andl(c_rarg1, 0x7fffffff);
801 __ cmpl(rax, c_rarg1); // NaN? -> 0
802 __ jcc(Assembler::negative, L);
803 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
804 __ mov64(c_rarg3, 0x8000000000000000);
805 __ mov64(rax, 0x7fffffffffffffff);
806 __ cmov(Assembler::positive, c_rarg3, rax);
807
808 __ bind(L);
809 __ movptr(inout, c_rarg3);
810
811 __ pop(c_rarg1);
812 __ pop(c_rarg2);
813 __ pop(c_rarg3);
814 __ pop(rax);
815
816 __ ret(0);
817
818 return start;
819 }
820
821 address generate_d2i_fixup() {
822 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
823 Address inout(rsp, 6 * wordSize); // return address + 5 saves
824
825 address start = __ pc();
826
827 Label L;
828
829 __ push(rax);
830 __ push(c_rarg3);
831 __ push(c_rarg2);
832 __ push(c_rarg1);
833 __ push(c_rarg0);
834
835 __ movl(rax, 0x7ff00000);
836 __ movq(c_rarg2, inout);
837 __ movl(c_rarg3, c_rarg2);
838 __ mov(c_rarg1, c_rarg2);
839 __ mov(c_rarg0, c_rarg2);
840 __ negl(c_rarg3);
841 __ shrptr(c_rarg1, 0x20);
842 __ orl(c_rarg3, c_rarg2);
843 __ andl(c_rarg1, 0x7fffffff);
844 __ xorl(c_rarg2, c_rarg2);
845 __ shrl(c_rarg3, 0x1f);
846 __ orl(c_rarg1, c_rarg3);
847 __ cmpl(rax, c_rarg1);
848 __ jcc(Assembler::negative, L); // NaN -> 0
849 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
850 __ movl(c_rarg2, 0x80000000);
851 __ movl(rax, 0x7fffffff);
852 __ cmov(Assembler::positive, c_rarg2, rax);
853
854 __ bind(L);
855 __ movptr(inout, c_rarg2);
856
857 __ pop(c_rarg0);
858 __ pop(c_rarg1);
859 __ pop(c_rarg2);
860 __ pop(c_rarg3);
861 __ pop(rax);
862
863 __ ret(0);
864
865 return start;
866 }
867
868 address generate_d2l_fixup() {
869 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
870 Address inout(rsp, 6 * wordSize); // return address + 5 saves
871
872 address start = __ pc();
873
874 Label L;
875
876 __ push(rax);
877 __ push(c_rarg3);
878 __ push(c_rarg2);
879 __ push(c_rarg1);
880 __ push(c_rarg0);
881
882 __ movl(rax, 0x7ff00000);
883 __ movq(c_rarg2, inout);
884 __ movl(c_rarg3, c_rarg2);
885 __ mov(c_rarg1, c_rarg2);
886 __ mov(c_rarg0, c_rarg2);
887 __ negl(c_rarg3);
888 __ shrptr(c_rarg1, 0x20);
889 __ orl(c_rarg3, c_rarg2);
890 __ andl(c_rarg1, 0x7fffffff);
891 __ xorl(c_rarg2, c_rarg2);
892 __ shrl(c_rarg3, 0x1f);
893 __ orl(c_rarg1, c_rarg3);
894 __ cmpl(rax, c_rarg1);
895 __ jcc(Assembler::negative, L); // NaN -> 0
896 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
897 __ mov64(c_rarg2, 0x8000000000000000);
898 __ mov64(rax, 0x7fffffffffffffff);
899 __ cmovq(Assembler::positive, c_rarg2, rax);
900
901 __ bind(L);
902 __ movq(inout, c_rarg2);
903
904 __ pop(c_rarg0);
905 __ pop(c_rarg1);
906 __ pop(c_rarg2);
907 __ pop(c_rarg3);
908 __ pop(rax);
909
910 __ ret(0);
911
912 return start;
913 }
914
915 address generate_fp_mask(const char *stub_name, int64_t mask) {
916 __ align(CodeEntryAlignment);
917 StubCodeMark mark(this, "StubRoutines", stub_name);
918 address start = __ pc();
919
920 __ emit_data64( mask, relocInfo::none );
921 __ emit_data64( mask, relocInfo::none );
922
923 return start;
924 }
925
926 // The following routine generates a subroutine to throw an
927 // asynchronous UnknownError when an unsafe access gets a fault that
928 // could not be reasonably prevented by the programmer. (Example:
929 // SIGBUS/OBJERR.)
930 address generate_handler_for_unsafe_access() {
931 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
932 address start = __ pc();
933
934 __ push(0); // hole for return address-to-be
935 __ pusha(); // push registers
936 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
937
938 __ subptr(rsp, frame::arg_reg_save_area_bytes);
939 BLOCK_COMMENT("call handle_unsafe_access");
940 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
941 __ addptr(rsp, frame::arg_reg_save_area_bytes);
942
943 __ movptr(next_pc, rax); // stuff next address
944 __ popa();
945 __ ret(0); // jump to next address
946
947 return start;
948 }
949
950 // Non-destructive plausibility checks for oops
951 //
952 // Arguments:
953 // all args on stack!
954 //
955 // Stack after saving c_rarg3:
956 // [tos + 0]: saved c_rarg3
957 // [tos + 1]: saved c_rarg2
958 // [tos + 2]: saved r12 (several TemplateTable methods use it)
959 // [tos + 3]: saved flags
960 // [tos + 4]: return address
961 // * [tos + 5]: error message (char*)
962 // * [tos + 6]: object to verify (oop)
963 // * [tos + 7]: saved rax - saved by caller and bashed
964 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
965 // * = popped on exit
966 address generate_verify_oop() {
967 StubCodeMark mark(this, "StubRoutines", "verify_oop");
968 address start = __ pc();
969
970 Label exit, error;
971
972 __ pushf();
973 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
974
975 __ push(r12);
976
977 // save c_rarg2 and c_rarg3
978 __ push(c_rarg2);
979 __ push(c_rarg3);
980
981 enum {
982 // After previous pushes.
983 oop_to_verify = 6 * wordSize,
984 saved_rax = 7 * wordSize,
985 saved_r10 = 8 * wordSize,
986
987 // Before the call to MacroAssembler::debug(), see below.
988 return_addr = 16 * wordSize,
989 error_msg = 17 * wordSize
990 };
991
992 // get object
993 __ movptr(rax, Address(rsp, oop_to_verify));
994
995 // make sure object is 'reasonable'
996 __ testptr(rax, rax);
997 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
998 // Check if the oop is in the right area of memory
999 __ movptr(c_rarg2, rax);
1000 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1001 __ andptr(c_rarg2, c_rarg3);
1002 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1003 __ cmpptr(c_rarg2, c_rarg3);
1004 __ jcc(Assembler::notZero, error);
1005
1006 // set r12 to heapbase for load_klass()
1007 __ reinit_heapbase();
1008
1009 // make sure klass is 'reasonable'
1010 __ load_klass(rax, rax); // get klass
1011 __ testptr(rax, rax);
1012 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1013 // Check if the klass is in the right area of memory
1014 __ mov(c_rarg2, rax);
1015 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
1016 __ andptr(c_rarg2, c_rarg3);
1017 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
1018 __ cmpptr(c_rarg2, c_rarg3);
1019 __ jcc(Assembler::notZero, error);
1020
1021 // make sure klass' klass is 'reasonable'
1022 __ load_klass(rax, rax);
1023 __ testptr(rax, rax);
1024 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
1025 // Check if the klass' klass is in the right area of memory
1026 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
1027 __ andptr(rax, c_rarg3);
1028 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
1029 __ cmpptr(rax, c_rarg3);
1030 __ jcc(Assembler::notZero, error);
1031
1032 // return if everything seems ok
1033 __ bind(exit);
1034 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1035 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1036 __ pop(c_rarg3); // restore c_rarg3
1037 __ pop(c_rarg2); // restore c_rarg2
1038 __ pop(r12); // restore r12
1039 __ popf(); // restore flags
1040 __ ret(4 * wordSize); // pop caller saved stuff
1041
1042 // handle errors
1043 __ bind(error);
1044 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1045 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1046 __ pop(c_rarg3); // get saved c_rarg3 back
1047 __ pop(c_rarg2); // get saved c_rarg2 back
1048 __ pop(r12); // get saved r12 back
1049 __ popf(); // get saved flags off stack --
1050 // will be ignored
1051
1052 __ pusha(); // push registers
1053 // (rip is already
1054 // already pushed)
1055 // debug(char* msg, int64_t pc, int64_t regs[])
1056 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1057 // pushed all the registers, so now the stack looks like:
1058 // [tos + 0] 16 saved registers
1059 // [tos + 16] return address
1060 // * [tos + 17] error message (char*)
1061 // * [tos + 18] object to verify (oop)
1062 // * [tos + 19] saved rax - saved by caller and bashed
1063 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1064 // * = popped on exit
1065
1066 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1067 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1068 __ movq(c_rarg2, rsp); // pass address of regs on stack
1069 __ mov(r12, rsp); // remember rsp
1070 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1071 __ andptr(rsp, -16); // align stack as required by ABI
1072 BLOCK_COMMENT("call MacroAssembler::debug");
1073 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1074 __ mov(rsp, r12); // restore rsp
1075 __ popa(); // pop registers (includes r12)
1076 __ ret(4 * wordSize); // pop caller saved stuff
1077
1078 return start;
1079 }
1080
1081 //
1082 // Verify that a register contains clean 32-bits positive value
1083 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1084 //
1085 // Input:
1086 // Rint - 32-bits value
1087 // Rtmp - scratch
1088 //
1089 void assert_clean_int(Register Rint, Register Rtmp) {
1090 #ifdef ASSERT
1091 Label L;
1092 assert_different_registers(Rtmp, Rint);
1093 __ movslq(Rtmp, Rint);
1094 __ cmpq(Rtmp, Rint);
1095 __ jcc(Assembler::equal, L);
1096 __ stop("high 32-bits of int value are not 0");
1097 __ bind(L);
1098 #endif
1099 }
1100
1101 // Generate overlap test for array copy stubs
1102 //
1103 // Input:
1104 // c_rarg0 - from
1105 // c_rarg1 - to
1106 // c_rarg2 - element count
1107 //
1108 // Output:
1109 // rax - &from[element count - 1]
1110 //
1111 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1112 assert(no_overlap_target != NULL, "must be generated");
1113 array_overlap_test(no_overlap_target, NULL, sf);
1114 }
1115 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1116 array_overlap_test(NULL, &L_no_overlap, sf);
1117 }
1118 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1119 const Register from = c_rarg0;
1120 const Register to = c_rarg1;
1121 const Register count = c_rarg2;
1122 const Register end_from = rax;
1123
1124 __ cmpptr(to, from);
1125 __ lea(end_from, Address(from, count, sf, 0));
1126 if (NOLp == NULL) {
1127 ExternalAddress no_overlap(no_overlap_target);
1128 __ jump_cc(Assembler::belowEqual, no_overlap);
1129 __ cmpptr(to, end_from);
1130 __ jump_cc(Assembler::aboveEqual, no_overlap);
1131 } else {
1132 __ jcc(Assembler::belowEqual, (*NOLp));
1133 __ cmpptr(to, end_from);
1134 __ jcc(Assembler::aboveEqual, (*NOLp));
1135 }
1136 }
1137
1138 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1139 //
1140 // Outputs:
1141 // rdi - rcx
1142 // rsi - rdx
1143 // rdx - r8
1144 // rcx - r9
1145 //
1146 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1147 // are non-volatile. r9 and r10 should not be used by the caller.
1148 //
1149 void setup_arg_regs(int nargs = 3) {
1150 const Register saved_rdi = r9;
1151 const Register saved_rsi = r10;
1152 assert(nargs == 3 || nargs == 4, "else fix");
1153 #ifdef _WIN64
1154 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1155 "unexpected argument registers");
1156 if (nargs >= 4)
1157 __ mov(rax, r9); // r9 is also saved_rdi
1158 __ movptr(saved_rdi, rdi);
1159 __ movptr(saved_rsi, rsi);
1160 __ mov(rdi, rcx); // c_rarg0
1161 __ mov(rsi, rdx); // c_rarg1
1162 __ mov(rdx, r8); // c_rarg2
1163 if (nargs >= 4)
1164 __ mov(rcx, rax); // c_rarg3 (via rax)
1165 #else
1166 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1167 "unexpected argument registers");
1168 #endif
1169 }
1170
1171 void restore_arg_regs() {
1172 const Register saved_rdi = r9;
1173 const Register saved_rsi = r10;
1174 #ifdef _WIN64
1175 __ movptr(rdi, saved_rdi);
1176 __ movptr(rsi, saved_rsi);
1177 #endif
1178 }
1179
1180 // Generate code for an array write pre barrier
1181 //
1182 // addr - starting address
1183 // count - element count
1184 // tmp - scratch register
1185 //
1186 // Destroy no registers!
1187 //
1188 void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) {
1189 BarrierSet* bs = Universe::heap()->barrier_set();
1190 switch (bs->kind()) {
1191 case BarrierSet::G1SATBCT:
1192 case BarrierSet::G1SATBCTLogging:
1193 // With G1, don't generate the call if we statically know that the target in uninitialized
1194 if (!dest_uninitialized) {
1195 __ pusha(); // push registers
1196 if (count == c_rarg0) {
1197 if (addr == c_rarg1) {
1198 // exactly backwards!!
1199 __ xchgptr(c_rarg1, c_rarg0);
1200 } else {
1201 __ movptr(c_rarg1, count);
1202 __ movptr(c_rarg0, addr);
1203 }
1204 } else {
1205 __ movptr(c_rarg0, addr);
1206 __ movptr(c_rarg1, count);
1207 }
1208 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1209 __ popa();
1210 }
1211 break;
1212 case BarrierSet::CardTableModRef:
1213 case BarrierSet::CardTableExtension:
1214 case BarrierSet::ModRef:
1215 break;
1216 default:
1217 ShouldNotReachHere();
1218
1219 }
1220 }
1221
1222 //
1223 // Generate code for an array write post barrier
1224 //
1225 // Input:
1226 // start - register containing starting address of destination array
1227 // end - register containing ending address of destination array
1228 // scratch - scratch register
1229 //
1230 // The input registers are overwritten.
1231 // The ending address is inclusive.
1232 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1233 assert_different_registers(start, end, scratch);
1234 BarrierSet* bs = Universe::heap()->barrier_set();
1235 switch (bs->kind()) {
1236 case BarrierSet::G1SATBCT:
1237 case BarrierSet::G1SATBCTLogging:
1238
1239 {
1240 __ pusha(); // push registers (overkill)
1241 // must compute element count unless barrier set interface is changed (other platforms supply count)
1242 assert_different_registers(start, end, scratch);
1243 __ lea(scratch, Address(end, BytesPerHeapOop));
1244 __ subptr(scratch, start); // subtract start to get #bytes
1245 __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count
1246 __ mov(c_rarg0, start);
1247 __ mov(c_rarg1, scratch);
1248 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1249 __ popa();
1250 }
1251 break;
1252 case BarrierSet::CardTableModRef:
1253 case BarrierSet::CardTableExtension:
1254 {
1255 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1256 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1257
1258 Label L_loop;
1259
1260 __ shrptr(start, CardTableModRefBS::card_shift);
1261 __ addptr(end, BytesPerHeapOop);
1262 __ shrptr(end, CardTableModRefBS::card_shift);
1263 __ subptr(end, start); // number of bytes to copy
1264
1265 intptr_t disp = (intptr_t) ct->byte_map_base;
1266 if (__ is_simm32(disp)) {
1267 Address cardtable(noreg, noreg, Address::no_scale, disp);
1268 __ lea(scratch, cardtable);
1269 } else {
1270 ExternalAddress cardtable((address)disp);
1271 __ lea(scratch, cardtable);
1272 }
1273
1274 const Register count = end; // 'end' register contains bytes count now
1275 __ addptr(start, scratch);
1276 __ BIND(L_loop);
1277 __ movb(Address(start, count, Address::times_1), 0);
1278 __ decrement(count);
1279 __ jcc(Assembler::greaterEqual, L_loop);
1280 }
1281 break;
1282 default:
1283 ShouldNotReachHere();
1284
1285 }
1286 }
1287
1288
1289 // Copy big chunks forward
1290 //
1291 // Inputs:
1292 // end_from - source arrays end address
1293 // end_to - destination array end address
1294 // qword_count - 64-bits element count, negative
1295 // to - scratch
1296 // L_copy_32_bytes - entry label
1297 // L_copy_8_bytes - exit label
1298 //
1299 void copy_32_bytes_forward(Register end_from, Register end_to,
1300 Register qword_count, Register to,
1301 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1302 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1303 Label L_loop;
1304 __ align(OptoLoopAlignment);
1305 __ BIND(L_loop);
1306 if(UseUnalignedLoadStores) {
1307 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1308 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1309 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1310 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1311
1312 } else {
1313 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1314 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1315 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1316 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1317 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1318 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1319 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1320 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1321 }
1322 __ BIND(L_copy_32_bytes);
1323 __ addptr(qword_count, 4);
1324 __ jcc(Assembler::lessEqual, L_loop);
1325 __ subptr(qword_count, 4);
1326 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1327 }
1328
1329
1330 // Copy big chunks backward
1331 //
1332 // Inputs:
1333 // from - source arrays address
1334 // dest - destination array address
1335 // qword_count - 64-bits element count
1336 // to - scratch
1337 // L_copy_32_bytes - entry label
1338 // L_copy_8_bytes - exit label
1339 //
1340 void copy_32_bytes_backward(Register from, Register dest,
1341 Register qword_count, Register to,
1342 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1343 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1344 Label L_loop;
1345 __ align(OptoLoopAlignment);
1346 __ BIND(L_loop);
1347 if(UseUnalignedLoadStores) {
1348 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1349 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1350 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1351 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1352
1353 } else {
1354 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1355 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1356 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1357 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1358 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1359 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1360 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1361 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1362 }
1363 __ BIND(L_copy_32_bytes);
1364 __ subptr(qword_count, 4);
1365 __ jcc(Assembler::greaterEqual, L_loop);
1366 __ addptr(qword_count, 4);
1367 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1368 }
1369
1370
1371 // Arguments:
1372 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1373 // ignored
1374 // name - stub name string
1375 //
1376 // Inputs:
1377 // c_rarg0 - source array address
1378 // c_rarg1 - destination array address
1379 // c_rarg2 - element count, treated as ssize_t, can be zero
1380 //
1381 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1382 // we let the hardware handle it. The one to eight bytes within words,
1383 // dwords or qwords that span cache line boundaries will still be loaded
1384 // and stored atomically.
1385 //
1386 // Side Effects:
1387 // disjoint_byte_copy_entry is set to the no-overlap entry point
1388 // used by generate_conjoint_byte_copy().
1389 //
1390 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1391 __ align(CodeEntryAlignment);
1392 StubCodeMark mark(this, "StubRoutines", name);
1393 address start = __ pc();
1394
1395 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1396 Label L_copy_byte, L_exit;
1397 const Register from = rdi; // source array address
1398 const Register to = rsi; // destination array address
1399 const Register count = rdx; // elements count
1400 const Register byte_count = rcx;
1401 const Register qword_count = count;
1402 const Register end_from = from; // source array end address
1403 const Register end_to = to; // destination array end address
1404 // End pointers are inclusive, and if count is not zero they point
1405 // to the last unit copied: end_to[0] := end_from[0]
1406
1407 __ enter(); // required for proper stackwalking of RuntimeStub frame
1408 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1409
1410 if (entry != NULL) {
1411 *entry = __ pc();
1412 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1413 BLOCK_COMMENT("Entry:");
1414 }
1415
1416 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1417 // r9 and r10 may be used to save non-volatile registers
1418
1419 // 'from', 'to' and 'count' are now valid
1420 __ movptr(byte_count, count);
1421 __ shrptr(count, 3); // count => qword_count
1422
1423 // Copy from low to high addresses. Use 'to' as scratch.
1424 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1425 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1426 __ negptr(qword_count); // make the count negative
1427 __ jmp(L_copy_32_bytes);
1428
1429 // Copy trailing qwords
1430 __ BIND(L_copy_8_bytes);
1431 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1432 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1433 __ increment(qword_count);
1434 __ jcc(Assembler::notZero, L_copy_8_bytes);
1435
1436 // Check for and copy trailing dword
1437 __ BIND(L_copy_4_bytes);
1438 __ testl(byte_count, 4);
1439 __ jccb(Assembler::zero, L_copy_2_bytes);
1440 __ movl(rax, Address(end_from, 8));
1441 __ movl(Address(end_to, 8), rax);
1442
1443 __ addptr(end_from, 4);
1444 __ addptr(end_to, 4);
1445
1446 // Check for and copy trailing word
1447 __ BIND(L_copy_2_bytes);
1448 __ testl(byte_count, 2);
1449 __ jccb(Assembler::zero, L_copy_byte);
1450 __ movw(rax, Address(end_from, 8));
1451 __ movw(Address(end_to, 8), rax);
1452
1453 __ addptr(end_from, 2);
1454 __ addptr(end_to, 2);
1455
1456 // Check for and copy trailing byte
1457 __ BIND(L_copy_byte);
1458 __ testl(byte_count, 1);
1459 __ jccb(Assembler::zero, L_exit);
1460 __ movb(rax, Address(end_from, 8));
1461 __ movb(Address(end_to, 8), rax);
1462
1463 __ BIND(L_exit);
1464 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1465 restore_arg_regs();
1466 __ xorptr(rax, rax); // return 0
1467 __ leave(); // required for proper stackwalking of RuntimeStub frame
1468 __ ret(0);
1469
1470 // Copy in 32-bytes chunks
1471 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1472 __ jmp(L_copy_4_bytes);
1473
1474 return start;
1475 }
1476
1477 // Arguments:
1478 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1479 // ignored
1480 // name - stub name string
1481 //
1482 // Inputs:
1483 // c_rarg0 - source array address
1484 // c_rarg1 - destination array address
1485 // c_rarg2 - element count, treated as ssize_t, can be zero
1486 //
1487 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1488 // we let the hardware handle it. The one to eight bytes within words,
1489 // dwords or qwords that span cache line boundaries will still be loaded
1490 // and stored atomically.
1491 //
1492 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1493 address* entry, const char *name) {
1494 __ align(CodeEntryAlignment);
1495 StubCodeMark mark(this, "StubRoutines", name);
1496 address start = __ pc();
1497
1498 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1499 const Register from = rdi; // source array address
1500 const Register to = rsi; // destination array address
1501 const Register count = rdx; // elements count
1502 const Register byte_count = rcx;
1503 const Register qword_count = count;
1504
1505 __ enter(); // required for proper stackwalking of RuntimeStub frame
1506 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1507
1508 if (entry != NULL) {
1509 *entry = __ pc();
1510 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1511 BLOCK_COMMENT("Entry:");
1512 }
1513
1514 array_overlap_test(nooverlap_target, Address::times_1);
1515 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1516 // r9 and r10 may be used to save non-volatile registers
1517
1518 // 'from', 'to' and 'count' are now valid
1519 __ movptr(byte_count, count);
1520 __ shrptr(count, 3); // count => qword_count
1521
1522 // Copy from high to low addresses.
1523
1524 // Check for and copy trailing byte
1525 __ testl(byte_count, 1);
1526 __ jcc(Assembler::zero, L_copy_2_bytes);
1527 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1528 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1529 __ decrement(byte_count); // Adjust for possible trailing word
1530
1531 // Check for and copy trailing word
1532 __ BIND(L_copy_2_bytes);
1533 __ testl(byte_count, 2);
1534 __ jcc(Assembler::zero, L_copy_4_bytes);
1535 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1536 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1537
1538 // Check for and copy trailing dword
1539 __ BIND(L_copy_4_bytes);
1540 __ testl(byte_count, 4);
1541 __ jcc(Assembler::zero, L_copy_32_bytes);
1542 __ movl(rax, Address(from, qword_count, Address::times_8));
1543 __ movl(Address(to, qword_count, Address::times_8), rax);
1544 __ jmp(L_copy_32_bytes);
1545
1546 // Copy trailing qwords
1547 __ BIND(L_copy_8_bytes);
1548 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1549 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1550 __ decrement(qword_count);
1551 __ jcc(Assembler::notZero, L_copy_8_bytes);
1552
1553 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1554 restore_arg_regs();
1555 __ xorptr(rax, rax); // return 0
1556 __ leave(); // required for proper stackwalking of RuntimeStub frame
1557 __ ret(0);
1558
1559 // Copy in 32-bytes chunks
1560 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1561
1562 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1563 restore_arg_regs();
1564 __ xorptr(rax, rax); // return 0
1565 __ leave(); // required for proper stackwalking of RuntimeStub frame
1566 __ ret(0);
1567
1568 return start;
1569 }
1570
1571 // Arguments:
1572 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1573 // ignored
1574 // name - stub name string
1575 //
1576 // Inputs:
1577 // c_rarg0 - source array address
1578 // c_rarg1 - destination array address
1579 // c_rarg2 - element count, treated as ssize_t, can be zero
1580 //
1581 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1582 // let the hardware handle it. The two or four words within dwords
1583 // or qwords that span cache line boundaries will still be loaded
1584 // and stored atomically.
1585 //
1586 // Side Effects:
1587 // disjoint_short_copy_entry is set to the no-overlap entry point
1588 // used by generate_conjoint_short_copy().
1589 //
1590 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1591 __ align(CodeEntryAlignment);
1592 StubCodeMark mark(this, "StubRoutines", name);
1593 address start = __ pc();
1594
1595 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1596 const Register from = rdi; // source array address
1597 const Register to = rsi; // destination array address
1598 const Register count = rdx; // elements count
1599 const Register word_count = rcx;
1600 const Register qword_count = count;
1601 const Register end_from = from; // source array end address
1602 const Register end_to = to; // destination array end address
1603 // End pointers are inclusive, and if count is not zero they point
1604 // to the last unit copied: end_to[0] := end_from[0]
1605
1606 __ enter(); // required for proper stackwalking of RuntimeStub frame
1607 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1608
1609 if (entry != NULL) {
1610 *entry = __ pc();
1611 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1612 BLOCK_COMMENT("Entry:");
1613 }
1614
1615 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1616 // r9 and r10 may be used to save non-volatile registers
1617
1618 // 'from', 'to' and 'count' are now valid
1619 __ movptr(word_count, count);
1620 __ shrptr(count, 2); // count => qword_count
1621
1622 // Copy from low to high addresses. Use 'to' as scratch.
1623 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1624 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1625 __ negptr(qword_count);
1626 __ jmp(L_copy_32_bytes);
1627
1628 // Copy trailing qwords
1629 __ BIND(L_copy_8_bytes);
1630 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1631 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1632 __ increment(qword_count);
1633 __ jcc(Assembler::notZero, L_copy_8_bytes);
1634
1635 // Original 'dest' is trashed, so we can't use it as a
1636 // base register for a possible trailing word copy
1637
1638 // Check for and copy trailing dword
1639 __ BIND(L_copy_4_bytes);
1640 __ testl(word_count, 2);
1641 __ jccb(Assembler::zero, L_copy_2_bytes);
1642 __ movl(rax, Address(end_from, 8));
1643 __ movl(Address(end_to, 8), rax);
1644
1645 __ addptr(end_from, 4);
1646 __ addptr(end_to, 4);
1647
1648 // Check for and copy trailing word
1649 __ BIND(L_copy_2_bytes);
1650 __ testl(word_count, 1);
1651 __ jccb(Assembler::zero, L_exit);
1652 __ movw(rax, Address(end_from, 8));
1653 __ movw(Address(end_to, 8), rax);
1654
1655 __ BIND(L_exit);
1656 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1657 restore_arg_regs();
1658 __ xorptr(rax, rax); // return 0
1659 __ leave(); // required for proper stackwalking of RuntimeStub frame
1660 __ ret(0);
1661
1662 // Copy in 32-bytes chunks
1663 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1664 __ jmp(L_copy_4_bytes);
1665
1666 return start;
1667 }
1668
1669 address generate_fill(BasicType t, bool aligned, const char *name) {
1670 __ align(CodeEntryAlignment);
1671 StubCodeMark mark(this, "StubRoutines", name);
1672 address start = __ pc();
1673
1674 BLOCK_COMMENT("Entry:");
1675
1676 const Register to = c_rarg0; // source array address
1677 const Register value = c_rarg1; // value
1678 const Register count = c_rarg2; // elements count
1679
1680 __ enter(); // required for proper stackwalking of RuntimeStub frame
1681
1682 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1683
1684 __ leave(); // required for proper stackwalking of RuntimeStub frame
1685 __ ret(0);
1686 return start;
1687 }
1688
1689 // Arguments:
1690 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1691 // ignored
1692 // name - stub name string
1693 //
1694 // Inputs:
1695 // c_rarg0 - source array address
1696 // c_rarg1 - destination array address
1697 // c_rarg2 - element count, treated as ssize_t, can be zero
1698 //
1699 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1700 // let the hardware handle it. The two or four words within dwords
1701 // or qwords that span cache line boundaries will still be loaded
1702 // and stored atomically.
1703 //
1704 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1705 address *entry, const char *name) {
1706 __ align(CodeEntryAlignment);
1707 StubCodeMark mark(this, "StubRoutines", name);
1708 address start = __ pc();
1709
1710 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1711 const Register from = rdi; // source array address
1712 const Register to = rsi; // destination array address
1713 const Register count = rdx; // elements count
1714 const Register word_count = rcx;
1715 const Register qword_count = count;
1716
1717 __ enter(); // required for proper stackwalking of RuntimeStub frame
1718 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1719
1720 if (entry != NULL) {
1721 *entry = __ pc();
1722 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1723 BLOCK_COMMENT("Entry:");
1724 }
1725
1726 array_overlap_test(nooverlap_target, Address::times_2);
1727 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1728 // r9 and r10 may be used to save non-volatile registers
1729
1730 // 'from', 'to' and 'count' are now valid
1731 __ movptr(word_count, count);
1732 __ shrptr(count, 2); // count => qword_count
1733
1734 // Copy from high to low addresses. Use 'to' as scratch.
1735
1736 // Check for and copy trailing word
1737 __ testl(word_count, 1);
1738 __ jccb(Assembler::zero, L_copy_4_bytes);
1739 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1740 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1741
1742 // Check for and copy trailing dword
1743 __ BIND(L_copy_4_bytes);
1744 __ testl(word_count, 2);
1745 __ jcc(Assembler::zero, L_copy_32_bytes);
1746 __ movl(rax, Address(from, qword_count, Address::times_8));
1747 __ movl(Address(to, qword_count, Address::times_8), rax);
1748 __ jmp(L_copy_32_bytes);
1749
1750 // Copy trailing qwords
1751 __ BIND(L_copy_8_bytes);
1752 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1753 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1754 __ decrement(qword_count);
1755 __ jcc(Assembler::notZero, L_copy_8_bytes);
1756
1757 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1758 restore_arg_regs();
1759 __ xorptr(rax, rax); // return 0
1760 __ leave(); // required for proper stackwalking of RuntimeStub frame
1761 __ ret(0);
1762
1763 // Copy in 32-bytes chunks
1764 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1765
1766 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1767 restore_arg_regs();
1768 __ xorptr(rax, rax); // return 0
1769 __ leave(); // required for proper stackwalking of RuntimeStub frame
1770 __ ret(0);
1771
1772 return start;
1773 }
1774
1775 // Arguments:
1776 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1777 // ignored
1778 // is_oop - true => oop array, so generate store check code
1779 // name - stub name string
1780 //
1781 // Inputs:
1782 // c_rarg0 - source array address
1783 // c_rarg1 - destination array address
1784 // c_rarg2 - element count, treated as ssize_t, can be zero
1785 //
1786 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1787 // the hardware handle it. The two dwords within qwords that span
1788 // cache line boundaries will still be loaded and stored atomicly.
1789 //
1790 // Side Effects:
1791 // disjoint_int_copy_entry is set to the no-overlap entry point
1792 // used by generate_conjoint_int_oop_copy().
1793 //
1794 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1795 const char *name, bool dest_uninitialized = false) {
1796 __ align(CodeEntryAlignment);
1797 StubCodeMark mark(this, "StubRoutines", name);
1798 address start = __ pc();
1799
1800 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1801 const Register from = rdi; // source array address
1802 const Register to = rsi; // destination array address
1803 const Register count = rdx; // elements count
1804 const Register dword_count = rcx;
1805 const Register qword_count = count;
1806 const Register end_from = from; // source array end address
1807 const Register end_to = to; // destination array end address
1808 const Register saved_to = r11; // saved destination array address
1809 // End pointers are inclusive, and if count is not zero they point
1810 // to the last unit copied: end_to[0] := end_from[0]
1811
1812 __ enter(); // required for proper stackwalking of RuntimeStub frame
1813 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1814
1815 if (entry != NULL) {
1816 *entry = __ pc();
1817 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1818 BLOCK_COMMENT("Entry:");
1819 }
1820
1821 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1822 // r9 and r10 may be used to save non-volatile registers
1823 if (is_oop) {
1824 __ movq(saved_to, to);
1825 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1826 }
1827
1828 // 'from', 'to' and 'count' are now valid
1829 __ movptr(dword_count, count);
1830 __ shrptr(count, 1); // count => qword_count
1831
1832 // Copy from low to high addresses. Use 'to' as scratch.
1833 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1834 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1835 __ negptr(qword_count);
1836 __ jmp(L_copy_32_bytes);
1837
1838 // Copy trailing qwords
1839 __ BIND(L_copy_8_bytes);
1840 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1841 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1842 __ increment(qword_count);
1843 __ jcc(Assembler::notZero, L_copy_8_bytes);
1844
1845 // Check for and copy trailing dword
1846 __ BIND(L_copy_4_bytes);
1847 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1848 __ jccb(Assembler::zero, L_exit);
1849 __ movl(rax, Address(end_from, 8));
1850 __ movl(Address(end_to, 8), rax);
1851
1852 __ BIND(L_exit);
1853 if (is_oop) {
1854 __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1855 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1856 }
1857 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1858 restore_arg_regs();
1859 __ xorptr(rax, rax); // return 0
1860 __ leave(); // required for proper stackwalking of RuntimeStub frame
1861 __ ret(0);
1862
1863 // Copy 32-bytes chunks
1864 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1865 __ jmp(L_copy_4_bytes);
1866
1867 return start;
1868 }
1869
1870 // Arguments:
1871 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1872 // ignored
1873 // is_oop - true => oop array, so generate store check code
1874 // name - stub name string
1875 //
1876 // Inputs:
1877 // c_rarg0 - source array address
1878 // c_rarg1 - destination array address
1879 // c_rarg2 - element count, treated as ssize_t, can be zero
1880 //
1881 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1882 // the hardware handle it. The two dwords within qwords that span
1883 // cache line boundaries will still be loaded and stored atomicly.
1884 //
1885 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1886 address *entry, const char *name,
1887 bool dest_uninitialized = false) {
1888 __ align(CodeEntryAlignment);
1889 StubCodeMark mark(this, "StubRoutines", name);
1890 address start = __ pc();
1891
1892 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1893 const Register from = rdi; // source array address
1894 const Register to = rsi; // destination array address
1895 const Register count = rdx; // elements count
1896 const Register dword_count = rcx;
1897 const Register qword_count = count;
1898
1899 __ enter(); // required for proper stackwalking of RuntimeStub frame
1900 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1901
1902 if (entry != NULL) {
1903 *entry = __ pc();
1904 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1905 BLOCK_COMMENT("Entry:");
1906 }
1907
1908 array_overlap_test(nooverlap_target, Address::times_4);
1909 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1910 // r9 and r10 may be used to save non-volatile registers
1911
1912 if (is_oop) {
1913 // no registers are destroyed by this call
1914 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1915 }
1916
1917 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1918 // 'from', 'to' and 'count' are now valid
1919 __ movptr(dword_count, count);
1920 __ shrptr(count, 1); // count => qword_count
1921
1922 // Copy from high to low addresses. Use 'to' as scratch.
1923
1924 // Check for and copy trailing dword
1925 __ testl(dword_count, 1);
1926 __ jcc(Assembler::zero, L_copy_32_bytes);
1927 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1928 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1929 __ jmp(L_copy_32_bytes);
1930
1931 // Copy trailing qwords
1932 __ BIND(L_copy_8_bytes);
1933 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1934 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1935 __ decrement(qword_count);
1936 __ jcc(Assembler::notZero, L_copy_8_bytes);
1937
1938 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1939 if (is_oop) {
1940 __ jmp(L_exit);
1941 }
1942 restore_arg_regs();
1943 __ xorptr(rax, rax); // return 0
1944 __ leave(); // required for proper stackwalking of RuntimeStub frame
1945 __ ret(0);
1946
1947 // Copy in 32-bytes chunks
1948 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1949
1950 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1951 __ bind(L_exit);
1952 if (is_oop) {
1953 Register end_to = rdx;
1954 __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1955 gen_write_ref_array_post_barrier(to, end_to, rax);
1956 }
1957 restore_arg_regs();
1958 __ xorptr(rax, rax); // return 0
1959 __ leave(); // required for proper stackwalking of RuntimeStub frame
1960 __ ret(0);
1961
1962 return start;
1963 }
1964
1965 // Arguments:
1966 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1967 // ignored
1968 // is_oop - true => oop array, so generate store check code
1969 // name - stub name string
1970 //
1971 // Inputs:
1972 // c_rarg0 - source array address
1973 // c_rarg1 - destination array address
1974 // c_rarg2 - element count, treated as ssize_t, can be zero
1975 //
1976 // Side Effects:
1977 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1978 // no-overlap entry point used by generate_conjoint_long_oop_copy().
1979 //
1980 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
1981 const char *name, bool dest_uninitialized = false) {
1982 __ align(CodeEntryAlignment);
1983 StubCodeMark mark(this, "StubRoutines", name);
1984 address start = __ pc();
1985
1986 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1987 const Register from = rdi; // source array address
1988 const Register to = rsi; // destination array address
1989 const Register qword_count = rdx; // elements count
1990 const Register end_from = from; // source array end address
1991 const Register end_to = rcx; // destination array end address
1992 const Register saved_to = to;
1993 // End pointers are inclusive, and if count is not zero they point
1994 // to the last unit copied: end_to[0] := end_from[0]
1995
1996 __ enter(); // required for proper stackwalking of RuntimeStub frame
1997 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1998 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1999
2000 if (entry != NULL) {
2001 *entry = __ pc();
2002 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2003 BLOCK_COMMENT("Entry:");
2004 }
2005
2006 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2007 // r9 and r10 may be used to save non-volatile registers
2008 // 'from', 'to' and 'qword_count' are now valid
2009 if (is_oop) {
2010 // no registers are destroyed by this call
2011 gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized);
2012 }
2013
2014 // Copy from low to high addresses. Use 'to' as scratch.
2015 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2016 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2017 __ negptr(qword_count);
2018 __ jmp(L_copy_32_bytes);
2019
2020 // Copy trailing qwords
2021 __ BIND(L_copy_8_bytes);
2022 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2023 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2024 __ increment(qword_count);
2025 __ jcc(Assembler::notZero, L_copy_8_bytes);
2026
2027 if (is_oop) {
2028 __ jmp(L_exit);
2029 } else {
2030 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2031 restore_arg_regs();
2032 __ xorptr(rax, rax); // return 0
2033 __ leave(); // required for proper stackwalking of RuntimeStub frame
2034 __ ret(0);
2035 }
2036
2037 // Copy 64-byte chunks
2038 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2039
2040 if (is_oop) {
2041 __ BIND(L_exit);
2042 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
2043 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2044 } else {
2045 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2046 }
2047 restore_arg_regs();
2048 __ xorptr(rax, rax); // return 0
2049 __ leave(); // required for proper stackwalking of RuntimeStub frame
2050 __ ret(0);
2051
2052 return start;
2053 }
2054
2055 // Arguments:
2056 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2057 // ignored
2058 // is_oop - true => oop array, so generate store check code
2059 // name - stub name string
2060 //
2061 // Inputs:
2062 // c_rarg0 - source array address
2063 // c_rarg1 - destination array address
2064 // c_rarg2 - element count, treated as ssize_t, can be zero
2065 //
2066 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2067 address nooverlap_target, address *entry,
2068 const char *name, bool dest_uninitialized = false) {
2069 __ align(CodeEntryAlignment);
2070 StubCodeMark mark(this, "StubRoutines", name);
2071 address start = __ pc();
2072
2073 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2074 const Register from = rdi; // source array address
2075 const Register to = rsi; // destination array address
2076 const Register qword_count = rdx; // elements count
2077 const Register saved_count = rcx;
2078
2079 __ enter(); // required for proper stackwalking of RuntimeStub frame
2080 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2081
2082 if (entry != NULL) {
2083 *entry = __ pc();
2084 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2085 BLOCK_COMMENT("Entry:");
2086 }
2087
2088 array_overlap_test(nooverlap_target, Address::times_8);
2089 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2090 // r9 and r10 may be used to save non-volatile registers
2091 // 'from', 'to' and 'qword_count' are now valid
2092 if (is_oop) {
2093 // Save to and count for store barrier
2094 __ movptr(saved_count, qword_count);
2095 // No registers are destroyed by this call
2096 gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized);
2097 }
2098
2099 __ jmp(L_copy_32_bytes);
2100
2101 // Copy trailing qwords
2102 __ BIND(L_copy_8_bytes);
2103 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2104 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2105 __ decrement(qword_count);
2106 __ jcc(Assembler::notZero, L_copy_8_bytes);
2107
2108 if (is_oop) {
2109 __ jmp(L_exit);
2110 } else {
2111 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2112 restore_arg_regs();
2113 __ xorptr(rax, rax); // return 0
2114 __ leave(); // required for proper stackwalking of RuntimeStub frame
2115 __ ret(0);
2116 }
2117
2118 // Copy in 32-bytes chunks
2119 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2120
2121 if (is_oop) {
2122 __ BIND(L_exit);
2123 __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2124 gen_write_ref_array_post_barrier(to, rcx, rax);
2125 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2126 } else {
2127 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2128 }
2129 restore_arg_regs();
2130 __ xorptr(rax, rax); // return 0
2131 __ leave(); // required for proper stackwalking of RuntimeStub frame
2132 __ ret(0);
2133
2134 return start;
2135 }
2136
2137
2138 // Helper for generating a dynamic type check.
2139 // Smashes no registers.
2140 void generate_type_check(Register sub_klass,
2141 Register super_check_offset,
2142 Register super_klass,
2143 Label& L_success) {
2144 assert_different_registers(sub_klass, super_check_offset, super_klass);
2145
2146 BLOCK_COMMENT("type_check:");
2147
2148 Label L_miss;
2149
2150 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2151 super_check_offset);
2152 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2153
2154 // Fall through on failure!
2155 __ BIND(L_miss);
2156 }
2157
2158 //
2159 // Generate checkcasting array copy stub
2160 //
2161 // Input:
2162 // c_rarg0 - source array address
2163 // c_rarg1 - destination array address
2164 // c_rarg2 - element count, treated as ssize_t, can be zero
2165 // c_rarg3 - size_t ckoff (super_check_offset)
2166 // not Win64
2167 // c_rarg4 - oop ckval (super_klass)
2168 // Win64
2169 // rsp+40 - oop ckval (super_klass)
2170 //
2171 // Output:
2172 // rax == 0 - success
2173 // rax == -1^K - failure, where K is partial transfer count
2174 //
2175 address generate_checkcast_copy(const char *name, address *entry,
2176 bool dest_uninitialized = false) {
2177
2178 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2179
2180 // Input registers (after setup_arg_regs)
2181 const Register from = rdi; // source array address
2182 const Register to = rsi; // destination array address
2183 const Register length = rdx; // elements count
2184 const Register ckoff = rcx; // super_check_offset
2185 const Register ckval = r8; // super_klass
2186
2187 // Registers used as temps (r13, r14 are save-on-entry)
2188 const Register end_from = from; // source array end address
2189 const Register end_to = r13; // destination array end address
2190 const Register count = rdx; // -(count_remaining)
2191 const Register r14_length = r14; // saved copy of length
2192 // End pointers are inclusive, and if length is not zero they point
2193 // to the last unit copied: end_to[0] := end_from[0]
2194
2195 const Register rax_oop = rax; // actual oop copied
2196 const Register r11_klass = r11; // oop._klass
2197
2198 //---------------------------------------------------------------
2199 // Assembler stub will be used for this call to arraycopy
2200 // if the two arrays are subtypes of Object[] but the
2201 // destination array type is not equal to or a supertype
2202 // of the source type. Each element must be separately
2203 // checked.
2204
2205 __ align(CodeEntryAlignment);
2206 StubCodeMark mark(this, "StubRoutines", name);
2207 address start = __ pc();
2208
2209 __ enter(); // required for proper stackwalking of RuntimeStub frame
2210
2211 #ifdef ASSERT
2212 // caller guarantees that the arrays really are different
2213 // otherwise, we would have to make conjoint checks
2214 { Label L;
2215 array_overlap_test(L, TIMES_OOP);
2216 __ stop("checkcast_copy within a single array");
2217 __ bind(L);
2218 }
2219 #endif //ASSERT
2220
2221 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2222 // ckoff => rcx, ckval => r8
2223 // r9 and r10 may be used to save non-volatile registers
2224 #ifdef _WIN64
2225 // last argument (#4) is on stack on Win64
2226 __ movptr(ckval, Address(rsp, 6 * wordSize));
2227 #endif
2228
2229 // Caller of this entry point must set up the argument registers.
2230 if (entry != NULL) {
2231 *entry = __ pc();
2232 BLOCK_COMMENT("Entry:");
2233 }
2234
2235 // allocate spill slots for r13, r14
2236 enum {
2237 saved_r13_offset,
2238 saved_r14_offset,
2239 saved_rbp_offset
2240 };
2241 __ subptr(rsp, saved_rbp_offset * wordSize);
2242 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2243 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2244
2245 // check that int operands are properly extended to size_t
2246 assert_clean_int(length, rax);
2247 assert_clean_int(ckoff, rax);
2248
2249 #ifdef ASSERT
2250 BLOCK_COMMENT("assert consistent ckoff/ckval");
2251 // The ckoff and ckval must be mutually consistent,
2252 // even though caller generates both.
2253 { Label L;
2254 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2255 Klass::super_check_offset_offset_in_bytes());
2256 __ cmpl(ckoff, Address(ckval, sco_offset));
2257 __ jcc(Assembler::equal, L);
2258 __ stop("super_check_offset inconsistent");
2259 __ bind(L);
2260 }
2261 #endif //ASSERT
2262
2263 // Loop-invariant addresses. They are exclusive end pointers.
2264 Address end_from_addr(from, length, TIMES_OOP, 0);
2265 Address end_to_addr(to, length, TIMES_OOP, 0);
2266 // Loop-variant addresses. They assume post-incremented count < 0.
2267 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2268 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2269
2270 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2271
2272 // Copy from low to high addresses, indexed from the end of each array.
2273 __ lea(end_from, end_from_addr);
2274 __ lea(end_to, end_to_addr);
2275 __ movptr(r14_length, length); // save a copy of the length
2276 assert(length == count, ""); // else fix next line:
2277 __ negptr(count); // negate and test the length
2278 __ jcc(Assembler::notZero, L_load_element);
2279
2280 // Empty array: Nothing to do.
2281 __ xorptr(rax, rax); // return 0 on (trivial) success
2282 __ jmp(L_done);
2283
2284 // ======== begin loop ========
2285 // (Loop is rotated; its entry is L_load_element.)
2286 // Loop control:
2287 // for (count = -count; count != 0; count++)
2288 // Base pointers src, dst are biased by 8*(count-1),to last element.
2289 __ align(OptoLoopAlignment);
2290
2291 __ BIND(L_store_element);
2292 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2293 __ increment(count); // increment the count toward zero
2294 __ jcc(Assembler::zero, L_do_card_marks);
2295
2296 // ======== loop entry is here ========
2297 __ BIND(L_load_element);
2298 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2299 __ testptr(rax_oop, rax_oop);
2300 __ jcc(Assembler::zero, L_store_element);
2301
2302 __ load_klass(r11_klass, rax_oop);// query the object klass
2303 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2304 // ======== end loop ========
2305
2306 // It was a real error; we must depend on the caller to finish the job.
2307 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2308 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2309 // and report their number to the caller.
2310 assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2311 __ lea(end_to, to_element_addr);
2312 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2313 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2314 __ movptr(rax, r14_length); // original oops
2315 __ addptr(rax, count); // K = (original - remaining) oops
2316 __ notptr(rax); // report (-1^K) to caller
2317 __ jmp(L_done);
2318
2319 // Come here on success only.
2320 __ BIND(L_do_card_marks);
2321 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2322 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2323 __ xorptr(rax, rax); // return 0 on success
2324
2325 // Common exit point (success or failure).
2326 __ BIND(L_done);
2327 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2328 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2329 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
2330 restore_arg_regs();
2331 __ leave(); // required for proper stackwalking of RuntimeStub frame
2332 __ ret(0);
2333
2334 return start;
2335 }
2336
2337 //
2338 // Generate 'unsafe' array copy stub
2339 // Though just as safe as the other stubs, it takes an unscaled
2340 // size_t argument instead of an element count.
2341 //
2342 // Input:
2343 // c_rarg0 - source array address
2344 // c_rarg1 - destination array address
2345 // c_rarg2 - byte count, treated as ssize_t, can be zero
2346 //
2347 // Examines the alignment of the operands and dispatches
2348 // to a long, int, short, or byte copy loop.
2349 //
2350 address generate_unsafe_copy(const char *name,
2351 address byte_copy_entry, address short_copy_entry,
2352 address int_copy_entry, address long_copy_entry) {
2353
2354 Label L_long_aligned, L_int_aligned, L_short_aligned;
2355
2356 // Input registers (before setup_arg_regs)
2357 const Register from = c_rarg0; // source array address
2358 const Register to = c_rarg1; // destination array address
2359 const Register size = c_rarg2; // byte count (size_t)
2360
2361 // Register used as a temp
2362 const Register bits = rax; // test copy of low bits
2363
2364 __ align(CodeEntryAlignment);
2365 StubCodeMark mark(this, "StubRoutines", name);
2366 address start = __ pc();
2367
2368 __ enter(); // required for proper stackwalking of RuntimeStub frame
2369
2370 // bump this on entry, not on exit:
2371 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2372
2373 __ mov(bits, from);
2374 __ orptr(bits, to);
2375 __ orptr(bits, size);
2376
2377 __ testb(bits, BytesPerLong-1);
2378 __ jccb(Assembler::zero, L_long_aligned);
2379
2380 __ testb(bits, BytesPerInt-1);
2381 __ jccb(Assembler::zero, L_int_aligned);
2382
2383 __ testb(bits, BytesPerShort-1);
2384 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2385
2386 __ BIND(L_short_aligned);
2387 __ shrptr(size, LogBytesPerShort); // size => short_count
2388 __ jump(RuntimeAddress(short_copy_entry));
2389
2390 __ BIND(L_int_aligned);
2391 __ shrptr(size, LogBytesPerInt); // size => int_count
2392 __ jump(RuntimeAddress(int_copy_entry));
2393
2394 __ BIND(L_long_aligned);
2395 __ shrptr(size, LogBytesPerLong); // size => qword_count
2396 __ jump(RuntimeAddress(long_copy_entry));
2397
2398 return start;
2399 }
2400
2401 // Perform range checks on the proposed arraycopy.
2402 // Kills temp, but nothing else.
2403 // Also, clean the sign bits of src_pos and dst_pos.
2404 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2405 Register src_pos, // source position (c_rarg1)
2406 Register dst, // destination array oo (c_rarg2)
2407 Register dst_pos, // destination position (c_rarg3)
2408 Register length,
2409 Register temp,
2410 Label& L_failed) {
2411 BLOCK_COMMENT("arraycopy_range_checks:");
2412
2413 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2414 __ movl(temp, length);
2415 __ addl(temp, src_pos); // src_pos + length
2416 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2417 __ jcc(Assembler::above, L_failed);
2418
2419 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2420 __ movl(temp, length);
2421 __ addl(temp, dst_pos); // dst_pos + length
2422 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2423 __ jcc(Assembler::above, L_failed);
2424
2425 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2426 // Move with sign extension can be used since they are positive.
2427 __ movslq(src_pos, src_pos);
2428 __ movslq(dst_pos, dst_pos);
2429
2430 BLOCK_COMMENT("arraycopy_range_checks done");
2431 }
2432
2433 //
2434 // Generate generic array copy stubs
2435 //
2436 // Input:
2437 // c_rarg0 - src oop
2438 // c_rarg1 - src_pos (32-bits)
2439 // c_rarg2 - dst oop
2440 // c_rarg3 - dst_pos (32-bits)
2441 // not Win64
2442 // c_rarg4 - element count (32-bits)
2443 // Win64
2444 // rsp+40 - element count (32-bits)
2445 //
2446 // Output:
2447 // rax == 0 - success
2448 // rax == -1^K - failure, where K is partial transfer count
2449 //
2450 address generate_generic_copy(const char *name,
2451 address byte_copy_entry, address short_copy_entry,
2452 address int_copy_entry, address long_copy_entry,
2453 address oop_copy_entry, address checkcast_copy_entry) {
2454
2455 Label L_failed, L_failed_0, L_objArray;
2456 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2457
2458 // Input registers
2459 const Register src = c_rarg0; // source array oop
2460 const Register src_pos = c_rarg1; // source position
2461 const Register dst = c_rarg2; // destination array oop
2462 const Register dst_pos = c_rarg3; // destination position
2463 #ifndef _WIN64
2464 const Register length = c_rarg4;
2465 #else
2466 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2467 #endif
2468
2469 { int modulus = CodeEntryAlignment;
2470 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2471 int advance = target - (__ offset() % modulus);
2472 if (advance < 0) advance += modulus;
2473 if (advance > 0) __ nop(advance);
2474 }
2475 StubCodeMark mark(this, "StubRoutines", name);
2476
2477 // Short-hop target to L_failed. Makes for denser prologue code.
2478 __ BIND(L_failed_0);
2479 __ jmp(L_failed);
2480 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2481
2482 __ align(CodeEntryAlignment);
2483 address start = __ pc();
2484
2485 __ enter(); // required for proper stackwalking of RuntimeStub frame
2486
2487 // bump this on entry, not on exit:
2488 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2489
2490 //-----------------------------------------------------------------------
2491 // Assembler stub will be used for this call to arraycopy
2492 // if the following conditions are met:
2493 //
2494 // (1) src and dst must not be null.
2495 // (2) src_pos must not be negative.
2496 // (3) dst_pos must not be negative.
2497 // (4) length must not be negative.
2498 // (5) src klass and dst klass should be the same and not NULL.
2499 // (6) src and dst should be arrays.
2500 // (7) src_pos + length must not exceed length of src.
2501 // (8) dst_pos + length must not exceed length of dst.
2502 //
2503
2504 // if (src == NULL) return -1;
2505 __ testptr(src, src); // src oop
2506 size_t j1off = __ offset();
2507 __ jccb(Assembler::zero, L_failed_0);
2508
2509 // if (src_pos < 0) return -1;
2510 __ testl(src_pos, src_pos); // src_pos (32-bits)
2511 __ jccb(Assembler::negative, L_failed_0);
2512
2513 // if (dst == NULL) return -1;
2514 __ testptr(dst, dst); // dst oop
2515 __ jccb(Assembler::zero, L_failed_0);
2516
2517 // if (dst_pos < 0) return -1;
2518 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2519 size_t j4off = __ offset();
2520 __ jccb(Assembler::negative, L_failed_0);
2521
2522 // The first four tests are very dense code,
2523 // but not quite dense enough to put four
2524 // jumps in a 16-byte instruction fetch buffer.
2525 // That's good, because some branch predicters
2526 // do not like jumps so close together.
2527 // Make sure of this.
2528 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2529
2530 // registers used as temp
2531 const Register r11_length = r11; // elements count to copy
2532 const Register r10_src_klass = r10; // array klass
2533
2534 // if (length < 0) return -1;
2535 __ movl(r11_length, length); // length (elements count, 32-bits value)
2536 __ testl(r11_length, r11_length);
2537 __ jccb(Assembler::negative, L_failed_0);
2538
2539 __ load_klass(r10_src_klass, src);
2540 #ifdef ASSERT
2541 // assert(src->klass() != NULL);
2542 {
2543 BLOCK_COMMENT("assert klasses not null {");
2544 Label L1, L2;
2545 __ testptr(r10_src_klass, r10_src_klass);
2546 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2547 __ bind(L1);
2548 __ stop("broken null klass");
2549 __ bind(L2);
2550 __ load_klass(rax, dst);
2551 __ cmpq(rax, 0);
2552 __ jcc(Assembler::equal, L1); // this would be broken also
2553 BLOCK_COMMENT("} assert klasses not null done");
2554 }
2555 #endif
2556
2557 // Load layout helper (32-bits)
2558 //
2559 // |array_tag| | header_size | element_type | |log2_element_size|
2560 // 32 30 24 16 8 2 0
2561 //
2562 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2563 //
2564
2565 const int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2566 Klass::layout_helper_offset_in_bytes();
2567
2568 // Handle objArrays completely differently...
2569 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2570 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2571 __ jcc(Assembler::equal, L_objArray);
2572
2573 // if (src->klass() != dst->klass()) return -1;
2574 __ load_klass(rax, dst);
2575 __ cmpq(r10_src_klass, rax);
2576 __ jcc(Assembler::notEqual, L_failed);
2577
2578 const Register rax_lh = rax; // layout helper
2579 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2580
2581 // if (!src->is_Array()) return -1;
2582 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2583 __ jcc(Assembler::greaterEqual, L_failed);
2584
2585 // At this point, it is known to be a typeArray (array_tag 0x3).
2586 #ifdef ASSERT
2587 {
2588 BLOCK_COMMENT("assert primitive array {");
2589 Label L;
2590 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2591 __ jcc(Assembler::greaterEqual, L);
2592 __ stop("must be a primitive array");
2593 __ bind(L);
2594 BLOCK_COMMENT("} assert primitive array done");
2595 }
2596 #endif
2597
2598 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2599 r10, L_failed);
2600
2601 // typeArrayKlass
2602 //
2603 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2604 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2605 //
2606
2607 const Register r10_offset = r10; // array offset
2608 const Register rax_elsize = rax_lh; // element size
2609
2610 __ movl(r10_offset, rax_lh);
2611 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2612 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2613 __ addptr(src, r10_offset); // src array offset
2614 __ addptr(dst, r10_offset); // dst array offset
2615 BLOCK_COMMENT("choose copy loop based on element size");
2616 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2617
2618 // next registers should be set before the jump to corresponding stub
2619 const Register from = c_rarg0; // source array address
2620 const Register to = c_rarg1; // destination array address
2621 const Register count = c_rarg2; // elements count
2622
2623 // 'from', 'to', 'count' registers should be set in such order
2624 // since they are the same as 'src', 'src_pos', 'dst'.
2625
2626 __ BIND(L_copy_bytes);
2627 __ cmpl(rax_elsize, 0);
2628 __ jccb(Assembler::notEqual, L_copy_shorts);
2629 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2630 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2631 __ movl2ptr(count, r11_length); // length
2632 __ jump(RuntimeAddress(byte_copy_entry));
2633
2634 __ BIND(L_copy_shorts);
2635 __ cmpl(rax_elsize, LogBytesPerShort);
2636 __ jccb(Assembler::notEqual, L_copy_ints);
2637 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2638 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2639 __ movl2ptr(count, r11_length); // length
2640 __ jump(RuntimeAddress(short_copy_entry));
2641
2642 __ BIND(L_copy_ints);
2643 __ cmpl(rax_elsize, LogBytesPerInt);
2644 __ jccb(Assembler::notEqual, L_copy_longs);
2645 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2646 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2647 __ movl2ptr(count, r11_length); // length
2648 __ jump(RuntimeAddress(int_copy_entry));
2649
2650 __ BIND(L_copy_longs);
2651 #ifdef ASSERT
2652 {
2653 BLOCK_COMMENT("assert long copy {");
2654 Label L;
2655 __ cmpl(rax_elsize, LogBytesPerLong);
2656 __ jcc(Assembler::equal, L);
2657 __ stop("must be long copy, but elsize is wrong");
2658 __ bind(L);
2659 BLOCK_COMMENT("} assert long copy done");
2660 }
2661 #endif
2662 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2663 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2664 __ movl2ptr(count, r11_length); // length
2665 __ jump(RuntimeAddress(long_copy_entry));
2666
2667 // objArrayKlass
2668 __ BIND(L_objArray);
2669 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2670
2671 Label L_plain_copy, L_checkcast_copy;
2672 // test array classes for subtyping
2673 __ load_klass(rax, dst);
2674 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2675 __ jcc(Assembler::notEqual, L_checkcast_copy);
2676
2677 // Identically typed arrays can be copied without element-wise checks.
2678 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2679 r10, L_failed);
2680
2681 __ lea(from, Address(src, src_pos, TIMES_OOP,
2682 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2683 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2684 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2685 __ movl2ptr(count, r11_length); // length
2686 __ BIND(L_plain_copy);
2687 __ jump(RuntimeAddress(oop_copy_entry));
2688
2689 __ BIND(L_checkcast_copy);
2690 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2691 {
2692 // Before looking at dst.length, make sure dst is also an objArray.
2693 __ cmpl(Address(rax, lh_offset), objArray_lh);
2694 __ jcc(Assembler::notEqual, L_failed);
2695
2696 // It is safe to examine both src.length and dst.length.
2697 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2698 rax, L_failed);
2699
2700 const Register r11_dst_klass = r11;
2701 __ load_klass(r11_dst_klass, dst); // reload
2702
2703 // Marshal the base address arguments now, freeing registers.
2704 __ lea(from, Address(src, src_pos, TIMES_OOP,
2705 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2706 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2707 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2708 __ movl(count, length); // length (reloaded)
2709 Register sco_temp = c_rarg3; // this register is free now
2710 assert_different_registers(from, to, count, sco_temp,
2711 r11_dst_klass, r10_src_klass);
2712 assert_clean_int(count, sco_temp);
2713
2714 // Generate the type check.
2715 const int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2716 Klass::super_check_offset_offset_in_bytes());
2717 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2718 assert_clean_int(sco_temp, rax);
2719 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2720
2721 // Fetch destination element klass from the objArrayKlass header.
2722 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2723 objArrayKlass::element_klass_offset_in_bytes());
2724 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2725 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2726 assert_clean_int(sco_temp, rax);
2727
2728 // the checkcast_copy loop needs two extra arguments:
2729 assert(c_rarg3 == sco_temp, "#3 already in place");
2730 // Set up arguments for checkcast_copy_entry.
2731 setup_arg_regs(4);
2732 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2733 __ jump(RuntimeAddress(checkcast_copy_entry));
2734 }
2735
2736 __ BIND(L_failed);
2737 __ xorptr(rax, rax);
2738 __ notptr(rax); // return -1
2739 __ leave(); // required for proper stackwalking of RuntimeStub frame
2740 __ ret(0);
2741
2742 return start;
2743 }
2744
2745 void generate_arraycopy_stubs() {
2746 address entry;
2747 address entry_jbyte_arraycopy;
2748 address entry_jshort_arraycopy;
2749 address entry_jint_arraycopy;
2750 address entry_oop_arraycopy;
2751 address entry_jlong_arraycopy;
2752 address entry_checkcast_arraycopy;
2753
2754 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2755 "jbyte_disjoint_arraycopy");
2756 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2757 "jbyte_arraycopy");
2758
2759 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2760 "jshort_disjoint_arraycopy");
2761 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2762 "jshort_arraycopy");
2763
2764 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2765 "jint_disjoint_arraycopy");
2766 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2767 &entry_jint_arraycopy, "jint_arraycopy");
2768
2769 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2770 "jlong_disjoint_arraycopy");
2771 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2772 &entry_jlong_arraycopy, "jlong_arraycopy");
2773
2774
2775 if (UseCompressedOops) {
2776 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2777 "oop_disjoint_arraycopy");
2778 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2779 &entry_oop_arraycopy, "oop_arraycopy");
2780 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2781 "oop_disjoint_arraycopy_uninit",
2782 /*dest_uninitialized*/true);
2783 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2784 NULL, "oop_arraycopy_uninit",
2785 /*dest_uninitialized*/true);
2786 } else {
2787 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2788 "oop_disjoint_arraycopy");
2789 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2790 &entry_oop_arraycopy, "oop_arraycopy");
2791 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2792 "oop_disjoint_arraycopy_uninit",
2793 /*dest_uninitialized*/true);
2794 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2795 NULL, "oop_arraycopy_uninit",
2796 /*dest_uninitialized*/true);
2797 }
2798
2799 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2800 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2801 /*dest_uninitialized*/true);
2802
2803 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2804 entry_jbyte_arraycopy,
2805 entry_jshort_arraycopy,
2806 entry_jint_arraycopy,
2807 entry_jlong_arraycopy);
2808 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2809 entry_jbyte_arraycopy,
2810 entry_jshort_arraycopy,
2811 entry_jint_arraycopy,
2812 entry_oop_arraycopy,
2813 entry_jlong_arraycopy,
2814 entry_checkcast_arraycopy);
2815
2816 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2817 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2818 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2819 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2820 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2821 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2822
2823 // We don't generate specialized code for HeapWord-aligned source
2824 // arrays, so just use the code we've already generated
2825 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2826 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2827
2828 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2829 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2830
2831 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2832 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2833
2834 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2835 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2836
2837 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2838 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2839
2840 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2841 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2842 }
2843
2844 void generate_math_stubs() {
2845 {
2846 StubCodeMark mark(this, "StubRoutines", "log");
2847 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2848
2849 __ subq(rsp, 8);
2850 __ movdbl(Address(rsp, 0), xmm0);
2851 __ fld_d(Address(rsp, 0));
2852 __ flog();
2853 __ fstp_d(Address(rsp, 0));
2854 __ movdbl(xmm0, Address(rsp, 0));
2855 __ addq(rsp, 8);
2856 __ ret(0);
2857 }
2858 {
2859 StubCodeMark mark(this, "StubRoutines", "log10");
2860 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2861
2862 __ subq(rsp, 8);
2863 __ movdbl(Address(rsp, 0), xmm0);
2864 __ fld_d(Address(rsp, 0));
2865 __ flog10();
2866 __ fstp_d(Address(rsp, 0));
2867 __ movdbl(xmm0, Address(rsp, 0));
2868 __ addq(rsp, 8);
2869 __ ret(0);
2870 }
2871 {
2872 StubCodeMark mark(this, "StubRoutines", "sin");
2873 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2874
2875 __ subq(rsp, 8);
2876 __ movdbl(Address(rsp, 0), xmm0);
2877 __ fld_d(Address(rsp, 0));
2878 __ trigfunc('s');
2879 __ fstp_d(Address(rsp, 0));
2880 __ movdbl(xmm0, Address(rsp, 0));
2881 __ addq(rsp, 8);
2882 __ ret(0);
2883 }
2884 {
2885 StubCodeMark mark(this, "StubRoutines", "cos");
2886 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2887
2888 __ subq(rsp, 8);
2889 __ movdbl(Address(rsp, 0), xmm0);
2890 __ fld_d(Address(rsp, 0));
2891 __ trigfunc('c');
2892 __ fstp_d(Address(rsp, 0));
2893 __ movdbl(xmm0, Address(rsp, 0));
2894 __ addq(rsp, 8);
2895 __ ret(0);
2896 }
2897 {
2898 StubCodeMark mark(this, "StubRoutines", "tan");
2899 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2900
2901 __ subq(rsp, 8);
2902 __ movdbl(Address(rsp, 0), xmm0);
2903 __ fld_d(Address(rsp, 0));
2904 __ trigfunc('t');
2905 __ fstp_d(Address(rsp, 0));
2906 __ movdbl(xmm0, Address(rsp, 0));
2907 __ addq(rsp, 8);
2908 __ ret(0);
2909 }
2910
2911 // The intrinsic version of these seem to return the same value as
2912 // the strict version.
2913 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2914 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2915 }
2916
2917 #undef __
2918 #define __ masm->
2919
2920 // Continuation point for throwing of implicit exceptions that are
2921 // not handled in the current activation. Fabricates an exception
2922 // oop and initiates normal exception dispatching in this
2923 // frame. Since we need to preserve callee-saved values (currently
2924 // only for C2, but done for C1 as well) we need a callee-saved oop
2925 // map and therefore have to make these stubs into RuntimeStubs
2926 // rather than BufferBlobs. If the compiler needs all registers to
2927 // be preserved between the fault point and the exception handler
2928 // then it must assume responsibility for that in
2929 // AbstractCompiler::continuation_for_implicit_null_exception or
2930 // continuation_for_implicit_division_by_zero_exception. All other
2931 // implicit exceptions (e.g., NullPointerException or
2932 // AbstractMethodError on entry) are either at call sites or
2933 // otherwise assume that stack unwinding will be initiated, so
2934 // caller saved registers were assumed volatile in the compiler.
2935 address generate_throw_exception(const char* name,
2936 address runtime_entry,
2937 bool restore_saved_exception_pc) {
2938 // Information about frame layout at time of blocking runtime call.
2939 // Note that we only have to preserve callee-saved registers since
2940 // the compilers are responsible for supplying a continuation point
2941 // if they expect all registers to be preserved.
2942 enum layout {
2943 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2944 rbp_off2,
2945 return_off,
2946 return_off2,
2947 framesize // inclusive of return address
2948 };
2949
2950 int insts_size = 512;
2951 int locs_size = 64;
2952
2953 CodeBuffer code(name, insts_size, locs_size);
2954 OopMapSet* oop_maps = new OopMapSet();
2955 MacroAssembler* masm = new MacroAssembler(&code);
2956
2957 address start = __ pc();
2958
2959 // This is an inlined and slightly modified version of call_VM
2960 // which has the ability to fetch the return PC out of
2961 // thread-local storage and also sets up last_Java_sp slightly
2962 // differently than the real call_VM
2963 if (restore_saved_exception_pc) {
2964 __ movptr(rax,
2965 Address(r15_thread,
2966 in_bytes(JavaThread::saved_exception_pc_offset())));
2967 __ push(rax);
2968 }
2969
2970 __ enter(); // required for proper stackwalking of RuntimeStub frame
2971
2972 assert(is_even(framesize/2), "sp not 16-byte aligned");
2973
2974 // return address and rbp are already in place
2975 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
2976
2977 int frame_complete = __ pc() - start;
2978
2979 // Set up last_Java_sp and last_Java_fp
2980 __ set_last_Java_frame(rsp, rbp, NULL);
2981
2982 // Call runtime
2983 __ movptr(c_rarg0, r15_thread);
2984 BLOCK_COMMENT("call runtime_entry");
2985 __ call(RuntimeAddress(runtime_entry));
2986
2987 // Generate oop map
2988 OopMap* map = new OopMap(framesize, 0);
2989
2990 oop_maps->add_gc_map(__ pc() - start, map);
2991
2992 __ reset_last_Java_frame(true, false);
2993
2994 __ leave(); // required for proper stackwalking of RuntimeStub frame
2995
2996 // check for pending exceptions
2997 #ifdef ASSERT
2998 Label L;
2999 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
3000 (int32_t) NULL_WORD);
3001 __ jcc(Assembler::notEqual, L);
3002 __ should_not_reach_here();
3003 __ bind(L);
3004 #endif // ASSERT
3005 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3006
3007
3008 // codeBlob framesize is in words (not VMRegImpl::slot_size)
3009 RuntimeStub* stub =
3010 RuntimeStub::new_runtime_stub(name,
3011 &code,
3012 frame_complete,
3013 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3014 oop_maps, false);
3015 return stub->entry_point();
3016 }
3017
3018 // Initialization
3019 void generate_initial() {
3020 // Generates all stubs and initializes the entry points
3021
3022 // This platform-specific stub is needed by generate_call_stub()
3023 StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
3024
3025 // entry points that exist in all platforms Note: This is code
3026 // that could be shared among different platforms - however the
3027 // benefit seems to be smaller than the disadvantage of having a
3028 // much more complicated generator structure. See also comment in
3029 // stubRoutines.hpp.
3030
3031 StubRoutines::_forward_exception_entry = generate_forward_exception();
3032
3033 StubRoutines::_call_stub_entry =
3034 generate_call_stub(StubRoutines::_call_stub_return_address);
3035
3036 // is referenced by megamorphic call
3037 StubRoutines::_catch_exception_entry = generate_catch_exception();
3038
3039 // atomic calls
3040 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3041 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
3042 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
3043 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
3044 StubRoutines::_atomic_add_entry = generate_atomic_add();
3045 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
3046 StubRoutines::_fence_entry = generate_orderaccess_fence();
3047
3048 StubRoutines::_handler_for_unsafe_access_entry =
3049 generate_handler_for_unsafe_access();
3050
3051 // platform dependent
3052 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
3053
3054 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3055 }
3056
3057 void generate_all() {
3058 // Generates all stubs and initializes the entry points
3059
3060 // These entry points require SharedInfo::stack0 to be set up in
3061 // non-core builds and need to be relocatable, so they each
3062 // fabricate a RuntimeStub internally.
3063 StubRoutines::_throw_AbstractMethodError_entry =
3064 generate_throw_exception("AbstractMethodError throw_exception",
3065 CAST_FROM_FN_PTR(address,
3066 SharedRuntime::
3067 throw_AbstractMethodError),
3068 false);
3069
3070 StubRoutines::_throw_IncompatibleClassChangeError_entry =
3071 generate_throw_exception("IncompatibleClassChangeError throw_exception",
3072 CAST_FROM_FN_PTR(address,
3073 SharedRuntime::
3074 throw_IncompatibleClassChangeError),
3075 false);
3076
3077 StubRoutines::_throw_ArithmeticException_entry =
3078 generate_throw_exception("ArithmeticException throw_exception",
3079 CAST_FROM_FN_PTR(address,
3080 SharedRuntime::
3081 throw_ArithmeticException),
3082 true);
3083
3084 StubRoutines::_throw_NullPointerException_entry =
3085 generate_throw_exception("NullPointerException throw_exception",
3086 CAST_FROM_FN_PTR(address,
3087 SharedRuntime::
3088 throw_NullPointerException),
3089 true);
3090
3091 StubRoutines::_throw_NullPointerException_at_call_entry =
3092 generate_throw_exception("NullPointerException at call throw_exception",
3093 CAST_FROM_FN_PTR(address,
3094 SharedRuntime::
3095 throw_NullPointerException_at_call),
3096 false);
3097
3098 StubRoutines::_throw_StackOverflowError_entry =
3099 generate_throw_exception("StackOverflowError throw_exception",
3100 CAST_FROM_FN_PTR(address,
3101 SharedRuntime::
3102 throw_StackOverflowError),
3103 false);
3104
3105 // entry points that are platform specific
3106 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
3107 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
3108 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
3109 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
3110
3111 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
3112 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
3113 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3114 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3115
3116 // support for verify_oop (must happen after universe_init)
3117 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3118
3119 // arraycopy stubs used by compilers
3120 generate_arraycopy_stubs();
3121
3122 generate_math_stubs();
3123 }
3124
3125 public:
3126 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3127 if (all) {
3128 generate_all();
3129 } else {
3130 generate_initial();
3131 }
3132 }
3133 }; // end class declaration
3134
3135 void StubGenerator_generate(CodeBuffer* code, bool all) {
3136 StubGenerator g(code, all);
3137 }