1 /*
2 * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "assembler_x86.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "nativeInst_x86.hpp"
30 #include "oops/instanceOop.hpp"
31 #include "oops/methodOop.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "oops/oop.inline.hpp"
34 #include "prims/methodHandles.hpp"
35 #include "runtime/frame.inline.hpp"
36 #include "runtime/handles.inline.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubCodeGenerator.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "utilities/top.hpp"
41 #ifdef TARGET_OS_FAMILY_linux
42 # include "thread_linux.inline.hpp"
43 #endif
44 #ifdef TARGET_OS_FAMILY_solaris
45 # include "thread_solaris.inline.hpp"
46 #endif
47 #ifdef TARGET_OS_FAMILY_windows
48 # include "thread_windows.inline.hpp"
49 #endif
50 #ifdef COMPILER2
51 #include "opto/runtime.hpp"
52 #endif
53
54 // Declaration and definition of StubGenerator (no .hpp file).
55 // For a more detailed description of the stub routine structure
56 // see the comment in stubRoutines.hpp
57
58 #define __ _masm->
59 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
60 #define a__ ((Assembler*)_masm)->
61
62 #ifdef PRODUCT
63 #define BLOCK_COMMENT(str) /* nothing */
64 #else
65 #define BLOCK_COMMENT(str) __ block_comment(str)
66 #endif
67
68 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
69 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
70
71 // Stub Code definitions
72
73 static address handle_unsafe_access() {
74 JavaThread* thread = JavaThread::current();
75 address pc = thread->saved_exception_pc();
76 // pc is the instruction which we must emulate
77 // doing a no-op is fine: return garbage from the load
78 // therefore, compute npc
79 address npc = Assembler::locate_next_instruction(pc);
80
81 // request an async exception
82 thread->set_pending_unsafe_access_error();
83
84 // return address of next instruction to execute
85 return npc;
86 }
87
88 class StubGenerator: public StubCodeGenerator {
89 private:
90
91 #ifdef PRODUCT
92 #define inc_counter_np(counter) (0)
93 #else
94 void inc_counter_np_(int& counter) {
95 __ incrementl(ExternalAddress((address)&counter));
96 }
97 #define inc_counter_np(counter) \
98 BLOCK_COMMENT("inc_counter " #counter); \
99 inc_counter_np_(counter);
100 #endif
101
102 // Call stubs are used to call Java from C
103 //
104 // Linux Arguments:
105 // c_rarg0: call wrapper address address
106 // c_rarg1: result address
107 // c_rarg2: result type BasicType
108 // c_rarg3: method methodOop
109 // c_rarg4: (interpreter) entry point address
110 // c_rarg5: parameters intptr_t*
111 // 16(rbp): parameter size (in words) int
112 // 24(rbp): thread Thread*
113 //
114 // [ return_from_Java ] <--- rsp
115 // [ argument word n ]
116 // ...
117 // -12 [ argument word 1 ]
118 // -11 [ saved r15 ] <--- rsp_after_call
119 // -10 [ saved r14 ]
120 // -9 [ saved r13 ]
121 // -8 [ saved r12 ]
122 // -7 [ saved rbx ]
123 // -6 [ call wrapper ]
124 // -5 [ result ]
125 // -4 [ result type ]
126 // -3 [ method ]
127 // -2 [ entry point ]
128 // -1 [ parameters ]
129 // 0 [ saved rbp ] <--- rbp
130 // 1 [ return address ]
131 // 2 [ parameter size ]
132 // 3 [ thread ]
133 //
134 // Windows Arguments:
135 // c_rarg0: call wrapper address address
136 // c_rarg1: result address
137 // c_rarg2: result type BasicType
138 // c_rarg3: method methodOop
139 // 48(rbp): (interpreter) entry point address
140 // 56(rbp): parameters intptr_t*
141 // 64(rbp): parameter size (in words) int
142 // 72(rbp): thread Thread*
143 //
144 // [ return_from_Java ] <--- rsp
145 // [ argument word n ]
146 // ...
147 // -8 [ argument word 1 ]
148 // -7 [ saved r15 ] <--- rsp_after_call
149 // -6 [ saved r14 ]
150 // -5 [ saved r13 ]
151 // -4 [ saved r12 ]
152 // -3 [ saved rdi ]
153 // -2 [ saved rsi ]
154 // -1 [ saved rbx ]
155 // 0 [ saved rbp ] <--- rbp
156 // 1 [ return address ]
157 // 2 [ call wrapper ]
158 // 3 [ result ]
159 // 4 [ result type ]
160 // 5 [ method ]
161 // 6 [ entry point ]
162 // 7 [ parameters ]
163 // 8 [ parameter size ]
164 // 9 [ thread ]
165 //
166 // Windows reserves the callers stack space for arguments 1-4.
167 // We spill c_rarg0-c_rarg3 to this space.
168
169 // Call stub stack layout word offsets from rbp
170 enum call_stub_layout {
171 #ifdef _WIN64
172 rsp_after_call_off = -7,
173 r15_off = rsp_after_call_off,
174 r14_off = -6,
175 r13_off = -5,
176 r12_off = -4,
177 rdi_off = -3,
178 rsi_off = -2,
179 rbx_off = -1,
180 rbp_off = 0,
181 retaddr_off = 1,
182 call_wrapper_off = 2,
183 result_off = 3,
184 result_type_off = 4,
185 method_off = 5,
186 entry_point_off = 6,
187 parameters_off = 7,
188 parameter_size_off = 8,
189 thread_off = 9
190 #else
191 rsp_after_call_off = -12,
192 mxcsr_off = rsp_after_call_off,
193 r15_off = -11,
194 r14_off = -10,
195 r13_off = -9,
196 r12_off = -8,
197 rbx_off = -7,
198 call_wrapper_off = -6,
199 result_off = -5,
200 result_type_off = -4,
201 method_off = -3,
202 entry_point_off = -2,
203 parameters_off = -1,
204 rbp_off = 0,
205 retaddr_off = 1,
206 parameter_size_off = 2,
207 thread_off = 3
208 #endif
209 };
210
211 address generate_call_stub(address& return_address) {
212 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
213 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
214 "adjust this code");
215 StubCodeMark mark(this, "StubRoutines", "call_stub");
216 address start = __ pc();
217
218 // same as in generate_catch_exception()!
219 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
220
221 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
222 const Address result (rbp, result_off * wordSize);
223 const Address result_type (rbp, result_type_off * wordSize);
224 const Address method (rbp, method_off * wordSize);
225 const Address entry_point (rbp, entry_point_off * wordSize);
226 const Address parameters (rbp, parameters_off * wordSize);
227 const Address parameter_size(rbp, parameter_size_off * wordSize);
228
229 // same as in generate_catch_exception()!
230 const Address thread (rbp, thread_off * wordSize);
231
232 const Address r15_save(rbp, r15_off * wordSize);
233 const Address r14_save(rbp, r14_off * wordSize);
234 const Address r13_save(rbp, r13_off * wordSize);
235 const Address r12_save(rbp, r12_off * wordSize);
236 const Address rbx_save(rbp, rbx_off * wordSize);
237
238 // stub code
239 __ enter();
240 __ subptr(rsp, -rsp_after_call_off * wordSize);
241
242 // save register parameters
243 #ifndef _WIN64
244 __ movptr(parameters, c_rarg5); // parameters
245 __ movptr(entry_point, c_rarg4); // entry_point
246 #endif
247
248 __ movptr(method, c_rarg3); // method
249 __ movl(result_type, c_rarg2); // result type
250 __ movptr(result, c_rarg1); // result
251 __ movptr(call_wrapper, c_rarg0); // call wrapper
252
253 // save regs belonging to calling function
254 __ movptr(rbx_save, rbx);
255 __ movptr(r12_save, r12);
256 __ movptr(r13_save, r13);
257 __ movptr(r14_save, r14);
258 __ movptr(r15_save, r15);
259
260 #ifdef _WIN64
261 const Address rdi_save(rbp, rdi_off * wordSize);
262 const Address rsi_save(rbp, rsi_off * wordSize);
263
264 __ movptr(rsi_save, rsi);
265 __ movptr(rdi_save, rdi);
266 #else
267 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
268 {
269 Label skip_ldmx;
270 __ stmxcsr(mxcsr_save);
271 __ movl(rax, mxcsr_save);
272 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
273 ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
274 __ cmp32(rax, mxcsr_std);
275 __ jcc(Assembler::equal, skip_ldmx);
276 __ ldmxcsr(mxcsr_std);
277 __ bind(skip_ldmx);
278 }
279 #endif
280
281 // Load up thread register
282 __ movptr(r15_thread, thread);
283 __ reinit_heapbase();
284
285 #ifdef ASSERT
286 // make sure we have no pending exceptions
287 {
288 Label L;
289 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
290 __ jcc(Assembler::equal, L);
291 __ stop("StubRoutines::call_stub: entered with pending exception");
292 __ bind(L);
293 }
294 #endif
295
296 // pass parameters if any
297 BLOCK_COMMENT("pass parameters if any");
298 Label parameters_done;
299 __ movl(c_rarg3, parameter_size);
300 __ testl(c_rarg3, c_rarg3);
301 __ jcc(Assembler::zero, parameters_done);
302
303 Label loop;
304 __ movptr(c_rarg2, parameters); // parameter pointer
305 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
306 __ BIND(loop);
307 __ movptr(rax, Address(c_rarg2, 0));// get parameter
308 __ addptr(c_rarg2, wordSize); // advance to next parameter
309 __ decrementl(c_rarg1); // decrement counter
310 __ push(rax); // pass parameter
311 __ jcc(Assembler::notZero, loop);
312
313 // call Java function
314 __ BIND(parameters_done);
315 __ movptr(rbx, method); // get methodOop
316 __ movptr(c_rarg1, entry_point); // get entry_point
317 __ mov(r13, rsp); // set sender sp
318 BLOCK_COMMENT("call Java function");
319 __ call(c_rarg1);
320
321 BLOCK_COMMENT("call_stub_return_address:");
322 return_address = __ pc();
323
324 // store result depending on type (everything that is not
325 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
326 __ movptr(c_rarg0, result);
327 Label is_long, is_float, is_double, exit;
328 __ movl(c_rarg1, result_type);
329 __ cmpl(c_rarg1, T_OBJECT);
330 __ jcc(Assembler::equal, is_long);
331 __ cmpl(c_rarg1, T_LONG);
332 __ jcc(Assembler::equal, is_long);
333 __ cmpl(c_rarg1, T_FLOAT);
334 __ jcc(Assembler::equal, is_float);
335 __ cmpl(c_rarg1, T_DOUBLE);
336 __ jcc(Assembler::equal, is_double);
337
338 // handle T_INT case
339 __ movl(Address(c_rarg0, 0), rax);
340
341 __ BIND(exit);
342
343 // pop parameters
344 __ lea(rsp, rsp_after_call);
345
346 #ifdef ASSERT
347 // verify that threads correspond
348 {
349 Label L, S;
350 __ cmpptr(r15_thread, thread);
351 __ jcc(Assembler::notEqual, S);
352 __ get_thread(rbx);
353 __ cmpptr(r15_thread, rbx);
354 __ jcc(Assembler::equal, L);
355 __ bind(S);
356 __ jcc(Assembler::equal, L);
357 __ stop("StubRoutines::call_stub: threads must correspond");
358 __ bind(L);
359 }
360 #endif
361
362 // restore regs belonging to calling function
363 __ movptr(r15, r15_save);
364 __ movptr(r14, r14_save);
365 __ movptr(r13, r13_save);
366 __ movptr(r12, r12_save);
367 __ movptr(rbx, rbx_save);
368
369 #ifdef _WIN64
370 __ movptr(rdi, rdi_save);
371 __ movptr(rsi, rsi_save);
372 #else
373 __ ldmxcsr(mxcsr_save);
374 #endif
375
376 // restore rsp
377 __ addptr(rsp, -rsp_after_call_off * wordSize);
378
379 // return
380 __ pop(rbp);
381 __ ret(0);
382
383 // handle return types different from T_INT
384 __ BIND(is_long);
385 __ movq(Address(c_rarg0, 0), rax);
386 __ jmp(exit);
387
388 __ BIND(is_float);
389 __ movflt(Address(c_rarg0, 0), xmm0);
390 __ jmp(exit);
391
392 __ BIND(is_double);
393 __ movdbl(Address(c_rarg0, 0), xmm0);
394 __ jmp(exit);
395
396 return start;
397 }
398
399 // Return point for a Java call if there's an exception thrown in
400 // Java code. The exception is caught and transformed into a
401 // pending exception stored in JavaThread that can be tested from
402 // within the VM.
403 //
404 // Note: Usually the parameters are removed by the callee. In case
405 // of an exception crossing an activation frame boundary, that is
406 // not the case if the callee is compiled code => need to setup the
407 // rsp.
408 //
409 // rax: exception oop
410
411 address generate_catch_exception() {
412 StubCodeMark mark(this, "StubRoutines", "catch_exception");
413 address start = __ pc();
414
415 // same as in generate_call_stub():
416 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
417 const Address thread (rbp, thread_off * wordSize);
418
419 #ifdef ASSERT
420 // verify that threads correspond
421 {
422 Label L, S;
423 __ cmpptr(r15_thread, thread);
424 __ jcc(Assembler::notEqual, S);
425 __ get_thread(rbx);
426 __ cmpptr(r15_thread, rbx);
427 __ jcc(Assembler::equal, L);
428 __ bind(S);
429 __ stop("StubRoutines::catch_exception: threads must correspond");
430 __ bind(L);
431 }
432 #endif
433
434 // set pending exception
435 __ verify_oop(rax);
436
437 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
438 __ lea(rscratch1, ExternalAddress((address)__FILE__));
439 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
440 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
441
442 // complete return to VM
443 assert(StubRoutines::_call_stub_return_address != NULL,
444 "_call_stub_return_address must have been generated before");
445 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
446
447 return start;
448 }
449
450 // Continuation point for runtime calls returning with a pending
451 // exception. The pending exception check happened in the runtime
452 // or native call stub. The pending exception in Thread is
453 // converted into a Java-level exception.
454 //
455 // Contract with Java-level exception handlers:
456 // rax: exception
457 // rdx: throwing pc
458 //
459 // NOTE: At entry of this stub, exception-pc must be on stack !!
460
461 address generate_forward_exception() {
462 StubCodeMark mark(this, "StubRoutines", "forward exception");
463 address start = __ pc();
464
465 // Upon entry, the sp points to the return address returning into
466 // Java (interpreted or compiled) code; i.e., the return address
467 // becomes the throwing pc.
468 //
469 // Arguments pushed before the runtime call are still on the stack
470 // but the exception handler will reset the stack pointer ->
471 // ignore them. A potential result in registers can be ignored as
472 // well.
473
474 #ifdef ASSERT
475 // make sure this code is only executed if there is a pending exception
476 {
477 Label L;
478 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
479 __ jcc(Assembler::notEqual, L);
480 __ stop("StubRoutines::forward exception: no pending exception (1)");
481 __ bind(L);
482 }
483 #endif
484
485 // compute exception handler into rbx
486 __ movptr(c_rarg0, Address(rsp, 0));
487 BLOCK_COMMENT("call exception_handler_for_return_address");
488 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
489 SharedRuntime::exception_handler_for_return_address),
490 r15_thread, c_rarg0);
491 __ mov(rbx, rax);
492
493 // setup rax & rdx, remove return address & clear pending exception
494 __ pop(rdx);
495 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
496 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
497
498 #ifdef ASSERT
499 // make sure exception is set
500 {
501 Label L;
502 __ testptr(rax, rax);
503 __ jcc(Assembler::notEqual, L);
504 __ stop("StubRoutines::forward exception: no pending exception (2)");
505 __ bind(L);
506 }
507 #endif
508
509 // continue at exception handler (return address removed)
510 // rax: exception
511 // rbx: exception handler
512 // rdx: throwing pc
513 __ verify_oop(rax);
514 __ jmp(rbx);
515
516 return start;
517 }
518
519 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
520 //
521 // Arguments :
522 // c_rarg0: exchange_value
523 // c_rarg0: dest
524 //
525 // Result:
526 // *dest <- ex, return (orig *dest)
527 address generate_atomic_xchg() {
528 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
529 address start = __ pc();
530
531 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
532 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
533 __ ret(0);
534
535 return start;
536 }
537
538 // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
539 //
540 // Arguments :
541 // c_rarg0: exchange_value
542 // c_rarg1: dest
543 //
544 // Result:
545 // *dest <- ex, return (orig *dest)
546 address generate_atomic_xchg_ptr() {
547 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
548 address start = __ pc();
549
550 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
551 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
552 __ ret(0);
553
554 return start;
555 }
556
557 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
558 // jint compare_value)
559 //
560 // Arguments :
561 // c_rarg0: exchange_value
562 // c_rarg1: dest
563 // c_rarg2: compare_value
564 //
565 // Result:
566 // if ( compare_value == *dest ) {
567 // *dest = exchange_value
568 // return compare_value;
569 // else
570 // return *dest;
571 address generate_atomic_cmpxchg() {
572 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
573 address start = __ pc();
574
575 __ movl(rax, c_rarg2);
576 if ( os::is_MP() ) __ lock();
577 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
578 __ ret(0);
579
580 return start;
581 }
582
583 // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
584 // volatile jlong* dest,
585 // jlong compare_value)
586 // Arguments :
587 // c_rarg0: exchange_value
588 // c_rarg1: dest
589 // c_rarg2: compare_value
590 //
591 // Result:
592 // if ( compare_value == *dest ) {
593 // *dest = exchange_value
594 // return compare_value;
595 // else
596 // return *dest;
597 address generate_atomic_cmpxchg_long() {
598 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
599 address start = __ pc();
600
601 __ movq(rax, c_rarg2);
602 if ( os::is_MP() ) __ lock();
603 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
604 __ ret(0);
605
606 return start;
607 }
608
609 // Support for jint atomic::add(jint add_value, volatile jint* dest)
610 //
611 // Arguments :
612 // c_rarg0: add_value
613 // c_rarg1: dest
614 //
615 // Result:
616 // *dest += add_value
617 // return *dest;
618 address generate_atomic_add() {
619 StubCodeMark mark(this, "StubRoutines", "atomic_add");
620 address start = __ pc();
621
622 __ movl(rax, c_rarg0);
623 if ( os::is_MP() ) __ lock();
624 __ xaddl(Address(c_rarg1, 0), c_rarg0);
625 __ addl(rax, c_rarg0);
626 __ ret(0);
627
628 return start;
629 }
630
631 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
632 //
633 // Arguments :
634 // c_rarg0: add_value
635 // c_rarg1: dest
636 //
637 // Result:
638 // *dest += add_value
639 // return *dest;
640 address generate_atomic_add_ptr() {
641 StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
642 address start = __ pc();
643
644 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
645 if ( os::is_MP() ) __ lock();
646 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
647 __ addptr(rax, c_rarg0);
648 __ ret(0);
649
650 return start;
651 }
652
653 // Support for intptr_t OrderAccess::fence()
654 //
655 // Arguments :
656 //
657 // Result:
658 address generate_orderaccess_fence() {
659 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
660 address start = __ pc();
661 __ membar(Assembler::StoreLoad);
662 __ ret(0);
663
664 return start;
665 }
666
667 // Support for intptr_t get_previous_fp()
668 //
669 // This routine is used to find the previous frame pointer for the
670 // caller (current_frame_guess). This is used as part of debugging
671 // ps() is seemingly lost trying to find frames.
672 // This code assumes that caller current_frame_guess) has a frame.
673 address generate_get_previous_fp() {
674 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
675 const Address old_fp(rbp, 0);
676 const Address older_fp(rax, 0);
677 address start = __ pc();
678
679 __ enter();
680 __ movptr(rax, old_fp); // callers fp
681 __ movptr(rax, older_fp); // the frame for ps()
682 __ pop(rbp);
683 __ ret(0);
684
685 return start;
686 }
687
688 //----------------------------------------------------------------------------------------------------
689 // Support for void verify_mxcsr()
690 //
691 // This routine is used with -Xcheck:jni to verify that native
692 // JNI code does not return to Java code without restoring the
693 // MXCSR register to our expected state.
694
695 address generate_verify_mxcsr() {
696 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
697 address start = __ pc();
698
699 const Address mxcsr_save(rsp, 0);
700
701 if (CheckJNICalls) {
702 Label ok_ret;
703 __ push(rax);
704 __ subptr(rsp, wordSize); // allocate a temp location
705 __ stmxcsr(mxcsr_save);
706 __ movl(rax, mxcsr_save);
707 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
708 __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
709 __ jcc(Assembler::equal, ok_ret);
710
711 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
712
713 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
714
715 __ bind(ok_ret);
716 __ addptr(rsp, wordSize);
717 __ pop(rax);
718 }
719
720 __ ret(0);
721
722 return start;
723 }
724
725 address generate_f2i_fixup() {
726 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
727 Address inout(rsp, 5 * wordSize); // return address + 4 saves
728
729 address start = __ pc();
730
731 Label L;
732
733 __ push(rax);
734 __ push(c_rarg3);
735 __ push(c_rarg2);
736 __ push(c_rarg1);
737
738 __ movl(rax, 0x7f800000);
739 __ xorl(c_rarg3, c_rarg3);
740 __ movl(c_rarg2, inout);
741 __ movl(c_rarg1, c_rarg2);
742 __ andl(c_rarg1, 0x7fffffff);
743 __ cmpl(rax, c_rarg1); // NaN? -> 0
744 __ jcc(Assembler::negative, L);
745 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
746 __ movl(c_rarg3, 0x80000000);
747 __ movl(rax, 0x7fffffff);
748 __ cmovl(Assembler::positive, c_rarg3, rax);
749
750 __ bind(L);
751 __ movptr(inout, c_rarg3);
752
753 __ pop(c_rarg1);
754 __ pop(c_rarg2);
755 __ pop(c_rarg3);
756 __ pop(rax);
757
758 __ ret(0);
759
760 return start;
761 }
762
763 address generate_f2l_fixup() {
764 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
765 Address inout(rsp, 5 * wordSize); // return address + 4 saves
766 address start = __ pc();
767
768 Label L;
769
770 __ push(rax);
771 __ push(c_rarg3);
772 __ push(c_rarg2);
773 __ push(c_rarg1);
774
775 __ movl(rax, 0x7f800000);
776 __ xorl(c_rarg3, c_rarg3);
777 __ movl(c_rarg2, inout);
778 __ movl(c_rarg1, c_rarg2);
779 __ andl(c_rarg1, 0x7fffffff);
780 __ cmpl(rax, c_rarg1); // NaN? -> 0
781 __ jcc(Assembler::negative, L);
782 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
783 __ mov64(c_rarg3, 0x8000000000000000);
784 __ mov64(rax, 0x7fffffffffffffff);
785 __ cmov(Assembler::positive, c_rarg3, rax);
786
787 __ bind(L);
788 __ movptr(inout, c_rarg3);
789
790 __ pop(c_rarg1);
791 __ pop(c_rarg2);
792 __ pop(c_rarg3);
793 __ pop(rax);
794
795 __ ret(0);
796
797 return start;
798 }
799
800 address generate_d2i_fixup() {
801 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
802 Address inout(rsp, 6 * wordSize); // return address + 5 saves
803
804 address start = __ pc();
805
806 Label L;
807
808 __ push(rax);
809 __ push(c_rarg3);
810 __ push(c_rarg2);
811 __ push(c_rarg1);
812 __ push(c_rarg0);
813
814 __ movl(rax, 0x7ff00000);
815 __ movq(c_rarg2, inout);
816 __ movl(c_rarg3, c_rarg2);
817 __ mov(c_rarg1, c_rarg2);
818 __ mov(c_rarg0, c_rarg2);
819 __ negl(c_rarg3);
820 __ shrptr(c_rarg1, 0x20);
821 __ orl(c_rarg3, c_rarg2);
822 __ andl(c_rarg1, 0x7fffffff);
823 __ xorl(c_rarg2, c_rarg2);
824 __ shrl(c_rarg3, 0x1f);
825 __ orl(c_rarg1, c_rarg3);
826 __ cmpl(rax, c_rarg1);
827 __ jcc(Assembler::negative, L); // NaN -> 0
828 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
829 __ movl(c_rarg2, 0x80000000);
830 __ movl(rax, 0x7fffffff);
831 __ cmov(Assembler::positive, c_rarg2, rax);
832
833 __ bind(L);
834 __ movptr(inout, c_rarg2);
835
836 __ pop(c_rarg0);
837 __ pop(c_rarg1);
838 __ pop(c_rarg2);
839 __ pop(c_rarg3);
840 __ pop(rax);
841
842 __ ret(0);
843
844 return start;
845 }
846
847 address generate_d2l_fixup() {
848 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
849 Address inout(rsp, 6 * wordSize); // return address + 5 saves
850
851 address start = __ pc();
852
853 Label L;
854
855 __ push(rax);
856 __ push(c_rarg3);
857 __ push(c_rarg2);
858 __ push(c_rarg1);
859 __ push(c_rarg0);
860
861 __ movl(rax, 0x7ff00000);
862 __ movq(c_rarg2, inout);
863 __ movl(c_rarg3, c_rarg2);
864 __ mov(c_rarg1, c_rarg2);
865 __ mov(c_rarg0, c_rarg2);
866 __ negl(c_rarg3);
867 __ shrptr(c_rarg1, 0x20);
868 __ orl(c_rarg3, c_rarg2);
869 __ andl(c_rarg1, 0x7fffffff);
870 __ xorl(c_rarg2, c_rarg2);
871 __ shrl(c_rarg3, 0x1f);
872 __ orl(c_rarg1, c_rarg3);
873 __ cmpl(rax, c_rarg1);
874 __ jcc(Assembler::negative, L); // NaN -> 0
875 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
876 __ mov64(c_rarg2, 0x8000000000000000);
877 __ mov64(rax, 0x7fffffffffffffff);
878 __ cmovq(Assembler::positive, c_rarg2, rax);
879
880 __ bind(L);
881 __ movq(inout, c_rarg2);
882
883 __ pop(c_rarg0);
884 __ pop(c_rarg1);
885 __ pop(c_rarg2);
886 __ pop(c_rarg3);
887 __ pop(rax);
888
889 __ ret(0);
890
891 return start;
892 }
893
894 address generate_fp_mask(const char *stub_name, int64_t mask) {
895 __ align(CodeEntryAlignment);
896 StubCodeMark mark(this, "StubRoutines", stub_name);
897 address start = __ pc();
898
899 __ emit_data64( mask, relocInfo::none );
900 __ emit_data64( mask, relocInfo::none );
901
902 return start;
903 }
904
905 // The following routine generates a subroutine to throw an
906 // asynchronous UnknownError when an unsafe access gets a fault that
907 // could not be reasonably prevented by the programmer. (Example:
908 // SIGBUS/OBJERR.)
909 address generate_handler_for_unsafe_access() {
910 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
911 address start = __ pc();
912
913 __ push(0); // hole for return address-to-be
914 __ pusha(); // push registers
915 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
916
917 __ subptr(rsp, frame::arg_reg_save_area_bytes);
918 BLOCK_COMMENT("call handle_unsafe_access");
919 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
920 __ addptr(rsp, frame::arg_reg_save_area_bytes);
921
922 __ movptr(next_pc, rax); // stuff next address
923 __ popa();
924 __ ret(0); // jump to next address
925
926 return start;
927 }
928
929 // Non-destructive plausibility checks for oops
930 //
931 // Arguments:
932 // all args on stack!
933 //
934 // Stack after saving c_rarg3:
935 // [tos + 0]: saved c_rarg3
936 // [tos + 1]: saved c_rarg2
937 // [tos + 2]: saved r12 (several TemplateTable methods use it)
938 // [tos + 3]: saved flags
939 // [tos + 4]: return address
940 // * [tos + 5]: error message (char*)
941 // * [tos + 6]: object to verify (oop)
942 // * [tos + 7]: saved rax - saved by caller and bashed
943 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
944 // * = popped on exit
945 address generate_verify_oop() {
946 StubCodeMark mark(this, "StubRoutines", "verify_oop");
947 address start = __ pc();
948
949 Label exit, error;
950
951 __ pushf();
952 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
953
954 __ push(r12);
955
956 // save c_rarg2 and c_rarg3
957 __ push(c_rarg2);
958 __ push(c_rarg3);
959
960 enum {
961 // After previous pushes.
962 oop_to_verify = 6 * wordSize,
963 saved_rax = 7 * wordSize,
964 saved_r10 = 8 * wordSize,
965
966 // Before the call to MacroAssembler::debug(), see below.
967 return_addr = 16 * wordSize,
968 error_msg = 17 * wordSize
969 };
970
971 // get object
972 __ movptr(rax, Address(rsp, oop_to_verify));
973
974 // make sure object is 'reasonable'
975 __ testptr(rax, rax);
976 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
977 // Check if the oop is in the right area of memory
978 __ movptr(c_rarg2, rax);
979 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
980 __ andptr(c_rarg2, c_rarg3);
981 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
982 __ cmpptr(c_rarg2, c_rarg3);
983 __ jcc(Assembler::notZero, error);
984
985 // set r12 to heapbase for load_klass()
986 __ reinit_heapbase();
987
988 // make sure klass is 'reasonable'
989 __ load_klass(rax, rax); // get klass
990 __ testptr(rax, rax);
991 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
992 // Check if the klass is in the right area of memory
993 __ mov(c_rarg2, rax);
994 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
995 __ andptr(c_rarg2, c_rarg3);
996 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
997 __ cmpptr(c_rarg2, c_rarg3);
998 __ jcc(Assembler::notZero, error);
999
1000 // make sure klass' klass is 'reasonable'
1001 __ load_klass(rax, rax);
1002 __ testptr(rax, rax);
1003 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
1004 // Check if the klass' klass is in the right area of memory
1005 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
1006 __ andptr(rax, c_rarg3);
1007 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
1008 __ cmpptr(rax, c_rarg3);
1009 __ jcc(Assembler::notZero, error);
1010
1011 // return if everything seems ok
1012 __ bind(exit);
1013 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1014 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1015 __ pop(c_rarg3); // restore c_rarg3
1016 __ pop(c_rarg2); // restore c_rarg2
1017 __ pop(r12); // restore r12
1018 __ popf(); // restore flags
1019 __ ret(4 * wordSize); // pop caller saved stuff
1020
1021 // handle errors
1022 __ bind(error);
1023 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1024 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1025 __ pop(c_rarg3); // get saved c_rarg3 back
1026 __ pop(c_rarg2); // get saved c_rarg2 back
1027 __ pop(r12); // get saved r12 back
1028 __ popf(); // get saved flags off stack --
1029 // will be ignored
1030
1031 __ pusha(); // push registers
1032 // (rip is already
1033 // already pushed)
1034 // debug(char* msg, int64_t pc, int64_t regs[])
1035 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1036 // pushed all the registers, so now the stack looks like:
1037 // [tos + 0] 16 saved registers
1038 // [tos + 16] return address
1039 // * [tos + 17] error message (char*)
1040 // * [tos + 18] object to verify (oop)
1041 // * [tos + 19] saved rax - saved by caller and bashed
1042 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1043 // * = popped on exit
1044
1045 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1046 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1047 __ movq(c_rarg2, rsp); // pass address of regs on stack
1048 __ mov(r12, rsp); // remember rsp
1049 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1050 __ andptr(rsp, -16); // align stack as required by ABI
1051 BLOCK_COMMENT("call MacroAssembler::debug");
1052 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1053 __ mov(rsp, r12); // restore rsp
1054 __ popa(); // pop registers (includes r12)
1055 __ ret(4 * wordSize); // pop caller saved stuff
1056
1057 return start;
1058 }
1059
1060 //
1061 // Verify that a register contains clean 32-bits positive value
1062 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1063 //
1064 // Input:
1065 // Rint - 32-bits value
1066 // Rtmp - scratch
1067 //
1068 void assert_clean_int(Register Rint, Register Rtmp) {
1069 #ifdef ASSERT
1070 Label L;
1071 assert_different_registers(Rtmp, Rint);
1072 __ movslq(Rtmp, Rint);
1073 __ cmpq(Rtmp, Rint);
1074 __ jcc(Assembler::equal, L);
1075 __ stop("high 32-bits of int value are not 0");
1076 __ bind(L);
1077 #endif
1078 }
1079
1080 // Generate overlap test for array copy stubs
1081 //
1082 // Input:
1083 // c_rarg0 - from
1084 // c_rarg1 - to
1085 // c_rarg2 - element count
1086 //
1087 // Output:
1088 // rax - &from[element count - 1]
1089 //
1090 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1091 assert(no_overlap_target != NULL, "must be generated");
1092 array_overlap_test(no_overlap_target, NULL, sf);
1093 }
1094 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1095 array_overlap_test(NULL, &L_no_overlap, sf);
1096 }
1097 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1098 const Register from = c_rarg0;
1099 const Register to = c_rarg1;
1100 const Register count = c_rarg2;
1101 const Register end_from = rax;
1102
1103 __ cmpptr(to, from);
1104 __ lea(end_from, Address(from, count, sf, 0));
1105 if (NOLp == NULL) {
1106 ExternalAddress no_overlap(no_overlap_target);
1107 __ jump_cc(Assembler::belowEqual, no_overlap);
1108 __ cmpptr(to, end_from);
1109 __ jump_cc(Assembler::aboveEqual, no_overlap);
1110 } else {
1111 __ jcc(Assembler::belowEqual, (*NOLp));
1112 __ cmpptr(to, end_from);
1113 __ jcc(Assembler::aboveEqual, (*NOLp));
1114 }
1115 }
1116
1117 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1118 //
1119 // Outputs:
1120 // rdi - rcx
1121 // rsi - rdx
1122 // rdx - r8
1123 // rcx - r9
1124 //
1125 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1126 // are non-volatile. r9 and r10 should not be used by the caller.
1127 //
1128 void setup_arg_regs(int nargs = 3) {
1129 const Register saved_rdi = r9;
1130 const Register saved_rsi = r10;
1131 assert(nargs == 3 || nargs == 4, "else fix");
1132 #ifdef _WIN64
1133 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1134 "unexpected argument registers");
1135 if (nargs >= 4)
1136 __ mov(rax, r9); // r9 is also saved_rdi
1137 __ movptr(saved_rdi, rdi);
1138 __ movptr(saved_rsi, rsi);
1139 __ mov(rdi, rcx); // c_rarg0
1140 __ mov(rsi, rdx); // c_rarg1
1141 __ mov(rdx, r8); // c_rarg2
1142 if (nargs >= 4)
1143 __ mov(rcx, rax); // c_rarg3 (via rax)
1144 #else
1145 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1146 "unexpected argument registers");
1147 #endif
1148 }
1149
1150 void restore_arg_regs() {
1151 const Register saved_rdi = r9;
1152 const Register saved_rsi = r10;
1153 #ifdef _WIN64
1154 __ movptr(rdi, saved_rdi);
1155 __ movptr(rsi, saved_rsi);
1156 #endif
1157 }
1158
1159 // Generate code for an array write pre barrier
1160 //
1161 // addr - starting address
1162 // count - element count
1163 //
1164 // Destroy no registers!
1165 //
1166 void gen_write_ref_array_pre_barrier(Register addr, Register count) {
1167 BarrierSet* bs = Universe::heap()->barrier_set();
1168 switch (bs->kind()) {
1169 case BarrierSet::G1SATBCT:
1170 case BarrierSet::G1SATBCTLogging:
1171 {
1172 __ pusha(); // push registers
1173 if (count == c_rarg0) {
1174 if (addr == c_rarg1) {
1175 // exactly backwards!!
1176 __ xchgptr(c_rarg1, c_rarg0);
1177 } else {
1178 __ movptr(c_rarg1, count);
1179 __ movptr(c_rarg0, addr);
1180 }
1181
1182 } else {
1183 __ movptr(c_rarg0, addr);
1184 __ movptr(c_rarg1, count);
1185 }
1186 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1187 __ popa();
1188 }
1189 break;
1190 case BarrierSet::CardTableModRef:
1191 case BarrierSet::CardTableExtension:
1192 case BarrierSet::ModRef:
1193 break;
1194 default:
1195 ShouldNotReachHere();
1196
1197 }
1198 }
1199
1200 //
1201 // Generate code for an array write post barrier
1202 //
1203 // Input:
1204 // start - register containing starting address of destination array
1205 // end - register containing ending address of destination array
1206 // scratch - scratch register
1207 //
1208 // The input registers are overwritten.
1209 // The ending address is inclusive.
1210 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1211 assert_different_registers(start, end, scratch);
1212 BarrierSet* bs = Universe::heap()->barrier_set();
1213 switch (bs->kind()) {
1214 case BarrierSet::G1SATBCT:
1215 case BarrierSet::G1SATBCTLogging:
1216
1217 {
1218 __ pusha(); // push registers (overkill)
1219 // must compute element count unless barrier set interface is changed (other platforms supply count)
1220 assert_different_registers(start, end, scratch);
1221 __ lea(scratch, Address(end, BytesPerHeapOop));
1222 __ subptr(scratch, start); // subtract start to get #bytes
1223 __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count
1224 __ mov(c_rarg0, start);
1225 __ mov(c_rarg1, scratch);
1226 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1227 __ popa();
1228 }
1229 break;
1230 case BarrierSet::CardTableModRef:
1231 case BarrierSet::CardTableExtension:
1232 {
1233 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1234 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1235
1236 Label L_loop;
1237
1238 __ shrptr(start, CardTableModRefBS::card_shift);
1239 __ addptr(end, BytesPerHeapOop);
1240 __ shrptr(end, CardTableModRefBS::card_shift);
1241 __ subptr(end, start); // number of bytes to copy
1242
1243 intptr_t disp = (intptr_t) ct->byte_map_base;
1244 if (__ is_simm32(disp)) {
1245 Address cardtable(noreg, noreg, Address::no_scale, disp);
1246 __ lea(scratch, cardtable);
1247 } else {
1248 ExternalAddress cardtable((address)disp);
1249 __ lea(scratch, cardtable);
1250 }
1251
1252 const Register count = end; // 'end' register contains bytes count now
1253 __ addptr(start, scratch);
1254 __ BIND(L_loop);
1255 __ movb(Address(start, count, Address::times_1), 0);
1256 __ decrement(count);
1257 __ jcc(Assembler::greaterEqual, L_loop);
1258 }
1259 break;
1260 default:
1261 ShouldNotReachHere();
1262
1263 }
1264 }
1265
1266
1267 // Copy big chunks forward
1268 //
1269 // Inputs:
1270 // end_from - source arrays end address
1271 // end_to - destination array end address
1272 // qword_count - 64-bits element count, negative
1273 // to - scratch
1274 // L_copy_32_bytes - entry label
1275 // L_copy_8_bytes - exit label
1276 //
1277 void copy_32_bytes_forward(Register end_from, Register end_to,
1278 Register qword_count, Register to,
1279 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1280 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1281 Label L_loop;
1282 __ align(OptoLoopAlignment);
1283 __ BIND(L_loop);
1284 if(UseUnalignedLoadStores) {
1285 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1286 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1287 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1288 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1289
1290 } else {
1291 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1292 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1293 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1294 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1295 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1296 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1297 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1298 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1299 }
1300 __ BIND(L_copy_32_bytes);
1301 __ addptr(qword_count, 4);
1302 __ jcc(Assembler::lessEqual, L_loop);
1303 __ subptr(qword_count, 4);
1304 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1305 }
1306
1307
1308 // Copy big chunks backward
1309 //
1310 // Inputs:
1311 // from - source arrays address
1312 // dest - destination array address
1313 // qword_count - 64-bits element count
1314 // to - scratch
1315 // L_copy_32_bytes - entry label
1316 // L_copy_8_bytes - exit label
1317 //
1318 void copy_32_bytes_backward(Register from, Register dest,
1319 Register qword_count, Register to,
1320 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1321 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1322 Label L_loop;
1323 __ align(OptoLoopAlignment);
1324 __ BIND(L_loop);
1325 if(UseUnalignedLoadStores) {
1326 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1327 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1328 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1329 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1330
1331 } else {
1332 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1333 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1334 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1335 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1336 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1337 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1338 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1339 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1340 }
1341 __ BIND(L_copy_32_bytes);
1342 __ subptr(qword_count, 4);
1343 __ jcc(Assembler::greaterEqual, L_loop);
1344 __ addptr(qword_count, 4);
1345 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1346 }
1347
1348
1349 // Arguments:
1350 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1351 // ignored
1352 // name - stub name string
1353 //
1354 // Inputs:
1355 // c_rarg0 - source array address
1356 // c_rarg1 - destination array address
1357 // c_rarg2 - element count, treated as ssize_t, can be zero
1358 //
1359 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1360 // we let the hardware handle it. The one to eight bytes within words,
1361 // dwords or qwords that span cache line boundaries will still be loaded
1362 // and stored atomically.
1363 //
1364 // Side Effects:
1365 // disjoint_byte_copy_entry is set to the no-overlap entry point
1366 // used by generate_conjoint_byte_copy().
1367 //
1368 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1369 __ align(CodeEntryAlignment);
1370 StubCodeMark mark(this, "StubRoutines", name);
1371 address start = __ pc();
1372
1373 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1374 Label L_copy_byte, L_exit;
1375 const Register from = rdi; // source array address
1376 const Register to = rsi; // destination array address
1377 const Register count = rdx; // elements count
1378 const Register byte_count = rcx;
1379 const Register qword_count = count;
1380 const Register end_from = from; // source array end address
1381 const Register end_to = to; // destination array end address
1382 // End pointers are inclusive, and if count is not zero they point
1383 // to the last unit copied: end_to[0] := end_from[0]
1384
1385 __ enter(); // required for proper stackwalking of RuntimeStub frame
1386 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1387
1388 if (entry != NULL) {
1389 *entry = __ pc();
1390 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1391 BLOCK_COMMENT("Entry:");
1392 }
1393
1394 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1395 // r9 and r10 may be used to save non-volatile registers
1396
1397 // 'from', 'to' and 'count' are now valid
1398 __ movptr(byte_count, count);
1399 __ shrptr(count, 3); // count => qword_count
1400
1401 // Copy from low to high addresses. Use 'to' as scratch.
1402 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1403 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1404 __ negptr(qword_count); // make the count negative
1405 __ jmp(L_copy_32_bytes);
1406
1407 // Copy trailing qwords
1408 __ BIND(L_copy_8_bytes);
1409 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1410 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1411 __ increment(qword_count);
1412 __ jcc(Assembler::notZero, L_copy_8_bytes);
1413
1414 // Check for and copy trailing dword
1415 __ BIND(L_copy_4_bytes);
1416 __ testl(byte_count, 4);
1417 __ jccb(Assembler::zero, L_copy_2_bytes);
1418 __ movl(rax, Address(end_from, 8));
1419 __ movl(Address(end_to, 8), rax);
1420
1421 __ addptr(end_from, 4);
1422 __ addptr(end_to, 4);
1423
1424 // Check for and copy trailing word
1425 __ BIND(L_copy_2_bytes);
1426 __ testl(byte_count, 2);
1427 __ jccb(Assembler::zero, L_copy_byte);
1428 __ movw(rax, Address(end_from, 8));
1429 __ movw(Address(end_to, 8), rax);
1430
1431 __ addptr(end_from, 2);
1432 __ addptr(end_to, 2);
1433
1434 // Check for and copy trailing byte
1435 __ BIND(L_copy_byte);
1436 __ testl(byte_count, 1);
1437 __ jccb(Assembler::zero, L_exit);
1438 __ movb(rax, Address(end_from, 8));
1439 __ movb(Address(end_to, 8), rax);
1440
1441 __ BIND(L_exit);
1442 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1443 restore_arg_regs();
1444 __ xorptr(rax, rax); // return 0
1445 __ leave(); // required for proper stackwalking of RuntimeStub frame
1446 __ ret(0);
1447
1448 // Copy in 32-bytes chunks
1449 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1450 __ jmp(L_copy_4_bytes);
1451
1452 return start;
1453 }
1454
1455 // Arguments:
1456 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1457 // ignored
1458 // name - stub name string
1459 //
1460 // Inputs:
1461 // c_rarg0 - source array address
1462 // c_rarg1 - destination array address
1463 // c_rarg2 - element count, treated as ssize_t, can be zero
1464 //
1465 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1466 // we let the hardware handle it. The one to eight bytes within words,
1467 // dwords or qwords that span cache line boundaries will still be loaded
1468 // and stored atomically.
1469 //
1470 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1471 address* entry, const char *name) {
1472 __ align(CodeEntryAlignment);
1473 StubCodeMark mark(this, "StubRoutines", name);
1474 address start = __ pc();
1475
1476 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1477 const Register from = rdi; // source array address
1478 const Register to = rsi; // destination array address
1479 const Register count = rdx; // elements count
1480 const Register byte_count = rcx;
1481 const Register qword_count = count;
1482
1483 __ enter(); // required for proper stackwalking of RuntimeStub frame
1484 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1485
1486 if (entry != NULL) {
1487 *entry = __ pc();
1488 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1489 BLOCK_COMMENT("Entry:");
1490 }
1491
1492 array_overlap_test(nooverlap_target, Address::times_1);
1493 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1494 // r9 and r10 may be used to save non-volatile registers
1495
1496 // 'from', 'to' and 'count' are now valid
1497 __ movptr(byte_count, count);
1498 __ shrptr(count, 3); // count => qword_count
1499
1500 // Copy from high to low addresses.
1501
1502 // Check for and copy trailing byte
1503 __ testl(byte_count, 1);
1504 __ jcc(Assembler::zero, L_copy_2_bytes);
1505 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1506 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1507 __ decrement(byte_count); // Adjust for possible trailing word
1508
1509 // Check for and copy trailing word
1510 __ BIND(L_copy_2_bytes);
1511 __ testl(byte_count, 2);
1512 __ jcc(Assembler::zero, L_copy_4_bytes);
1513 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1514 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1515
1516 // Check for and copy trailing dword
1517 __ BIND(L_copy_4_bytes);
1518 __ testl(byte_count, 4);
1519 __ jcc(Assembler::zero, L_copy_32_bytes);
1520 __ movl(rax, Address(from, qword_count, Address::times_8));
1521 __ movl(Address(to, qword_count, Address::times_8), rax);
1522 __ jmp(L_copy_32_bytes);
1523
1524 // Copy trailing qwords
1525 __ BIND(L_copy_8_bytes);
1526 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1527 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1528 __ decrement(qword_count);
1529 __ jcc(Assembler::notZero, L_copy_8_bytes);
1530
1531 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1532 restore_arg_regs();
1533 __ xorptr(rax, rax); // return 0
1534 __ leave(); // required for proper stackwalking of RuntimeStub frame
1535 __ ret(0);
1536
1537 // Copy in 32-bytes chunks
1538 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1539
1540 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1541 restore_arg_regs();
1542 __ xorptr(rax, rax); // return 0
1543 __ leave(); // required for proper stackwalking of RuntimeStub frame
1544 __ ret(0);
1545
1546 return start;
1547 }
1548
1549 // Arguments:
1550 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1551 // ignored
1552 // name - stub name string
1553 //
1554 // Inputs:
1555 // c_rarg0 - source array address
1556 // c_rarg1 - destination array address
1557 // c_rarg2 - element count, treated as ssize_t, can be zero
1558 //
1559 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1560 // let the hardware handle it. The two or four words within dwords
1561 // or qwords that span cache line boundaries will still be loaded
1562 // and stored atomically.
1563 //
1564 // Side Effects:
1565 // disjoint_short_copy_entry is set to the no-overlap entry point
1566 // used by generate_conjoint_short_copy().
1567 //
1568 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1569 __ align(CodeEntryAlignment);
1570 StubCodeMark mark(this, "StubRoutines", name);
1571 address start = __ pc();
1572
1573 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1574 const Register from = rdi; // source array address
1575 const Register to = rsi; // destination array address
1576 const Register count = rdx; // elements count
1577 const Register word_count = rcx;
1578 const Register qword_count = count;
1579 const Register end_from = from; // source array end address
1580 const Register end_to = to; // destination array end address
1581 // End pointers are inclusive, and if count is not zero they point
1582 // to the last unit copied: end_to[0] := end_from[0]
1583
1584 __ enter(); // required for proper stackwalking of RuntimeStub frame
1585 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1586
1587 if (entry != NULL) {
1588 *entry = __ pc();
1589 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1590 BLOCK_COMMENT("Entry:");
1591 }
1592
1593 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1594 // r9 and r10 may be used to save non-volatile registers
1595
1596 // 'from', 'to' and 'count' are now valid
1597 __ movptr(word_count, count);
1598 __ shrptr(count, 2); // count => qword_count
1599
1600 // Copy from low to high addresses. Use 'to' as scratch.
1601 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1602 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1603 __ negptr(qword_count);
1604 __ jmp(L_copy_32_bytes);
1605
1606 // Copy trailing qwords
1607 __ BIND(L_copy_8_bytes);
1608 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1609 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1610 __ increment(qword_count);
1611 __ jcc(Assembler::notZero, L_copy_8_bytes);
1612
1613 // Original 'dest' is trashed, so we can't use it as a
1614 // base register for a possible trailing word copy
1615
1616 // Check for and copy trailing dword
1617 __ BIND(L_copy_4_bytes);
1618 __ testl(word_count, 2);
1619 __ jccb(Assembler::zero, L_copy_2_bytes);
1620 __ movl(rax, Address(end_from, 8));
1621 __ movl(Address(end_to, 8), rax);
1622
1623 __ addptr(end_from, 4);
1624 __ addptr(end_to, 4);
1625
1626 // Check for and copy trailing word
1627 __ BIND(L_copy_2_bytes);
1628 __ testl(word_count, 1);
1629 __ jccb(Assembler::zero, L_exit);
1630 __ movw(rax, Address(end_from, 8));
1631 __ movw(Address(end_to, 8), rax);
1632
1633 __ BIND(L_exit);
1634 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1635 restore_arg_regs();
1636 __ xorptr(rax, rax); // return 0
1637 __ leave(); // required for proper stackwalking of RuntimeStub frame
1638 __ ret(0);
1639
1640 // Copy in 32-bytes chunks
1641 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1642 __ jmp(L_copy_4_bytes);
1643
1644 return start;
1645 }
1646
1647 address generate_fill(BasicType t, bool aligned, const char *name) {
1648 __ align(CodeEntryAlignment);
1649 StubCodeMark mark(this, "StubRoutines", name);
1650 address start = __ pc();
1651
1652 BLOCK_COMMENT("Entry:");
1653
1654 const Register to = c_rarg0; // source array address
1655 const Register value = c_rarg1; // value
1656 const Register count = c_rarg2; // elements count
1657
1658 __ enter(); // required for proper stackwalking of RuntimeStub frame
1659
1660 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1661
1662 __ leave(); // required for proper stackwalking of RuntimeStub frame
1663 __ ret(0);
1664 return start;
1665 }
1666
1667 // Arguments:
1668 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1669 // ignored
1670 // name - stub name string
1671 //
1672 // Inputs:
1673 // c_rarg0 - source array address
1674 // c_rarg1 - destination array address
1675 // c_rarg2 - element count, treated as ssize_t, can be zero
1676 //
1677 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1678 // let the hardware handle it. The two or four words within dwords
1679 // or qwords that span cache line boundaries will still be loaded
1680 // and stored atomically.
1681 //
1682 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1683 address *entry, const char *name) {
1684 __ align(CodeEntryAlignment);
1685 StubCodeMark mark(this, "StubRoutines", name);
1686 address start = __ pc();
1687
1688 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1689 const Register from = rdi; // source array address
1690 const Register to = rsi; // destination array address
1691 const Register count = rdx; // elements count
1692 const Register word_count = rcx;
1693 const Register qword_count = count;
1694
1695 __ enter(); // required for proper stackwalking of RuntimeStub frame
1696 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1697
1698 if (entry != NULL) {
1699 *entry = __ pc();
1700 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1701 BLOCK_COMMENT("Entry:");
1702 }
1703
1704 array_overlap_test(nooverlap_target, Address::times_2);
1705 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1706 // r9 and r10 may be used to save non-volatile registers
1707
1708 // 'from', 'to' and 'count' are now valid
1709 __ movptr(word_count, count);
1710 __ shrptr(count, 2); // count => qword_count
1711
1712 // Copy from high to low addresses. Use 'to' as scratch.
1713
1714 // Check for and copy trailing word
1715 __ testl(word_count, 1);
1716 __ jccb(Assembler::zero, L_copy_4_bytes);
1717 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1718 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1719
1720 // Check for and copy trailing dword
1721 __ BIND(L_copy_4_bytes);
1722 __ testl(word_count, 2);
1723 __ jcc(Assembler::zero, L_copy_32_bytes);
1724 __ movl(rax, Address(from, qword_count, Address::times_8));
1725 __ movl(Address(to, qword_count, Address::times_8), rax);
1726 __ jmp(L_copy_32_bytes);
1727
1728 // Copy trailing qwords
1729 __ BIND(L_copy_8_bytes);
1730 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1731 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1732 __ decrement(qword_count);
1733 __ jcc(Assembler::notZero, L_copy_8_bytes);
1734
1735 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1736 restore_arg_regs();
1737 __ xorptr(rax, rax); // return 0
1738 __ leave(); // required for proper stackwalking of RuntimeStub frame
1739 __ ret(0);
1740
1741 // Copy in 32-bytes chunks
1742 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1743
1744 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1745 restore_arg_regs();
1746 __ xorptr(rax, rax); // return 0
1747 __ leave(); // required for proper stackwalking of RuntimeStub frame
1748 __ ret(0);
1749
1750 return start;
1751 }
1752
1753 // Arguments:
1754 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1755 // ignored
1756 // is_oop - true => oop array, so generate store check code
1757 // name - stub name string
1758 //
1759 // Inputs:
1760 // c_rarg0 - source array address
1761 // c_rarg1 - destination array address
1762 // c_rarg2 - element count, treated as ssize_t, can be zero
1763 //
1764 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1765 // the hardware handle it. The two dwords within qwords that span
1766 // cache line boundaries will still be loaded and stored atomicly.
1767 //
1768 // Side Effects:
1769 // disjoint_int_copy_entry is set to the no-overlap entry point
1770 // used by generate_conjoint_int_oop_copy().
1771 //
1772 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry, const char *name) {
1773 __ align(CodeEntryAlignment);
1774 StubCodeMark mark(this, "StubRoutines", name);
1775 address start = __ pc();
1776
1777 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1778 const Register from = rdi; // source array address
1779 const Register to = rsi; // destination array address
1780 const Register count = rdx; // elements count
1781 const Register dword_count = rcx;
1782 const Register qword_count = count;
1783 const Register end_from = from; // source array end address
1784 const Register end_to = to; // destination array end address
1785 const Register saved_to = r11; // saved destination array address
1786 // End pointers are inclusive, and if count is not zero they point
1787 // to the last unit copied: end_to[0] := end_from[0]
1788
1789 __ enter(); // required for proper stackwalking of RuntimeStub frame
1790 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1791
1792 if (entry != NULL) {
1793 *entry = __ pc();
1794 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1795 BLOCK_COMMENT("Entry:");
1796 }
1797
1798 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1799 // r9 and r10 may be used to save non-volatile registers
1800 if (is_oop) {
1801 __ movq(saved_to, to);
1802 gen_write_ref_array_pre_barrier(to, count);
1803 }
1804
1805 // 'from', 'to' and 'count' are now valid
1806 __ movptr(dword_count, count);
1807 __ shrptr(count, 1); // count => qword_count
1808
1809 // Copy from low to high addresses. Use 'to' as scratch.
1810 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1811 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1812 __ negptr(qword_count);
1813 __ jmp(L_copy_32_bytes);
1814
1815 // Copy trailing qwords
1816 __ BIND(L_copy_8_bytes);
1817 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1818 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1819 __ increment(qword_count);
1820 __ jcc(Assembler::notZero, L_copy_8_bytes);
1821
1822 // Check for and copy trailing dword
1823 __ BIND(L_copy_4_bytes);
1824 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1825 __ jccb(Assembler::zero, L_exit);
1826 __ movl(rax, Address(end_from, 8));
1827 __ movl(Address(end_to, 8), rax);
1828
1829 __ BIND(L_exit);
1830 if (is_oop) {
1831 __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1832 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1833 }
1834 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1835 restore_arg_regs();
1836 __ xorptr(rax, rax); // return 0
1837 __ leave(); // required for proper stackwalking of RuntimeStub frame
1838 __ ret(0);
1839
1840 // Copy 32-bytes chunks
1841 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1842 __ jmp(L_copy_4_bytes);
1843
1844 return start;
1845 }
1846
1847 // Arguments:
1848 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1849 // ignored
1850 // is_oop - true => oop array, so generate store check code
1851 // name - stub name string
1852 //
1853 // Inputs:
1854 // c_rarg0 - source array address
1855 // c_rarg1 - destination array address
1856 // c_rarg2 - element count, treated as ssize_t, can be zero
1857 //
1858 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1859 // the hardware handle it. The two dwords within qwords that span
1860 // cache line boundaries will still be loaded and stored atomicly.
1861 //
1862 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1863 address *entry, const char *name) {
1864 __ align(CodeEntryAlignment);
1865 StubCodeMark mark(this, "StubRoutines", name);
1866 address start = __ pc();
1867
1868 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1869 const Register from = rdi; // source array address
1870 const Register to = rsi; // destination array address
1871 const Register count = rdx; // elements count
1872 const Register dword_count = rcx;
1873 const Register qword_count = count;
1874
1875 __ enter(); // required for proper stackwalking of RuntimeStub frame
1876 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1877
1878 if (entry != NULL) {
1879 *entry = __ pc();
1880 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1881 BLOCK_COMMENT("Entry:");
1882 }
1883
1884 array_overlap_test(nooverlap_target, Address::times_4);
1885 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1886 // r9 and r10 may be used to save non-volatile registers
1887
1888 if (is_oop) {
1889 // no registers are destroyed by this call
1890 gen_write_ref_array_pre_barrier(to, count);
1891 }
1892
1893 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1894 // 'from', 'to' and 'count' are now valid
1895 __ movptr(dword_count, count);
1896 __ shrptr(count, 1); // count => qword_count
1897
1898 // Copy from high to low addresses. Use 'to' as scratch.
1899
1900 // Check for and copy trailing dword
1901 __ testl(dword_count, 1);
1902 __ jcc(Assembler::zero, L_copy_32_bytes);
1903 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1904 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1905 __ jmp(L_copy_32_bytes);
1906
1907 // Copy trailing qwords
1908 __ BIND(L_copy_8_bytes);
1909 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1910 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1911 __ decrement(qword_count);
1912 __ jcc(Assembler::notZero, L_copy_8_bytes);
1913
1914 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1915 if (is_oop) {
1916 __ jmp(L_exit);
1917 }
1918 restore_arg_regs();
1919 __ xorptr(rax, rax); // return 0
1920 __ leave(); // required for proper stackwalking of RuntimeStub frame
1921 __ ret(0);
1922
1923 // Copy in 32-bytes chunks
1924 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1925
1926 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1927 __ bind(L_exit);
1928 if (is_oop) {
1929 Register end_to = rdx;
1930 __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1931 gen_write_ref_array_post_barrier(to, end_to, rax);
1932 }
1933 restore_arg_regs();
1934 __ xorptr(rax, rax); // return 0
1935 __ leave(); // required for proper stackwalking of RuntimeStub frame
1936 __ ret(0);
1937
1938 return start;
1939 }
1940
1941 // Arguments:
1942 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1943 // ignored
1944 // is_oop - true => oop array, so generate store check code
1945 // name - stub name string
1946 //
1947 // Inputs:
1948 // c_rarg0 - source array address
1949 // c_rarg1 - destination array address
1950 // c_rarg2 - element count, treated as ssize_t, can be zero
1951 //
1952 // Side Effects:
1953 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1954 // no-overlap entry point used by generate_conjoint_long_oop_copy().
1955 //
1956 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry, const char *name) {
1957 __ align(CodeEntryAlignment);
1958 StubCodeMark mark(this, "StubRoutines", name);
1959 address start = __ pc();
1960
1961 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1962 const Register from = rdi; // source array address
1963 const Register to = rsi; // destination array address
1964 const Register qword_count = rdx; // elements count
1965 const Register end_from = from; // source array end address
1966 const Register end_to = rcx; // destination array end address
1967 const Register saved_to = to;
1968 // End pointers are inclusive, and if count is not zero they point
1969 // to the last unit copied: end_to[0] := end_from[0]
1970
1971 __ enter(); // required for proper stackwalking of RuntimeStub frame
1972 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1973 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1974
1975 if (entry != NULL) {
1976 *entry = __ pc();
1977 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1978 BLOCK_COMMENT("Entry:");
1979 }
1980
1981 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1982 // r9 and r10 may be used to save non-volatile registers
1983 // 'from', 'to' and 'qword_count' are now valid
1984 if (is_oop) {
1985 // no registers are destroyed by this call
1986 gen_write_ref_array_pre_barrier(to, qword_count);
1987 }
1988
1989 // Copy from low to high addresses. Use 'to' as scratch.
1990 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1991 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1992 __ negptr(qword_count);
1993 __ jmp(L_copy_32_bytes);
1994
1995 // Copy trailing qwords
1996 __ BIND(L_copy_8_bytes);
1997 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1998 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1999 __ increment(qword_count);
2000 __ jcc(Assembler::notZero, L_copy_8_bytes);
2001
2002 if (is_oop) {
2003 __ jmp(L_exit);
2004 } else {
2005 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2006 restore_arg_regs();
2007 __ xorptr(rax, rax); // return 0
2008 __ leave(); // required for proper stackwalking of RuntimeStub frame
2009 __ ret(0);
2010 }
2011
2012 // Copy 64-byte chunks
2013 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2014
2015 if (is_oop) {
2016 __ BIND(L_exit);
2017 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
2018 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2019 } else {
2020 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2021 }
2022 restore_arg_regs();
2023 __ xorptr(rax, rax); // return 0
2024 __ leave(); // required for proper stackwalking of RuntimeStub frame
2025 __ ret(0);
2026
2027 return start;
2028 }
2029
2030 // Arguments:
2031 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2032 // ignored
2033 // is_oop - true => oop array, so generate store check code
2034 // name - stub name string
2035 //
2036 // Inputs:
2037 // c_rarg0 - source array address
2038 // c_rarg1 - destination array address
2039 // c_rarg2 - element count, treated as ssize_t, can be zero
2040 //
2041 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
2042 address *entry, const char *name) {
2043 __ align(CodeEntryAlignment);
2044 StubCodeMark mark(this, "StubRoutines", name);
2045 address start = __ pc();
2046
2047 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2048 const Register from = rdi; // source array address
2049 const Register to = rsi; // destination array address
2050 const Register qword_count = rdx; // elements count
2051 const Register saved_count = rcx;
2052
2053 __ enter(); // required for proper stackwalking of RuntimeStub frame
2054 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2055
2056 if (entry != NULL) {
2057 *entry = __ pc();
2058 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2059 BLOCK_COMMENT("Entry:");
2060 }
2061
2062 array_overlap_test(nooverlap_target, Address::times_8);
2063 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2064 // r9 and r10 may be used to save non-volatile registers
2065 // 'from', 'to' and 'qword_count' are now valid
2066 if (is_oop) {
2067 // Save to and count for store barrier
2068 __ movptr(saved_count, qword_count);
2069 // No registers are destroyed by this call
2070 gen_write_ref_array_pre_barrier(to, saved_count);
2071 }
2072
2073 __ jmp(L_copy_32_bytes);
2074
2075 // Copy trailing qwords
2076 __ BIND(L_copy_8_bytes);
2077 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2078 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2079 __ decrement(qword_count);
2080 __ jcc(Assembler::notZero, L_copy_8_bytes);
2081
2082 if (is_oop) {
2083 __ jmp(L_exit);
2084 } else {
2085 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2086 restore_arg_regs();
2087 __ xorptr(rax, rax); // return 0
2088 __ leave(); // required for proper stackwalking of RuntimeStub frame
2089 __ ret(0);
2090 }
2091
2092 // Copy in 32-bytes chunks
2093 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2094
2095 if (is_oop) {
2096 __ BIND(L_exit);
2097 __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2098 gen_write_ref_array_post_barrier(to, rcx, rax);
2099 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2100 } else {
2101 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2102 }
2103 restore_arg_regs();
2104 __ xorptr(rax, rax); // return 0
2105 __ leave(); // required for proper stackwalking of RuntimeStub frame
2106 __ ret(0);
2107
2108 return start;
2109 }
2110
2111
2112 // Helper for generating a dynamic type check.
2113 // Smashes no registers.
2114 void generate_type_check(Register sub_klass,
2115 Register super_check_offset,
2116 Register super_klass,
2117 Label& L_success) {
2118 assert_different_registers(sub_klass, super_check_offset, super_klass);
2119
2120 BLOCK_COMMENT("type_check:");
2121
2122 Label L_miss;
2123
2124 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2125 super_check_offset);
2126 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2127
2128 // Fall through on failure!
2129 __ BIND(L_miss);
2130 }
2131
2132 //
2133 // Generate checkcasting array copy stub
2134 //
2135 // Input:
2136 // c_rarg0 - source array address
2137 // c_rarg1 - destination array address
2138 // c_rarg2 - element count, treated as ssize_t, can be zero
2139 // c_rarg3 - size_t ckoff (super_check_offset)
2140 // not Win64
2141 // c_rarg4 - oop ckval (super_klass)
2142 // Win64
2143 // rsp+40 - oop ckval (super_klass)
2144 //
2145 // Output:
2146 // rax == 0 - success
2147 // rax == -1^K - failure, where K is partial transfer count
2148 //
2149 address generate_checkcast_copy(const char *name, address *entry) {
2150
2151 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2152
2153 // Input registers (after setup_arg_regs)
2154 const Register from = rdi; // source array address
2155 const Register to = rsi; // destination array address
2156 const Register length = rdx; // elements count
2157 const Register ckoff = rcx; // super_check_offset
2158 const Register ckval = r8; // super_klass
2159
2160 // Registers used as temps (r13, r14 are save-on-entry)
2161 const Register end_from = from; // source array end address
2162 const Register end_to = r13; // destination array end address
2163 const Register count = rdx; // -(count_remaining)
2164 const Register r14_length = r14; // saved copy of length
2165 // End pointers are inclusive, and if length is not zero they point
2166 // to the last unit copied: end_to[0] := end_from[0]
2167
2168 const Register rax_oop = rax; // actual oop copied
2169 const Register r11_klass = r11; // oop._klass
2170
2171 //---------------------------------------------------------------
2172 // Assembler stub will be used for this call to arraycopy
2173 // if the two arrays are subtypes of Object[] but the
2174 // destination array type is not equal to or a supertype
2175 // of the source type. Each element must be separately
2176 // checked.
2177
2178 __ align(CodeEntryAlignment);
2179 StubCodeMark mark(this, "StubRoutines", name);
2180 address start = __ pc();
2181
2182 __ enter(); // required for proper stackwalking of RuntimeStub frame
2183
2184 #ifdef ASSERT
2185 // caller guarantees that the arrays really are different
2186 // otherwise, we would have to make conjoint checks
2187 { Label L;
2188 array_overlap_test(L, TIMES_OOP);
2189 __ stop("checkcast_copy within a single array");
2190 __ bind(L);
2191 }
2192 #endif //ASSERT
2193
2194 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2195 // ckoff => rcx, ckval => r8
2196 // r9 and r10 may be used to save non-volatile registers
2197 #ifdef _WIN64
2198 // last argument (#4) is on stack on Win64
2199 __ movptr(ckval, Address(rsp, 6 * wordSize));
2200 #endif
2201
2202 // Caller of this entry point must set up the argument registers.
2203 if (entry != NULL) {
2204 *entry = __ pc();
2205 BLOCK_COMMENT("Entry:");
2206 }
2207
2208 // allocate spill slots for r13, r14
2209 enum {
2210 saved_r13_offset,
2211 saved_r14_offset,
2212 saved_rbp_offset
2213 };
2214 __ subptr(rsp, saved_rbp_offset * wordSize);
2215 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2216 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2217
2218 // check that int operands are properly extended to size_t
2219 assert_clean_int(length, rax);
2220 assert_clean_int(ckoff, rax);
2221
2222 #ifdef ASSERT
2223 BLOCK_COMMENT("assert consistent ckoff/ckval");
2224 // The ckoff and ckval must be mutually consistent,
2225 // even though caller generates both.
2226 { Label L;
2227 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2228 Klass::super_check_offset_offset_in_bytes());
2229 __ cmpl(ckoff, Address(ckval, sco_offset));
2230 __ jcc(Assembler::equal, L);
2231 __ stop("super_check_offset inconsistent");
2232 __ bind(L);
2233 }
2234 #endif //ASSERT
2235
2236 // Loop-invariant addresses. They are exclusive end pointers.
2237 Address end_from_addr(from, length, TIMES_OOP, 0);
2238 Address end_to_addr(to, length, TIMES_OOP, 0);
2239 // Loop-variant addresses. They assume post-incremented count < 0.
2240 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2241 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2242
2243 gen_write_ref_array_pre_barrier(to, count);
2244
2245 // Copy from low to high addresses, indexed from the end of each array.
2246 __ lea(end_from, end_from_addr);
2247 __ lea(end_to, end_to_addr);
2248 __ movptr(r14_length, length); // save a copy of the length
2249 assert(length == count, ""); // else fix next line:
2250 __ negptr(count); // negate and test the length
2251 __ jcc(Assembler::notZero, L_load_element);
2252
2253 // Empty array: Nothing to do.
2254 __ xorptr(rax, rax); // return 0 on (trivial) success
2255 __ jmp(L_done);
2256
2257 // ======== begin loop ========
2258 // (Loop is rotated; its entry is L_load_element.)
2259 // Loop control:
2260 // for (count = -count; count != 0; count++)
2261 // Base pointers src, dst are biased by 8*(count-1),to last element.
2262 __ align(OptoLoopAlignment);
2263
2264 __ BIND(L_store_element);
2265 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2266 __ increment(count); // increment the count toward zero
2267 __ jcc(Assembler::zero, L_do_card_marks);
2268
2269 // ======== loop entry is here ========
2270 __ BIND(L_load_element);
2271 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2272 __ testptr(rax_oop, rax_oop);
2273 __ jcc(Assembler::zero, L_store_element);
2274
2275 __ load_klass(r11_klass, rax_oop);// query the object klass
2276 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2277 // ======== end loop ========
2278
2279 // It was a real error; we must depend on the caller to finish the job.
2280 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2281 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2282 // and report their number to the caller.
2283 assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2284 __ lea(end_to, to_element_addr);
2285 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2286 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2287 __ movptr(rax, r14_length); // original oops
2288 __ addptr(rax, count); // K = (original - remaining) oops
2289 __ notptr(rax); // report (-1^K) to caller
2290 __ jmp(L_done);
2291
2292 // Come here on success only.
2293 __ BIND(L_do_card_marks);
2294 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2295 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2296 __ xorptr(rax, rax); // return 0 on success
2297
2298 // Common exit point (success or failure).
2299 __ BIND(L_done);
2300 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2301 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2302 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
2303 restore_arg_regs();
2304 __ leave(); // required for proper stackwalking of RuntimeStub frame
2305 __ ret(0);
2306
2307 return start;
2308 }
2309
2310 //
2311 // Generate 'unsafe' array copy stub
2312 // Though just as safe as the other stubs, it takes an unscaled
2313 // size_t argument instead of an element count.
2314 //
2315 // Input:
2316 // c_rarg0 - source array address
2317 // c_rarg1 - destination array address
2318 // c_rarg2 - byte count, treated as ssize_t, can be zero
2319 //
2320 // Examines the alignment of the operands and dispatches
2321 // to a long, int, short, or byte copy loop.
2322 //
2323 address generate_unsafe_copy(const char *name,
2324 address byte_copy_entry, address short_copy_entry,
2325 address int_copy_entry, address long_copy_entry) {
2326
2327 Label L_long_aligned, L_int_aligned, L_short_aligned;
2328
2329 // Input registers (before setup_arg_regs)
2330 const Register from = c_rarg0; // source array address
2331 const Register to = c_rarg1; // destination array address
2332 const Register size = c_rarg2; // byte count (size_t)
2333
2334 // Register used as a temp
2335 const Register bits = rax; // test copy of low bits
2336
2337 __ align(CodeEntryAlignment);
2338 StubCodeMark mark(this, "StubRoutines", name);
2339 address start = __ pc();
2340
2341 __ enter(); // required for proper stackwalking of RuntimeStub frame
2342
2343 // bump this on entry, not on exit:
2344 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2345
2346 __ mov(bits, from);
2347 __ orptr(bits, to);
2348 __ orptr(bits, size);
2349
2350 __ testb(bits, BytesPerLong-1);
2351 __ jccb(Assembler::zero, L_long_aligned);
2352
2353 __ testb(bits, BytesPerInt-1);
2354 __ jccb(Assembler::zero, L_int_aligned);
2355
2356 __ testb(bits, BytesPerShort-1);
2357 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2358
2359 __ BIND(L_short_aligned);
2360 __ shrptr(size, LogBytesPerShort); // size => short_count
2361 __ jump(RuntimeAddress(short_copy_entry));
2362
2363 __ BIND(L_int_aligned);
2364 __ shrptr(size, LogBytesPerInt); // size => int_count
2365 __ jump(RuntimeAddress(int_copy_entry));
2366
2367 __ BIND(L_long_aligned);
2368 __ shrptr(size, LogBytesPerLong); // size => qword_count
2369 __ jump(RuntimeAddress(long_copy_entry));
2370
2371 return start;
2372 }
2373
2374 // Perform range checks on the proposed arraycopy.
2375 // Kills temp, but nothing else.
2376 // Also, clean the sign bits of src_pos and dst_pos.
2377 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2378 Register src_pos, // source position (c_rarg1)
2379 Register dst, // destination array oo (c_rarg2)
2380 Register dst_pos, // destination position (c_rarg3)
2381 Register length,
2382 Register temp,
2383 Label& L_failed) {
2384 BLOCK_COMMENT("arraycopy_range_checks:");
2385
2386 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2387 __ movl(temp, length);
2388 __ addl(temp, src_pos); // src_pos + length
2389 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2390 __ jcc(Assembler::above, L_failed);
2391
2392 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2393 __ movl(temp, length);
2394 __ addl(temp, dst_pos); // dst_pos + length
2395 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2396 __ jcc(Assembler::above, L_failed);
2397
2398 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2399 // Move with sign extension can be used since they are positive.
2400 __ movslq(src_pos, src_pos);
2401 __ movslq(dst_pos, dst_pos);
2402
2403 BLOCK_COMMENT("arraycopy_range_checks done");
2404 }
2405
2406 //
2407 // Generate generic array copy stubs
2408 //
2409 // Input:
2410 // c_rarg0 - src oop
2411 // c_rarg1 - src_pos (32-bits)
2412 // c_rarg2 - dst oop
2413 // c_rarg3 - dst_pos (32-bits)
2414 // not Win64
2415 // c_rarg4 - element count (32-bits)
2416 // Win64
2417 // rsp+40 - element count (32-bits)
2418 //
2419 // Output:
2420 // rax == 0 - success
2421 // rax == -1^K - failure, where K is partial transfer count
2422 //
2423 address generate_generic_copy(const char *name,
2424 address byte_copy_entry, address short_copy_entry,
2425 address int_copy_entry, address long_copy_entry,
2426 address oop_copy_entry, address checkcast_copy_entry) {
2427
2428 Label L_failed, L_failed_0, L_objArray;
2429 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2430
2431 // Input registers
2432 const Register src = c_rarg0; // source array oop
2433 const Register src_pos = c_rarg1; // source position
2434 const Register dst = c_rarg2; // destination array oop
2435 const Register dst_pos = c_rarg3; // destination position
2436 #ifndef _WIN64
2437 const Register length = c_rarg4;
2438 #else
2439 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2440 #endif
2441
2442 { int modulus = CodeEntryAlignment;
2443 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2444 int advance = target - (__ offset() % modulus);
2445 if (advance < 0) advance += modulus;
2446 if (advance > 0) __ nop(advance);
2447 }
2448 StubCodeMark mark(this, "StubRoutines", name);
2449
2450 // Short-hop target to L_failed. Makes for denser prologue code.
2451 __ BIND(L_failed_0);
2452 __ jmp(L_failed);
2453 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2454
2455 __ align(CodeEntryAlignment);
2456 address start = __ pc();
2457
2458 __ enter(); // required for proper stackwalking of RuntimeStub frame
2459
2460 // bump this on entry, not on exit:
2461 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2462
2463 //-----------------------------------------------------------------------
2464 // Assembler stub will be used for this call to arraycopy
2465 // if the following conditions are met:
2466 //
2467 // (1) src and dst must not be null.
2468 // (2) src_pos must not be negative.
2469 // (3) dst_pos must not be negative.
2470 // (4) length must not be negative.
2471 // (5) src klass and dst klass should be the same and not NULL.
2472 // (6) src and dst should be arrays.
2473 // (7) src_pos + length must not exceed length of src.
2474 // (8) dst_pos + length must not exceed length of dst.
2475 //
2476
2477 // if (src == NULL) return -1;
2478 __ testptr(src, src); // src oop
2479 size_t j1off = __ offset();
2480 __ jccb(Assembler::zero, L_failed_0);
2481
2482 // if (src_pos < 0) return -1;
2483 __ testl(src_pos, src_pos); // src_pos (32-bits)
2484 __ jccb(Assembler::negative, L_failed_0);
2485
2486 // if (dst == NULL) return -1;
2487 __ testptr(dst, dst); // dst oop
2488 __ jccb(Assembler::zero, L_failed_0);
2489
2490 // if (dst_pos < 0) return -1;
2491 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2492 size_t j4off = __ offset();
2493 __ jccb(Assembler::negative, L_failed_0);
2494
2495 // The first four tests are very dense code,
2496 // but not quite dense enough to put four
2497 // jumps in a 16-byte instruction fetch buffer.
2498 // That's good, because some branch predicters
2499 // do not like jumps so close together.
2500 // Make sure of this.
2501 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2502
2503 // registers used as temp
2504 const Register r11_length = r11; // elements count to copy
2505 const Register r10_src_klass = r10; // array klass
2506
2507 // if (length < 0) return -1;
2508 __ movl(r11_length, length); // length (elements count, 32-bits value)
2509 __ testl(r11_length, r11_length);
2510 __ jccb(Assembler::negative, L_failed_0);
2511
2512 __ load_klass(r10_src_klass, src);
2513 #ifdef ASSERT
2514 // assert(src->klass() != NULL);
2515 {
2516 BLOCK_COMMENT("assert klasses not null {");
2517 Label L1, L2;
2518 __ testptr(r10_src_klass, r10_src_klass);
2519 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2520 __ bind(L1);
2521 __ stop("broken null klass");
2522 __ bind(L2);
2523 __ load_klass(rax, dst);
2524 __ cmpq(rax, 0);
2525 __ jcc(Assembler::equal, L1); // this would be broken also
2526 BLOCK_COMMENT("} assert klasses not null done");
2527 }
2528 #endif
2529
2530 // Load layout helper (32-bits)
2531 //
2532 // |array_tag| | header_size | element_type | |log2_element_size|
2533 // 32 30 24 16 8 2 0
2534 //
2535 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2536 //
2537
2538 const int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2539 Klass::layout_helper_offset_in_bytes();
2540
2541 // Handle objArrays completely differently...
2542 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2543 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2544 __ jcc(Assembler::equal, L_objArray);
2545
2546 // if (src->klass() != dst->klass()) return -1;
2547 __ load_klass(rax, dst);
2548 __ cmpq(r10_src_klass, rax);
2549 __ jcc(Assembler::notEqual, L_failed);
2550
2551 const Register rax_lh = rax; // layout helper
2552 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2553
2554 // if (!src->is_Array()) return -1;
2555 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2556 __ jcc(Assembler::greaterEqual, L_failed);
2557
2558 // At this point, it is known to be a typeArray (array_tag 0x3).
2559 #ifdef ASSERT
2560 {
2561 BLOCK_COMMENT("assert primitive array {");
2562 Label L;
2563 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2564 __ jcc(Assembler::greaterEqual, L);
2565 __ stop("must be a primitive array");
2566 __ bind(L);
2567 BLOCK_COMMENT("} assert primitive array done");
2568 }
2569 #endif
2570
2571 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2572 r10, L_failed);
2573
2574 // typeArrayKlass
2575 //
2576 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2577 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2578 //
2579
2580 const Register r10_offset = r10; // array offset
2581 const Register rax_elsize = rax_lh; // element size
2582
2583 __ movl(r10_offset, rax_lh);
2584 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2585 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2586 __ addptr(src, r10_offset); // src array offset
2587 __ addptr(dst, r10_offset); // dst array offset
2588 BLOCK_COMMENT("choose copy loop based on element size");
2589 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2590
2591 // next registers should be set before the jump to corresponding stub
2592 const Register from = c_rarg0; // source array address
2593 const Register to = c_rarg1; // destination array address
2594 const Register count = c_rarg2; // elements count
2595
2596 // 'from', 'to', 'count' registers should be set in such order
2597 // since they are the same as 'src', 'src_pos', 'dst'.
2598
2599 __ BIND(L_copy_bytes);
2600 __ cmpl(rax_elsize, 0);
2601 __ jccb(Assembler::notEqual, L_copy_shorts);
2602 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2603 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2604 __ movl2ptr(count, r11_length); // length
2605 __ jump(RuntimeAddress(byte_copy_entry));
2606
2607 __ BIND(L_copy_shorts);
2608 __ cmpl(rax_elsize, LogBytesPerShort);
2609 __ jccb(Assembler::notEqual, L_copy_ints);
2610 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2611 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2612 __ movl2ptr(count, r11_length); // length
2613 __ jump(RuntimeAddress(short_copy_entry));
2614
2615 __ BIND(L_copy_ints);
2616 __ cmpl(rax_elsize, LogBytesPerInt);
2617 __ jccb(Assembler::notEqual, L_copy_longs);
2618 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2619 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2620 __ movl2ptr(count, r11_length); // length
2621 __ jump(RuntimeAddress(int_copy_entry));
2622
2623 __ BIND(L_copy_longs);
2624 #ifdef ASSERT
2625 {
2626 BLOCK_COMMENT("assert long copy {");
2627 Label L;
2628 __ cmpl(rax_elsize, LogBytesPerLong);
2629 __ jcc(Assembler::equal, L);
2630 __ stop("must be long copy, but elsize is wrong");
2631 __ bind(L);
2632 BLOCK_COMMENT("} assert long copy done");
2633 }
2634 #endif
2635 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2636 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2637 __ movl2ptr(count, r11_length); // length
2638 __ jump(RuntimeAddress(long_copy_entry));
2639
2640 // objArrayKlass
2641 __ BIND(L_objArray);
2642 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2643
2644 Label L_plain_copy, L_checkcast_copy;
2645 // test array classes for subtyping
2646 __ load_klass(rax, dst);
2647 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2648 __ jcc(Assembler::notEqual, L_checkcast_copy);
2649
2650 // Identically typed arrays can be copied without element-wise checks.
2651 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2652 r10, L_failed);
2653
2654 __ lea(from, Address(src, src_pos, TIMES_OOP,
2655 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2656 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2657 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2658 __ movl2ptr(count, r11_length); // length
2659 __ BIND(L_plain_copy);
2660 __ jump(RuntimeAddress(oop_copy_entry));
2661
2662 __ BIND(L_checkcast_copy);
2663 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2664 {
2665 // Before looking at dst.length, make sure dst is also an objArray.
2666 __ cmpl(Address(rax, lh_offset), objArray_lh);
2667 __ jcc(Assembler::notEqual, L_failed);
2668
2669 // It is safe to examine both src.length and dst.length.
2670 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2671 rax, L_failed);
2672
2673 const Register r11_dst_klass = r11;
2674 __ load_klass(r11_dst_klass, dst); // reload
2675
2676 // Marshal the base address arguments now, freeing registers.
2677 __ lea(from, Address(src, src_pos, TIMES_OOP,
2678 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2679 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2680 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2681 __ movl(count, length); // length (reloaded)
2682 Register sco_temp = c_rarg3; // this register is free now
2683 assert_different_registers(from, to, count, sco_temp,
2684 r11_dst_klass, r10_src_klass);
2685 assert_clean_int(count, sco_temp);
2686
2687 // Generate the type check.
2688 const int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2689 Klass::super_check_offset_offset_in_bytes());
2690 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2691 assert_clean_int(sco_temp, rax);
2692 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2693
2694 // Fetch destination element klass from the objArrayKlass header.
2695 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2696 objArrayKlass::element_klass_offset_in_bytes());
2697 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2698 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2699 assert_clean_int(sco_temp, rax);
2700
2701 // the checkcast_copy loop needs two extra arguments:
2702 assert(c_rarg3 == sco_temp, "#3 already in place");
2703 // Set up arguments for checkcast_copy_entry.
2704 setup_arg_regs(4);
2705 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2706 __ jump(RuntimeAddress(checkcast_copy_entry));
2707 }
2708
2709 __ BIND(L_failed);
2710 __ xorptr(rax, rax);
2711 __ notptr(rax); // return -1
2712 __ leave(); // required for proper stackwalking of RuntimeStub frame
2713 __ ret(0);
2714
2715 return start;
2716 }
2717
2718 void generate_arraycopy_stubs() {
2719 address entry;
2720 address entry_jbyte_arraycopy;
2721 address entry_jshort_arraycopy;
2722 address entry_jint_arraycopy;
2723 address entry_oop_arraycopy;
2724 address entry_jlong_arraycopy;
2725 address entry_checkcast_arraycopy;
2726
2727 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2728 "jbyte_disjoint_arraycopy");
2729 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2730 "jbyte_arraycopy");
2731
2732 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2733 "jshort_disjoint_arraycopy");
2734 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2735 "jshort_arraycopy");
2736
2737 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2738 "jint_disjoint_arraycopy");
2739 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2740 &entry_jint_arraycopy, "jint_arraycopy");
2741
2742 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2743 "jlong_disjoint_arraycopy");
2744 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2745 &entry_jlong_arraycopy, "jlong_arraycopy");
2746
2747
2748 if (UseCompressedOops) {
2749 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2750 "oop_disjoint_arraycopy");
2751 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2752 &entry_oop_arraycopy, "oop_arraycopy");
2753 } else {
2754 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2755 "oop_disjoint_arraycopy");
2756 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2757 &entry_oop_arraycopy, "oop_arraycopy");
2758 }
2759
2760 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2761 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2762 entry_jbyte_arraycopy,
2763 entry_jshort_arraycopy,
2764 entry_jint_arraycopy,
2765 entry_jlong_arraycopy);
2766 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2767 entry_jbyte_arraycopy,
2768 entry_jshort_arraycopy,
2769 entry_jint_arraycopy,
2770 entry_oop_arraycopy,
2771 entry_jlong_arraycopy,
2772 entry_checkcast_arraycopy);
2773
2774 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2775 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2776 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2777 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2778 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2779 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2780
2781 // We don't generate specialized code for HeapWord-aligned source
2782 // arrays, so just use the code we've already generated
2783 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2784 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2785
2786 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2787 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2788
2789 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2790 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2791
2792 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2793 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2794
2795 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2796 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2797 }
2798
2799 void generate_math_stubs() {
2800 {
2801 StubCodeMark mark(this, "StubRoutines", "log");
2802 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2803
2804 __ subq(rsp, 8);
2805 __ movdbl(Address(rsp, 0), xmm0);
2806 __ fld_d(Address(rsp, 0));
2807 __ flog();
2808 __ fstp_d(Address(rsp, 0));
2809 __ movdbl(xmm0, Address(rsp, 0));
2810 __ addq(rsp, 8);
2811 __ ret(0);
2812 }
2813 {
2814 StubCodeMark mark(this, "StubRoutines", "log10");
2815 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2816
2817 __ subq(rsp, 8);
2818 __ movdbl(Address(rsp, 0), xmm0);
2819 __ fld_d(Address(rsp, 0));
2820 __ flog10();
2821 __ fstp_d(Address(rsp, 0));
2822 __ movdbl(xmm0, Address(rsp, 0));
2823 __ addq(rsp, 8);
2824 __ ret(0);
2825 }
2826 {
2827 StubCodeMark mark(this, "StubRoutines", "sin");
2828 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2829
2830 __ subq(rsp, 8);
2831 __ movdbl(Address(rsp, 0), xmm0);
2832 __ fld_d(Address(rsp, 0));
2833 __ trigfunc('s');
2834 __ fstp_d(Address(rsp, 0));
2835 __ movdbl(xmm0, Address(rsp, 0));
2836 __ addq(rsp, 8);
2837 __ ret(0);
2838 }
2839 {
2840 StubCodeMark mark(this, "StubRoutines", "cos");
2841 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2842
2843 __ subq(rsp, 8);
2844 __ movdbl(Address(rsp, 0), xmm0);
2845 __ fld_d(Address(rsp, 0));
2846 __ trigfunc('c');
2847 __ fstp_d(Address(rsp, 0));
2848 __ movdbl(xmm0, Address(rsp, 0));
2849 __ addq(rsp, 8);
2850 __ ret(0);
2851 }
2852 {
2853 StubCodeMark mark(this, "StubRoutines", "tan");
2854 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2855
2856 __ subq(rsp, 8);
2857 __ movdbl(Address(rsp, 0), xmm0);
2858 __ fld_d(Address(rsp, 0));
2859 __ trigfunc('t');
2860 __ fstp_d(Address(rsp, 0));
2861 __ movdbl(xmm0, Address(rsp, 0));
2862 __ addq(rsp, 8);
2863 __ ret(0);
2864 }
2865
2866 // The intrinsic version of these seem to return the same value as
2867 // the strict version.
2868 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2869 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2870 }
2871
2872 #undef __
2873 #define __ masm->
2874
2875 // Continuation point for throwing of implicit exceptions that are
2876 // not handled in the current activation. Fabricates an exception
2877 // oop and initiates normal exception dispatching in this
2878 // frame. Since we need to preserve callee-saved values (currently
2879 // only for C2, but done for C1 as well) we need a callee-saved oop
2880 // map and therefore have to make these stubs into RuntimeStubs
2881 // rather than BufferBlobs. If the compiler needs all registers to
2882 // be preserved between the fault point and the exception handler
2883 // then it must assume responsibility for that in
2884 // AbstractCompiler::continuation_for_implicit_null_exception or
2885 // continuation_for_implicit_division_by_zero_exception. All other
2886 // implicit exceptions (e.g., NullPointerException or
2887 // AbstractMethodError on entry) are either at call sites or
2888 // otherwise assume that stack unwinding will be initiated, so
2889 // caller saved registers were assumed volatile in the compiler.
2890 address generate_throw_exception(const char* name,
2891 address runtime_entry,
2892 bool restore_saved_exception_pc) {
2893 // Information about frame layout at time of blocking runtime call.
2894 // Note that we only have to preserve callee-saved registers since
2895 // the compilers are responsible for supplying a continuation point
2896 // if they expect all registers to be preserved.
2897 enum layout {
2898 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2899 rbp_off2,
2900 return_off,
2901 return_off2,
2902 framesize // inclusive of return address
2903 };
2904
2905 int insts_size = 512;
2906 int locs_size = 64;
2907
2908 CodeBuffer code(name, insts_size, locs_size);
2909 OopMapSet* oop_maps = new OopMapSet();
2910 MacroAssembler* masm = new MacroAssembler(&code);
2911
2912 address start = __ pc();
2913
2914 // This is an inlined and slightly modified version of call_VM
2915 // which has the ability to fetch the return PC out of
2916 // thread-local storage and also sets up last_Java_sp slightly
2917 // differently than the real call_VM
2918 if (restore_saved_exception_pc) {
2919 __ movptr(rax,
2920 Address(r15_thread,
2921 in_bytes(JavaThread::saved_exception_pc_offset())));
2922 __ push(rax);
2923 }
2924
2925 __ enter(); // required for proper stackwalking of RuntimeStub frame
2926
2927 assert(is_even(framesize/2), "sp not 16-byte aligned");
2928
2929 // return address and rbp are already in place
2930 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
2931
2932 int frame_complete = __ pc() - start;
2933
2934 // Set up last_Java_sp and last_Java_fp
2935 __ set_last_Java_frame(rsp, rbp, NULL);
2936
2937 // Call runtime
2938 __ movptr(c_rarg0, r15_thread);
2939 BLOCK_COMMENT("call runtime_entry");
2940 __ call(RuntimeAddress(runtime_entry));
2941
2942 // Generate oop map
2943 OopMap* map = new OopMap(framesize, 0);
2944
2945 oop_maps->add_gc_map(__ pc() - start, map);
2946
2947 __ reset_last_Java_frame(true, false);
2948
2949 __ leave(); // required for proper stackwalking of RuntimeStub frame
2950
2951 // check for pending exceptions
2952 #ifdef ASSERT
2953 Label L;
2954 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
2955 (int32_t) NULL_WORD);
2956 __ jcc(Assembler::notEqual, L);
2957 __ should_not_reach_here();
2958 __ bind(L);
2959 #endif // ASSERT
2960 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2961
2962
2963 // codeBlob framesize is in words (not VMRegImpl::slot_size)
2964 RuntimeStub* stub =
2965 RuntimeStub::new_runtime_stub(name,
2966 &code,
2967 frame_complete,
2968 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2969 oop_maps, false);
2970 return stub->entry_point();
2971 }
2972
2973 // Initialization
2974 void generate_initial() {
2975 // Generates all stubs and initializes the entry points
2976
2977 // This platform-specific stub is needed by generate_call_stub()
2978 StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
2979
2980 // entry points that exist in all platforms Note: This is code
2981 // that could be shared among different platforms - however the
2982 // benefit seems to be smaller than the disadvantage of having a
2983 // much more complicated generator structure. See also comment in
2984 // stubRoutines.hpp.
2985
2986 StubRoutines::_forward_exception_entry = generate_forward_exception();
2987
2988 StubRoutines::_call_stub_entry =
2989 generate_call_stub(StubRoutines::_call_stub_return_address);
2990
2991 // is referenced by megamorphic call
2992 StubRoutines::_catch_exception_entry = generate_catch_exception();
2993
2994 // atomic calls
2995 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2996 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
2997 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
2998 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2999 StubRoutines::_atomic_add_entry = generate_atomic_add();
3000 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
3001 StubRoutines::_fence_entry = generate_orderaccess_fence();
3002
3003 StubRoutines::_handler_for_unsafe_access_entry =
3004 generate_handler_for_unsafe_access();
3005
3006 // platform dependent
3007 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
3008
3009 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3010 }
3011
3012 void generate_all() {
3013 // Generates all stubs and initializes the entry points
3014
3015 // These entry points require SharedInfo::stack0 to be set up in
3016 // non-core builds and need to be relocatable, so they each
3017 // fabricate a RuntimeStub internally.
3018 StubRoutines::_throw_AbstractMethodError_entry =
3019 generate_throw_exception("AbstractMethodError throw_exception",
3020 CAST_FROM_FN_PTR(address,
3021 SharedRuntime::
3022 throw_AbstractMethodError),
3023 false);
3024
3025 StubRoutines::_throw_IncompatibleClassChangeError_entry =
3026 generate_throw_exception("IncompatibleClassChangeError throw_exception",
3027 CAST_FROM_FN_PTR(address,
3028 SharedRuntime::
3029 throw_IncompatibleClassChangeError),
3030 false);
3031
3032 StubRoutines::_throw_ArithmeticException_entry =
3033 generate_throw_exception("ArithmeticException throw_exception",
3034 CAST_FROM_FN_PTR(address,
3035 SharedRuntime::
3036 throw_ArithmeticException),
3037 true);
3038
3039 StubRoutines::_throw_NullPointerException_entry =
3040 generate_throw_exception("NullPointerException throw_exception",
3041 CAST_FROM_FN_PTR(address,
3042 SharedRuntime::
3043 throw_NullPointerException),
3044 true);
3045
3046 StubRoutines::_throw_NullPointerException_at_call_entry =
3047 generate_throw_exception("NullPointerException at call throw_exception",
3048 CAST_FROM_FN_PTR(address,
3049 SharedRuntime::
3050 throw_NullPointerException_at_call),
3051 false);
3052
3053 StubRoutines::_throw_StackOverflowError_entry =
3054 generate_throw_exception("StackOverflowError throw_exception",
3055 CAST_FROM_FN_PTR(address,
3056 SharedRuntime::
3057 throw_StackOverflowError),
3058 false);
3059
3060 // entry points that are platform specific
3061 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
3062 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
3063 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
3064 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
3065
3066 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
3067 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
3068 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3069 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3070
3071 // support for verify_oop (must happen after universe_init)
3072 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3073
3074 // arraycopy stubs used by compilers
3075 generate_arraycopy_stubs();
3076
3077 generate_math_stubs();
3078 }
3079
3080 public:
3081 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3082 if (all) {
3083 generate_all();
3084 } else {
3085 generate_initial();
3086 }
3087 }
3088 }; // end class declaration
3089
3090 void StubGenerator_generate(CodeBuffer* code, bool all) {
3091 StubGenerator g(code, all);
3092 }