1 /*
2 * Copyright (c) 2003, 2010, 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 "incls/_precompiled.incl"
26 #include "incls/_stubGenerator_x86_64.cpp.incl"
27
28 // Declaration and definition of StubGenerator (no .hpp file).
29 // For a more detailed description of the stub routine structure
30 // see the comment in stubRoutines.hpp
31
32 #define __ _masm->
33 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
34 #define a__ ((Assembler*)_masm)->
35
36 #ifdef PRODUCT
37 #define BLOCK_COMMENT(str) /* nothing */
38 #else
39 #define BLOCK_COMMENT(str) __ block_comment(str)
40 #endif
41
42 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
43 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
44
45 // Stub Code definitions
46
47 static address handle_unsafe_access() {
48 JavaThread* thread = JavaThread::current();
49 address pc = thread->saved_exception_pc();
50 // pc is the instruction which we must emulate
51 // doing a no-op is fine: return garbage from the load
52 // therefore, compute npc
53 address npc = Assembler::locate_next_instruction(pc);
54
55 // request an async exception
56 thread->set_pending_unsafe_access_error();
57
58 // return address of next instruction to execute
59 return npc;
60 }
61
62 class StubGenerator: public StubCodeGenerator {
63 private:
64
65 #ifdef PRODUCT
66 #define inc_counter_np(counter) (0)
67 #else
68 void inc_counter_np_(int& counter) {
69 __ incrementl(ExternalAddress((address)&counter));
70 }
71 #define inc_counter_np(counter) \
72 BLOCK_COMMENT("inc_counter " #counter); \
73 inc_counter_np_(counter);
74 #endif
75
76 // Call stubs are used to call Java from C
77 //
78 // Linux Arguments:
79 // c_rarg0: call wrapper address address
80 // c_rarg1: result address
81 // c_rarg2: result type BasicType
82 // c_rarg3: method methodOop
83 // c_rarg4: (interpreter) entry point address
84 // c_rarg5: parameters intptr_t*
85 // 16(rbp): parameter size (in words) int
86 // 24(rbp): thread Thread*
87 //
88 // [ return_from_Java ] <--- rsp
89 // [ argument word n ]
90 // ...
91 // -12 [ argument word 1 ]
92 // -11 [ saved r15 ] <--- rsp_after_call
93 // -10 [ saved r14 ]
94 // -9 [ saved r13 ]
95 // -8 [ saved r12 ]
96 // -7 [ saved rbx ]
97 // -6 [ call wrapper ]
98 // -5 [ result ]
99 // -4 [ result type ]
100 // -3 [ method ]
101 // -2 [ entry point ]
102 // -1 [ parameters ]
103 // 0 [ saved rbp ] <--- rbp
104 // 1 [ return address ]
105 // 2 [ parameter size ]
106 // 3 [ thread ]
107 //
108 // Windows Arguments:
109 // c_rarg0: call wrapper address address
110 // c_rarg1: result address
111 // c_rarg2: result type BasicType
112 // c_rarg3: method methodOop
113 // 48(rbp): (interpreter) entry point address
114 // 56(rbp): parameters intptr_t*
115 // 64(rbp): parameter size (in words) int
116 // 72(rbp): thread Thread*
117 //
118 // [ return_from_Java ] <--- rsp
119 // [ argument word n ]
120 // ...
121 // -8 [ argument word 1 ]
122 // -7 [ saved r15 ] <--- rsp_after_call
123 // -6 [ saved r14 ]
124 // -5 [ saved r13 ]
125 // -4 [ saved r12 ]
126 // -3 [ saved rdi ]
127 // -2 [ saved rsi ]
128 // -1 [ saved rbx ]
129 // 0 [ saved rbp ] <--- rbp
130 // 1 [ return address ]
131 // 2 [ call wrapper ]
132 // 3 [ result ]
133 // 4 [ result type ]
134 // 5 [ method ]
135 // 6 [ entry point ]
136 // 7 [ parameters ]
137 // 8 [ parameter size ]
138 // 9 [ thread ]
139 //
140 // Windows reserves the callers stack space for arguments 1-4.
141 // We spill c_rarg0-c_rarg3 to this space.
142
143 // Call stub stack layout word offsets from rbp
144 enum call_stub_layout {
145 #ifdef _WIN64
146 rsp_after_call_off = -7,
147 r15_off = rsp_after_call_off,
148 r14_off = -6,
149 r13_off = -5,
150 r12_off = -4,
151 rdi_off = -3,
152 rsi_off = -2,
153 rbx_off = -1,
154 rbp_off = 0,
155 retaddr_off = 1,
156 call_wrapper_off = 2,
157 result_off = 3,
158 result_type_off = 4,
159 method_off = 5,
160 entry_point_off = 6,
161 parameters_off = 7,
162 parameter_size_off = 8,
163 thread_off = 9
164 #else
165 rsp_after_call_off = -12,
166 mxcsr_off = rsp_after_call_off,
167 r15_off = -11,
168 r14_off = -10,
169 r13_off = -9,
170 r12_off = -8,
171 rbx_off = -7,
172 call_wrapper_off = -6,
173 result_off = -5,
174 result_type_off = -4,
175 method_off = -3,
176 entry_point_off = -2,
177 parameters_off = -1,
178 rbp_off = 0,
179 retaddr_off = 1,
180 parameter_size_off = 2,
181 thread_off = 3
182 #endif
183 };
184
185 address generate_call_stub(address& return_address) {
186 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
187 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
188 "adjust this code");
189 StubCodeMark mark(this, "StubRoutines", "call_stub");
190 address start = __ pc();
191
192 // same as in generate_catch_exception()!
193 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
194
195 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
196 const Address result (rbp, result_off * wordSize);
197 const Address result_type (rbp, result_type_off * wordSize);
198 const Address method (rbp, method_off * wordSize);
199 const Address entry_point (rbp, entry_point_off * wordSize);
200 const Address parameters (rbp, parameters_off * wordSize);
201 const Address parameter_size(rbp, parameter_size_off * wordSize);
202
203 // same as in generate_catch_exception()!
204 const Address thread (rbp, thread_off * wordSize);
205
206 const Address r15_save(rbp, r15_off * wordSize);
207 const Address r14_save(rbp, r14_off * wordSize);
208 const Address r13_save(rbp, r13_off * wordSize);
209 const Address r12_save(rbp, r12_off * wordSize);
210 const Address rbx_save(rbp, rbx_off * wordSize);
211
212 // stub code
213 __ enter();
214 __ subptr(rsp, -rsp_after_call_off * wordSize);
215
216 // save register parameters
217 #ifndef _WIN64
218 __ movptr(parameters, c_rarg5); // parameters
219 __ movptr(entry_point, c_rarg4); // entry_point
220 #endif
221
222 __ movptr(method, c_rarg3); // method
223 __ movl(result_type, c_rarg2); // result type
224 __ movptr(result, c_rarg1); // result
225 __ movptr(call_wrapper, c_rarg0); // call wrapper
226
227 // save regs belonging to calling function
228 __ movptr(rbx_save, rbx);
229 __ movptr(r12_save, r12);
230 __ movptr(r13_save, r13);
231 __ movptr(r14_save, r14);
232 __ movptr(r15_save, r15);
233
234 #ifdef _WIN64
235 const Address rdi_save(rbp, rdi_off * wordSize);
236 const Address rsi_save(rbp, rsi_off * wordSize);
237
238 __ movptr(rsi_save, rsi);
239 __ movptr(rdi_save, rdi);
240 #else
241 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
242 {
243 Label skip_ldmx;
244 __ stmxcsr(mxcsr_save);
245 __ movl(rax, mxcsr_save);
246 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
247 ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
248 __ cmp32(rax, mxcsr_std);
249 __ jcc(Assembler::equal, skip_ldmx);
250 __ ldmxcsr(mxcsr_std);
251 __ bind(skip_ldmx);
252 }
253 #endif
254
255 // Load up thread register
256 __ movptr(r15_thread, thread);
257 __ reinit_heapbase();
258
259 #ifdef ASSERT
260 // make sure we have no pending exceptions
261 {
262 Label L;
263 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
264 __ jcc(Assembler::equal, L);
265 __ stop("StubRoutines::call_stub: entered with pending exception");
266 __ bind(L);
267 }
268 #endif
269
270 // pass parameters if any
271 BLOCK_COMMENT("pass parameters if any");
272 Label parameters_done;
273 __ movl(c_rarg3, parameter_size);
274 __ testl(c_rarg3, c_rarg3);
275 __ jcc(Assembler::zero, parameters_done);
276
277 Label loop;
278 __ movptr(c_rarg2, parameters); // parameter pointer
279 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
280 __ BIND(loop);
281 __ movptr(rax, Address(c_rarg2, 0));// get parameter
282 __ addptr(c_rarg2, wordSize); // advance to next parameter
283 __ decrementl(c_rarg1); // decrement counter
284 __ push(rax); // pass parameter
285 __ jcc(Assembler::notZero, loop);
286
287 // call Java function
288 __ BIND(parameters_done);
289 __ movptr(rbx, method); // get methodOop
290 __ movptr(c_rarg1, entry_point); // get entry_point
291 __ mov(r13, rsp); // set sender sp
292 BLOCK_COMMENT("call Java function");
293 __ call(c_rarg1);
294
295 BLOCK_COMMENT("call_stub_return_address:");
296 return_address = __ pc();
297
298 // store result depending on type (everything that is not
299 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
300 __ movptr(c_rarg0, result);
301 Label is_long, is_float, is_double, exit;
302 __ movl(c_rarg1, result_type);
303 __ cmpl(c_rarg1, T_OBJECT);
304 __ jcc(Assembler::equal, is_long);
305 __ cmpl(c_rarg1, T_LONG);
306 __ jcc(Assembler::equal, is_long);
307 __ cmpl(c_rarg1, T_FLOAT);
308 __ jcc(Assembler::equal, is_float);
309 __ cmpl(c_rarg1, T_DOUBLE);
310 __ jcc(Assembler::equal, is_double);
311
312 // handle T_INT case
313 __ movl(Address(c_rarg0, 0), rax);
314
315 __ BIND(exit);
316
317 // pop parameters
318 __ lea(rsp, rsp_after_call);
319
320 #ifdef ASSERT
321 // verify that threads correspond
322 {
323 Label L, S;
324 __ cmpptr(r15_thread, thread);
325 __ jcc(Assembler::notEqual, S);
326 __ get_thread(rbx);
327 __ cmpptr(r15_thread, rbx);
328 __ jcc(Assembler::equal, L);
329 __ bind(S);
330 __ jcc(Assembler::equal, L);
331 __ stop("StubRoutines::call_stub: threads must correspond");
332 __ bind(L);
333 }
334 #endif
335
336 // restore regs belonging to calling function
337 __ movptr(r15, r15_save);
338 __ movptr(r14, r14_save);
339 __ movptr(r13, r13_save);
340 __ movptr(r12, r12_save);
341 __ movptr(rbx, rbx_save);
342
343 #ifdef _WIN64
344 __ movptr(rdi, rdi_save);
345 __ movptr(rsi, rsi_save);
346 #else
347 __ ldmxcsr(mxcsr_save);
348 #endif
349
350 // restore rsp
351 __ addptr(rsp, -rsp_after_call_off * wordSize);
352
353 // return
354 __ pop(rbp);
355 __ ret(0);
356
357 // handle return types different from T_INT
358 __ BIND(is_long);
359 __ movq(Address(c_rarg0, 0), rax);
360 __ jmp(exit);
361
362 __ BIND(is_float);
363 __ movflt(Address(c_rarg0, 0), xmm0);
364 __ jmp(exit);
365
366 __ BIND(is_double);
367 __ movdbl(Address(c_rarg0, 0), xmm0);
368 __ jmp(exit);
369
370 return start;
371 }
372
373 // Return point for a Java call if there's an exception thrown in
374 // Java code. The exception is caught and transformed into a
375 // pending exception stored in JavaThread that can be tested from
376 // within the VM.
377 //
378 // Note: Usually the parameters are removed by the callee. In case
379 // of an exception crossing an activation frame boundary, that is
380 // not the case if the callee is compiled code => need to setup the
381 // rsp.
382 //
383 // rax: exception oop
384
385 address generate_catch_exception() {
386 StubCodeMark mark(this, "StubRoutines", "catch_exception");
387 address start = __ pc();
388
389 // same as in generate_call_stub():
390 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
391 const Address thread (rbp, thread_off * wordSize);
392
393 #ifdef ASSERT
394 // verify that threads correspond
395 {
396 Label L, S;
397 __ cmpptr(r15_thread, thread);
398 __ jcc(Assembler::notEqual, S);
399 __ get_thread(rbx);
400 __ cmpptr(r15_thread, rbx);
401 __ jcc(Assembler::equal, L);
402 __ bind(S);
403 __ stop("StubRoutines::catch_exception: threads must correspond");
404 __ bind(L);
405 }
406 #endif
407
408 // set pending exception
409 __ verify_oop(rax);
410
411 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
412 __ lea(rscratch1, ExternalAddress((address)__FILE__));
413 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
414 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
415
416 // complete return to VM
417 assert(StubRoutines::_call_stub_return_address != NULL,
418 "_call_stub_return_address must have been generated before");
419 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
420
421 return start;
422 }
423
424 // Continuation point for runtime calls returning with a pending
425 // exception. The pending exception check happened in the runtime
426 // or native call stub. The pending exception in Thread is
427 // converted into a Java-level exception.
428 //
429 // Contract with Java-level exception handlers:
430 // rax: exception
431 // rdx: throwing pc
432 //
433 // NOTE: At entry of this stub, exception-pc must be on stack !!
434
435 address generate_forward_exception() {
436 StubCodeMark mark(this, "StubRoutines", "forward exception");
437 address start = __ pc();
438
439 // Upon entry, the sp points to the return address returning into
440 // Java (interpreted or compiled) code; i.e., the return address
441 // becomes the throwing pc.
442 //
443 // Arguments pushed before the runtime call are still on the stack
444 // but the exception handler will reset the stack pointer ->
445 // ignore them. A potential result in registers can be ignored as
446 // well.
447
448 #ifdef ASSERT
449 // make sure this code is only executed if there is a pending exception
450 {
451 Label L;
452 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
453 __ jcc(Assembler::notEqual, L);
454 __ stop("StubRoutines::forward exception: no pending exception (1)");
455 __ bind(L);
456 }
457 #endif
458
459 // compute exception handler into rbx
460 __ movptr(c_rarg0, Address(rsp, 0));
461 BLOCK_COMMENT("call exception_handler_for_return_address");
462 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
463 SharedRuntime::exception_handler_for_return_address),
464 r15_thread, c_rarg0);
465 __ mov(rbx, rax);
466
467 // setup rax & rdx, remove return address & clear pending exception
468 __ pop(rdx);
469 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
470 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
471
472 #ifdef ASSERT
473 // make sure exception is set
474 {
475 Label L;
476 __ testptr(rax, rax);
477 __ jcc(Assembler::notEqual, L);
478 __ stop("StubRoutines::forward exception: no pending exception (2)");
479 __ bind(L);
480 }
481 #endif
482
483 // continue at exception handler (return address removed)
484 // rax: exception
485 // rbx: exception handler
486 // rdx: throwing pc
487 __ verify_oop(rax);
488 __ jmp(rbx);
489
490 return start;
491 }
492
493 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
494 //
495 // Arguments :
496 // c_rarg0: exchange_value
497 // c_rarg0: dest
498 //
499 // Result:
500 // *dest <- ex, return (orig *dest)
501 address generate_atomic_xchg() {
502 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
503 address start = __ pc();
504
505 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
506 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
507 __ ret(0);
508
509 return start;
510 }
511
512 // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
513 //
514 // Arguments :
515 // c_rarg0: exchange_value
516 // c_rarg1: dest
517 //
518 // Result:
519 // *dest <- ex, return (orig *dest)
520 address generate_atomic_xchg_ptr() {
521 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
522 address start = __ pc();
523
524 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
525 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
526 __ ret(0);
527
528 return start;
529 }
530
531 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
532 // jint compare_value)
533 //
534 // Arguments :
535 // c_rarg0: exchange_value
536 // c_rarg1: dest
537 // c_rarg2: compare_value
538 //
539 // Result:
540 // if ( compare_value == *dest ) {
541 // *dest = exchange_value
542 // return compare_value;
543 // else
544 // return *dest;
545 address generate_atomic_cmpxchg() {
546 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
547 address start = __ pc();
548
549 __ movl(rax, c_rarg2);
550 if ( os::is_MP() ) __ lock();
551 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
552 __ ret(0);
553
554 return start;
555 }
556
557 // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
558 // volatile jlong* dest,
559 // jlong compare_value)
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_long() {
572 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
573 address start = __ pc();
574
575 __ movq(rax, c_rarg2);
576 if ( os::is_MP() ) __ lock();
577 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
578 __ ret(0);
579
580 return start;
581 }
582
583 // Support for jint atomic::add(jint add_value, volatile jint* dest)
584 //
585 // Arguments :
586 // c_rarg0: add_value
587 // c_rarg1: dest
588 //
589 // Result:
590 // *dest += add_value
591 // return *dest;
592 address generate_atomic_add() {
593 StubCodeMark mark(this, "StubRoutines", "atomic_add");
594 address start = __ pc();
595
596 __ movl(rax, c_rarg0);
597 if ( os::is_MP() ) __ lock();
598 __ xaddl(Address(c_rarg1, 0), c_rarg0);
599 __ addl(rax, c_rarg0);
600 __ ret(0);
601
602 return start;
603 }
604
605 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
606 //
607 // Arguments :
608 // c_rarg0: add_value
609 // c_rarg1: dest
610 //
611 // Result:
612 // *dest += add_value
613 // return *dest;
614 address generate_atomic_add_ptr() {
615 StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
616 address start = __ pc();
617
618 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
619 if ( os::is_MP() ) __ lock();
620 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
621 __ addptr(rax, c_rarg0);
622 __ ret(0);
623
624 return start;
625 }
626
627 // Support for intptr_t OrderAccess::fence()
628 //
629 // Arguments :
630 //
631 // Result:
632 address generate_orderaccess_fence() {
633 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
634 address start = __ pc();
635 __ membar(Assembler::StoreLoad);
636 __ ret(0);
637
638 return start;
639 }
640
641 // Support for intptr_t get_previous_fp()
642 //
643 // This routine is used to find the previous frame pointer for the
644 // caller (current_frame_guess). This is used as part of debugging
645 // ps() is seemingly lost trying to find frames.
646 // This code assumes that caller current_frame_guess) has a frame.
647 address generate_get_previous_fp() {
648 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
649 const Address old_fp(rbp, 0);
650 const Address older_fp(rax, 0);
651 address start = __ pc();
652
653 __ enter();
654 __ movptr(rax, old_fp); // callers fp
655 __ movptr(rax, older_fp); // the frame for ps()
656 __ pop(rbp);
657 __ ret(0);
658
659 return start;
660 }
661
662 //----------------------------------------------------------------------------------------------------
663 // Support for void verify_mxcsr()
664 //
665 // This routine is used with -Xcheck:jni to verify that native
666 // JNI code does not return to Java code without restoring the
667 // MXCSR register to our expected state.
668
669 address generate_verify_mxcsr() {
670 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
671 address start = __ pc();
672
673 const Address mxcsr_save(rsp, 0);
674
675 if (CheckJNICalls) {
676 Label ok_ret;
677 __ push(rax);
678 __ subptr(rsp, wordSize); // allocate a temp location
679 __ stmxcsr(mxcsr_save);
680 __ movl(rax, mxcsr_save);
681 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
682 __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
683 __ jcc(Assembler::equal, ok_ret);
684
685 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
686
687 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
688
689 __ bind(ok_ret);
690 __ addptr(rsp, wordSize);
691 __ pop(rax);
692 }
693
694 __ ret(0);
695
696 return start;
697 }
698
699 address generate_f2i_fixup() {
700 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
701 Address inout(rsp, 5 * wordSize); // return address + 4 saves
702
703 address start = __ pc();
704
705 Label L;
706
707 __ push(rax);
708 __ push(c_rarg3);
709 __ push(c_rarg2);
710 __ push(c_rarg1);
711
712 __ movl(rax, 0x7f800000);
713 __ xorl(c_rarg3, c_rarg3);
714 __ movl(c_rarg2, inout);
715 __ movl(c_rarg1, c_rarg2);
716 __ andl(c_rarg1, 0x7fffffff);
717 __ cmpl(rax, c_rarg1); // NaN? -> 0
718 __ jcc(Assembler::negative, L);
719 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
720 __ movl(c_rarg3, 0x80000000);
721 __ movl(rax, 0x7fffffff);
722 __ cmovl(Assembler::positive, c_rarg3, rax);
723
724 __ bind(L);
725 __ movptr(inout, c_rarg3);
726
727 __ pop(c_rarg1);
728 __ pop(c_rarg2);
729 __ pop(c_rarg3);
730 __ pop(rax);
731
732 __ ret(0);
733
734 return start;
735 }
736
737 address generate_f2l_fixup() {
738 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
739 Address inout(rsp, 5 * wordSize); // return address + 4 saves
740 address start = __ pc();
741
742 Label L;
743
744 __ push(rax);
745 __ push(c_rarg3);
746 __ push(c_rarg2);
747 __ push(c_rarg1);
748
749 __ movl(rax, 0x7f800000);
750 __ xorl(c_rarg3, c_rarg3);
751 __ movl(c_rarg2, inout);
752 __ movl(c_rarg1, c_rarg2);
753 __ andl(c_rarg1, 0x7fffffff);
754 __ cmpl(rax, c_rarg1); // NaN? -> 0
755 __ jcc(Assembler::negative, L);
756 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
757 __ mov64(c_rarg3, 0x8000000000000000);
758 __ mov64(rax, 0x7fffffffffffffff);
759 __ cmov(Assembler::positive, c_rarg3, rax);
760
761 __ bind(L);
762 __ movptr(inout, c_rarg3);
763
764 __ pop(c_rarg1);
765 __ pop(c_rarg2);
766 __ pop(c_rarg3);
767 __ pop(rax);
768
769 __ ret(0);
770
771 return start;
772 }
773
774 address generate_d2i_fixup() {
775 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
776 Address inout(rsp, 6 * wordSize); // return address + 5 saves
777
778 address start = __ pc();
779
780 Label L;
781
782 __ push(rax);
783 __ push(c_rarg3);
784 __ push(c_rarg2);
785 __ push(c_rarg1);
786 __ push(c_rarg0);
787
788 __ movl(rax, 0x7ff00000);
789 __ movq(c_rarg2, inout);
790 __ movl(c_rarg3, c_rarg2);
791 __ mov(c_rarg1, c_rarg2);
792 __ mov(c_rarg0, c_rarg2);
793 __ negl(c_rarg3);
794 __ shrptr(c_rarg1, 0x20);
795 __ orl(c_rarg3, c_rarg2);
796 __ andl(c_rarg1, 0x7fffffff);
797 __ xorl(c_rarg2, c_rarg2);
798 __ shrl(c_rarg3, 0x1f);
799 __ orl(c_rarg1, c_rarg3);
800 __ cmpl(rax, c_rarg1);
801 __ jcc(Assembler::negative, L); // NaN -> 0
802 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
803 __ movl(c_rarg2, 0x80000000);
804 __ movl(rax, 0x7fffffff);
805 __ cmov(Assembler::positive, c_rarg2, rax);
806
807 __ bind(L);
808 __ movptr(inout, c_rarg2);
809
810 __ pop(c_rarg0);
811 __ pop(c_rarg1);
812 __ pop(c_rarg2);
813 __ pop(c_rarg3);
814 __ pop(rax);
815
816 __ ret(0);
817
818 return start;
819 }
820
821 address generate_d2l_fixup() {
822 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
823 Address inout(rsp, 6 * wordSize); // return address + 5 saves
824
825 address start = __ pc();
826
827 Label L;
828
829 __ push(rax);
830 __ push(c_rarg3);
831 __ push(c_rarg2);
832 __ push(c_rarg1);
833 __ push(c_rarg0);
834
835 __ movl(rax, 0x7ff00000);
836 __ movq(c_rarg2, inout);
837 __ movl(c_rarg3, c_rarg2);
838 __ mov(c_rarg1, c_rarg2);
839 __ mov(c_rarg0, c_rarg2);
840 __ negl(c_rarg3);
841 __ shrptr(c_rarg1, 0x20);
842 __ orl(c_rarg3, c_rarg2);
843 __ andl(c_rarg1, 0x7fffffff);
844 __ xorl(c_rarg2, c_rarg2);
845 __ shrl(c_rarg3, 0x1f);
846 __ orl(c_rarg1, c_rarg3);
847 __ cmpl(rax, c_rarg1);
848 __ jcc(Assembler::negative, L); // NaN -> 0
849 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
850 __ mov64(c_rarg2, 0x8000000000000000);
851 __ mov64(rax, 0x7fffffffffffffff);
852 __ cmovq(Assembler::positive, c_rarg2, rax);
853
854 __ bind(L);
855 __ movq(inout, c_rarg2);
856
857 __ pop(c_rarg0);
858 __ pop(c_rarg1);
859 __ pop(c_rarg2);
860 __ pop(c_rarg3);
861 __ pop(rax);
862
863 __ ret(0);
864
865 return start;
866 }
867
868 address generate_fp_mask(const char *stub_name, int64_t mask) {
869 __ align(CodeEntryAlignment);
870 StubCodeMark mark(this, "StubRoutines", stub_name);
871 address start = __ pc();
872
873 __ emit_data64( mask, relocInfo::none );
874 __ emit_data64( mask, relocInfo::none );
875
876 return start;
877 }
878
879 // The following routine generates a subroutine to throw an
880 // asynchronous UnknownError when an unsafe access gets a fault that
881 // could not be reasonably prevented by the programmer. (Example:
882 // SIGBUS/OBJERR.)
883 address generate_handler_for_unsafe_access() {
884 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
885 address start = __ pc();
886
887 __ push(0); // hole for return address-to-be
888 __ pusha(); // push registers
889 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
890
891 __ subptr(rsp, frame::arg_reg_save_area_bytes);
892 BLOCK_COMMENT("call handle_unsafe_access");
893 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
894 __ addptr(rsp, frame::arg_reg_save_area_bytes);
895
896 __ movptr(next_pc, rax); // stuff next address
897 __ popa();
898 __ ret(0); // jump to next address
899
900 return start;
901 }
902
903 // Non-destructive plausibility checks for oops
904 //
905 // Arguments:
906 // all args on stack!
907 //
908 // Stack after saving c_rarg3:
909 // [tos + 0]: saved c_rarg3
910 // [tos + 1]: saved c_rarg2
911 // [tos + 2]: saved r12 (several TemplateTable methods use it)
912 // [tos + 3]: saved flags
913 // [tos + 4]: return address
914 // * [tos + 5]: error message (char*)
915 // * [tos + 6]: object to verify (oop)
916 // [tos + 7]: saved rax - saved by caller
917 // [tos + 8]: saved r10 (rscratch1) - saved by caller
918 // * = popped on exit
919 address generate_verify_oop() {
920 StubCodeMark mark(this, "StubRoutines", "verify_oop");
921 address start = __ pc();
922
923 Label exit, error;
924
925 __ pushf();
926 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
927
928 __ push(r12);
929
930 // save c_rarg2 and c_rarg3
931 __ push(c_rarg2);
932 __ push(c_rarg3);
933
934 enum {
935 // After previous pushes.
936 oop_to_verify = 6 * wordSize,
937 saved_rax = 7 * wordSize,
938
939 // Before the call to MacroAssembler::debug(), see below.
940 return_addr = 16 * wordSize,
941 error_msg = 17 * wordSize
942 };
943
944 // get object
945 __ movptr(rax, Address(rsp, oop_to_verify));
946
947 // make sure object is 'reasonable'
948 __ testptr(rax, rax);
949 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
950 // Check if the oop is in the right area of memory
951 __ movptr(c_rarg2, rax);
952 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
953 __ andptr(c_rarg2, c_rarg3);
954 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
955 __ cmpptr(c_rarg2, c_rarg3);
956 __ jcc(Assembler::notZero, error);
957
958 // set r12 to heapbase for load_klass()
959 __ reinit_heapbase();
960
961 // make sure klass is 'reasonable'
962 __ load_klass(rax, rax); // get klass
963 __ testptr(rax, rax);
964 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
965 // Check if the klass is in the right area of memory
966 __ mov(c_rarg2, rax);
967 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
968 __ andptr(c_rarg2, c_rarg3);
969 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
970 __ cmpptr(c_rarg2, c_rarg3);
971 __ jcc(Assembler::notZero, error);
972
973 // make sure klass' klass is 'reasonable'
974 __ load_klass(rax, rax);
975 __ testptr(rax, rax);
976 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
977 // Check if the klass' klass is in the right area of memory
978 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
979 __ andptr(rax, c_rarg3);
980 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
981 __ cmpptr(rax, c_rarg3);
982 __ jcc(Assembler::notZero, error);
983
984 // return if everything seems ok
985 __ bind(exit);
986 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
987 __ pop(c_rarg3); // restore c_rarg3
988 __ pop(c_rarg2); // restore c_rarg2
989 __ pop(r12); // restore r12
990 __ popf(); // restore flags
991 __ ret(2 * wordSize); // pop caller saved stuff
992
993 // handle errors
994 __ bind(error);
995 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
996 __ pop(c_rarg3); // get saved c_rarg3 back
997 __ pop(c_rarg2); // get saved c_rarg2 back
998 __ pop(r12); // get saved r12 back
999 __ popf(); // get saved flags off stack --
1000 // will be ignored
1001
1002 __ pusha(); // push registers
1003 // (rip is already
1004 // already pushed)
1005 // debug(char* msg, int64_t pc, int64_t regs[])
1006 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1007 // pushed all the registers, so now the stack looks like:
1008 // [tos + 0] 16 saved registers
1009 // [tos + 16] return address
1010 // * [tos + 17] error message (char*)
1011 // * [tos + 18] object to verify (oop)
1012 // [tos + 19] saved rax - saved by caller
1013 // [tos + 20] saved r10 (rscratch1) - saved by caller
1014 // * = popped on exit
1015
1016 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1017 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1018 __ movq(c_rarg2, rsp); // pass address of regs on stack
1019 __ mov(r12, rsp); // remember rsp
1020 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1021 __ andptr(rsp, -16); // align stack as required by ABI
1022 BLOCK_COMMENT("call MacroAssembler::debug");
1023 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1024 __ mov(rsp, r12); // restore rsp
1025 __ popa(); // pop registers (includes r12)
1026 __ ret(2 * wordSize); // pop caller saved stuff
1027
1028 return start;
1029 }
1030
1031 static address disjoint_byte_copy_entry;
1032 static address disjoint_short_copy_entry;
1033 static address disjoint_int_copy_entry;
1034 static address disjoint_long_copy_entry;
1035 static address disjoint_oop_copy_entry;
1036
1037 static address byte_copy_entry;
1038 static address short_copy_entry;
1039 static address int_copy_entry;
1040 static address long_copy_entry;
1041 static address oop_copy_entry;
1042
1043 static address checkcast_copy_entry;
1044
1045 //
1046 // Verify that a register contains clean 32-bits positive value
1047 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1048 //
1049 // Input:
1050 // Rint - 32-bits value
1051 // Rtmp - scratch
1052 //
1053 void assert_clean_int(Register Rint, Register Rtmp) {
1054 #ifdef ASSERT
1055 Label L;
1056 assert_different_registers(Rtmp, Rint);
1057 __ movslq(Rtmp, Rint);
1058 __ cmpq(Rtmp, Rint);
1059 __ jcc(Assembler::equal, L);
1060 __ stop("high 32-bits of int value are not 0");
1061 __ bind(L);
1062 #endif
1063 }
1064
1065 // Generate overlap test for array copy stubs
1066 //
1067 // Input:
1068 // c_rarg0 - from
1069 // c_rarg1 - to
1070 // c_rarg2 - element count
1071 //
1072 // Output:
1073 // rax - &from[element count - 1]
1074 //
1075 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1076 assert(no_overlap_target != NULL, "must be generated");
1077 array_overlap_test(no_overlap_target, NULL, sf);
1078 }
1079 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1080 array_overlap_test(NULL, &L_no_overlap, sf);
1081 }
1082 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1083 const Register from = c_rarg0;
1084 const Register to = c_rarg1;
1085 const Register count = c_rarg2;
1086 const Register end_from = rax;
1087
1088 __ cmpptr(to, from);
1089 __ lea(end_from, Address(from, count, sf, 0));
1090 if (NOLp == NULL) {
1091 ExternalAddress no_overlap(no_overlap_target);
1092 __ jump_cc(Assembler::belowEqual, no_overlap);
1093 __ cmpptr(to, end_from);
1094 __ jump_cc(Assembler::aboveEqual, no_overlap);
1095 } else {
1096 __ jcc(Assembler::belowEqual, (*NOLp));
1097 __ cmpptr(to, end_from);
1098 __ jcc(Assembler::aboveEqual, (*NOLp));
1099 }
1100 }
1101
1102 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1103 //
1104 // Outputs:
1105 // rdi - rcx
1106 // rsi - rdx
1107 // rdx - r8
1108 // rcx - r9
1109 //
1110 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1111 // are non-volatile. r9 and r10 should not be used by the caller.
1112 //
1113 void setup_arg_regs(int nargs = 3) {
1114 const Register saved_rdi = r9;
1115 const Register saved_rsi = r10;
1116 assert(nargs == 3 || nargs == 4, "else fix");
1117 #ifdef _WIN64
1118 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1119 "unexpected argument registers");
1120 if (nargs >= 4)
1121 __ mov(rax, r9); // r9 is also saved_rdi
1122 __ movptr(saved_rdi, rdi);
1123 __ movptr(saved_rsi, rsi);
1124 __ mov(rdi, rcx); // c_rarg0
1125 __ mov(rsi, rdx); // c_rarg1
1126 __ mov(rdx, r8); // c_rarg2
1127 if (nargs >= 4)
1128 __ mov(rcx, rax); // c_rarg3 (via rax)
1129 #else
1130 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1131 "unexpected argument registers");
1132 #endif
1133 }
1134
1135 void restore_arg_regs() {
1136 const Register saved_rdi = r9;
1137 const Register saved_rsi = r10;
1138 #ifdef _WIN64
1139 __ movptr(rdi, saved_rdi);
1140 __ movptr(rsi, saved_rsi);
1141 #endif
1142 }
1143
1144 // Generate code for an array write pre barrier
1145 //
1146 // addr - starting address
1147 // count - element count
1148 //
1149 // Destroy no registers!
1150 //
1151 void gen_write_ref_array_pre_barrier(Register addr, Register count) {
1152 BarrierSet* bs = Universe::heap()->barrier_set();
1153 switch (bs->kind()) {
1154 case BarrierSet::G1SATBCT:
1155 case BarrierSet::G1SATBCTLogging:
1156 {
1157 __ pusha(); // push registers
1158 if (count == c_rarg0) {
1159 if (addr == c_rarg1) {
1160 // exactly backwards!!
1161 __ xchgptr(c_rarg1, c_rarg0);
1162 } else {
1163 __ movptr(c_rarg1, count);
1164 __ movptr(c_rarg0, addr);
1165 }
1166
1167 } else {
1168 __ movptr(c_rarg0, addr);
1169 __ movptr(c_rarg1, count);
1170 }
1171 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1172 __ popa();
1173 }
1174 break;
1175 case BarrierSet::CardTableModRef:
1176 case BarrierSet::CardTableExtension:
1177 case BarrierSet::ModRef:
1178 break;
1179 default:
1180 ShouldNotReachHere();
1181
1182 }
1183 }
1184
1185 //
1186 // Generate code for an array write post barrier
1187 //
1188 // Input:
1189 // start - register containing starting address of destination array
1190 // end - register containing ending address of destination array
1191 // scratch - scratch register
1192 //
1193 // The input registers are overwritten.
1194 // The ending address is inclusive.
1195 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1196 assert_different_registers(start, end, scratch);
1197 BarrierSet* bs = Universe::heap()->barrier_set();
1198 switch (bs->kind()) {
1199 case BarrierSet::G1SATBCT:
1200 case BarrierSet::G1SATBCTLogging:
1201
1202 {
1203 __ pusha(); // push registers (overkill)
1204 // must compute element count unless barrier set interface is changed (other platforms supply count)
1205 assert_different_registers(start, end, scratch);
1206 __ lea(scratch, Address(end, BytesPerHeapOop));
1207 __ subptr(scratch, start); // subtract start to get #bytes
1208 __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count
1209 __ mov(c_rarg0, start);
1210 __ mov(c_rarg1, scratch);
1211 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1212 __ popa();
1213 }
1214 break;
1215 case BarrierSet::CardTableModRef:
1216 case BarrierSet::CardTableExtension:
1217 {
1218 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1219 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1220
1221 Label L_loop;
1222
1223 __ shrptr(start, CardTableModRefBS::card_shift);
1224 __ addptr(end, BytesPerHeapOop);
1225 __ shrptr(end, CardTableModRefBS::card_shift);
1226 __ subptr(end, start); // number of bytes to copy
1227
1228 intptr_t disp = (intptr_t) ct->byte_map_base;
1229 if (__ is_simm32(disp)) {
1230 Address cardtable(noreg, noreg, Address::no_scale, disp);
1231 __ lea(scratch, cardtable);
1232 } else {
1233 ExternalAddress cardtable((address)disp);
1234 __ lea(scratch, cardtable);
1235 }
1236
1237 const Register count = end; // 'end' register contains bytes count now
1238 __ addptr(start, scratch);
1239 __ BIND(L_loop);
1240 __ movb(Address(start, count, Address::times_1), 0);
1241 __ decrement(count);
1242 __ jcc(Assembler::greaterEqual, L_loop);
1243 }
1244 break;
1245 default:
1246 ShouldNotReachHere();
1247
1248 }
1249 }
1250
1251
1252 // Copy big chunks forward
1253 //
1254 // Inputs:
1255 // end_from - source arrays end address
1256 // end_to - destination array end address
1257 // qword_count - 64-bits element count, negative
1258 // to - scratch
1259 // L_copy_32_bytes - entry label
1260 // L_copy_8_bytes - exit label
1261 //
1262 void copy_32_bytes_forward(Register end_from, Register end_to,
1263 Register qword_count, Register to,
1264 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1265 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1266 Label L_loop;
1267 __ align(OptoLoopAlignment);
1268 __ BIND(L_loop);
1269 if(UseUnalignedLoadStores) {
1270 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1271 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1272 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1273 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1274
1275 } else {
1276 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1277 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1278 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1279 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1280 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1281 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1282 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1283 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1284 }
1285 __ BIND(L_copy_32_bytes);
1286 __ addptr(qword_count, 4);
1287 __ jcc(Assembler::lessEqual, L_loop);
1288 __ subptr(qword_count, 4);
1289 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1290 }
1291
1292
1293 // Copy big chunks backward
1294 //
1295 // Inputs:
1296 // from - source arrays address
1297 // dest - destination array address
1298 // qword_count - 64-bits element count
1299 // to - scratch
1300 // L_copy_32_bytes - entry label
1301 // L_copy_8_bytes - exit label
1302 //
1303 void copy_32_bytes_backward(Register from, Register dest,
1304 Register qword_count, Register to,
1305 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1306 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1307 Label L_loop;
1308 __ align(OptoLoopAlignment);
1309 __ BIND(L_loop);
1310 if(UseUnalignedLoadStores) {
1311 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1312 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1313 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1314 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1315
1316 } else {
1317 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1318 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1319 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1320 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1321 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1322 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1323 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1324 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1325 }
1326 __ BIND(L_copy_32_bytes);
1327 __ subptr(qword_count, 4);
1328 __ jcc(Assembler::greaterEqual, L_loop);
1329 __ addptr(qword_count, 4);
1330 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1331 }
1332
1333
1334 // Arguments:
1335 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1336 // ignored
1337 // name - stub name string
1338 //
1339 // Inputs:
1340 // c_rarg0 - source array address
1341 // c_rarg1 - destination array address
1342 // c_rarg2 - element count, treated as ssize_t, can be zero
1343 //
1344 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1345 // we let the hardware handle it. The one to eight bytes within words,
1346 // dwords or qwords that span cache line boundaries will still be loaded
1347 // and stored atomically.
1348 //
1349 // Side Effects:
1350 // disjoint_byte_copy_entry is set to the no-overlap entry point
1351 // used by generate_conjoint_byte_copy().
1352 //
1353 address generate_disjoint_byte_copy(bool aligned, const char *name) {
1354 __ align(CodeEntryAlignment);
1355 StubCodeMark mark(this, "StubRoutines", name);
1356 address start = __ pc();
1357
1358 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1359 Label L_copy_byte, L_exit;
1360 const Register from = rdi; // source array address
1361 const Register to = rsi; // destination array address
1362 const Register count = rdx; // elements count
1363 const Register byte_count = rcx;
1364 const Register qword_count = count;
1365 const Register end_from = from; // source array end address
1366 const Register end_to = to; // destination array end address
1367 // End pointers are inclusive, and if count is not zero they point
1368 // to the last unit copied: end_to[0] := end_from[0]
1369
1370 __ enter(); // required for proper stackwalking of RuntimeStub frame
1371 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1372
1373 disjoint_byte_copy_entry = __ pc();
1374 BLOCK_COMMENT("Entry:");
1375 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1376
1377 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1378 // r9 and r10 may be used to save non-volatile registers
1379
1380 // 'from', 'to' and 'count' are now valid
1381 __ movptr(byte_count, count);
1382 __ shrptr(count, 3); // count => qword_count
1383
1384 // Copy from low to high addresses. Use 'to' as scratch.
1385 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1386 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1387 __ negptr(qword_count); // make the count negative
1388 __ jmp(L_copy_32_bytes);
1389
1390 // Copy trailing qwords
1391 __ BIND(L_copy_8_bytes);
1392 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1393 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1394 __ increment(qword_count);
1395 __ jcc(Assembler::notZero, L_copy_8_bytes);
1396
1397 // Check for and copy trailing dword
1398 __ BIND(L_copy_4_bytes);
1399 __ testl(byte_count, 4);
1400 __ jccb(Assembler::zero, L_copy_2_bytes);
1401 __ movl(rax, Address(end_from, 8));
1402 __ movl(Address(end_to, 8), rax);
1403
1404 __ addptr(end_from, 4);
1405 __ addptr(end_to, 4);
1406
1407 // Check for and copy trailing word
1408 __ BIND(L_copy_2_bytes);
1409 __ testl(byte_count, 2);
1410 __ jccb(Assembler::zero, L_copy_byte);
1411 __ movw(rax, Address(end_from, 8));
1412 __ movw(Address(end_to, 8), rax);
1413
1414 __ addptr(end_from, 2);
1415 __ addptr(end_to, 2);
1416
1417 // Check for and copy trailing byte
1418 __ BIND(L_copy_byte);
1419 __ testl(byte_count, 1);
1420 __ jccb(Assembler::zero, L_exit);
1421 __ movb(rax, Address(end_from, 8));
1422 __ movb(Address(end_to, 8), rax);
1423
1424 __ BIND(L_exit);
1425 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1426 restore_arg_regs();
1427 __ xorptr(rax, rax); // return 0
1428 __ leave(); // required for proper stackwalking of RuntimeStub frame
1429 __ ret(0);
1430
1431 // Copy in 32-bytes chunks
1432 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1433 __ jmp(L_copy_4_bytes);
1434
1435 return start;
1436 }
1437
1438 // Arguments:
1439 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1440 // ignored
1441 // name - stub name string
1442 //
1443 // Inputs:
1444 // c_rarg0 - source array address
1445 // c_rarg1 - destination array address
1446 // c_rarg2 - element count, treated as ssize_t, can be zero
1447 //
1448 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1449 // we let the hardware handle it. The one to eight bytes within words,
1450 // dwords or qwords that span cache line boundaries will still be loaded
1451 // and stored atomically.
1452 //
1453 address generate_conjoint_byte_copy(bool aligned, const char *name) {
1454 __ align(CodeEntryAlignment);
1455 StubCodeMark mark(this, "StubRoutines", name);
1456 address start = __ pc();
1457
1458 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1459 const Register from = rdi; // source array address
1460 const Register to = rsi; // destination array address
1461 const Register count = rdx; // elements count
1462 const Register byte_count = rcx;
1463 const Register qword_count = count;
1464
1465 __ enter(); // required for proper stackwalking of RuntimeStub frame
1466 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1467
1468 byte_copy_entry = __ pc();
1469 BLOCK_COMMENT("Entry:");
1470 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1471
1472 array_overlap_test(disjoint_byte_copy_entry, Address::times_1);
1473 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1474 // r9 and r10 may be used to save non-volatile registers
1475
1476 // 'from', 'to' and 'count' are now valid
1477 __ movptr(byte_count, count);
1478 __ shrptr(count, 3); // count => qword_count
1479
1480 // Copy from high to low addresses.
1481
1482 // Check for and copy trailing byte
1483 __ testl(byte_count, 1);
1484 __ jcc(Assembler::zero, L_copy_2_bytes);
1485 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1486 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1487 __ decrement(byte_count); // Adjust for possible trailing word
1488
1489 // Check for and copy trailing word
1490 __ BIND(L_copy_2_bytes);
1491 __ testl(byte_count, 2);
1492 __ jcc(Assembler::zero, L_copy_4_bytes);
1493 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1494 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1495
1496 // Check for and copy trailing dword
1497 __ BIND(L_copy_4_bytes);
1498 __ testl(byte_count, 4);
1499 __ jcc(Assembler::zero, L_copy_32_bytes);
1500 __ movl(rax, Address(from, qword_count, Address::times_8));
1501 __ movl(Address(to, qword_count, Address::times_8), rax);
1502 __ jmp(L_copy_32_bytes);
1503
1504 // Copy trailing qwords
1505 __ BIND(L_copy_8_bytes);
1506 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1507 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1508 __ decrement(qword_count);
1509 __ jcc(Assembler::notZero, L_copy_8_bytes);
1510
1511 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1512 restore_arg_regs();
1513 __ xorptr(rax, rax); // return 0
1514 __ leave(); // required for proper stackwalking of RuntimeStub frame
1515 __ ret(0);
1516
1517 // Copy in 32-bytes chunks
1518 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1519
1520 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1521 restore_arg_regs();
1522 __ xorptr(rax, rax); // return 0
1523 __ leave(); // required for proper stackwalking of RuntimeStub frame
1524 __ ret(0);
1525
1526 return start;
1527 }
1528
1529 // Arguments:
1530 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1531 // ignored
1532 // name - stub name string
1533 //
1534 // Inputs:
1535 // c_rarg0 - source array address
1536 // c_rarg1 - destination array address
1537 // c_rarg2 - element count, treated as ssize_t, can be zero
1538 //
1539 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1540 // let the hardware handle it. The two or four words within dwords
1541 // or qwords that span cache line boundaries will still be loaded
1542 // and stored atomically.
1543 //
1544 // Side Effects:
1545 // disjoint_short_copy_entry is set to the no-overlap entry point
1546 // used by generate_conjoint_short_copy().
1547 //
1548 address generate_disjoint_short_copy(bool aligned, const char *name) {
1549 __ align(CodeEntryAlignment);
1550 StubCodeMark mark(this, "StubRoutines", name);
1551 address start = __ pc();
1552
1553 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1554 const Register from = rdi; // source array address
1555 const Register to = rsi; // destination array address
1556 const Register count = rdx; // elements count
1557 const Register word_count = rcx;
1558 const Register qword_count = count;
1559 const Register end_from = from; // source array end address
1560 const Register end_to = to; // destination array end address
1561 // End pointers are inclusive, and if count is not zero they point
1562 // to the last unit copied: end_to[0] := end_from[0]
1563
1564 __ enter(); // required for proper stackwalking of RuntimeStub frame
1565 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1566
1567 disjoint_short_copy_entry = __ pc();
1568 BLOCK_COMMENT("Entry:");
1569 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1570
1571 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1572 // r9 and r10 may be used to save non-volatile registers
1573
1574 // 'from', 'to' and 'count' are now valid
1575 __ movptr(word_count, count);
1576 __ shrptr(count, 2); // count => qword_count
1577
1578 // Copy from low to high addresses. Use 'to' as scratch.
1579 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1580 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1581 __ negptr(qword_count);
1582 __ jmp(L_copy_32_bytes);
1583
1584 // Copy trailing qwords
1585 __ BIND(L_copy_8_bytes);
1586 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1587 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1588 __ increment(qword_count);
1589 __ jcc(Assembler::notZero, L_copy_8_bytes);
1590
1591 // Original 'dest' is trashed, so we can't use it as a
1592 // base register for a possible trailing word copy
1593
1594 // Check for and copy trailing dword
1595 __ BIND(L_copy_4_bytes);
1596 __ testl(word_count, 2);
1597 __ jccb(Assembler::zero, L_copy_2_bytes);
1598 __ movl(rax, Address(end_from, 8));
1599 __ movl(Address(end_to, 8), rax);
1600
1601 __ addptr(end_from, 4);
1602 __ addptr(end_to, 4);
1603
1604 // Check for and copy trailing word
1605 __ BIND(L_copy_2_bytes);
1606 __ testl(word_count, 1);
1607 __ jccb(Assembler::zero, L_exit);
1608 __ movw(rax, Address(end_from, 8));
1609 __ movw(Address(end_to, 8), rax);
1610
1611 __ BIND(L_exit);
1612 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1613 restore_arg_regs();
1614 __ xorptr(rax, rax); // return 0
1615 __ leave(); // required for proper stackwalking of RuntimeStub frame
1616 __ ret(0);
1617
1618 // Copy in 32-bytes chunks
1619 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1620 __ jmp(L_copy_4_bytes);
1621
1622 return start;
1623 }
1624
1625 // Arguments:
1626 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1627 // ignored
1628 // name - stub name string
1629 //
1630 // Inputs:
1631 // c_rarg0 - source array address
1632 // c_rarg1 - destination array address
1633 // c_rarg2 - element count, treated as ssize_t, can be zero
1634 //
1635 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1636 // let the hardware handle it. The two or four words within dwords
1637 // or qwords that span cache line boundaries will still be loaded
1638 // and stored atomically.
1639 //
1640 address generate_conjoint_short_copy(bool aligned, const char *name) {
1641 __ align(CodeEntryAlignment);
1642 StubCodeMark mark(this, "StubRoutines", name);
1643 address start = __ pc();
1644
1645 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1646 const Register from = rdi; // source array address
1647 const Register to = rsi; // destination array address
1648 const Register count = rdx; // elements count
1649 const Register word_count = rcx;
1650 const Register qword_count = count;
1651
1652 __ enter(); // required for proper stackwalking of RuntimeStub frame
1653 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1654
1655 short_copy_entry = __ pc();
1656 BLOCK_COMMENT("Entry:");
1657 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1658
1659 array_overlap_test(disjoint_short_copy_entry, Address::times_2);
1660 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1661 // r9 and r10 may be used to save non-volatile registers
1662
1663 // 'from', 'to' and 'count' are now valid
1664 __ movptr(word_count, count);
1665 __ shrptr(count, 2); // count => qword_count
1666
1667 // Copy from high to low addresses. Use 'to' as scratch.
1668
1669 // Check for and copy trailing word
1670 __ testl(word_count, 1);
1671 __ jccb(Assembler::zero, L_copy_4_bytes);
1672 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1673 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1674
1675 // Check for and copy trailing dword
1676 __ BIND(L_copy_4_bytes);
1677 __ testl(word_count, 2);
1678 __ jcc(Assembler::zero, L_copy_32_bytes);
1679 __ movl(rax, Address(from, qword_count, Address::times_8));
1680 __ movl(Address(to, qword_count, Address::times_8), rax);
1681 __ jmp(L_copy_32_bytes);
1682
1683 // Copy trailing qwords
1684 __ BIND(L_copy_8_bytes);
1685 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1686 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1687 __ decrement(qword_count);
1688 __ jcc(Assembler::notZero, L_copy_8_bytes);
1689
1690 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1691 restore_arg_regs();
1692 __ xorptr(rax, rax); // return 0
1693 __ leave(); // required for proper stackwalking of RuntimeStub frame
1694 __ ret(0);
1695
1696 // Copy in 32-bytes chunks
1697 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1698
1699 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1700 restore_arg_regs();
1701 __ xorptr(rax, rax); // return 0
1702 __ leave(); // required for proper stackwalking of RuntimeStub frame
1703 __ ret(0);
1704
1705 return start;
1706 }
1707
1708 // Arguments:
1709 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1710 // ignored
1711 // is_oop - true => oop array, so generate store check code
1712 // name - stub name string
1713 //
1714 // Inputs:
1715 // c_rarg0 - source array address
1716 // c_rarg1 - destination array address
1717 // c_rarg2 - element count, treated as ssize_t, can be zero
1718 //
1719 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1720 // the hardware handle it. The two dwords within qwords that span
1721 // cache line boundaries will still be loaded and stored atomicly.
1722 //
1723 // Side Effects:
1724 // disjoint_int_copy_entry is set to the no-overlap entry point
1725 // used by generate_conjoint_int_oop_copy().
1726 //
1727 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1728 __ align(CodeEntryAlignment);
1729 StubCodeMark mark(this, "StubRoutines", name);
1730 address start = __ pc();
1731
1732 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1733 const Register from = rdi; // source array address
1734 const Register to = rsi; // destination array address
1735 const Register count = rdx; // elements count
1736 const Register dword_count = rcx;
1737 const Register qword_count = count;
1738 const Register end_from = from; // source array end address
1739 const Register end_to = to; // destination array end address
1740 const Register saved_to = r11; // saved destination array address
1741 // End pointers are inclusive, and if count is not zero they point
1742 // to the last unit copied: end_to[0] := end_from[0]
1743
1744 __ enter(); // required for proper stackwalking of RuntimeStub frame
1745 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1746
1747 (is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry) = __ pc();
1748
1749 if (is_oop) {
1750 // no registers are destroyed by this call
1751 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1752 }
1753
1754 BLOCK_COMMENT("Entry:");
1755 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1756
1757 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1758 // r9 and r10 may be used to save non-volatile registers
1759
1760 if (is_oop) {
1761 __ movq(saved_to, to);
1762 }
1763
1764 // 'from', 'to' and 'count' are now valid
1765 __ movptr(dword_count, count);
1766 __ shrptr(count, 1); // count => qword_count
1767
1768 // Copy from low to high addresses. Use 'to' as scratch.
1769 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1770 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1771 __ negptr(qword_count);
1772 __ jmp(L_copy_32_bytes);
1773
1774 // Copy trailing qwords
1775 __ BIND(L_copy_8_bytes);
1776 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1777 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1778 __ increment(qword_count);
1779 __ jcc(Assembler::notZero, L_copy_8_bytes);
1780
1781 // Check for and copy trailing dword
1782 __ BIND(L_copy_4_bytes);
1783 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1784 __ jccb(Assembler::zero, L_exit);
1785 __ movl(rax, Address(end_from, 8));
1786 __ movl(Address(end_to, 8), rax);
1787
1788 __ BIND(L_exit);
1789 if (is_oop) {
1790 __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1791 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1792 }
1793 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1794 restore_arg_regs();
1795 __ xorptr(rax, rax); // return 0
1796 __ leave(); // required for proper stackwalking of RuntimeStub frame
1797 __ ret(0);
1798
1799 // Copy 32-bytes chunks
1800 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1801 __ jmp(L_copy_4_bytes);
1802
1803 return start;
1804 }
1805
1806 // Arguments:
1807 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1808 // ignored
1809 // is_oop - true => oop array, so generate store check code
1810 // name - stub name string
1811 //
1812 // Inputs:
1813 // c_rarg0 - source array address
1814 // c_rarg1 - destination array address
1815 // c_rarg2 - element count, treated as ssize_t, can be zero
1816 //
1817 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1818 // the hardware handle it. The two dwords within qwords that span
1819 // cache line boundaries will still be loaded and stored atomicly.
1820 //
1821 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1822 __ align(CodeEntryAlignment);
1823 StubCodeMark mark(this, "StubRoutines", name);
1824 address start = __ pc();
1825
1826 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1827 const Register from = rdi; // source array address
1828 const Register to = rsi; // destination array address
1829 const Register count = rdx; // elements count
1830 const Register dword_count = rcx;
1831 const Register qword_count = count;
1832
1833 __ enter(); // required for proper stackwalking of RuntimeStub frame
1834 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1835
1836 if (is_oop) {
1837 // no registers are destroyed by this call
1838 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1839 }
1840
1841 (is_oop ? oop_copy_entry : int_copy_entry) = __ pc();
1842 BLOCK_COMMENT("Entry:");
1843 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1844
1845 array_overlap_test(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry,
1846 Address::times_4);
1847 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1848 // r9 and r10 may be used to save non-volatile registers
1849
1850 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1851 // 'from', 'to' and 'count' are now valid
1852 __ movptr(dword_count, count);
1853 __ shrptr(count, 1); // count => qword_count
1854
1855 // Copy from high to low addresses. Use 'to' as scratch.
1856
1857 // Check for and copy trailing dword
1858 __ testl(dword_count, 1);
1859 __ jcc(Assembler::zero, L_copy_32_bytes);
1860 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1861 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1862 __ jmp(L_copy_32_bytes);
1863
1864 // Copy trailing qwords
1865 __ BIND(L_copy_8_bytes);
1866 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1867 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1868 __ decrement(qword_count);
1869 __ jcc(Assembler::notZero, L_copy_8_bytes);
1870
1871 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1872 if (is_oop) {
1873 __ jmp(L_exit);
1874 }
1875 restore_arg_regs();
1876 __ xorptr(rax, rax); // return 0
1877 __ leave(); // required for proper stackwalking of RuntimeStub frame
1878 __ ret(0);
1879
1880 // Copy in 32-bytes chunks
1881 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1882
1883 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1884 __ bind(L_exit);
1885 if (is_oop) {
1886 Register end_to = rdx;
1887 __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1888 gen_write_ref_array_post_barrier(to, end_to, rax);
1889 }
1890 restore_arg_regs();
1891 __ xorptr(rax, rax); // return 0
1892 __ leave(); // required for proper stackwalking of RuntimeStub frame
1893 __ ret(0);
1894
1895 return start;
1896 }
1897
1898 // Arguments:
1899 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1900 // ignored
1901 // is_oop - true => oop array, so generate store check code
1902 // name - stub name string
1903 //
1904 // Inputs:
1905 // c_rarg0 - source array address
1906 // c_rarg1 - destination array address
1907 // c_rarg2 - element count, treated as ssize_t, can be zero
1908 //
1909 // Side Effects:
1910 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1911 // no-overlap entry point used by generate_conjoint_long_oop_copy().
1912 //
1913 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1914 __ align(CodeEntryAlignment);
1915 StubCodeMark mark(this, "StubRoutines", name);
1916 address start = __ pc();
1917
1918 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1919 const Register from = rdi; // source array address
1920 const Register to = rsi; // destination array address
1921 const Register qword_count = rdx; // elements count
1922 const Register end_from = from; // source array end address
1923 const Register end_to = rcx; // destination array end address
1924 const Register saved_to = to;
1925 // End pointers are inclusive, and if count is not zero they point
1926 // to the last unit copied: end_to[0] := end_from[0]
1927
1928 __ enter(); // required for proper stackwalking of RuntimeStub frame
1929 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1930 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1931
1932 if (is_oop) {
1933 disjoint_oop_copy_entry = __ pc();
1934 // no registers are destroyed by this call
1935 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1936 } else {
1937 disjoint_long_copy_entry = __ pc();
1938 }
1939 BLOCK_COMMENT("Entry:");
1940 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1941
1942 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1943 // r9 and r10 may be used to save non-volatile registers
1944
1945 // 'from', 'to' and 'qword_count' are now valid
1946
1947 // Copy from low to high addresses. Use 'to' as scratch.
1948 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1949 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1950 __ negptr(qword_count);
1951 __ jmp(L_copy_32_bytes);
1952
1953 // Copy trailing qwords
1954 __ BIND(L_copy_8_bytes);
1955 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1956 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1957 __ increment(qword_count);
1958 __ jcc(Assembler::notZero, L_copy_8_bytes);
1959
1960 if (is_oop) {
1961 __ jmp(L_exit);
1962 } else {
1963 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1964 restore_arg_regs();
1965 __ xorptr(rax, rax); // return 0
1966 __ leave(); // required for proper stackwalking of RuntimeStub frame
1967 __ ret(0);
1968 }
1969
1970 // Copy 64-byte chunks
1971 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1972
1973 if (is_oop) {
1974 __ BIND(L_exit);
1975 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1976 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
1977 } else {
1978 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1979 }
1980 restore_arg_regs();
1981 __ xorptr(rax, rax); // return 0
1982 __ leave(); // required for proper stackwalking of RuntimeStub frame
1983 __ ret(0);
1984
1985 return start;
1986 }
1987
1988 // Arguments:
1989 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1990 // ignored
1991 // is_oop - true => oop array, so generate store check code
1992 // name - stub name string
1993 //
1994 // Inputs:
1995 // c_rarg0 - source array address
1996 // c_rarg1 - destination array address
1997 // c_rarg2 - element count, treated as ssize_t, can be zero
1998 //
1999 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
2000 __ align(CodeEntryAlignment);
2001 StubCodeMark mark(this, "StubRoutines", name);
2002 address start = __ pc();
2003
2004 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2005 const Register from = rdi; // source array address
2006 const Register to = rsi; // destination array address
2007 const Register qword_count = rdx; // elements count
2008 const Register saved_count = rcx;
2009
2010 __ enter(); // required for proper stackwalking of RuntimeStub frame
2011 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2012
2013 address disjoint_copy_entry = NULL;
2014 if (is_oop) {
2015 assert(!UseCompressedOops, "shouldn't be called for compressed oops");
2016 disjoint_copy_entry = disjoint_oop_copy_entry;
2017 oop_copy_entry = __ pc();
2018 array_overlap_test(disjoint_oop_copy_entry, Address::times_8);
2019 } else {
2020 disjoint_copy_entry = disjoint_long_copy_entry;
2021 long_copy_entry = __ pc();
2022 array_overlap_test(disjoint_long_copy_entry, Address::times_8);
2023 }
2024 BLOCK_COMMENT("Entry:");
2025 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2026
2027 array_overlap_test(disjoint_copy_entry, Address::times_8);
2028 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2029 // r9 and r10 may be used to save non-volatile registers
2030
2031 // 'from', 'to' and 'qword_count' are now valid
2032
2033 if (is_oop) {
2034 // Save to and count for store barrier
2035 __ movptr(saved_count, qword_count);
2036 // No registers are destroyed by this call
2037 gen_write_ref_array_pre_barrier(to, saved_count);
2038 }
2039
2040 __ jmp(L_copy_32_bytes);
2041
2042 // Copy trailing qwords
2043 __ BIND(L_copy_8_bytes);
2044 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2045 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2046 __ decrement(qword_count);
2047 __ jcc(Assembler::notZero, L_copy_8_bytes);
2048
2049 if (is_oop) {
2050 __ jmp(L_exit);
2051 } else {
2052 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2053 restore_arg_regs();
2054 __ xorptr(rax, rax); // return 0
2055 __ leave(); // required for proper stackwalking of RuntimeStub frame
2056 __ ret(0);
2057 }
2058
2059 // Copy in 32-bytes chunks
2060 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2061
2062 if (is_oop) {
2063 __ BIND(L_exit);
2064 __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2065 gen_write_ref_array_post_barrier(to, rcx, rax);
2066 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2067 } else {
2068 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2069 }
2070 restore_arg_regs();
2071 __ xorptr(rax, rax); // return 0
2072 __ leave(); // required for proper stackwalking of RuntimeStub frame
2073 __ ret(0);
2074
2075 return start;
2076 }
2077
2078
2079 // Helper for generating a dynamic type check.
2080 // Smashes no registers.
2081 void generate_type_check(Register sub_klass,
2082 Register super_check_offset,
2083 Register super_klass,
2084 Label& L_success) {
2085 assert_different_registers(sub_klass, super_check_offset, super_klass);
2086
2087 BLOCK_COMMENT("type_check:");
2088
2089 Label L_miss;
2090
2091 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2092 super_check_offset);
2093 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2094
2095 // Fall through on failure!
2096 __ BIND(L_miss);
2097 }
2098
2099 //
2100 // Generate checkcasting array copy stub
2101 //
2102 // Input:
2103 // c_rarg0 - source array address
2104 // c_rarg1 - destination array address
2105 // c_rarg2 - element count, treated as ssize_t, can be zero
2106 // c_rarg3 - size_t ckoff (super_check_offset)
2107 // not Win64
2108 // c_rarg4 - oop ckval (super_klass)
2109 // Win64
2110 // rsp+40 - oop ckval (super_klass)
2111 //
2112 // Output:
2113 // rax == 0 - success
2114 // rax == -1^K - failure, where K is partial transfer count
2115 //
2116 address generate_checkcast_copy(const char *name) {
2117
2118 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2119
2120 // Input registers (after setup_arg_regs)
2121 const Register from = rdi; // source array address
2122 const Register to = rsi; // destination array address
2123 const Register length = rdx; // elements count
2124 const Register ckoff = rcx; // super_check_offset
2125 const Register ckval = r8; // super_klass
2126
2127 // Registers used as temps (r13, r14 are save-on-entry)
2128 const Register end_from = from; // source array end address
2129 const Register end_to = r13; // destination array end address
2130 const Register count = rdx; // -(count_remaining)
2131 const Register r14_length = r14; // saved copy of length
2132 // End pointers are inclusive, and if length is not zero they point
2133 // to the last unit copied: end_to[0] := end_from[0]
2134
2135 const Register rax_oop = rax; // actual oop copied
2136 const Register r11_klass = r11; // oop._klass
2137
2138 //---------------------------------------------------------------
2139 // Assembler stub will be used for this call to arraycopy
2140 // if the two arrays are subtypes of Object[] but the
2141 // destination array type is not equal to or a supertype
2142 // of the source type. Each element must be separately
2143 // checked.
2144
2145 __ align(CodeEntryAlignment);
2146 StubCodeMark mark(this, "StubRoutines", name);
2147 address start = __ pc();
2148
2149 __ enter(); // required for proper stackwalking of RuntimeStub frame
2150
2151 checkcast_copy_entry = __ pc();
2152 BLOCK_COMMENT("Entry:");
2153
2154 #ifdef ASSERT
2155 // caller guarantees that the arrays really are different
2156 // otherwise, we would have to make conjoint checks
2157 { Label L;
2158 array_overlap_test(L, TIMES_OOP);
2159 __ stop("checkcast_copy within a single array");
2160 __ bind(L);
2161 }
2162 #endif //ASSERT
2163
2164 // allocate spill slots for r13, r14
2165 enum {
2166 saved_r13_offset,
2167 saved_r14_offset,
2168 saved_rbp_offset,
2169 saved_rip_offset,
2170 saved_rarg0_offset
2171 };
2172 __ subptr(rsp, saved_rbp_offset * wordSize);
2173 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2174 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2175 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2176 // ckoff => rcx, ckval => r8
2177 // r9 and r10 may be used to save non-volatile registers
2178 #ifdef _WIN64
2179 // last argument (#4) is on stack on Win64
2180 const int ckval_offset = saved_rarg0_offset + 4;
2181 __ movptr(ckval, Address(rsp, ckval_offset * wordSize));
2182 #endif
2183
2184 // check that int operands are properly extended to size_t
2185 assert_clean_int(length, rax);
2186 assert_clean_int(ckoff, rax);
2187
2188 #ifdef ASSERT
2189 BLOCK_COMMENT("assert consistent ckoff/ckval");
2190 // The ckoff and ckval must be mutually consistent,
2191 // even though caller generates both.
2192 { Label L;
2193 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2194 Klass::super_check_offset_offset_in_bytes());
2195 __ cmpl(ckoff, Address(ckval, sco_offset));
2196 __ jcc(Assembler::equal, L);
2197 __ stop("super_check_offset inconsistent");
2198 __ bind(L);
2199 }
2200 #endif //ASSERT
2201
2202 // Loop-invariant addresses. They are exclusive end pointers.
2203 Address end_from_addr(from, length, TIMES_OOP, 0);
2204 Address end_to_addr(to, length, TIMES_OOP, 0);
2205 // Loop-variant addresses. They assume post-incremented count < 0.
2206 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2207 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2208
2209 gen_write_ref_array_pre_barrier(to, count);
2210
2211 // Copy from low to high addresses, indexed from the end of each array.
2212 __ lea(end_from, end_from_addr);
2213 __ lea(end_to, end_to_addr);
2214 __ movptr(r14_length, length); // save a copy of the length
2215 assert(length == count, ""); // else fix next line:
2216 __ negptr(count); // negate and test the length
2217 __ jcc(Assembler::notZero, L_load_element);
2218
2219 // Empty array: Nothing to do.
2220 __ xorptr(rax, rax); // return 0 on (trivial) success
2221 __ jmp(L_done);
2222
2223 // ======== begin loop ========
2224 // (Loop is rotated; its entry is L_load_element.)
2225 // Loop control:
2226 // for (count = -count; count != 0; count++)
2227 // Base pointers src, dst are biased by 8*(count-1),to last element.
2228 __ align(OptoLoopAlignment);
2229
2230 __ BIND(L_store_element);
2231 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2232 __ increment(count); // increment the count toward zero
2233 __ jcc(Assembler::zero, L_do_card_marks);
2234
2235 // ======== loop entry is here ========
2236 __ BIND(L_load_element);
2237 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2238 __ testptr(rax_oop, rax_oop);
2239 __ jcc(Assembler::zero, L_store_element);
2240
2241 __ load_klass(r11_klass, rax_oop);// query the object klass
2242 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2243 // ======== end loop ========
2244
2245 // It was a real error; we must depend on the caller to finish the job.
2246 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2247 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2248 // and report their number to the caller.
2249 assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2250 __ lea(end_to, to_element_addr);
2251 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2252 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2253 __ movptr(rax, r14_length); // original oops
2254 __ addptr(rax, count); // K = (original - remaining) oops
2255 __ notptr(rax); // report (-1^K) to caller
2256 __ jmp(L_done);
2257
2258 // Come here on success only.
2259 __ BIND(L_do_card_marks);
2260 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2261 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2262 __ xorptr(rax, rax); // return 0 on success
2263
2264 // Common exit point (success or failure).
2265 __ BIND(L_done);
2266 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2267 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2268 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
2269 restore_arg_regs();
2270 __ leave(); // required for proper stackwalking of RuntimeStub frame
2271 __ ret(0);
2272
2273 return start;
2274 }
2275
2276 //
2277 // Generate 'unsafe' array copy stub
2278 // Though just as safe as the other stubs, it takes an unscaled
2279 // size_t argument instead of an element count.
2280 //
2281 // Input:
2282 // c_rarg0 - source array address
2283 // c_rarg1 - destination array address
2284 // c_rarg2 - byte count, treated as ssize_t, can be zero
2285 //
2286 // Examines the alignment of the operands and dispatches
2287 // to a long, int, short, or byte copy loop.
2288 //
2289 address generate_unsafe_copy(const char *name) {
2290
2291 Label L_long_aligned, L_int_aligned, L_short_aligned;
2292
2293 // Input registers (before setup_arg_regs)
2294 const Register from = c_rarg0; // source array address
2295 const Register to = c_rarg1; // destination array address
2296 const Register size = c_rarg2; // byte count (size_t)
2297
2298 // Register used as a temp
2299 const Register bits = rax; // test copy of low bits
2300
2301 __ align(CodeEntryAlignment);
2302 StubCodeMark mark(this, "StubRoutines", name);
2303 address start = __ pc();
2304
2305 __ enter(); // required for proper stackwalking of RuntimeStub frame
2306
2307 // bump this on entry, not on exit:
2308 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2309
2310 __ mov(bits, from);
2311 __ orptr(bits, to);
2312 __ orptr(bits, size);
2313
2314 __ testb(bits, BytesPerLong-1);
2315 __ jccb(Assembler::zero, L_long_aligned);
2316
2317 __ testb(bits, BytesPerInt-1);
2318 __ jccb(Assembler::zero, L_int_aligned);
2319
2320 __ testb(bits, BytesPerShort-1);
2321 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2322
2323 __ BIND(L_short_aligned);
2324 __ shrptr(size, LogBytesPerShort); // size => short_count
2325 __ jump(RuntimeAddress(short_copy_entry));
2326
2327 __ BIND(L_int_aligned);
2328 __ shrptr(size, LogBytesPerInt); // size => int_count
2329 __ jump(RuntimeAddress(int_copy_entry));
2330
2331 __ BIND(L_long_aligned);
2332 __ shrptr(size, LogBytesPerLong); // size => qword_count
2333 __ jump(RuntimeAddress(long_copy_entry));
2334
2335 return start;
2336 }
2337
2338 // Perform range checks on the proposed arraycopy.
2339 // Kills temp, but nothing else.
2340 // Also, clean the sign bits of src_pos and dst_pos.
2341 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2342 Register src_pos, // source position (c_rarg1)
2343 Register dst, // destination array oo (c_rarg2)
2344 Register dst_pos, // destination position (c_rarg3)
2345 Register length,
2346 Register temp,
2347 Label& L_failed) {
2348 BLOCK_COMMENT("arraycopy_range_checks:");
2349
2350 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2351 __ movl(temp, length);
2352 __ addl(temp, src_pos); // src_pos + length
2353 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2354 __ jcc(Assembler::above, L_failed);
2355
2356 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2357 __ movl(temp, length);
2358 __ addl(temp, dst_pos); // dst_pos + length
2359 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2360 __ jcc(Assembler::above, L_failed);
2361
2362 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2363 // Move with sign extension can be used since they are positive.
2364 __ movslq(src_pos, src_pos);
2365 __ movslq(dst_pos, dst_pos);
2366
2367 BLOCK_COMMENT("arraycopy_range_checks done");
2368 }
2369
2370 //
2371 // Generate generic array copy stubs
2372 //
2373 // Input:
2374 // c_rarg0 - src oop
2375 // c_rarg1 - src_pos (32-bits)
2376 // c_rarg2 - dst oop
2377 // c_rarg3 - dst_pos (32-bits)
2378 // not Win64
2379 // c_rarg4 - element count (32-bits)
2380 // Win64
2381 // rsp+40 - element count (32-bits)
2382 //
2383 // Output:
2384 // rax == 0 - success
2385 // rax == -1^K - failure, where K is partial transfer count
2386 //
2387 address generate_generic_copy(const char *name) {
2388
2389 Label L_failed, L_failed_0, L_objArray;
2390 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2391
2392 // Input registers
2393 const Register src = c_rarg0; // source array oop
2394 const Register src_pos = c_rarg1; // source position
2395 const Register dst = c_rarg2; // destination array oop
2396 const Register dst_pos = c_rarg3; // destination position
2397 // elements count is on stack on Win64
2398 #ifdef _WIN64
2399 #define C_RARG4 Address(rsp, 6 * wordSize)
2400 #else
2401 #define C_RARG4 c_rarg4
2402 #endif
2403
2404 { int modulus = CodeEntryAlignment;
2405 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2406 int advance = target - (__ offset() % modulus);
2407 if (advance < 0) advance += modulus;
2408 if (advance > 0) __ nop(advance);
2409 }
2410 StubCodeMark mark(this, "StubRoutines", name);
2411
2412 // Short-hop target to L_failed. Makes for denser prologue code.
2413 __ BIND(L_failed_0);
2414 __ jmp(L_failed);
2415 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2416
2417 __ align(CodeEntryAlignment);
2418 address start = __ pc();
2419
2420 __ enter(); // required for proper stackwalking of RuntimeStub frame
2421
2422 // bump this on entry, not on exit:
2423 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2424
2425 //-----------------------------------------------------------------------
2426 // Assembler stub will be used for this call to arraycopy
2427 // if the following conditions are met:
2428 //
2429 // (1) src and dst must not be null.
2430 // (2) src_pos must not be negative.
2431 // (3) dst_pos must not be negative.
2432 // (4) length must not be negative.
2433 // (5) src klass and dst klass should be the same and not NULL.
2434 // (6) src and dst should be arrays.
2435 // (7) src_pos + length must not exceed length of src.
2436 // (8) dst_pos + length must not exceed length of dst.
2437 //
2438
2439 // if (src == NULL) return -1;
2440 __ testptr(src, src); // src oop
2441 size_t j1off = __ offset();
2442 __ jccb(Assembler::zero, L_failed_0);
2443
2444 // if (src_pos < 0) return -1;
2445 __ testl(src_pos, src_pos); // src_pos (32-bits)
2446 __ jccb(Assembler::negative, L_failed_0);
2447
2448 // if (dst == NULL) return -1;
2449 __ testptr(dst, dst); // dst oop
2450 __ jccb(Assembler::zero, L_failed_0);
2451
2452 // if (dst_pos < 0) return -1;
2453 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2454 size_t j4off = __ offset();
2455 __ jccb(Assembler::negative, L_failed_0);
2456
2457 // The first four tests are very dense code,
2458 // but not quite dense enough to put four
2459 // jumps in a 16-byte instruction fetch buffer.
2460 // That's good, because some branch predicters
2461 // do not like jumps so close together.
2462 // Make sure of this.
2463 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2464
2465 // registers used as temp
2466 const Register r11_length = r11; // elements count to copy
2467 const Register r10_src_klass = r10; // array klass
2468 const Register r9_dst_klass = r9; // dest array klass
2469
2470 // if (length < 0) return -1;
2471 __ movl(r11_length, C_RARG4); // length (elements count, 32-bits value)
2472 __ testl(r11_length, r11_length);
2473 __ jccb(Assembler::negative, L_failed_0);
2474
2475 __ load_klass(r10_src_klass, src);
2476 #ifdef ASSERT
2477 // assert(src->klass() != NULL);
2478 BLOCK_COMMENT("assert klasses not null");
2479 { Label L1, L2;
2480 __ testptr(r10_src_klass, r10_src_klass);
2481 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2482 __ bind(L1);
2483 __ stop("broken null klass");
2484 __ bind(L2);
2485 __ load_klass(r9_dst_klass, dst);
2486 __ cmpq(r9_dst_klass, 0);
2487 __ jcc(Assembler::equal, L1); // this would be broken also
2488 BLOCK_COMMENT("assert done");
2489 }
2490 #endif
2491
2492 // Load layout helper (32-bits)
2493 //
2494 // |array_tag| | header_size | element_type | |log2_element_size|
2495 // 32 30 24 16 8 2 0
2496 //
2497 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2498 //
2499
2500 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2501 Klass::layout_helper_offset_in_bytes();
2502
2503 const Register rax_lh = rax; // layout helper
2504
2505 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2506
2507 // Handle objArrays completely differently...
2508 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2509 __ cmpl(rax_lh, objArray_lh);
2510 __ jcc(Assembler::equal, L_objArray);
2511
2512 // if (src->klass() != dst->klass()) return -1;
2513 __ load_klass(r9_dst_klass, dst);
2514 __ cmpq(r10_src_klass, r9_dst_klass);
2515 __ jcc(Assembler::notEqual, L_failed);
2516
2517 // if (!src->is_Array()) return -1;
2518 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2519 __ jcc(Assembler::greaterEqual, L_failed);
2520
2521 // At this point, it is known to be a typeArray (array_tag 0x3).
2522 #ifdef ASSERT
2523 { Label L;
2524 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2525 __ jcc(Assembler::greaterEqual, L);
2526 __ stop("must be a primitive array");
2527 __ bind(L);
2528 }
2529 #endif
2530
2531 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2532 r10, L_failed);
2533
2534 // typeArrayKlass
2535 //
2536 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2537 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2538 //
2539
2540 const Register r10_offset = r10; // array offset
2541 const Register rax_elsize = rax_lh; // element size
2542
2543 __ movl(r10_offset, rax_lh);
2544 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2545 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2546 __ addptr(src, r10_offset); // src array offset
2547 __ addptr(dst, r10_offset); // dst array offset
2548 BLOCK_COMMENT("choose copy loop based on element size");
2549 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2550
2551 // next registers should be set before the jump to corresponding stub
2552 const Register from = c_rarg0; // source array address
2553 const Register to = c_rarg1; // destination array address
2554 const Register count = c_rarg2; // elements count
2555
2556 // 'from', 'to', 'count' registers should be set in such order
2557 // since they are the same as 'src', 'src_pos', 'dst'.
2558
2559 __ BIND(L_copy_bytes);
2560 __ cmpl(rax_elsize, 0);
2561 __ jccb(Assembler::notEqual, L_copy_shorts);
2562 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2563 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2564 __ movl2ptr(count, r11_length); // length
2565 __ jump(RuntimeAddress(byte_copy_entry));
2566
2567 __ BIND(L_copy_shorts);
2568 __ cmpl(rax_elsize, LogBytesPerShort);
2569 __ jccb(Assembler::notEqual, L_copy_ints);
2570 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2571 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2572 __ movl2ptr(count, r11_length); // length
2573 __ jump(RuntimeAddress(short_copy_entry));
2574
2575 __ BIND(L_copy_ints);
2576 __ cmpl(rax_elsize, LogBytesPerInt);
2577 __ jccb(Assembler::notEqual, L_copy_longs);
2578 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2579 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2580 __ movl2ptr(count, r11_length); // length
2581 __ jump(RuntimeAddress(int_copy_entry));
2582
2583 __ BIND(L_copy_longs);
2584 #ifdef ASSERT
2585 { Label L;
2586 __ cmpl(rax_elsize, LogBytesPerLong);
2587 __ jcc(Assembler::equal, L);
2588 __ stop("must be long copy, but elsize is wrong");
2589 __ bind(L);
2590 }
2591 #endif
2592 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2593 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2594 __ movl2ptr(count, r11_length); // length
2595 __ jump(RuntimeAddress(long_copy_entry));
2596
2597 // objArrayKlass
2598 __ BIND(L_objArray);
2599 // live at this point: r10_src_klass, src[_pos], dst[_pos]
2600
2601 Label L_plain_copy, L_checkcast_copy;
2602 // test array classes for subtyping
2603 __ load_klass(r9_dst_klass, dst);
2604 __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality
2605 __ jcc(Assembler::notEqual, L_checkcast_copy);
2606
2607 // Identically typed arrays can be copied without element-wise checks.
2608 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2609 r10, L_failed);
2610
2611 __ lea(from, Address(src, src_pos, TIMES_OOP,
2612 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2613 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2614 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2615 __ movl2ptr(count, r11_length); // length
2616 __ BIND(L_plain_copy);
2617 __ jump(RuntimeAddress(oop_copy_entry));
2618
2619 __ BIND(L_checkcast_copy);
2620 // live at this point: r10_src_klass, !r11_length
2621 {
2622 // assert(r11_length == C_RARG4); // will reload from here
2623 Register r11_dst_klass = r11;
2624 __ load_klass(r11_dst_klass, dst);
2625
2626 // Before looking at dst.length, make sure dst is also an objArray.
2627 __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh);
2628 __ jcc(Assembler::notEqual, L_failed);
2629
2630 // It is safe to examine both src.length and dst.length.
2631 #ifndef _WIN64
2632 arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4,
2633 rax, L_failed);
2634 #else
2635 __ movl(r11_length, C_RARG4); // reload
2636 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2637 rax, L_failed);
2638 __ load_klass(r11_dst_klass, dst); // reload
2639 #endif
2640
2641 // Marshal the base address arguments now, freeing registers.
2642 __ lea(from, Address(src, src_pos, TIMES_OOP,
2643 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2644 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2645 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2646 __ movl(count, C_RARG4); // length (reloaded)
2647 Register sco_temp = c_rarg3; // this register is free now
2648 assert_different_registers(from, to, count, sco_temp,
2649 r11_dst_klass, r10_src_klass);
2650 assert_clean_int(count, sco_temp);
2651
2652 // Generate the type check.
2653 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2654 Klass::super_check_offset_offset_in_bytes());
2655 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2656 assert_clean_int(sco_temp, rax);
2657 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2658
2659 // Fetch destination element klass from the objArrayKlass header.
2660 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2661 objArrayKlass::element_klass_offset_in_bytes());
2662 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2663 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2664 assert_clean_int(sco_temp, rax);
2665
2666 // the checkcast_copy loop needs two extra arguments:
2667 assert(c_rarg3 == sco_temp, "#3 already in place");
2668 __ movptr(C_RARG4, r11_dst_klass); // dst.klass.element_klass
2669 __ jump(RuntimeAddress(checkcast_copy_entry));
2670 }
2671
2672 __ BIND(L_failed);
2673 __ xorptr(rax, rax);
2674 __ notptr(rax); // return -1
2675 __ leave(); // required for proper stackwalking of RuntimeStub frame
2676 __ ret(0);
2677
2678 return start;
2679 }
2680
2681 #undef length_arg
2682
2683 void generate_arraycopy_stubs() {
2684 // Call the conjoint generation methods immediately after
2685 // the disjoint ones so that short branches from the former
2686 // to the latter can be generated.
2687 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
2688 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
2689
2690 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
2691 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
2692
2693 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy");
2694 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy");
2695
2696 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy");
2697 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy");
2698
2699
2700 if (UseCompressedOops) {
2701 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy");
2702 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy");
2703 } else {
2704 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy");
2705 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy");
2706 }
2707
2708 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
2709 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy");
2710 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy");
2711
2712 // We don't generate specialized code for HeapWord-aligned source
2713 // arrays, so just use the code we've already generated
2714 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2715 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2716
2717 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2718 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2719
2720 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2721 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2722
2723 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2724 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2725
2726 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2727 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2728 }
2729
2730 void generate_math_stubs() {
2731 {
2732 StubCodeMark mark(this, "StubRoutines", "log");
2733 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2734
2735 __ subq(rsp, 8);
2736 __ movdbl(Address(rsp, 0), xmm0);
2737 __ fld_d(Address(rsp, 0));
2738 __ flog();
2739 __ fstp_d(Address(rsp, 0));
2740 __ movdbl(xmm0, Address(rsp, 0));
2741 __ addq(rsp, 8);
2742 __ ret(0);
2743 }
2744 {
2745 StubCodeMark mark(this, "StubRoutines", "log10");
2746 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2747
2748 __ subq(rsp, 8);
2749 __ movdbl(Address(rsp, 0), xmm0);
2750 __ fld_d(Address(rsp, 0));
2751 __ flog10();
2752 __ fstp_d(Address(rsp, 0));
2753 __ movdbl(xmm0, Address(rsp, 0));
2754 __ addq(rsp, 8);
2755 __ ret(0);
2756 }
2757 {
2758 StubCodeMark mark(this, "StubRoutines", "sin");
2759 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2760
2761 __ subq(rsp, 8);
2762 __ movdbl(Address(rsp, 0), xmm0);
2763 __ fld_d(Address(rsp, 0));
2764 __ trigfunc('s');
2765 __ fstp_d(Address(rsp, 0));
2766 __ movdbl(xmm0, Address(rsp, 0));
2767 __ addq(rsp, 8);
2768 __ ret(0);
2769 }
2770 {
2771 StubCodeMark mark(this, "StubRoutines", "cos");
2772 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2773
2774 __ subq(rsp, 8);
2775 __ movdbl(Address(rsp, 0), xmm0);
2776 __ fld_d(Address(rsp, 0));
2777 __ trigfunc('c');
2778 __ fstp_d(Address(rsp, 0));
2779 __ movdbl(xmm0, Address(rsp, 0));
2780 __ addq(rsp, 8);
2781 __ ret(0);
2782 }
2783 {
2784 StubCodeMark mark(this, "StubRoutines", "tan");
2785 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2786
2787 __ subq(rsp, 8);
2788 __ movdbl(Address(rsp, 0), xmm0);
2789 __ fld_d(Address(rsp, 0));
2790 __ trigfunc('t');
2791 __ fstp_d(Address(rsp, 0));
2792 __ movdbl(xmm0, Address(rsp, 0));
2793 __ addq(rsp, 8);
2794 __ ret(0);
2795 }
2796
2797 // The intrinsic version of these seem to return the same value as
2798 // the strict version.
2799 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2800 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2801 }
2802
2803 #undef __
2804 #define __ masm->
2805
2806 // Continuation point for throwing of implicit exceptions that are
2807 // not handled in the current activation. Fabricates an exception
2808 // oop and initiates normal exception dispatching in this
2809 // frame. Since we need to preserve callee-saved values (currently
2810 // only for C2, but done for C1 as well) we need a callee-saved oop
2811 // map and therefore have to make these stubs into RuntimeStubs
2812 // rather than BufferBlobs. If the compiler needs all registers to
2813 // be preserved between the fault point and the exception handler
2814 // then it must assume responsibility for that in
2815 // AbstractCompiler::continuation_for_implicit_null_exception or
2816 // continuation_for_implicit_division_by_zero_exception. All other
2817 // implicit exceptions (e.g., NullPointerException or
2818 // AbstractMethodError on entry) are either at call sites or
2819 // otherwise assume that stack unwinding will be initiated, so
2820 // caller saved registers were assumed volatile in the compiler.
2821 address generate_throw_exception(const char* name,
2822 address runtime_entry,
2823 bool restore_saved_exception_pc) {
2824 // Information about frame layout at time of blocking runtime call.
2825 // Note that we only have to preserve callee-saved registers since
2826 // the compilers are responsible for supplying a continuation point
2827 // if they expect all registers to be preserved.
2828 enum layout {
2829 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2830 rbp_off2,
2831 return_off,
2832 return_off2,
2833 framesize // inclusive of return address
2834 };
2835
2836 int insts_size = 512;
2837 int locs_size = 64;
2838
2839 CodeBuffer code(name, insts_size, locs_size);
2840 OopMapSet* oop_maps = new OopMapSet();
2841 MacroAssembler* masm = new MacroAssembler(&code);
2842
2843 address start = __ pc();
2844
2845 // This is an inlined and slightly modified version of call_VM
2846 // which has the ability to fetch the return PC out of
2847 // thread-local storage and also sets up last_Java_sp slightly
2848 // differently than the real call_VM
2849 if (restore_saved_exception_pc) {
2850 __ movptr(rax,
2851 Address(r15_thread,
2852 in_bytes(JavaThread::saved_exception_pc_offset())));
2853 __ push(rax);
2854 }
2855
2856 __ enter(); // required for proper stackwalking of RuntimeStub frame
2857
2858 assert(is_even(framesize/2), "sp not 16-byte aligned");
2859
2860 // return address and rbp are already in place
2861 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
2862
2863 int frame_complete = __ pc() - start;
2864
2865 // Set up last_Java_sp and last_Java_fp
2866 __ set_last_Java_frame(rsp, rbp, NULL);
2867
2868 // Call runtime
2869 __ movptr(c_rarg0, r15_thread);
2870 BLOCK_COMMENT("call runtime_entry");
2871 __ call(RuntimeAddress(runtime_entry));
2872
2873 // Generate oop map
2874 OopMap* map = new OopMap(framesize, 0);
2875
2876 oop_maps->add_gc_map(__ pc() - start, map);
2877
2878 __ reset_last_Java_frame(true, false);
2879
2880 __ leave(); // required for proper stackwalking of RuntimeStub frame
2881
2882 // check for pending exceptions
2883 #ifdef ASSERT
2884 Label L;
2885 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
2886 (int32_t) NULL_WORD);
2887 __ jcc(Assembler::notEqual, L);
2888 __ should_not_reach_here();
2889 __ bind(L);
2890 #endif // ASSERT
2891 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2892
2893
2894 // codeBlob framesize is in words (not VMRegImpl::slot_size)
2895 RuntimeStub* stub =
2896 RuntimeStub::new_runtime_stub(name,
2897 &code,
2898 frame_complete,
2899 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2900 oop_maps, false);
2901 return stub->entry_point();
2902 }
2903
2904 // Initialization
2905 void generate_initial() {
2906 // Generates all stubs and initializes the entry points
2907
2908 // This platform-specific stub is needed by generate_call_stub()
2909 StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
2910
2911 // entry points that exist in all platforms Note: This is code
2912 // that could be shared among different platforms - however the
2913 // benefit seems to be smaller than the disadvantage of having a
2914 // much more complicated generator structure. See also comment in
2915 // stubRoutines.hpp.
2916
2917 StubRoutines::_forward_exception_entry = generate_forward_exception();
2918
2919 StubRoutines::_call_stub_entry =
2920 generate_call_stub(StubRoutines::_call_stub_return_address);
2921
2922 // is referenced by megamorphic call
2923 StubRoutines::_catch_exception_entry = generate_catch_exception();
2924
2925 // atomic calls
2926 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2927 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
2928 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
2929 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2930 StubRoutines::_atomic_add_entry = generate_atomic_add();
2931 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
2932 StubRoutines::_fence_entry = generate_orderaccess_fence();
2933
2934 StubRoutines::_handler_for_unsafe_access_entry =
2935 generate_handler_for_unsafe_access();
2936
2937 // platform dependent
2938 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
2939
2940 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2941 }
2942
2943 void generate_all() {
2944 // Generates all stubs and initializes the entry points
2945
2946 // These entry points require SharedInfo::stack0 to be set up in
2947 // non-core builds and need to be relocatable, so they each
2948 // fabricate a RuntimeStub internally.
2949 StubRoutines::_throw_AbstractMethodError_entry =
2950 generate_throw_exception("AbstractMethodError throw_exception",
2951 CAST_FROM_FN_PTR(address,
2952 SharedRuntime::
2953 throw_AbstractMethodError),
2954 false);
2955
2956 StubRoutines::_throw_IncompatibleClassChangeError_entry =
2957 generate_throw_exception("IncompatibleClassChangeError throw_exception",
2958 CAST_FROM_FN_PTR(address,
2959 SharedRuntime::
2960 throw_IncompatibleClassChangeError),
2961 false);
2962
2963 StubRoutines::_throw_ArithmeticException_entry =
2964 generate_throw_exception("ArithmeticException throw_exception",
2965 CAST_FROM_FN_PTR(address,
2966 SharedRuntime::
2967 throw_ArithmeticException),
2968 true);
2969
2970 StubRoutines::_throw_NullPointerException_entry =
2971 generate_throw_exception("NullPointerException throw_exception",
2972 CAST_FROM_FN_PTR(address,
2973 SharedRuntime::
2974 throw_NullPointerException),
2975 true);
2976
2977 StubRoutines::_throw_NullPointerException_at_call_entry =
2978 generate_throw_exception("NullPointerException at call throw_exception",
2979 CAST_FROM_FN_PTR(address,
2980 SharedRuntime::
2981 throw_NullPointerException_at_call),
2982 false);
2983
2984 StubRoutines::_throw_StackOverflowError_entry =
2985 generate_throw_exception("StackOverflowError throw_exception",
2986 CAST_FROM_FN_PTR(address,
2987 SharedRuntime::
2988 throw_StackOverflowError),
2989 false);
2990
2991 // entry points that are platform specific
2992 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
2993 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
2994 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
2995 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
2996
2997 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
2998 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
2999 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3000 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3001
3002 // support for verify_oop (must happen after universe_init)
3003 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3004
3005 // arraycopy stubs used by compilers
3006 generate_arraycopy_stubs();
3007
3008 generate_math_stubs();
3009 }
3010
3011 public:
3012 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3013 if (all) {
3014 generate_all();
3015 } else {
3016 generate_initial();
3017 }
3018 }
3019 }; // end class declaration
3020
3021 address StubGenerator::disjoint_byte_copy_entry = NULL;
3022 address StubGenerator::disjoint_short_copy_entry = NULL;
3023 address StubGenerator::disjoint_int_copy_entry = NULL;
3024 address StubGenerator::disjoint_long_copy_entry = NULL;
3025 address StubGenerator::disjoint_oop_copy_entry = NULL;
3026
3027 address StubGenerator::byte_copy_entry = NULL;
3028 address StubGenerator::short_copy_entry = NULL;
3029 address StubGenerator::int_copy_entry = NULL;
3030 address StubGenerator::long_copy_entry = NULL;
3031 address StubGenerator::oop_copy_entry = NULL;
3032
3033 address StubGenerator::checkcast_copy_entry = NULL;
3034
3035 void StubGenerator_generate(CodeBuffer* code, bool all) {
3036 StubGenerator g(code, all);
3037 }