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 and bashed
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 saved_r10 = 8 * wordSize,
939
940 // Before the call to MacroAssembler::debug(), see below.
941 return_addr = 16 * wordSize,
942 error_msg = 17 * wordSize
943 };
944
945 // get object
946 __ movptr(rax, Address(rsp, oop_to_verify));
947
948 // make sure object is 'reasonable'
949 __ testptr(rax, rax);
950 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
951 // Check if the oop is in the right area of memory
952 __ movptr(c_rarg2, rax);
953 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
954 __ andptr(c_rarg2, c_rarg3);
955 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
956 __ cmpptr(c_rarg2, c_rarg3);
957 __ jcc(Assembler::notZero, error);
958
959 // set r12 to heapbase for load_klass()
960 __ reinit_heapbase();
961
962 // make sure klass is 'reasonable'
963 __ load_klass(rax, rax); // get klass
964 __ testptr(rax, rax);
965 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
966 // Check if the klass is in the right area of memory
967 __ mov(c_rarg2, rax);
968 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
969 __ andptr(c_rarg2, c_rarg3);
970 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
971 __ cmpptr(c_rarg2, c_rarg3);
972 __ jcc(Assembler::notZero, error);
973
974 // make sure klass' klass is 'reasonable'
975 __ load_klass(rax, rax);
976 __ testptr(rax, rax);
977 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
978 // Check if the klass' klass is in the right area of memory
979 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
980 __ andptr(rax, c_rarg3);
981 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
982 __ cmpptr(rax, c_rarg3);
983 __ jcc(Assembler::notZero, error);
984
985 // return if everything seems ok
986 __ bind(exit);
987 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
988 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
989 __ pop(c_rarg3); // restore c_rarg3
990 __ pop(c_rarg2); // restore c_rarg2
991 __ pop(r12); // restore r12
992 __ popf(); // restore flags
993 __ ret(4 * wordSize); // pop caller saved stuff
994
995 // handle errors
996 __ bind(error);
997 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
998 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
999 __ pop(c_rarg3); // get saved c_rarg3 back
1000 __ pop(c_rarg2); // get saved c_rarg2 back
1001 __ pop(r12); // get saved r12 back
1002 __ popf(); // get saved flags off stack --
1003 // will be ignored
1004
1005 __ pusha(); // push registers
1006 // (rip is already
1007 // already pushed)
1008 // debug(char* msg, int64_t pc, int64_t regs[])
1009 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1010 // pushed all the registers, so now the stack looks like:
1011 // [tos + 0] 16 saved registers
1012 // [tos + 16] return address
1013 // * [tos + 17] error message (char*)
1014 // * [tos + 18] object to verify (oop)
1015 // * [tos + 19] saved rax - saved by caller and bashed
1016 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1017 // * = popped on exit
1018
1019 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1020 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1021 __ movq(c_rarg2, rsp); // pass address of regs on stack
1022 __ mov(r12, rsp); // remember rsp
1023 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1024 __ andptr(rsp, -16); // align stack as required by ABI
1025 BLOCK_COMMENT("call MacroAssembler::debug");
1026 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1027 __ mov(rsp, r12); // restore rsp
1028 __ popa(); // pop registers (includes r12)
1029 __ ret(4 * wordSize); // pop caller saved stuff
1030
1031 return start;
1032 }
1033
1034 static address disjoint_byte_copy_entry;
1035 static address disjoint_short_copy_entry;
1036 static address disjoint_int_copy_entry;
1037 static address disjoint_long_copy_entry;
1038 static address disjoint_oop_copy_entry;
1039
1040 static address byte_copy_entry;
1041 static address short_copy_entry;
1042 static address int_copy_entry;
1043 static address long_copy_entry;
1044 static address oop_copy_entry;
1045
1046 static address checkcast_copy_entry;
1047
1048 //
1049 // Verify that a register contains clean 32-bits positive value
1050 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1051 //
1052 // Input:
1053 // Rint - 32-bits value
1054 // Rtmp - scratch
1055 //
1056 void assert_clean_int(Register Rint, Register Rtmp) {
1057 #ifdef ASSERT
1058 Label L;
1059 assert_different_registers(Rtmp, Rint);
1060 __ movslq(Rtmp, Rint);
1061 __ cmpq(Rtmp, Rint);
1062 __ jcc(Assembler::equal, L);
1063 __ stop("high 32-bits of int value are not 0");
1064 __ bind(L);
1065 #endif
1066 }
1067
1068 // Generate overlap test for array copy stubs
1069 //
1070 // Input:
1071 // c_rarg0 - from
1072 // c_rarg1 - to
1073 // c_rarg2 - element count
1074 //
1075 // Output:
1076 // rax - &from[element count - 1]
1077 //
1078 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1079 assert(no_overlap_target != NULL, "must be generated");
1080 array_overlap_test(no_overlap_target, NULL, sf);
1081 }
1082 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1083 array_overlap_test(NULL, &L_no_overlap, sf);
1084 }
1085 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1086 const Register from = c_rarg0;
1087 const Register to = c_rarg1;
1088 const Register count = c_rarg2;
1089 const Register end_from = rax;
1090
1091 __ cmpptr(to, from);
1092 __ lea(end_from, Address(from, count, sf, 0));
1093 if (NOLp == NULL) {
1094 ExternalAddress no_overlap(no_overlap_target);
1095 __ jump_cc(Assembler::belowEqual, no_overlap);
1096 __ cmpptr(to, end_from);
1097 __ jump_cc(Assembler::aboveEqual, no_overlap);
1098 } else {
1099 __ jcc(Assembler::belowEqual, (*NOLp));
1100 __ cmpptr(to, end_from);
1101 __ jcc(Assembler::aboveEqual, (*NOLp));
1102 }
1103 }
1104
1105 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1106 //
1107 // Outputs:
1108 // rdi - rcx
1109 // rsi - rdx
1110 // rdx - r8
1111 // rcx - r9
1112 //
1113 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1114 // are non-volatile. r9 and r10 should not be used by the caller.
1115 //
1116 void setup_arg_regs(int nargs = 3) {
1117 const Register saved_rdi = r9;
1118 const Register saved_rsi = r10;
1119 assert(nargs == 3 || nargs == 4, "else fix");
1120 #ifdef _WIN64
1121 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1122 "unexpected argument registers");
1123 if (nargs >= 4)
1124 __ mov(rax, r9); // r9 is also saved_rdi
1125 __ movptr(saved_rdi, rdi);
1126 __ movptr(saved_rsi, rsi);
1127 __ mov(rdi, rcx); // c_rarg0
1128 __ mov(rsi, rdx); // c_rarg1
1129 __ mov(rdx, r8); // c_rarg2
1130 if (nargs >= 4)
1131 __ mov(rcx, rax); // c_rarg3 (via rax)
1132 #else
1133 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1134 "unexpected argument registers");
1135 #endif
1136 }
1137
1138 void restore_arg_regs() {
1139 const Register saved_rdi = r9;
1140 const Register saved_rsi = r10;
1141 #ifdef _WIN64
1142 __ movptr(rdi, saved_rdi);
1143 __ movptr(rsi, saved_rsi);
1144 #endif
1145 }
1146
1147 // Generate code for an array write pre barrier
1148 //
1149 // addr - starting address
1150 // count - element count
1151 //
1152 // Destroy no registers!
1153 //
1154 void gen_write_ref_array_pre_barrier(Register addr, Register count) {
1155 BarrierSet* bs = Universe::heap()->barrier_set();
1156 switch (bs->kind()) {
1157 case BarrierSet::G1SATBCT:
1158 case BarrierSet::G1SATBCTLogging:
1159 {
1160 __ pusha(); // push registers
1161 if (count == c_rarg0) {
1162 if (addr == c_rarg1) {
1163 // exactly backwards!!
1164 __ xchgptr(c_rarg1, c_rarg0);
1165 } else {
1166 __ movptr(c_rarg1, count);
1167 __ movptr(c_rarg0, addr);
1168 }
1169
1170 } else {
1171 __ movptr(c_rarg0, addr);
1172 __ movptr(c_rarg1, count);
1173 }
1174 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1175 __ popa();
1176 }
1177 break;
1178 case BarrierSet::CardTableModRef:
1179 case BarrierSet::CardTableExtension:
1180 case BarrierSet::ModRef:
1181 break;
1182 default:
1183 ShouldNotReachHere();
1184
1185 }
1186 }
1187
1188 //
1189 // Generate code for an array write post barrier
1190 //
1191 // Input:
1192 // start - register containing starting address of destination array
1193 // end - register containing ending address of destination array
1194 // scratch - scratch register
1195 //
1196 // The input registers are overwritten.
1197 // The ending address is inclusive.
1198 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1199 assert_different_registers(start, end, scratch);
1200 BarrierSet* bs = Universe::heap()->barrier_set();
1201 switch (bs->kind()) {
1202 case BarrierSet::G1SATBCT:
1203 case BarrierSet::G1SATBCTLogging:
1204
1205 {
1206 __ pusha(); // push registers (overkill)
1207 // must compute element count unless barrier set interface is changed (other platforms supply count)
1208 assert_different_registers(start, end, scratch);
1209 __ lea(scratch, Address(end, BytesPerHeapOop));
1210 __ subptr(scratch, start); // subtract start to get #bytes
1211 __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count
1212 __ mov(c_rarg0, start);
1213 __ mov(c_rarg1, scratch);
1214 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1215 __ popa();
1216 }
1217 break;
1218 case BarrierSet::CardTableModRef:
1219 case BarrierSet::CardTableExtension:
1220 {
1221 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1222 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1223
1224 Label L_loop;
1225
1226 __ shrptr(start, CardTableModRefBS::card_shift);
1227 __ addptr(end, BytesPerHeapOop);
1228 __ shrptr(end, CardTableModRefBS::card_shift);
1229 __ subptr(end, start); // number of bytes to copy
1230
1231 intptr_t disp = (intptr_t) ct->byte_map_base;
1232 if (__ is_simm32(disp)) {
1233 Address cardtable(noreg, noreg, Address::no_scale, disp);
1234 __ lea(scratch, cardtable);
1235 } else {
1236 ExternalAddress cardtable((address)disp);
1237 __ lea(scratch, cardtable);
1238 }
1239
1240 const Register count = end; // 'end' register contains bytes count now
1241 __ addptr(start, scratch);
1242 __ BIND(L_loop);
1243 __ movb(Address(start, count, Address::times_1), 0);
1244 __ decrement(count);
1245 __ jcc(Assembler::greaterEqual, L_loop);
1246 }
1247 break;
1248 default:
1249 ShouldNotReachHere();
1250
1251 }
1252 }
1253
1254
1255 // Copy big chunks forward
1256 //
1257 // Inputs:
1258 // end_from - source arrays end address
1259 // end_to - destination array end address
1260 // qword_count - 64-bits element count, negative
1261 // to - scratch
1262 // L_copy_32_bytes - entry label
1263 // L_copy_8_bytes - exit label
1264 //
1265 void copy_32_bytes_forward(Register end_from, Register end_to,
1266 Register qword_count, Register to,
1267 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1268 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1269 Label L_loop;
1270 __ align(OptoLoopAlignment);
1271 __ BIND(L_loop);
1272 if(UseUnalignedLoadStores) {
1273 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1274 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1275 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1276 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1277
1278 } else {
1279 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1280 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1281 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1282 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1283 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1284 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1285 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1286 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1287 }
1288 __ BIND(L_copy_32_bytes);
1289 __ addptr(qword_count, 4);
1290 __ jcc(Assembler::lessEqual, L_loop);
1291 __ subptr(qword_count, 4);
1292 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1293 }
1294
1295
1296 // Copy big chunks backward
1297 //
1298 // Inputs:
1299 // from - source arrays address
1300 // dest - destination array address
1301 // qword_count - 64-bits element count
1302 // to - scratch
1303 // L_copy_32_bytes - entry label
1304 // L_copy_8_bytes - exit label
1305 //
1306 void copy_32_bytes_backward(Register from, Register dest,
1307 Register qword_count, Register to,
1308 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1309 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1310 Label L_loop;
1311 __ align(OptoLoopAlignment);
1312 __ BIND(L_loop);
1313 if(UseUnalignedLoadStores) {
1314 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1315 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1316 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1317 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1318
1319 } else {
1320 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1321 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1322 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1323 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1324 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1325 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1326 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1327 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1328 }
1329 __ BIND(L_copy_32_bytes);
1330 __ subptr(qword_count, 4);
1331 __ jcc(Assembler::greaterEqual, L_loop);
1332 __ addptr(qword_count, 4);
1333 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1334 }
1335
1336
1337 // Arguments:
1338 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1339 // ignored
1340 // name - stub name string
1341 //
1342 // Inputs:
1343 // c_rarg0 - source array address
1344 // c_rarg1 - destination array address
1345 // c_rarg2 - element count, treated as ssize_t, can be zero
1346 //
1347 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1348 // we let the hardware handle it. The one to eight bytes within words,
1349 // dwords or qwords that span cache line boundaries will still be loaded
1350 // and stored atomically.
1351 //
1352 // Side Effects:
1353 // disjoint_byte_copy_entry is set to the no-overlap entry point
1354 // used by generate_conjoint_byte_copy().
1355 //
1356 address generate_disjoint_byte_copy(bool aligned, const char *name) {
1357 __ align(CodeEntryAlignment);
1358 StubCodeMark mark(this, "StubRoutines", name);
1359 address start = __ pc();
1360
1361 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1362 Label L_copy_byte, L_exit;
1363 const Register from = rdi; // source array address
1364 const Register to = rsi; // destination array address
1365 const Register count = rdx; // elements count
1366 const Register byte_count = rcx;
1367 const Register qword_count = count;
1368 const Register end_from = from; // source array end address
1369 const Register end_to = to; // destination array end address
1370 // End pointers are inclusive, and if count is not zero they point
1371 // to the last unit copied: end_to[0] := end_from[0]
1372
1373 __ enter(); // required for proper stackwalking of RuntimeStub frame
1374 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1375
1376 disjoint_byte_copy_entry = __ pc();
1377 BLOCK_COMMENT("Entry:");
1378 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1379
1380 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1381 // r9 and r10 may be used to save non-volatile registers
1382
1383 // 'from', 'to' and 'count' are now valid
1384 __ movptr(byte_count, count);
1385 __ shrptr(count, 3); // count => qword_count
1386
1387 // Copy from low to high addresses. Use 'to' as scratch.
1388 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1389 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1390 __ negptr(qword_count); // make the count negative
1391 __ jmp(L_copy_32_bytes);
1392
1393 // Copy trailing qwords
1394 __ BIND(L_copy_8_bytes);
1395 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1396 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1397 __ increment(qword_count);
1398 __ jcc(Assembler::notZero, L_copy_8_bytes);
1399
1400 // Check for and copy trailing dword
1401 __ BIND(L_copy_4_bytes);
1402 __ testl(byte_count, 4);
1403 __ jccb(Assembler::zero, L_copy_2_bytes);
1404 __ movl(rax, Address(end_from, 8));
1405 __ movl(Address(end_to, 8), rax);
1406
1407 __ addptr(end_from, 4);
1408 __ addptr(end_to, 4);
1409
1410 // Check for and copy trailing word
1411 __ BIND(L_copy_2_bytes);
1412 __ testl(byte_count, 2);
1413 __ jccb(Assembler::zero, L_copy_byte);
1414 __ movw(rax, Address(end_from, 8));
1415 __ movw(Address(end_to, 8), rax);
1416
1417 __ addptr(end_from, 2);
1418 __ addptr(end_to, 2);
1419
1420 // Check for and copy trailing byte
1421 __ BIND(L_copy_byte);
1422 __ testl(byte_count, 1);
1423 __ jccb(Assembler::zero, L_exit);
1424 __ movb(rax, Address(end_from, 8));
1425 __ movb(Address(end_to, 8), rax);
1426
1427 __ BIND(L_exit);
1428 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1429 restore_arg_regs();
1430 __ xorptr(rax, rax); // return 0
1431 __ leave(); // required for proper stackwalking of RuntimeStub frame
1432 __ ret(0);
1433
1434 // Copy in 32-bytes chunks
1435 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1436 __ jmp(L_copy_4_bytes);
1437
1438 return start;
1439 }
1440
1441 // Arguments:
1442 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1443 // ignored
1444 // name - stub name string
1445 //
1446 // Inputs:
1447 // c_rarg0 - source array address
1448 // c_rarg1 - destination array address
1449 // c_rarg2 - element count, treated as ssize_t, can be zero
1450 //
1451 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1452 // we let the hardware handle it. The one to eight bytes within words,
1453 // dwords or qwords that span cache line boundaries will still be loaded
1454 // and stored atomically.
1455 //
1456 address generate_conjoint_byte_copy(bool aligned, const char *name) {
1457 __ align(CodeEntryAlignment);
1458 StubCodeMark mark(this, "StubRoutines", name);
1459 address start = __ pc();
1460
1461 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1462 const Register from = rdi; // source array address
1463 const Register to = rsi; // destination array address
1464 const Register count = rdx; // elements count
1465 const Register byte_count = rcx;
1466 const Register qword_count = count;
1467
1468 __ enter(); // required for proper stackwalking of RuntimeStub frame
1469 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1470
1471 byte_copy_entry = __ pc();
1472 BLOCK_COMMENT("Entry:");
1473 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1474
1475 array_overlap_test(disjoint_byte_copy_entry, Address::times_1);
1476 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1477 // r9 and r10 may be used to save non-volatile registers
1478
1479 // 'from', 'to' and 'count' are now valid
1480 __ movptr(byte_count, count);
1481 __ shrptr(count, 3); // count => qword_count
1482
1483 // Copy from high to low addresses.
1484
1485 // Check for and copy trailing byte
1486 __ testl(byte_count, 1);
1487 __ jcc(Assembler::zero, L_copy_2_bytes);
1488 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1489 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1490 __ decrement(byte_count); // Adjust for possible trailing word
1491
1492 // Check for and copy trailing word
1493 __ BIND(L_copy_2_bytes);
1494 __ testl(byte_count, 2);
1495 __ jcc(Assembler::zero, L_copy_4_bytes);
1496 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1497 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1498
1499 // Check for and copy trailing dword
1500 __ BIND(L_copy_4_bytes);
1501 __ testl(byte_count, 4);
1502 __ jcc(Assembler::zero, L_copy_32_bytes);
1503 __ movl(rax, Address(from, qword_count, Address::times_8));
1504 __ movl(Address(to, qword_count, Address::times_8), rax);
1505 __ jmp(L_copy_32_bytes);
1506
1507 // Copy trailing qwords
1508 __ BIND(L_copy_8_bytes);
1509 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1510 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1511 __ decrement(qword_count);
1512 __ jcc(Assembler::notZero, L_copy_8_bytes);
1513
1514 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1515 restore_arg_regs();
1516 __ xorptr(rax, rax); // return 0
1517 __ leave(); // required for proper stackwalking of RuntimeStub frame
1518 __ ret(0);
1519
1520 // Copy in 32-bytes chunks
1521 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1522
1523 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1524 restore_arg_regs();
1525 __ xorptr(rax, rax); // return 0
1526 __ leave(); // required for proper stackwalking of RuntimeStub frame
1527 __ ret(0);
1528
1529 return start;
1530 }
1531
1532 // Arguments:
1533 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1534 // ignored
1535 // name - stub name string
1536 //
1537 // Inputs:
1538 // c_rarg0 - source array address
1539 // c_rarg1 - destination array address
1540 // c_rarg2 - element count, treated as ssize_t, can be zero
1541 //
1542 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1543 // let the hardware handle it. The two or four words within dwords
1544 // or qwords that span cache line boundaries will still be loaded
1545 // and stored atomically.
1546 //
1547 // Side Effects:
1548 // disjoint_short_copy_entry is set to the no-overlap entry point
1549 // used by generate_conjoint_short_copy().
1550 //
1551 address generate_disjoint_short_copy(bool aligned, const char *name) {
1552 __ align(CodeEntryAlignment);
1553 StubCodeMark mark(this, "StubRoutines", name);
1554 address start = __ pc();
1555
1556 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1557 const Register from = rdi; // source array address
1558 const Register to = rsi; // destination array address
1559 const Register count = rdx; // elements count
1560 const Register word_count = rcx;
1561 const Register qword_count = count;
1562 const Register end_from = from; // source array end address
1563 const Register end_to = to; // destination array end address
1564 // End pointers are inclusive, and if count is not zero they point
1565 // to the last unit copied: end_to[0] := end_from[0]
1566
1567 __ enter(); // required for proper stackwalking of RuntimeStub frame
1568 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1569
1570 disjoint_short_copy_entry = __ pc();
1571 BLOCK_COMMENT("Entry:");
1572 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1573
1574 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1575 // r9 and r10 may be used to save non-volatile registers
1576
1577 // 'from', 'to' and 'count' are now valid
1578 __ movptr(word_count, count);
1579 __ shrptr(count, 2); // count => qword_count
1580
1581 // Copy from low to high addresses. Use 'to' as scratch.
1582 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1583 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1584 __ negptr(qword_count);
1585 __ jmp(L_copy_32_bytes);
1586
1587 // Copy trailing qwords
1588 __ BIND(L_copy_8_bytes);
1589 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1590 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1591 __ increment(qword_count);
1592 __ jcc(Assembler::notZero, L_copy_8_bytes);
1593
1594 // Original 'dest' is trashed, so we can't use it as a
1595 // base register for a possible trailing word copy
1596
1597 // Check for and copy trailing dword
1598 __ BIND(L_copy_4_bytes);
1599 __ testl(word_count, 2);
1600 __ jccb(Assembler::zero, L_copy_2_bytes);
1601 __ movl(rax, Address(end_from, 8));
1602 __ movl(Address(end_to, 8), rax);
1603
1604 __ addptr(end_from, 4);
1605 __ addptr(end_to, 4);
1606
1607 // Check for and copy trailing word
1608 __ BIND(L_copy_2_bytes);
1609 __ testl(word_count, 1);
1610 __ jccb(Assembler::zero, L_exit);
1611 __ movw(rax, Address(end_from, 8));
1612 __ movw(Address(end_to, 8), rax);
1613
1614 __ BIND(L_exit);
1615 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1616 restore_arg_regs();
1617 __ xorptr(rax, rax); // return 0
1618 __ leave(); // required for proper stackwalking of RuntimeStub frame
1619 __ ret(0);
1620
1621 // Copy in 32-bytes chunks
1622 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1623 __ jmp(L_copy_4_bytes);
1624
1625 return start;
1626 }
1627
1628 // Arguments:
1629 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1630 // ignored
1631 // name - stub name string
1632 //
1633 // Inputs:
1634 // c_rarg0 - source array address
1635 // c_rarg1 - destination array address
1636 // c_rarg2 - element count, treated as ssize_t, can be zero
1637 //
1638 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1639 // let the hardware handle it. The two or four words within dwords
1640 // or qwords that span cache line boundaries will still be loaded
1641 // and stored atomically.
1642 //
1643 address generate_conjoint_short_copy(bool aligned, const char *name) {
1644 __ align(CodeEntryAlignment);
1645 StubCodeMark mark(this, "StubRoutines", name);
1646 address start = __ pc();
1647
1648 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1649 const Register from = rdi; // source array address
1650 const Register to = rsi; // destination array address
1651 const Register count = rdx; // elements count
1652 const Register word_count = rcx;
1653 const Register qword_count = count;
1654
1655 __ enter(); // required for proper stackwalking of RuntimeStub frame
1656 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1657
1658 short_copy_entry = __ pc();
1659 BLOCK_COMMENT("Entry:");
1660 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1661
1662 array_overlap_test(disjoint_short_copy_entry, Address::times_2);
1663 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1664 // r9 and r10 may be used to save non-volatile registers
1665
1666 // 'from', 'to' and 'count' are now valid
1667 __ movptr(word_count, count);
1668 __ shrptr(count, 2); // count => qword_count
1669
1670 // Copy from high to low addresses. Use 'to' as scratch.
1671
1672 // Check for and copy trailing word
1673 __ testl(word_count, 1);
1674 __ jccb(Assembler::zero, L_copy_4_bytes);
1675 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1676 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1677
1678 // Check for and copy trailing dword
1679 __ BIND(L_copy_4_bytes);
1680 __ testl(word_count, 2);
1681 __ jcc(Assembler::zero, L_copy_32_bytes);
1682 __ movl(rax, Address(from, qword_count, Address::times_8));
1683 __ movl(Address(to, qword_count, Address::times_8), rax);
1684 __ jmp(L_copy_32_bytes);
1685
1686 // Copy trailing qwords
1687 __ BIND(L_copy_8_bytes);
1688 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1689 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1690 __ decrement(qword_count);
1691 __ jcc(Assembler::notZero, L_copy_8_bytes);
1692
1693 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1694 restore_arg_regs();
1695 __ xorptr(rax, rax); // return 0
1696 __ leave(); // required for proper stackwalking of RuntimeStub frame
1697 __ ret(0);
1698
1699 // Copy in 32-bytes chunks
1700 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1701
1702 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1703 restore_arg_regs();
1704 __ xorptr(rax, rax); // return 0
1705 __ leave(); // required for proper stackwalking of RuntimeStub frame
1706 __ ret(0);
1707
1708 return start;
1709 }
1710
1711 // Arguments:
1712 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1713 // ignored
1714 // is_oop - true => oop array, so generate store check code
1715 // name - stub name string
1716 //
1717 // Inputs:
1718 // c_rarg0 - source array address
1719 // c_rarg1 - destination array address
1720 // c_rarg2 - element count, treated as ssize_t, can be zero
1721 //
1722 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1723 // the hardware handle it. The two dwords within qwords that span
1724 // cache line boundaries will still be loaded and stored atomicly.
1725 //
1726 // Side Effects:
1727 // disjoint_int_copy_entry is set to the no-overlap entry point
1728 // used by generate_conjoint_int_oop_copy().
1729 //
1730 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1731 __ align(CodeEntryAlignment);
1732 StubCodeMark mark(this, "StubRoutines", name);
1733 address start = __ pc();
1734
1735 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1736 const Register from = rdi; // source array address
1737 const Register to = rsi; // destination array address
1738 const Register count = rdx; // elements count
1739 const Register dword_count = rcx;
1740 const Register qword_count = count;
1741 const Register end_from = from; // source array end address
1742 const Register end_to = to; // destination array end address
1743 const Register saved_to = r11; // saved destination array address
1744 // End pointers are inclusive, and if count is not zero they point
1745 // to the last unit copied: end_to[0] := end_from[0]
1746
1747 __ enter(); // required for proper stackwalking of RuntimeStub frame
1748 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1749
1750 (is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry) = __ pc();
1751
1752 if (is_oop) {
1753 // no registers are destroyed by this call
1754 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1755 }
1756
1757 BLOCK_COMMENT("Entry:");
1758 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1759
1760 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1761 // r9 and r10 may be used to save non-volatile registers
1762
1763 if (is_oop) {
1764 __ movq(saved_to, to);
1765 }
1766
1767 // 'from', 'to' and 'count' are now valid
1768 __ movptr(dword_count, count);
1769 __ shrptr(count, 1); // count => qword_count
1770
1771 // Copy from low to high addresses. Use 'to' as scratch.
1772 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1773 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1774 __ negptr(qword_count);
1775 __ jmp(L_copy_32_bytes);
1776
1777 // Copy trailing qwords
1778 __ BIND(L_copy_8_bytes);
1779 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1780 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1781 __ increment(qword_count);
1782 __ jcc(Assembler::notZero, L_copy_8_bytes);
1783
1784 // Check for and copy trailing dword
1785 __ BIND(L_copy_4_bytes);
1786 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1787 __ jccb(Assembler::zero, L_exit);
1788 __ movl(rax, Address(end_from, 8));
1789 __ movl(Address(end_to, 8), rax);
1790
1791 __ BIND(L_exit);
1792 if (is_oop) {
1793 __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1794 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1795 }
1796 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1797 restore_arg_regs();
1798 __ xorptr(rax, rax); // return 0
1799 __ leave(); // required for proper stackwalking of RuntimeStub frame
1800 __ ret(0);
1801
1802 // Copy 32-bytes chunks
1803 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1804 __ jmp(L_copy_4_bytes);
1805
1806 return start;
1807 }
1808
1809 // Arguments:
1810 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1811 // ignored
1812 // is_oop - true => oop array, so generate store check code
1813 // name - stub name string
1814 //
1815 // Inputs:
1816 // c_rarg0 - source array address
1817 // c_rarg1 - destination array address
1818 // c_rarg2 - element count, treated as ssize_t, can be zero
1819 //
1820 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1821 // the hardware handle it. The two dwords within qwords that span
1822 // cache line boundaries will still be loaded and stored atomicly.
1823 //
1824 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1825 __ align(CodeEntryAlignment);
1826 StubCodeMark mark(this, "StubRoutines", name);
1827 address start = __ pc();
1828
1829 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1830 const Register from = rdi; // source array address
1831 const Register to = rsi; // destination array address
1832 const Register count = rdx; // elements count
1833 const Register dword_count = rcx;
1834 const Register qword_count = count;
1835
1836 __ enter(); // required for proper stackwalking of RuntimeStub frame
1837 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1838
1839 if (is_oop) {
1840 // no registers are destroyed by this call
1841 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1842 }
1843
1844 (is_oop ? oop_copy_entry : int_copy_entry) = __ pc();
1845 BLOCK_COMMENT("Entry:");
1846 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1847
1848 array_overlap_test(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry,
1849 Address::times_4);
1850 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1851 // r9 and r10 may be used to save non-volatile registers
1852
1853 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1854 // 'from', 'to' and 'count' are now valid
1855 __ movptr(dword_count, count);
1856 __ shrptr(count, 1); // count => qword_count
1857
1858 // Copy from high to low addresses. Use 'to' as scratch.
1859
1860 // Check for and copy trailing dword
1861 __ testl(dword_count, 1);
1862 __ jcc(Assembler::zero, L_copy_32_bytes);
1863 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1864 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1865 __ jmp(L_copy_32_bytes);
1866
1867 // Copy trailing qwords
1868 __ BIND(L_copy_8_bytes);
1869 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1870 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1871 __ decrement(qword_count);
1872 __ jcc(Assembler::notZero, L_copy_8_bytes);
1873
1874 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1875 if (is_oop) {
1876 __ jmp(L_exit);
1877 }
1878 restore_arg_regs();
1879 __ xorptr(rax, rax); // return 0
1880 __ leave(); // required for proper stackwalking of RuntimeStub frame
1881 __ ret(0);
1882
1883 // Copy in 32-bytes chunks
1884 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1885
1886 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1887 __ bind(L_exit);
1888 if (is_oop) {
1889 Register end_to = rdx;
1890 __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1891 gen_write_ref_array_post_barrier(to, end_to, rax);
1892 }
1893 restore_arg_regs();
1894 __ xorptr(rax, rax); // return 0
1895 __ leave(); // required for proper stackwalking of RuntimeStub frame
1896 __ ret(0);
1897
1898 return start;
1899 }
1900
1901 // Arguments:
1902 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1903 // ignored
1904 // is_oop - true => oop array, so generate store check code
1905 // name - stub name string
1906 //
1907 // Inputs:
1908 // c_rarg0 - source array address
1909 // c_rarg1 - destination array address
1910 // c_rarg2 - element count, treated as ssize_t, can be zero
1911 //
1912 // Side Effects:
1913 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1914 // no-overlap entry point used by generate_conjoint_long_oop_copy().
1915 //
1916 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1917 __ align(CodeEntryAlignment);
1918 StubCodeMark mark(this, "StubRoutines", name);
1919 address start = __ pc();
1920
1921 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1922 const Register from = rdi; // source array address
1923 const Register to = rsi; // destination array address
1924 const Register qword_count = rdx; // elements count
1925 const Register end_from = from; // source array end address
1926 const Register end_to = rcx; // destination array end address
1927 const Register saved_to = to;
1928 // End pointers are inclusive, and if count is not zero they point
1929 // to the last unit copied: end_to[0] := end_from[0]
1930
1931 __ enter(); // required for proper stackwalking of RuntimeStub frame
1932 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1933 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1934
1935 if (is_oop) {
1936 disjoint_oop_copy_entry = __ pc();
1937 // no registers are destroyed by this call
1938 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1939 } else {
1940 disjoint_long_copy_entry = __ pc();
1941 }
1942 BLOCK_COMMENT("Entry:");
1943 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1944
1945 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1946 // r9 and r10 may be used to save non-volatile registers
1947
1948 // 'from', 'to' and 'qword_count' are now valid
1949
1950 // Copy from low to high addresses. Use 'to' as scratch.
1951 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1952 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1953 __ negptr(qword_count);
1954 __ jmp(L_copy_32_bytes);
1955
1956 // Copy trailing qwords
1957 __ BIND(L_copy_8_bytes);
1958 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1959 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1960 __ increment(qword_count);
1961 __ jcc(Assembler::notZero, L_copy_8_bytes);
1962
1963 if (is_oop) {
1964 __ jmp(L_exit);
1965 } else {
1966 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1967 restore_arg_regs();
1968 __ xorptr(rax, rax); // return 0
1969 __ leave(); // required for proper stackwalking of RuntimeStub frame
1970 __ ret(0);
1971 }
1972
1973 // Copy 64-byte chunks
1974 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1975
1976 if (is_oop) {
1977 __ BIND(L_exit);
1978 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1979 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
1980 } else {
1981 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1982 }
1983 restore_arg_regs();
1984 __ xorptr(rax, rax); // return 0
1985 __ leave(); // required for proper stackwalking of RuntimeStub frame
1986 __ ret(0);
1987
1988 return start;
1989 }
1990
1991 // Arguments:
1992 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1993 // ignored
1994 // is_oop - true => oop array, so generate store check code
1995 // name - stub name string
1996 //
1997 // Inputs:
1998 // c_rarg0 - source array address
1999 // c_rarg1 - destination array address
2000 // c_rarg2 - element count, treated as ssize_t, can be zero
2001 //
2002 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
2003 __ align(CodeEntryAlignment);
2004 StubCodeMark mark(this, "StubRoutines", name);
2005 address start = __ pc();
2006
2007 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2008 const Register from = rdi; // source array address
2009 const Register to = rsi; // destination array address
2010 const Register qword_count = rdx; // elements count
2011 const Register saved_count = rcx;
2012
2013 __ enter(); // required for proper stackwalking of RuntimeStub frame
2014 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2015
2016 address disjoint_copy_entry = NULL;
2017 if (is_oop) {
2018 assert(!UseCompressedOops, "shouldn't be called for compressed oops");
2019 disjoint_copy_entry = disjoint_oop_copy_entry;
2020 oop_copy_entry = __ pc();
2021 array_overlap_test(disjoint_oop_copy_entry, Address::times_8);
2022 } else {
2023 disjoint_copy_entry = disjoint_long_copy_entry;
2024 long_copy_entry = __ pc();
2025 array_overlap_test(disjoint_long_copy_entry, Address::times_8);
2026 }
2027 BLOCK_COMMENT("Entry:");
2028 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2029
2030 array_overlap_test(disjoint_copy_entry, Address::times_8);
2031 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2032 // r9 and r10 may be used to save non-volatile registers
2033
2034 // 'from', 'to' and 'qword_count' are now valid
2035
2036 if (is_oop) {
2037 // Save to and count for store barrier
2038 __ movptr(saved_count, qword_count);
2039 // No registers are destroyed by this call
2040 gen_write_ref_array_pre_barrier(to, saved_count);
2041 }
2042
2043 __ jmp(L_copy_32_bytes);
2044
2045 // Copy trailing qwords
2046 __ BIND(L_copy_8_bytes);
2047 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2048 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2049 __ decrement(qword_count);
2050 __ jcc(Assembler::notZero, L_copy_8_bytes);
2051
2052 if (is_oop) {
2053 __ jmp(L_exit);
2054 } else {
2055 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2056 restore_arg_regs();
2057 __ xorptr(rax, rax); // return 0
2058 __ leave(); // required for proper stackwalking of RuntimeStub frame
2059 __ ret(0);
2060 }
2061
2062 // Copy in 32-bytes chunks
2063 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2064
2065 if (is_oop) {
2066 __ BIND(L_exit);
2067 __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2068 gen_write_ref_array_post_barrier(to, rcx, rax);
2069 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2070 } else {
2071 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2072 }
2073 restore_arg_regs();
2074 __ xorptr(rax, rax); // return 0
2075 __ leave(); // required for proper stackwalking of RuntimeStub frame
2076 __ ret(0);
2077
2078 return start;
2079 }
2080
2081
2082 // Helper for generating a dynamic type check.
2083 // Smashes no registers.
2084 void generate_type_check(Register sub_klass,
2085 Register super_check_offset,
2086 Register super_klass,
2087 Label& L_success) {
2088 assert_different_registers(sub_klass, super_check_offset, super_klass);
2089
2090 BLOCK_COMMENT("type_check:");
2091
2092 Label L_miss;
2093
2094 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2095 super_check_offset);
2096 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2097
2098 // Fall through on failure!
2099 __ BIND(L_miss);
2100 }
2101
2102 //
2103 // Generate checkcasting array copy stub
2104 //
2105 // Input:
2106 // c_rarg0 - source array address
2107 // c_rarg1 - destination array address
2108 // c_rarg2 - element count, treated as ssize_t, can be zero
2109 // c_rarg3 - size_t ckoff (super_check_offset)
2110 // not Win64
2111 // c_rarg4 - oop ckval (super_klass)
2112 // Win64
2113 // rsp+40 - oop ckval (super_klass)
2114 //
2115 // Output:
2116 // rax == 0 - success
2117 // rax == -1^K - failure, where K is partial transfer count
2118 //
2119 address generate_checkcast_copy(const char *name) {
2120
2121 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2122
2123 // Input registers (after setup_arg_regs)
2124 const Register from = rdi; // source array address
2125 const Register to = rsi; // destination array address
2126 const Register length = rdx; // elements count
2127 const Register ckoff = rcx; // super_check_offset
2128 const Register ckval = r8; // super_klass
2129
2130 // Registers used as temps (r13, r14 are save-on-entry)
2131 const Register end_from = from; // source array end address
2132 const Register end_to = r13; // destination array end address
2133 const Register count = rdx; // -(count_remaining)
2134 const Register r14_length = r14; // saved copy of length
2135 // End pointers are inclusive, and if length is not zero they point
2136 // to the last unit copied: end_to[0] := end_from[0]
2137
2138 const Register rax_oop = rax; // actual oop copied
2139 const Register r11_klass = r11; // oop._klass
2140
2141 //---------------------------------------------------------------
2142 // Assembler stub will be used for this call to arraycopy
2143 // if the two arrays are subtypes of Object[] but the
2144 // destination array type is not equal to or a supertype
2145 // of the source type. Each element must be separately
2146 // checked.
2147
2148 __ align(CodeEntryAlignment);
2149 StubCodeMark mark(this, "StubRoutines", name);
2150 address start = __ pc();
2151
2152 __ enter(); // required for proper stackwalking of RuntimeStub frame
2153
2154 checkcast_copy_entry = __ pc();
2155 BLOCK_COMMENT("Entry:");
2156
2157 #ifdef ASSERT
2158 // caller guarantees that the arrays really are different
2159 // otherwise, we would have to make conjoint checks
2160 { Label L;
2161 array_overlap_test(L, TIMES_OOP);
2162 __ stop("checkcast_copy within a single array");
2163 __ bind(L);
2164 }
2165 #endif //ASSERT
2166
2167 // allocate spill slots for r13, r14
2168 enum {
2169 saved_r13_offset,
2170 saved_r14_offset,
2171 saved_rbp_offset,
2172 saved_rip_offset,
2173 saved_rarg0_offset
2174 };
2175 __ subptr(rsp, saved_rbp_offset * wordSize);
2176 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2177 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2178 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2179 // ckoff => rcx, ckval => r8
2180 // r9 and r10 may be used to save non-volatile registers
2181 #ifdef _WIN64
2182 // last argument (#4) is on stack on Win64
2183 const int ckval_offset = saved_rarg0_offset + 4;
2184 __ movptr(ckval, Address(rsp, ckval_offset * wordSize));
2185 #endif
2186
2187 // check that int operands are properly extended to size_t
2188 assert_clean_int(length, rax);
2189 assert_clean_int(ckoff, rax);
2190
2191 #ifdef ASSERT
2192 BLOCK_COMMENT("assert consistent ckoff/ckval");
2193 // The ckoff and ckval must be mutually consistent,
2194 // even though caller generates both.
2195 { Label L;
2196 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2197 Klass::super_check_offset_offset_in_bytes());
2198 __ cmpl(ckoff, Address(ckval, sco_offset));
2199 __ jcc(Assembler::equal, L);
2200 __ stop("super_check_offset inconsistent");
2201 __ bind(L);
2202 }
2203 #endif //ASSERT
2204
2205 // Loop-invariant addresses. They are exclusive end pointers.
2206 Address end_from_addr(from, length, TIMES_OOP, 0);
2207 Address end_to_addr(to, length, TIMES_OOP, 0);
2208 // Loop-variant addresses. They assume post-incremented count < 0.
2209 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2210 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2211
2212 gen_write_ref_array_pre_barrier(to, count);
2213
2214 // Copy from low to high addresses, indexed from the end of each array.
2215 __ lea(end_from, end_from_addr);
2216 __ lea(end_to, end_to_addr);
2217 __ movptr(r14_length, length); // save a copy of the length
2218 assert(length == count, ""); // else fix next line:
2219 __ negptr(count); // negate and test the length
2220 __ jcc(Assembler::notZero, L_load_element);
2221
2222 // Empty array: Nothing to do.
2223 __ xorptr(rax, rax); // return 0 on (trivial) success
2224 __ jmp(L_done);
2225
2226 // ======== begin loop ========
2227 // (Loop is rotated; its entry is L_load_element.)
2228 // Loop control:
2229 // for (count = -count; count != 0; count++)
2230 // Base pointers src, dst are biased by 8*(count-1),to last element.
2231 __ align(OptoLoopAlignment);
2232
2233 __ BIND(L_store_element);
2234 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2235 __ increment(count); // increment the count toward zero
2236 __ jcc(Assembler::zero, L_do_card_marks);
2237
2238 // ======== loop entry is here ========
2239 __ BIND(L_load_element);
2240 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2241 __ testptr(rax_oop, rax_oop);
2242 __ jcc(Assembler::zero, L_store_element);
2243
2244 __ load_klass(r11_klass, rax_oop);// query the object klass
2245 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2246 // ======== end loop ========
2247
2248 // It was a real error; we must depend on the caller to finish the job.
2249 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2250 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2251 // and report their number to the caller.
2252 assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2253 __ lea(end_to, to_element_addr);
2254 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2255 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2256 __ movptr(rax, r14_length); // original oops
2257 __ addptr(rax, count); // K = (original - remaining) oops
2258 __ notptr(rax); // report (-1^K) to caller
2259 __ jmp(L_done);
2260
2261 // Come here on success only.
2262 __ BIND(L_do_card_marks);
2263 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2264 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2265 __ xorptr(rax, rax); // return 0 on success
2266
2267 // Common exit point (success or failure).
2268 __ BIND(L_done);
2269 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2270 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2271 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
2272 restore_arg_regs();
2273 __ leave(); // required for proper stackwalking of RuntimeStub frame
2274 __ ret(0);
2275
2276 return start;
2277 }
2278
2279 //
2280 // Generate 'unsafe' array copy stub
2281 // Though just as safe as the other stubs, it takes an unscaled
2282 // size_t argument instead of an element count.
2283 //
2284 // Input:
2285 // c_rarg0 - source array address
2286 // c_rarg1 - destination array address
2287 // c_rarg2 - byte count, treated as ssize_t, can be zero
2288 //
2289 // Examines the alignment of the operands and dispatches
2290 // to a long, int, short, or byte copy loop.
2291 //
2292 address generate_unsafe_copy(const char *name) {
2293
2294 Label L_long_aligned, L_int_aligned, L_short_aligned;
2295
2296 // Input registers (before setup_arg_regs)
2297 const Register from = c_rarg0; // source array address
2298 const Register to = c_rarg1; // destination array address
2299 const Register size = c_rarg2; // byte count (size_t)
2300
2301 // Register used as a temp
2302 const Register bits = rax; // test copy of low bits
2303
2304 __ align(CodeEntryAlignment);
2305 StubCodeMark mark(this, "StubRoutines", name);
2306 address start = __ pc();
2307
2308 __ enter(); // required for proper stackwalking of RuntimeStub frame
2309
2310 // bump this on entry, not on exit:
2311 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2312
2313 __ mov(bits, from);
2314 __ orptr(bits, to);
2315 __ orptr(bits, size);
2316
2317 __ testb(bits, BytesPerLong-1);
2318 __ jccb(Assembler::zero, L_long_aligned);
2319
2320 __ testb(bits, BytesPerInt-1);
2321 __ jccb(Assembler::zero, L_int_aligned);
2322
2323 __ testb(bits, BytesPerShort-1);
2324 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2325
2326 __ BIND(L_short_aligned);
2327 __ shrptr(size, LogBytesPerShort); // size => short_count
2328 __ jump(RuntimeAddress(short_copy_entry));
2329
2330 __ BIND(L_int_aligned);
2331 __ shrptr(size, LogBytesPerInt); // size => int_count
2332 __ jump(RuntimeAddress(int_copy_entry));
2333
2334 __ BIND(L_long_aligned);
2335 __ shrptr(size, LogBytesPerLong); // size => qword_count
2336 __ jump(RuntimeAddress(long_copy_entry));
2337
2338 return start;
2339 }
2340
2341 // Perform range checks on the proposed arraycopy.
2342 // Kills temp, but nothing else.
2343 // Also, clean the sign bits of src_pos and dst_pos.
2344 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2345 Register src_pos, // source position (c_rarg1)
2346 Register dst, // destination array oo (c_rarg2)
2347 Register dst_pos, // destination position (c_rarg3)
2348 Register length,
2349 Register temp,
2350 Label& L_failed) {
2351 BLOCK_COMMENT("arraycopy_range_checks:");
2352
2353 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2354 __ movl(temp, length);
2355 __ addl(temp, src_pos); // src_pos + length
2356 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2357 __ jcc(Assembler::above, L_failed);
2358
2359 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2360 __ movl(temp, length);
2361 __ addl(temp, dst_pos); // dst_pos + length
2362 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2363 __ jcc(Assembler::above, L_failed);
2364
2365 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2366 // Move with sign extension can be used since they are positive.
2367 __ movslq(src_pos, src_pos);
2368 __ movslq(dst_pos, dst_pos);
2369
2370 BLOCK_COMMENT("arraycopy_range_checks done");
2371 }
2372
2373 //
2374 // Generate generic array copy stubs
2375 //
2376 // Input:
2377 // c_rarg0 - src oop
2378 // c_rarg1 - src_pos (32-bits)
2379 // c_rarg2 - dst oop
2380 // c_rarg3 - dst_pos (32-bits)
2381 // not Win64
2382 // c_rarg4 - element count (32-bits)
2383 // Win64
2384 // rsp+40 - element count (32-bits)
2385 //
2386 // Output:
2387 // rax == 0 - success
2388 // rax == -1^K - failure, where K is partial transfer count
2389 //
2390 address generate_generic_copy(const char *name) {
2391
2392 Label L_failed, L_failed_0, L_objArray;
2393 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2394
2395 // Input registers
2396 const Register src = c_rarg0; // source array oop
2397 const Register src_pos = c_rarg1; // source position
2398 const Register dst = c_rarg2; // destination array oop
2399 const Register dst_pos = c_rarg3; // destination position
2400 // elements count is on stack on Win64
2401 #ifdef _WIN64
2402 #define C_RARG4 Address(rsp, 6 * wordSize)
2403 #else
2404 #define C_RARG4 c_rarg4
2405 #endif
2406
2407 { int modulus = CodeEntryAlignment;
2408 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2409 int advance = target - (__ offset() % modulus);
2410 if (advance < 0) advance += modulus;
2411 if (advance > 0) __ nop(advance);
2412 }
2413 StubCodeMark mark(this, "StubRoutines", name);
2414
2415 // Short-hop target to L_failed. Makes for denser prologue code.
2416 __ BIND(L_failed_0);
2417 __ jmp(L_failed);
2418 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2419
2420 __ align(CodeEntryAlignment);
2421 address start = __ pc();
2422
2423 __ enter(); // required for proper stackwalking of RuntimeStub frame
2424
2425 // bump this on entry, not on exit:
2426 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2427
2428 //-----------------------------------------------------------------------
2429 // Assembler stub will be used for this call to arraycopy
2430 // if the following conditions are met:
2431 //
2432 // (1) src and dst must not be null.
2433 // (2) src_pos must not be negative.
2434 // (3) dst_pos must not be negative.
2435 // (4) length must not be negative.
2436 // (5) src klass and dst klass should be the same and not NULL.
2437 // (6) src and dst should be arrays.
2438 // (7) src_pos + length must not exceed length of src.
2439 // (8) dst_pos + length must not exceed length of dst.
2440 //
2441
2442 // if (src == NULL) return -1;
2443 __ testptr(src, src); // src oop
2444 size_t j1off = __ offset();
2445 __ jccb(Assembler::zero, L_failed_0);
2446
2447 // if (src_pos < 0) return -1;
2448 __ testl(src_pos, src_pos); // src_pos (32-bits)
2449 __ jccb(Assembler::negative, L_failed_0);
2450
2451 // if (dst == NULL) return -1;
2452 __ testptr(dst, dst); // dst oop
2453 __ jccb(Assembler::zero, L_failed_0);
2454
2455 // if (dst_pos < 0) return -1;
2456 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2457 size_t j4off = __ offset();
2458 __ jccb(Assembler::negative, L_failed_0);
2459
2460 // The first four tests are very dense code,
2461 // but not quite dense enough to put four
2462 // jumps in a 16-byte instruction fetch buffer.
2463 // That's good, because some branch predicters
2464 // do not like jumps so close together.
2465 // Make sure of this.
2466 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2467
2468 // registers used as temp
2469 const Register r11_length = r11; // elements count to copy
2470 const Register r10_src_klass = r10; // array klass
2471 const Register r9_dst_klass = r9; // dest array klass
2472
2473 // if (length < 0) return -1;
2474 __ movl(r11_length, C_RARG4); // length (elements count, 32-bits value)
2475 __ testl(r11_length, r11_length);
2476 __ jccb(Assembler::negative, L_failed_0);
2477
2478 __ load_klass(r10_src_klass, src);
2479 #ifdef ASSERT
2480 // assert(src->klass() != NULL);
2481 BLOCK_COMMENT("assert klasses not null");
2482 { Label L1, L2;
2483 __ testptr(r10_src_klass, r10_src_klass);
2484 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2485 __ bind(L1);
2486 __ stop("broken null klass");
2487 __ bind(L2);
2488 __ load_klass(r9_dst_klass, dst);
2489 __ cmpq(r9_dst_klass, 0);
2490 __ jcc(Assembler::equal, L1); // this would be broken also
2491 BLOCK_COMMENT("assert done");
2492 }
2493 #endif
2494
2495 // Load layout helper (32-bits)
2496 //
2497 // |array_tag| | header_size | element_type | |log2_element_size|
2498 // 32 30 24 16 8 2 0
2499 //
2500 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2501 //
2502
2503 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2504 Klass::layout_helper_offset_in_bytes();
2505
2506 const Register rax_lh = rax; // layout helper
2507
2508 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2509
2510 // Handle objArrays completely differently...
2511 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2512 __ cmpl(rax_lh, objArray_lh);
2513 __ jcc(Assembler::equal, L_objArray);
2514
2515 // if (src->klass() != dst->klass()) return -1;
2516 __ load_klass(r9_dst_klass, dst);
2517 __ cmpq(r10_src_klass, r9_dst_klass);
2518 __ jcc(Assembler::notEqual, L_failed);
2519
2520 // if (!src->is_Array()) return -1;
2521 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2522 __ jcc(Assembler::greaterEqual, L_failed);
2523
2524 // At this point, it is known to be a typeArray (array_tag 0x3).
2525 #ifdef ASSERT
2526 { Label L;
2527 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2528 __ jcc(Assembler::greaterEqual, L);
2529 __ stop("must be a primitive array");
2530 __ bind(L);
2531 }
2532 #endif
2533
2534 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2535 r10, L_failed);
2536
2537 // typeArrayKlass
2538 //
2539 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2540 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2541 //
2542
2543 const Register r10_offset = r10; // array offset
2544 const Register rax_elsize = rax_lh; // element size
2545
2546 __ movl(r10_offset, rax_lh);
2547 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2548 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2549 __ addptr(src, r10_offset); // src array offset
2550 __ addptr(dst, r10_offset); // dst array offset
2551 BLOCK_COMMENT("choose copy loop based on element size");
2552 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2553
2554 // next registers should be set before the jump to corresponding stub
2555 const Register from = c_rarg0; // source array address
2556 const Register to = c_rarg1; // destination array address
2557 const Register count = c_rarg2; // elements count
2558
2559 // 'from', 'to', 'count' registers should be set in such order
2560 // since they are the same as 'src', 'src_pos', 'dst'.
2561
2562 __ BIND(L_copy_bytes);
2563 __ cmpl(rax_elsize, 0);
2564 __ jccb(Assembler::notEqual, L_copy_shorts);
2565 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2566 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2567 __ movl2ptr(count, r11_length); // length
2568 __ jump(RuntimeAddress(byte_copy_entry));
2569
2570 __ BIND(L_copy_shorts);
2571 __ cmpl(rax_elsize, LogBytesPerShort);
2572 __ jccb(Assembler::notEqual, L_copy_ints);
2573 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2574 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2575 __ movl2ptr(count, r11_length); // length
2576 __ jump(RuntimeAddress(short_copy_entry));
2577
2578 __ BIND(L_copy_ints);
2579 __ cmpl(rax_elsize, LogBytesPerInt);
2580 __ jccb(Assembler::notEqual, L_copy_longs);
2581 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2582 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2583 __ movl2ptr(count, r11_length); // length
2584 __ jump(RuntimeAddress(int_copy_entry));
2585
2586 __ BIND(L_copy_longs);
2587 #ifdef ASSERT
2588 { Label L;
2589 __ cmpl(rax_elsize, LogBytesPerLong);
2590 __ jcc(Assembler::equal, L);
2591 __ stop("must be long copy, but elsize is wrong");
2592 __ bind(L);
2593 }
2594 #endif
2595 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2596 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2597 __ movl2ptr(count, r11_length); // length
2598 __ jump(RuntimeAddress(long_copy_entry));
2599
2600 // objArrayKlass
2601 __ BIND(L_objArray);
2602 // live at this point: r10_src_klass, src[_pos], dst[_pos]
2603
2604 Label L_plain_copy, L_checkcast_copy;
2605 // test array classes for subtyping
2606 __ load_klass(r9_dst_klass, dst);
2607 __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality
2608 __ jcc(Assembler::notEqual, L_checkcast_copy);
2609
2610 // Identically typed arrays can be copied without element-wise checks.
2611 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2612 r10, L_failed);
2613
2614 __ lea(from, Address(src, src_pos, TIMES_OOP,
2615 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2616 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2617 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2618 __ movl2ptr(count, r11_length); // length
2619 __ BIND(L_plain_copy);
2620 __ jump(RuntimeAddress(oop_copy_entry));
2621
2622 __ BIND(L_checkcast_copy);
2623 // live at this point: r10_src_klass, !r11_length
2624 {
2625 // assert(r11_length == C_RARG4); // will reload from here
2626 Register r11_dst_klass = r11;
2627 __ load_klass(r11_dst_klass, dst);
2628
2629 // Before looking at dst.length, make sure dst is also an objArray.
2630 __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh);
2631 __ jcc(Assembler::notEqual, L_failed);
2632
2633 // It is safe to examine both src.length and dst.length.
2634 #ifndef _WIN64
2635 arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4,
2636 rax, L_failed);
2637 #else
2638 __ movl(r11_length, C_RARG4); // reload
2639 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2640 rax, L_failed);
2641 __ load_klass(r11_dst_klass, dst); // reload
2642 #endif
2643
2644 // Marshal the base address arguments now, freeing registers.
2645 __ lea(from, Address(src, src_pos, TIMES_OOP,
2646 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2647 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2648 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2649 __ movl(count, C_RARG4); // length (reloaded)
2650 Register sco_temp = c_rarg3; // this register is free now
2651 assert_different_registers(from, to, count, sco_temp,
2652 r11_dst_klass, r10_src_klass);
2653 assert_clean_int(count, sco_temp);
2654
2655 // Generate the type check.
2656 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2657 Klass::super_check_offset_offset_in_bytes());
2658 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2659 assert_clean_int(sco_temp, rax);
2660 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2661
2662 // Fetch destination element klass from the objArrayKlass header.
2663 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2664 objArrayKlass::element_klass_offset_in_bytes());
2665 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2666 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2667 assert_clean_int(sco_temp, rax);
2668
2669 // the checkcast_copy loop needs two extra arguments:
2670 assert(c_rarg3 == sco_temp, "#3 already in place");
2671 __ movptr(C_RARG4, r11_dst_klass); // dst.klass.element_klass
2672 __ jump(RuntimeAddress(checkcast_copy_entry));
2673 }
2674
2675 __ BIND(L_failed);
2676 __ xorptr(rax, rax);
2677 __ notptr(rax); // return -1
2678 __ leave(); // required for proper stackwalking of RuntimeStub frame
2679 __ ret(0);
2680
2681 return start;
2682 }
2683
2684 #undef length_arg
2685
2686 void generate_arraycopy_stubs() {
2687 // Call the conjoint generation methods immediately after
2688 // the disjoint ones so that short branches from the former
2689 // to the latter can be generated.
2690 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
2691 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
2692
2693 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
2694 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
2695
2696 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy");
2697 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy");
2698
2699 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy");
2700 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy");
2701
2702
2703 if (UseCompressedOops) {
2704 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy");
2705 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy");
2706 } else {
2707 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy");
2708 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy");
2709 }
2710
2711 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
2712 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy");
2713 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy");
2714
2715 // We don't generate specialized code for HeapWord-aligned source
2716 // arrays, so just use the code we've already generated
2717 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2718 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2719
2720 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2721 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2722
2723 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2724 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2725
2726 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2727 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2728
2729 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2730 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2731 }
2732
2733 void generate_math_stubs() {
2734 {
2735 StubCodeMark mark(this, "StubRoutines", "log");
2736 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2737
2738 __ subq(rsp, 8);
2739 __ movdbl(Address(rsp, 0), xmm0);
2740 __ fld_d(Address(rsp, 0));
2741 __ flog();
2742 __ fstp_d(Address(rsp, 0));
2743 __ movdbl(xmm0, Address(rsp, 0));
2744 __ addq(rsp, 8);
2745 __ ret(0);
2746 }
2747 {
2748 StubCodeMark mark(this, "StubRoutines", "log10");
2749 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2750
2751 __ subq(rsp, 8);
2752 __ movdbl(Address(rsp, 0), xmm0);
2753 __ fld_d(Address(rsp, 0));
2754 __ flog10();
2755 __ fstp_d(Address(rsp, 0));
2756 __ movdbl(xmm0, Address(rsp, 0));
2757 __ addq(rsp, 8);
2758 __ ret(0);
2759 }
2760 {
2761 StubCodeMark mark(this, "StubRoutines", "sin");
2762 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2763
2764 __ subq(rsp, 8);
2765 __ movdbl(Address(rsp, 0), xmm0);
2766 __ fld_d(Address(rsp, 0));
2767 __ trigfunc('s');
2768 __ fstp_d(Address(rsp, 0));
2769 __ movdbl(xmm0, Address(rsp, 0));
2770 __ addq(rsp, 8);
2771 __ ret(0);
2772 }
2773 {
2774 StubCodeMark mark(this, "StubRoutines", "cos");
2775 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2776
2777 __ subq(rsp, 8);
2778 __ movdbl(Address(rsp, 0), xmm0);
2779 __ fld_d(Address(rsp, 0));
2780 __ trigfunc('c');
2781 __ fstp_d(Address(rsp, 0));
2782 __ movdbl(xmm0, Address(rsp, 0));
2783 __ addq(rsp, 8);
2784 __ ret(0);
2785 }
2786 {
2787 StubCodeMark mark(this, "StubRoutines", "tan");
2788 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2789
2790 __ subq(rsp, 8);
2791 __ movdbl(Address(rsp, 0), xmm0);
2792 __ fld_d(Address(rsp, 0));
2793 __ trigfunc('t');
2794 __ fstp_d(Address(rsp, 0));
2795 __ movdbl(xmm0, Address(rsp, 0));
2796 __ addq(rsp, 8);
2797 __ ret(0);
2798 }
2799
2800 // The intrinsic version of these seem to return the same value as
2801 // the strict version.
2802 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2803 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2804 }
2805
2806 #undef __
2807 #define __ masm->
2808
2809 // Continuation point for throwing of implicit exceptions that are
2810 // not handled in the current activation. Fabricates an exception
2811 // oop and initiates normal exception dispatching in this
2812 // frame. Since we need to preserve callee-saved values (currently
2813 // only for C2, but done for C1 as well) we need a callee-saved oop
2814 // map and therefore have to make these stubs into RuntimeStubs
2815 // rather than BufferBlobs. If the compiler needs all registers to
2816 // be preserved between the fault point and the exception handler
2817 // then it must assume responsibility for that in
2818 // AbstractCompiler::continuation_for_implicit_null_exception or
2819 // continuation_for_implicit_division_by_zero_exception. All other
2820 // implicit exceptions (e.g., NullPointerException or
2821 // AbstractMethodError on entry) are either at call sites or
2822 // otherwise assume that stack unwinding will be initiated, so
2823 // caller saved registers were assumed volatile in the compiler.
2824 address generate_throw_exception(const char* name,
2825 address runtime_entry,
2826 bool restore_saved_exception_pc) {
2827 // Information about frame layout at time of blocking runtime call.
2828 // Note that we only have to preserve callee-saved registers since
2829 // the compilers are responsible for supplying a continuation point
2830 // if they expect all registers to be preserved.
2831 enum layout {
2832 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2833 rbp_off2,
2834 return_off,
2835 return_off2,
2836 framesize // inclusive of return address
2837 };
2838
2839 int insts_size = 512;
2840 int locs_size = 64;
2841
2842 CodeBuffer code(name, insts_size, locs_size);
2843 OopMapSet* oop_maps = new OopMapSet();
2844 MacroAssembler* masm = new MacroAssembler(&code);
2845
2846 address start = __ pc();
2847
2848 // This is an inlined and slightly modified version of call_VM
2849 // which has the ability to fetch the return PC out of
2850 // thread-local storage and also sets up last_Java_sp slightly
2851 // differently than the real call_VM
2852 if (restore_saved_exception_pc) {
2853 __ movptr(rax,
2854 Address(r15_thread,
2855 in_bytes(JavaThread::saved_exception_pc_offset())));
2856 __ push(rax);
2857 }
2858
2859 __ enter(); // required for proper stackwalking of RuntimeStub frame
2860
2861 assert(is_even(framesize/2), "sp not 16-byte aligned");
2862
2863 // return address and rbp are already in place
2864 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
2865
2866 int frame_complete = __ pc() - start;
2867
2868 // Set up last_Java_sp and last_Java_fp
2869 __ set_last_Java_frame(rsp, rbp, NULL);
2870
2871 // Call runtime
2872 __ movptr(c_rarg0, r15_thread);
2873 BLOCK_COMMENT("call runtime_entry");
2874 __ call(RuntimeAddress(runtime_entry));
2875
2876 // Generate oop map
2877 OopMap* map = new OopMap(framesize, 0);
2878
2879 oop_maps->add_gc_map(__ pc() - start, map);
2880
2881 __ reset_last_Java_frame(true, false);
2882
2883 __ leave(); // required for proper stackwalking of RuntimeStub frame
2884
2885 // check for pending exceptions
2886 #ifdef ASSERT
2887 Label L;
2888 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
2889 (int32_t) NULL_WORD);
2890 __ jcc(Assembler::notEqual, L);
2891 __ should_not_reach_here();
2892 __ bind(L);
2893 #endif // ASSERT
2894 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2895
2896
2897 // codeBlob framesize is in words (not VMRegImpl::slot_size)
2898 RuntimeStub* stub =
2899 RuntimeStub::new_runtime_stub(name,
2900 &code,
2901 frame_complete,
2902 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2903 oop_maps, false);
2904 return stub->entry_point();
2905 }
2906
2907 // Initialization
2908 void generate_initial() {
2909 // Generates all stubs and initializes the entry points
2910
2911 // This platform-specific stub is needed by generate_call_stub()
2912 StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
2913
2914 // entry points that exist in all platforms Note: This is code
2915 // that could be shared among different platforms - however the
2916 // benefit seems to be smaller than the disadvantage of having a
2917 // much more complicated generator structure. See also comment in
2918 // stubRoutines.hpp.
2919
2920 StubRoutines::_forward_exception_entry = generate_forward_exception();
2921
2922 StubRoutines::_call_stub_entry =
2923 generate_call_stub(StubRoutines::_call_stub_return_address);
2924
2925 // is referenced by megamorphic call
2926 StubRoutines::_catch_exception_entry = generate_catch_exception();
2927
2928 // atomic calls
2929 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2930 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
2931 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
2932 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2933 StubRoutines::_atomic_add_entry = generate_atomic_add();
2934 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
2935 StubRoutines::_fence_entry = generate_orderaccess_fence();
2936
2937 StubRoutines::_handler_for_unsafe_access_entry =
2938 generate_handler_for_unsafe_access();
2939
2940 // platform dependent
2941 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
2942
2943 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2944 }
2945
2946 void generate_all() {
2947 // Generates all stubs and initializes the entry points
2948
2949 // These entry points require SharedInfo::stack0 to be set up in
2950 // non-core builds and need to be relocatable, so they each
2951 // fabricate a RuntimeStub internally.
2952 StubRoutines::_throw_AbstractMethodError_entry =
2953 generate_throw_exception("AbstractMethodError throw_exception",
2954 CAST_FROM_FN_PTR(address,
2955 SharedRuntime::
2956 throw_AbstractMethodError),
2957 false);
2958
2959 StubRoutines::_throw_IncompatibleClassChangeError_entry =
2960 generate_throw_exception("IncompatibleClassChangeError throw_exception",
2961 CAST_FROM_FN_PTR(address,
2962 SharedRuntime::
2963 throw_IncompatibleClassChangeError),
2964 false);
2965
2966 StubRoutines::_throw_ArithmeticException_entry =
2967 generate_throw_exception("ArithmeticException throw_exception",
2968 CAST_FROM_FN_PTR(address,
2969 SharedRuntime::
2970 throw_ArithmeticException),
2971 true);
2972
2973 StubRoutines::_throw_NullPointerException_entry =
2974 generate_throw_exception("NullPointerException throw_exception",
2975 CAST_FROM_FN_PTR(address,
2976 SharedRuntime::
2977 throw_NullPointerException),
2978 true);
2979
2980 StubRoutines::_throw_NullPointerException_at_call_entry =
2981 generate_throw_exception("NullPointerException at call throw_exception",
2982 CAST_FROM_FN_PTR(address,
2983 SharedRuntime::
2984 throw_NullPointerException_at_call),
2985 false);
2986
2987 StubRoutines::_throw_StackOverflowError_entry =
2988 generate_throw_exception("StackOverflowError throw_exception",
2989 CAST_FROM_FN_PTR(address,
2990 SharedRuntime::
2991 throw_StackOverflowError),
2992 false);
2993
2994 // entry points that are platform specific
2995 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
2996 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
2997 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
2998 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
2999
3000 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
3001 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
3002 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3003 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3004
3005 // support for verify_oop (must happen after universe_init)
3006 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3007
3008 // arraycopy stubs used by compilers
3009 generate_arraycopy_stubs();
3010
3011 generate_math_stubs();
3012 }
3013
3014 public:
3015 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3016 if (all) {
3017 generate_all();
3018 } else {
3019 generate_initial();
3020 }
3021 }
3022 }; // end class declaration
3023
3024 address StubGenerator::disjoint_byte_copy_entry = NULL;
3025 address StubGenerator::disjoint_short_copy_entry = NULL;
3026 address StubGenerator::disjoint_int_copy_entry = NULL;
3027 address StubGenerator::disjoint_long_copy_entry = NULL;
3028 address StubGenerator::disjoint_oop_copy_entry = NULL;
3029
3030 address StubGenerator::byte_copy_entry = NULL;
3031 address StubGenerator::short_copy_entry = NULL;
3032 address StubGenerator::int_copy_entry = NULL;
3033 address StubGenerator::long_copy_entry = NULL;
3034 address StubGenerator::oop_copy_entry = NULL;
3035
3036 address StubGenerator::checkcast_copy_entry = NULL;
3037
3038 void StubGenerator_generate(CodeBuffer* code, bool all) {
3039 StubGenerator g(code, all);
3040 }