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