Thu, 26 Mar 2009 14:31:45 -0700
6822204: volatile fences should prefer lock:addl to actual mfence instructions
Reviewed-by: kvn, phh
1 /*
2 * Copyright 2003-2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_stubGenerator_x86_64.cpp.incl"
28 // Declaration and definition of StubGenerator (no .hpp file).
29 // For a more detailed description of the stub routine structure
30 // see the comment in stubRoutines.hpp
32 #define __ _masm->
33 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
34 #define a__ ((Assembler*)_masm)->
36 #ifdef PRODUCT
37 #define BLOCK_COMMENT(str) /* nothing */
38 #else
39 #define BLOCK_COMMENT(str) __ block_comment(str)
40 #endif
42 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
43 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
45 // Stub Code definitions
47 static address handle_unsafe_access() {
48 JavaThread* thread = JavaThread::current();
49 address pc = thread->saved_exception_pc();
50 // pc is the instruction which we must emulate
51 // doing a no-op is fine: return garbage from the load
52 // therefore, compute npc
53 address npc = Assembler::locate_next_instruction(pc);
55 // request an async exception
56 thread->set_pending_unsafe_access_error();
58 // return address of next instruction to execute
59 return npc;
60 }
62 class StubGenerator: public StubCodeGenerator {
63 private:
65 #ifdef PRODUCT
66 #define inc_counter_np(counter) (0)
67 #else
68 void inc_counter_np_(int& counter) {
69 __ incrementl(ExternalAddress((address)&counter));
70 }
71 #define inc_counter_np(counter) \
72 BLOCK_COMMENT("inc_counter " #counter); \
73 inc_counter_np_(counter);
74 #endif
76 // Call stubs are used to call Java from C
77 //
78 // Linux Arguments:
79 // c_rarg0: call wrapper address address
80 // c_rarg1: result address
81 // c_rarg2: result type BasicType
82 // c_rarg3: method methodOop
83 // c_rarg4: (interpreter) entry point address
84 // c_rarg5: parameters intptr_t*
85 // 16(rbp): parameter size (in words) int
86 // 24(rbp): thread Thread*
87 //
88 // [ return_from_Java ] <--- rsp
89 // [ argument word n ]
90 // ...
91 // -12 [ argument word 1 ]
92 // -11 [ saved r15 ] <--- rsp_after_call
93 // -10 [ saved r14 ]
94 // -9 [ saved r13 ]
95 // -8 [ saved r12 ]
96 // -7 [ saved rbx ]
97 // -6 [ call wrapper ]
98 // -5 [ result ]
99 // -4 [ result type ]
100 // -3 [ method ]
101 // -2 [ entry point ]
102 // -1 [ parameters ]
103 // 0 [ saved rbp ] <--- rbp
104 // 1 [ return address ]
105 // 2 [ parameter size ]
106 // 3 [ thread ]
107 //
108 // Windows Arguments:
109 // c_rarg0: call wrapper address address
110 // c_rarg1: result address
111 // c_rarg2: result type BasicType
112 // c_rarg3: method methodOop
113 // 48(rbp): (interpreter) entry point address
114 // 56(rbp): parameters intptr_t*
115 // 64(rbp): parameter size (in words) int
116 // 72(rbp): thread Thread*
117 //
118 // [ return_from_Java ] <--- rsp
119 // [ argument word n ]
120 // ...
121 // -8 [ argument word 1 ]
122 // -7 [ saved r15 ] <--- rsp_after_call
123 // -6 [ saved r14 ]
124 // -5 [ saved r13 ]
125 // -4 [ saved r12 ]
126 // -3 [ saved rdi ]
127 // -2 [ saved rsi ]
128 // -1 [ saved rbx ]
129 // 0 [ saved rbp ] <--- rbp
130 // 1 [ return address ]
131 // 2 [ call wrapper ]
132 // 3 [ result ]
133 // 4 [ result type ]
134 // 5 [ method ]
135 // 6 [ entry point ]
136 // 7 [ parameters ]
137 // 8 [ parameter size ]
138 // 9 [ thread ]
139 //
140 // Windows reserves the callers stack space for arguments 1-4.
141 // We spill c_rarg0-c_rarg3 to this space.
143 // Call stub stack layout word offsets from rbp
144 enum call_stub_layout {
145 #ifdef _WIN64
146 rsp_after_call_off = -7,
147 r15_off = rsp_after_call_off,
148 r14_off = -6,
149 r13_off = -5,
150 r12_off = -4,
151 rdi_off = -3,
152 rsi_off = -2,
153 rbx_off = -1,
154 rbp_off = 0,
155 retaddr_off = 1,
156 call_wrapper_off = 2,
157 result_off = 3,
158 result_type_off = 4,
159 method_off = 5,
160 entry_point_off = 6,
161 parameters_off = 7,
162 parameter_size_off = 8,
163 thread_off = 9
164 #else
165 rsp_after_call_off = -12,
166 mxcsr_off = rsp_after_call_off,
167 r15_off = -11,
168 r14_off = -10,
169 r13_off = -9,
170 r12_off = -8,
171 rbx_off = -7,
172 call_wrapper_off = -6,
173 result_off = -5,
174 result_type_off = -4,
175 method_off = -3,
176 entry_point_off = -2,
177 parameters_off = -1,
178 rbp_off = 0,
179 retaddr_off = 1,
180 parameter_size_off = 2,
181 thread_off = 3
182 #endif
183 };
185 address generate_call_stub(address& return_address) {
186 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
187 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
188 "adjust this code");
189 StubCodeMark mark(this, "StubRoutines", "call_stub");
190 address start = __ pc();
192 // same as in generate_catch_exception()!
193 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
195 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
196 const Address result (rbp, result_off * wordSize);
197 const Address result_type (rbp, result_type_off * wordSize);
198 const Address method (rbp, method_off * wordSize);
199 const Address entry_point (rbp, entry_point_off * wordSize);
200 const Address parameters (rbp, parameters_off * wordSize);
201 const Address parameter_size(rbp, parameter_size_off * wordSize);
203 // same as in generate_catch_exception()!
204 const Address thread (rbp, thread_off * wordSize);
206 const Address r15_save(rbp, r15_off * wordSize);
207 const Address r14_save(rbp, r14_off * wordSize);
208 const Address r13_save(rbp, r13_off * wordSize);
209 const Address r12_save(rbp, r12_off * wordSize);
210 const Address rbx_save(rbp, rbx_off * wordSize);
212 // stub code
213 __ enter();
214 __ subptr(rsp, -rsp_after_call_off * wordSize);
216 // save register parameters
217 #ifndef _WIN64
218 __ movptr(parameters, c_rarg5); // parameters
219 __ movptr(entry_point, c_rarg4); // entry_point
220 #endif
222 __ movptr(method, c_rarg3); // method
223 __ movl(result_type, c_rarg2); // result type
224 __ movptr(result, c_rarg1); // result
225 __ movptr(call_wrapper, c_rarg0); // call wrapper
227 // save regs belonging to calling function
228 __ movptr(rbx_save, rbx);
229 __ movptr(r12_save, r12);
230 __ movptr(r13_save, r13);
231 __ movptr(r14_save, r14);
232 __ movptr(r15_save, r15);
234 #ifdef _WIN64
235 const Address rdi_save(rbp, rdi_off * wordSize);
236 const Address rsi_save(rbp, rsi_off * wordSize);
238 __ movptr(rsi_save, rsi);
239 __ movptr(rdi_save, rdi);
240 #else
241 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
242 {
243 Label skip_ldmx;
244 __ stmxcsr(mxcsr_save);
245 __ movl(rax, mxcsr_save);
246 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
247 ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
248 __ cmp32(rax, mxcsr_std);
249 __ jcc(Assembler::equal, skip_ldmx);
250 __ ldmxcsr(mxcsr_std);
251 __ bind(skip_ldmx);
252 }
253 #endif
255 // Load up thread register
256 __ movptr(r15_thread, thread);
257 __ reinit_heapbase();
259 #ifdef ASSERT
260 // make sure we have no pending exceptions
261 {
262 Label L;
263 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
264 __ jcc(Assembler::equal, L);
265 __ stop("StubRoutines::call_stub: entered with pending exception");
266 __ bind(L);
267 }
268 #endif
270 // pass parameters if any
271 BLOCK_COMMENT("pass parameters if any");
272 Label parameters_done;
273 __ movl(c_rarg3, parameter_size);
274 __ testl(c_rarg3, c_rarg3);
275 __ jcc(Assembler::zero, parameters_done);
277 Label loop;
278 __ movptr(c_rarg2, parameters); // parameter pointer
279 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
280 __ BIND(loop);
281 if (TaggedStackInterpreter) {
282 __ movl(rax, Address(c_rarg2, 0)); // get tag
283 __ addptr(c_rarg2, wordSize); // advance to next tag
284 __ push(rax); // pass tag
285 }
286 __ movptr(rax, Address(c_rarg2, 0));// get parameter
287 __ addptr(c_rarg2, wordSize); // advance to next parameter
288 __ decrementl(c_rarg1); // decrement counter
289 __ push(rax); // pass parameter
290 __ jcc(Assembler::notZero, loop);
292 // call Java function
293 __ BIND(parameters_done);
294 __ movptr(rbx, method); // get methodOop
295 __ movptr(c_rarg1, entry_point); // get entry_point
296 __ mov(r13, rsp); // set sender sp
297 BLOCK_COMMENT("call Java function");
298 __ call(c_rarg1);
300 BLOCK_COMMENT("call_stub_return_address:");
301 return_address = __ pc();
303 // store result depending on type (everything that is not
304 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
305 __ movptr(c_rarg0, result);
306 Label is_long, is_float, is_double, exit;
307 __ movl(c_rarg1, result_type);
308 __ cmpl(c_rarg1, T_OBJECT);
309 __ jcc(Assembler::equal, is_long);
310 __ cmpl(c_rarg1, T_LONG);
311 __ jcc(Assembler::equal, is_long);
312 __ cmpl(c_rarg1, T_FLOAT);
313 __ jcc(Assembler::equal, is_float);
314 __ cmpl(c_rarg1, T_DOUBLE);
315 __ jcc(Assembler::equal, is_double);
317 // handle T_INT case
318 __ movl(Address(c_rarg0, 0), rax);
320 __ BIND(exit);
322 // pop parameters
323 __ lea(rsp, rsp_after_call);
325 #ifdef ASSERT
326 // verify that threads correspond
327 {
328 Label L, S;
329 __ cmpptr(r15_thread, thread);
330 __ jcc(Assembler::notEqual, S);
331 __ get_thread(rbx);
332 __ cmpptr(r15_thread, rbx);
333 __ jcc(Assembler::equal, L);
334 __ bind(S);
335 __ jcc(Assembler::equal, L);
336 __ stop("StubRoutines::call_stub: threads must correspond");
337 __ bind(L);
338 }
339 #endif
341 // restore regs belonging to calling function
342 __ movptr(r15, r15_save);
343 __ movptr(r14, r14_save);
344 __ movptr(r13, r13_save);
345 __ movptr(r12, r12_save);
346 __ movptr(rbx, rbx_save);
348 #ifdef _WIN64
349 __ movptr(rdi, rdi_save);
350 __ movptr(rsi, rsi_save);
351 #else
352 __ ldmxcsr(mxcsr_save);
353 #endif
355 // restore rsp
356 __ addptr(rsp, -rsp_after_call_off * wordSize);
358 // return
359 __ pop(rbp);
360 __ ret(0);
362 // handle return types different from T_INT
363 __ BIND(is_long);
364 __ movq(Address(c_rarg0, 0), rax);
365 __ jmp(exit);
367 __ BIND(is_float);
368 __ movflt(Address(c_rarg0, 0), xmm0);
369 __ jmp(exit);
371 __ BIND(is_double);
372 __ movdbl(Address(c_rarg0, 0), xmm0);
373 __ jmp(exit);
375 return start;
376 }
378 // Return point for a Java call if there's an exception thrown in
379 // Java code. The exception is caught and transformed into a
380 // pending exception stored in JavaThread that can be tested from
381 // within the VM.
382 //
383 // Note: Usually the parameters are removed by the callee. In case
384 // of an exception crossing an activation frame boundary, that is
385 // not the case if the callee is compiled code => need to setup the
386 // rsp.
387 //
388 // rax: exception oop
390 address generate_catch_exception() {
391 StubCodeMark mark(this, "StubRoutines", "catch_exception");
392 address start = __ pc();
394 // same as in generate_call_stub():
395 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
396 const Address thread (rbp, thread_off * wordSize);
398 #ifdef ASSERT
399 // verify that threads correspond
400 {
401 Label L, S;
402 __ cmpptr(r15_thread, thread);
403 __ jcc(Assembler::notEqual, S);
404 __ get_thread(rbx);
405 __ cmpptr(r15_thread, rbx);
406 __ jcc(Assembler::equal, L);
407 __ bind(S);
408 __ stop("StubRoutines::catch_exception: threads must correspond");
409 __ bind(L);
410 }
411 #endif
413 // set pending exception
414 __ verify_oop(rax);
416 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
417 __ lea(rscratch1, ExternalAddress((address)__FILE__));
418 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
419 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
421 // complete return to VM
422 assert(StubRoutines::_call_stub_return_address != NULL,
423 "_call_stub_return_address must have been generated before");
424 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
426 return start;
427 }
429 // Continuation point for runtime calls returning with a pending
430 // exception. The pending exception check happened in the runtime
431 // or native call stub. The pending exception in Thread is
432 // converted into a Java-level exception.
433 //
434 // Contract with Java-level exception handlers:
435 // rax: exception
436 // rdx: throwing pc
437 //
438 // NOTE: At entry of this stub, exception-pc must be on stack !!
440 address generate_forward_exception() {
441 StubCodeMark mark(this, "StubRoutines", "forward exception");
442 address start = __ pc();
444 // Upon entry, the sp points to the return address returning into
445 // Java (interpreted or compiled) code; i.e., the return address
446 // becomes the throwing pc.
447 //
448 // Arguments pushed before the runtime call are still on the stack
449 // but the exception handler will reset the stack pointer ->
450 // ignore them. A potential result in registers can be ignored as
451 // well.
453 #ifdef ASSERT
454 // make sure this code is only executed if there is a pending exception
455 {
456 Label L;
457 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
458 __ jcc(Assembler::notEqual, L);
459 __ stop("StubRoutines::forward exception: no pending exception (1)");
460 __ bind(L);
461 }
462 #endif
464 // compute exception handler into rbx
465 __ movptr(c_rarg0, Address(rsp, 0));
466 BLOCK_COMMENT("call exception_handler_for_return_address");
467 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
468 SharedRuntime::exception_handler_for_return_address),
469 c_rarg0);
470 __ mov(rbx, rax);
472 // setup rax & rdx, remove return address & clear pending exception
473 __ pop(rdx);
474 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
475 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
477 #ifdef ASSERT
478 // make sure exception is set
479 {
480 Label L;
481 __ testptr(rax, rax);
482 __ jcc(Assembler::notEqual, L);
483 __ stop("StubRoutines::forward exception: no pending exception (2)");
484 __ bind(L);
485 }
486 #endif
488 // continue at exception handler (return address removed)
489 // rax: exception
490 // rbx: exception handler
491 // rdx: throwing pc
492 __ verify_oop(rax);
493 __ jmp(rbx);
495 return start;
496 }
498 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
499 //
500 // Arguments :
501 // c_rarg0: exchange_value
502 // c_rarg0: dest
503 //
504 // Result:
505 // *dest <- ex, return (orig *dest)
506 address generate_atomic_xchg() {
507 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
508 address start = __ pc();
510 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
511 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
512 __ ret(0);
514 return start;
515 }
517 // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
518 //
519 // Arguments :
520 // c_rarg0: exchange_value
521 // c_rarg1: dest
522 //
523 // Result:
524 // *dest <- ex, return (orig *dest)
525 address generate_atomic_xchg_ptr() {
526 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
527 address start = __ pc();
529 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
530 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
531 __ ret(0);
533 return start;
534 }
536 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
537 // jint compare_value)
538 //
539 // Arguments :
540 // c_rarg0: exchange_value
541 // c_rarg1: dest
542 // c_rarg2: compare_value
543 //
544 // Result:
545 // if ( compare_value == *dest ) {
546 // *dest = exchange_value
547 // return compare_value;
548 // else
549 // return *dest;
550 address generate_atomic_cmpxchg() {
551 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
552 address start = __ pc();
554 __ movl(rax, c_rarg2);
555 if ( os::is_MP() ) __ lock();
556 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
557 __ ret(0);
559 return start;
560 }
562 // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
563 // volatile jlong* dest,
564 // jlong compare_value)
565 // Arguments :
566 // c_rarg0: exchange_value
567 // c_rarg1: dest
568 // c_rarg2: compare_value
569 //
570 // Result:
571 // if ( compare_value == *dest ) {
572 // *dest = exchange_value
573 // return compare_value;
574 // else
575 // return *dest;
576 address generate_atomic_cmpxchg_long() {
577 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
578 address start = __ pc();
580 __ movq(rax, c_rarg2);
581 if ( os::is_MP() ) __ lock();
582 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
583 __ ret(0);
585 return start;
586 }
588 // Support for jint atomic::add(jint add_value, volatile jint* dest)
589 //
590 // Arguments :
591 // c_rarg0: add_value
592 // c_rarg1: dest
593 //
594 // Result:
595 // *dest += add_value
596 // return *dest;
597 address generate_atomic_add() {
598 StubCodeMark mark(this, "StubRoutines", "atomic_add");
599 address start = __ pc();
601 __ movl(rax, c_rarg0);
602 if ( os::is_MP() ) __ lock();
603 __ xaddl(Address(c_rarg1, 0), c_rarg0);
604 __ addl(rax, c_rarg0);
605 __ ret(0);
607 return start;
608 }
610 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
611 //
612 // Arguments :
613 // c_rarg0: add_value
614 // c_rarg1: dest
615 //
616 // Result:
617 // *dest += add_value
618 // return *dest;
619 address generate_atomic_add_ptr() {
620 StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
621 address start = __ pc();
623 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
624 if ( os::is_MP() ) __ lock();
625 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
626 __ addptr(rax, c_rarg0);
627 __ ret(0);
629 return start;
630 }
632 // Support for intptr_t OrderAccess::fence()
633 //
634 // Arguments :
635 //
636 // Result:
637 address generate_orderaccess_fence() {
638 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
639 address start = __ pc();
640 __ membar(Assembler::StoreLoad);
641 __ ret(0);
643 return start;
644 }
646 // Support for intptr_t get_previous_fp()
647 //
648 // This routine is used to find the previous frame pointer for the
649 // caller (current_frame_guess). This is used as part of debugging
650 // ps() is seemingly lost trying to find frames.
651 // This code assumes that caller current_frame_guess) has a frame.
652 address generate_get_previous_fp() {
653 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
654 const Address old_fp(rbp, 0);
655 const Address older_fp(rax, 0);
656 address start = __ pc();
658 __ enter();
659 __ movptr(rax, old_fp); // callers fp
660 __ movptr(rax, older_fp); // the frame for ps()
661 __ pop(rbp);
662 __ ret(0);
664 return start;
665 }
667 //----------------------------------------------------------------------------------------------------
668 // Support for void verify_mxcsr()
669 //
670 // This routine is used with -Xcheck:jni to verify that native
671 // JNI code does not return to Java code without restoring the
672 // MXCSR register to our expected state.
674 address generate_verify_mxcsr() {
675 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
676 address start = __ pc();
678 const Address mxcsr_save(rsp, 0);
680 if (CheckJNICalls) {
681 Label ok_ret;
682 __ push(rax);
683 __ subptr(rsp, wordSize); // allocate a temp location
684 __ stmxcsr(mxcsr_save);
685 __ movl(rax, mxcsr_save);
686 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
687 __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
688 __ jcc(Assembler::equal, ok_ret);
690 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
692 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
694 __ bind(ok_ret);
695 __ addptr(rsp, wordSize);
696 __ pop(rax);
697 }
699 __ ret(0);
701 return start;
702 }
704 address generate_f2i_fixup() {
705 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
706 Address inout(rsp, 5 * wordSize); // return address + 4 saves
708 address start = __ pc();
710 Label L;
712 __ push(rax);
713 __ push(c_rarg3);
714 __ push(c_rarg2);
715 __ push(c_rarg1);
717 __ movl(rax, 0x7f800000);
718 __ xorl(c_rarg3, c_rarg3);
719 __ movl(c_rarg2, inout);
720 __ movl(c_rarg1, c_rarg2);
721 __ andl(c_rarg1, 0x7fffffff);
722 __ cmpl(rax, c_rarg1); // NaN? -> 0
723 __ jcc(Assembler::negative, L);
724 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
725 __ movl(c_rarg3, 0x80000000);
726 __ movl(rax, 0x7fffffff);
727 __ cmovl(Assembler::positive, c_rarg3, rax);
729 __ bind(L);
730 __ movptr(inout, c_rarg3);
732 __ pop(c_rarg1);
733 __ pop(c_rarg2);
734 __ pop(c_rarg3);
735 __ pop(rax);
737 __ ret(0);
739 return start;
740 }
742 address generate_f2l_fixup() {
743 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
744 Address inout(rsp, 5 * wordSize); // return address + 4 saves
745 address start = __ pc();
747 Label L;
749 __ push(rax);
750 __ push(c_rarg3);
751 __ push(c_rarg2);
752 __ push(c_rarg1);
754 __ movl(rax, 0x7f800000);
755 __ xorl(c_rarg3, c_rarg3);
756 __ movl(c_rarg2, inout);
757 __ movl(c_rarg1, c_rarg2);
758 __ andl(c_rarg1, 0x7fffffff);
759 __ cmpl(rax, c_rarg1); // NaN? -> 0
760 __ jcc(Assembler::negative, L);
761 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
762 __ mov64(c_rarg3, 0x8000000000000000);
763 __ mov64(rax, 0x7fffffffffffffff);
764 __ cmov(Assembler::positive, c_rarg3, rax);
766 __ bind(L);
767 __ movptr(inout, c_rarg3);
769 __ pop(c_rarg1);
770 __ pop(c_rarg2);
771 __ pop(c_rarg3);
772 __ pop(rax);
774 __ ret(0);
776 return start;
777 }
779 address generate_d2i_fixup() {
780 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
781 Address inout(rsp, 6 * wordSize); // return address + 5 saves
783 address start = __ pc();
785 Label L;
787 __ push(rax);
788 __ push(c_rarg3);
789 __ push(c_rarg2);
790 __ push(c_rarg1);
791 __ push(c_rarg0);
793 __ movl(rax, 0x7ff00000);
794 __ movq(c_rarg2, inout);
795 __ movl(c_rarg3, c_rarg2);
796 __ mov(c_rarg1, c_rarg2);
797 __ mov(c_rarg0, c_rarg2);
798 __ negl(c_rarg3);
799 __ shrptr(c_rarg1, 0x20);
800 __ orl(c_rarg3, c_rarg2);
801 __ andl(c_rarg1, 0x7fffffff);
802 __ xorl(c_rarg2, c_rarg2);
803 __ shrl(c_rarg3, 0x1f);
804 __ orl(c_rarg1, c_rarg3);
805 __ cmpl(rax, c_rarg1);
806 __ jcc(Assembler::negative, L); // NaN -> 0
807 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
808 __ movl(c_rarg2, 0x80000000);
809 __ movl(rax, 0x7fffffff);
810 __ cmov(Assembler::positive, c_rarg2, rax);
812 __ bind(L);
813 __ movptr(inout, c_rarg2);
815 __ pop(c_rarg0);
816 __ pop(c_rarg1);
817 __ pop(c_rarg2);
818 __ pop(c_rarg3);
819 __ pop(rax);
821 __ ret(0);
823 return start;
824 }
826 address generate_d2l_fixup() {
827 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
828 Address inout(rsp, 6 * wordSize); // return address + 5 saves
830 address start = __ pc();
832 Label L;
834 __ push(rax);
835 __ push(c_rarg3);
836 __ push(c_rarg2);
837 __ push(c_rarg1);
838 __ push(c_rarg0);
840 __ movl(rax, 0x7ff00000);
841 __ movq(c_rarg2, inout);
842 __ movl(c_rarg3, c_rarg2);
843 __ mov(c_rarg1, c_rarg2);
844 __ mov(c_rarg0, c_rarg2);
845 __ negl(c_rarg3);
846 __ shrptr(c_rarg1, 0x20);
847 __ orl(c_rarg3, c_rarg2);
848 __ andl(c_rarg1, 0x7fffffff);
849 __ xorl(c_rarg2, c_rarg2);
850 __ shrl(c_rarg3, 0x1f);
851 __ orl(c_rarg1, c_rarg3);
852 __ cmpl(rax, c_rarg1);
853 __ jcc(Assembler::negative, L); // NaN -> 0
854 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
855 __ mov64(c_rarg2, 0x8000000000000000);
856 __ mov64(rax, 0x7fffffffffffffff);
857 __ cmovq(Assembler::positive, c_rarg2, rax);
859 __ bind(L);
860 __ movq(inout, c_rarg2);
862 __ pop(c_rarg0);
863 __ pop(c_rarg1);
864 __ pop(c_rarg2);
865 __ pop(c_rarg3);
866 __ pop(rax);
868 __ ret(0);
870 return start;
871 }
873 address generate_fp_mask(const char *stub_name, int64_t mask) {
874 StubCodeMark mark(this, "StubRoutines", stub_name);
876 __ align(16);
877 address start = __ pc();
879 __ emit_data64( mask, relocInfo::none );
880 __ emit_data64( mask, relocInfo::none );
882 return start;
883 }
885 // The following routine generates a subroutine to throw an
886 // asynchronous UnknownError when an unsafe access gets a fault that
887 // could not be reasonably prevented by the programmer. (Example:
888 // SIGBUS/OBJERR.)
889 address generate_handler_for_unsafe_access() {
890 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
891 address start = __ pc();
893 __ push(0); // hole for return address-to-be
894 __ pusha(); // push registers
895 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
897 __ subptr(rsp, frame::arg_reg_save_area_bytes);
898 BLOCK_COMMENT("call handle_unsafe_access");
899 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
900 __ addptr(rsp, frame::arg_reg_save_area_bytes);
902 __ movptr(next_pc, rax); // stuff next address
903 __ popa();
904 __ ret(0); // jump to next address
906 return start;
907 }
909 // Non-destructive plausibility checks for oops
910 //
911 // Arguments:
912 // all args on stack!
913 //
914 // Stack after saving c_rarg3:
915 // [tos + 0]: saved c_rarg3
916 // [tos + 1]: saved c_rarg2
917 // [tos + 2]: saved r12 (several TemplateTable methods use it)
918 // [tos + 3]: saved flags
919 // [tos + 4]: return address
920 // * [tos + 5]: error message (char*)
921 // * [tos + 6]: object to verify (oop)
922 // * [tos + 7]: saved rax - saved by caller and bashed
923 // * = popped on exit
924 address generate_verify_oop() {
925 StubCodeMark mark(this, "StubRoutines", "verify_oop");
926 address start = __ pc();
928 Label exit, error;
930 __ pushf();
931 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
933 __ push(r12);
935 // save c_rarg2 and c_rarg3
936 __ push(c_rarg2);
937 __ push(c_rarg3);
939 enum {
940 // After previous pushes.
941 oop_to_verify = 6 * wordSize,
942 saved_rax = 7 * wordSize,
944 // Before the call to MacroAssembler::debug(), see below.
945 return_addr = 16 * wordSize,
946 error_msg = 17 * wordSize
947 };
949 // get object
950 __ movptr(rax, Address(rsp, oop_to_verify));
952 // make sure object is 'reasonable'
953 __ testptr(rax, rax);
954 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
955 // Check if the oop is in the right area of memory
956 __ movptr(c_rarg2, rax);
957 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
958 __ andptr(c_rarg2, c_rarg3);
959 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
960 __ cmpptr(c_rarg2, c_rarg3);
961 __ jcc(Assembler::notZero, error);
963 // set r12 to heapbase for load_klass()
964 __ reinit_heapbase();
966 // make sure klass is 'reasonable'
967 __ load_klass(rax, rax); // get klass
968 __ testptr(rax, rax);
969 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
970 // Check if the klass is in the right area of memory
971 __ mov(c_rarg2, rax);
972 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
973 __ andptr(c_rarg2, c_rarg3);
974 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
975 __ cmpptr(c_rarg2, c_rarg3);
976 __ jcc(Assembler::notZero, error);
978 // make sure klass' klass is 'reasonable'
979 __ load_klass(rax, rax);
980 __ testptr(rax, rax);
981 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
982 // Check if the klass' klass is in the right area of memory
983 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
984 __ andptr(rax, c_rarg3);
985 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
986 __ cmpptr(rax, c_rarg3);
987 __ jcc(Assembler::notZero, error);
989 // return if everything seems ok
990 __ bind(exit);
991 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
992 __ pop(c_rarg3); // restore c_rarg3
993 __ pop(c_rarg2); // restore c_rarg2
994 __ pop(r12); // restore r12
995 __ popf(); // restore flags
996 __ ret(3 * wordSize); // pop caller saved stuff
998 // handle errors
999 __ bind(error);
1000 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1001 __ pop(c_rarg3); // get saved c_rarg3 back
1002 __ pop(c_rarg2); // get saved c_rarg2 back
1003 __ pop(r12); // get saved r12 back
1004 __ popf(); // get saved flags off stack --
1005 // will be ignored
1007 __ pusha(); // push registers
1008 // (rip is already
1009 // already pushed)
1010 // debug(char* msg, int64_t pc, int64_t regs[])
1011 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1012 // pushed all the registers, so now the stack looks like:
1013 // [tos + 0] 16 saved registers
1014 // [tos + 16] return address
1015 // * [tos + 17] error message (char*)
1016 // * [tos + 18] object to verify (oop)
1017 // * [tos + 19] saved rax - saved by caller and bashed
1018 // * = popped on exit
1020 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1021 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1022 __ movq(c_rarg2, rsp); // pass address of regs on stack
1023 __ mov(r12, rsp); // remember rsp
1024 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1025 __ andptr(rsp, -16); // align stack as required by ABI
1026 BLOCK_COMMENT("call MacroAssembler::debug");
1027 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1028 __ mov(rsp, r12); // restore rsp
1029 __ popa(); // pop registers (includes r12)
1030 __ ret(3 * wordSize); // pop caller saved stuff
1032 return start;
1033 }
1035 static address disjoint_byte_copy_entry;
1036 static address disjoint_short_copy_entry;
1037 static address disjoint_int_copy_entry;
1038 static address disjoint_long_copy_entry;
1039 static address disjoint_oop_copy_entry;
1041 static address byte_copy_entry;
1042 static address short_copy_entry;
1043 static address int_copy_entry;
1044 static address long_copy_entry;
1045 static address oop_copy_entry;
1047 static address checkcast_copy_entry;
1049 //
1050 // Verify that a register contains clean 32-bits positive value
1051 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1052 //
1053 // Input:
1054 // Rint - 32-bits value
1055 // Rtmp - scratch
1056 //
1057 void assert_clean_int(Register Rint, Register Rtmp) {
1058 #ifdef ASSERT
1059 Label L;
1060 assert_different_registers(Rtmp, Rint);
1061 __ movslq(Rtmp, Rint);
1062 __ cmpq(Rtmp, Rint);
1063 __ jcc(Assembler::equal, L);
1064 __ stop("high 32-bits of int value are not 0");
1065 __ bind(L);
1066 #endif
1067 }
1069 // Generate overlap test for array copy stubs
1070 //
1071 // Input:
1072 // c_rarg0 - from
1073 // c_rarg1 - to
1074 // c_rarg2 - element count
1075 //
1076 // Output:
1077 // rax - &from[element count - 1]
1078 //
1079 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1080 assert(no_overlap_target != NULL, "must be generated");
1081 array_overlap_test(no_overlap_target, NULL, sf);
1082 }
1083 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1084 array_overlap_test(NULL, &L_no_overlap, sf);
1085 }
1086 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1087 const Register from = c_rarg0;
1088 const Register to = c_rarg1;
1089 const Register count = c_rarg2;
1090 const Register end_from = rax;
1092 __ cmpptr(to, from);
1093 __ lea(end_from, Address(from, count, sf, 0));
1094 if (NOLp == NULL) {
1095 ExternalAddress no_overlap(no_overlap_target);
1096 __ jump_cc(Assembler::belowEqual, no_overlap);
1097 __ cmpptr(to, end_from);
1098 __ jump_cc(Assembler::aboveEqual, no_overlap);
1099 } else {
1100 __ jcc(Assembler::belowEqual, (*NOLp));
1101 __ cmpptr(to, end_from);
1102 __ jcc(Assembler::aboveEqual, (*NOLp));
1103 }
1104 }
1106 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1107 //
1108 // Outputs:
1109 // rdi - rcx
1110 // rsi - rdx
1111 // rdx - r8
1112 // rcx - r9
1113 //
1114 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1115 // are non-volatile. r9 and r10 should not be used by the caller.
1116 //
1117 void setup_arg_regs(int nargs = 3) {
1118 const Register saved_rdi = r9;
1119 const Register saved_rsi = r10;
1120 assert(nargs == 3 || nargs == 4, "else fix");
1121 #ifdef _WIN64
1122 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1123 "unexpected argument registers");
1124 if (nargs >= 4)
1125 __ mov(rax, r9); // r9 is also saved_rdi
1126 __ movptr(saved_rdi, rdi);
1127 __ movptr(saved_rsi, rsi);
1128 __ mov(rdi, rcx); // c_rarg0
1129 __ mov(rsi, rdx); // c_rarg1
1130 __ mov(rdx, r8); // c_rarg2
1131 if (nargs >= 4)
1132 __ mov(rcx, rax); // c_rarg3 (via rax)
1133 #else
1134 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1135 "unexpected argument registers");
1136 #endif
1137 }
1139 void restore_arg_regs() {
1140 const Register saved_rdi = r9;
1141 const Register saved_rsi = r10;
1142 #ifdef _WIN64
1143 __ movptr(rdi, saved_rdi);
1144 __ movptr(rsi, saved_rsi);
1145 #endif
1146 }
1148 // Generate code for an array write pre barrier
1149 //
1150 // addr - starting address
1151 // count - element count
1152 //
1153 // Destroy no registers!
1154 //
1155 void gen_write_ref_array_pre_barrier(Register addr, Register count) {
1156 BarrierSet* bs = Universe::heap()->barrier_set();
1157 switch (bs->kind()) {
1158 case BarrierSet::G1SATBCT:
1159 case BarrierSet::G1SATBCTLogging:
1160 {
1161 __ pusha(); // push registers
1162 if (count == c_rarg0) {
1163 if (addr == c_rarg1) {
1164 // exactly backwards!!
1165 __ xchgptr(c_rarg1, c_rarg0);
1166 } else {
1167 __ movptr(c_rarg1, count);
1168 __ movptr(c_rarg0, addr);
1169 }
1171 } else {
1172 __ movptr(c_rarg0, addr);
1173 __ movptr(c_rarg1, count);
1174 }
1175 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre)));
1176 __ popa();
1177 }
1178 break;
1179 case BarrierSet::CardTableModRef:
1180 case BarrierSet::CardTableExtension:
1181 case BarrierSet::ModRef:
1182 break;
1183 default:
1184 ShouldNotReachHere();
1186 }
1187 }
1189 //
1190 // Generate code for an array write post barrier
1191 //
1192 // Input:
1193 // start - register containing starting address of destination array
1194 // end - register containing ending address of destination array
1195 // scratch - scratch register
1196 //
1197 // The input registers are overwritten.
1198 // The ending address is inclusive.
1199 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1200 assert_different_registers(start, end, scratch);
1201 BarrierSet* bs = Universe::heap()->barrier_set();
1202 switch (bs->kind()) {
1203 case BarrierSet::G1SATBCT:
1204 case BarrierSet::G1SATBCTLogging:
1206 {
1207 __ pusha(); // push registers (overkill)
1208 // must compute element count unless barrier set interface is changed (other platforms supply count)
1209 assert_different_registers(start, end, scratch);
1210 __ lea(scratch, Address(end, wordSize));
1211 __ subptr(scratch, start);
1212 __ shrptr(scratch, LogBytesPerWord);
1213 __ mov(c_rarg0, start);
1214 __ mov(c_rarg1, scratch);
1215 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)));
1216 __ popa();
1217 }
1218 break;
1219 case BarrierSet::CardTableModRef:
1220 case BarrierSet::CardTableExtension:
1221 {
1222 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1223 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1225 Label L_loop;
1227 __ shrptr(start, CardTableModRefBS::card_shift);
1228 __ shrptr(end, CardTableModRefBS::card_shift);
1229 __ subptr(end, start); // number of bytes to copy
1231 intptr_t disp = (intptr_t) ct->byte_map_base;
1232 if (__ is_simm32(disp)) {
1233 Address cardtable(noreg, noreg, Address::no_scale, disp);
1234 __ lea(scratch, cardtable);
1235 } else {
1236 ExternalAddress cardtable((address)disp);
1237 __ lea(scratch, cardtable);
1238 }
1240 const Register count = end; // 'end' register contains bytes count now
1241 __ addptr(start, scratch);
1242 __ BIND(L_loop);
1243 __ movb(Address(start, count, Address::times_1), 0);
1244 __ decrement(count);
1245 __ jcc(Assembler::greaterEqual, L_loop);
1246 }
1247 break;
1248 default:
1249 ShouldNotReachHere();
1251 }
1252 }
1255 // Copy big chunks forward
1256 //
1257 // Inputs:
1258 // end_from - source arrays end address
1259 // end_to - destination array end address
1260 // qword_count - 64-bits element count, negative
1261 // to - scratch
1262 // L_copy_32_bytes - entry label
1263 // L_copy_8_bytes - exit label
1264 //
1265 void copy_32_bytes_forward(Register end_from, Register end_to,
1266 Register qword_count, Register to,
1267 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1268 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1269 Label L_loop;
1270 __ align(16);
1271 __ BIND(L_loop);
1272 if(UseUnalignedLoadStores) {
1273 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1274 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1275 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1276 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1278 } else {
1279 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1280 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1281 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1282 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1283 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1284 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1285 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1286 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1287 }
1288 __ BIND(L_copy_32_bytes);
1289 __ addptr(qword_count, 4);
1290 __ jcc(Assembler::lessEqual, L_loop);
1291 __ subptr(qword_count, 4);
1292 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1293 }
1296 // Copy big chunks backward
1297 //
1298 // Inputs:
1299 // from - source arrays address
1300 // dest - destination array address
1301 // qword_count - 64-bits element count
1302 // to - scratch
1303 // L_copy_32_bytes - entry label
1304 // L_copy_8_bytes - exit label
1305 //
1306 void copy_32_bytes_backward(Register from, Register dest,
1307 Register qword_count, Register to,
1308 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1309 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1310 Label L_loop;
1311 __ align(16);
1312 __ BIND(L_loop);
1313 if(UseUnalignedLoadStores) {
1314 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1315 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1316 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1317 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1319 } else {
1320 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1321 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1322 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1323 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1324 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1325 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1326 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1327 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1328 }
1329 __ BIND(L_copy_32_bytes);
1330 __ subptr(qword_count, 4);
1331 __ jcc(Assembler::greaterEqual, L_loop);
1332 __ addptr(qword_count, 4);
1333 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1334 }
1337 // Arguments:
1338 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1339 // ignored
1340 // name - stub name string
1341 //
1342 // Inputs:
1343 // c_rarg0 - source array address
1344 // c_rarg1 - destination array address
1345 // c_rarg2 - element count, treated as ssize_t, can be zero
1346 //
1347 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1348 // we let the hardware handle it. The one to eight bytes within words,
1349 // dwords or qwords that span cache line boundaries will still be loaded
1350 // and stored atomically.
1351 //
1352 // Side Effects:
1353 // disjoint_byte_copy_entry is set to the no-overlap entry point
1354 // used by generate_conjoint_byte_copy().
1355 //
1356 address generate_disjoint_byte_copy(bool aligned, const char *name) {
1357 __ align(CodeEntryAlignment);
1358 StubCodeMark mark(this, "StubRoutines", name);
1359 address start = __ pc();
1361 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1362 Label L_copy_byte, L_exit;
1363 const Register from = rdi; // source array address
1364 const Register to = rsi; // destination array address
1365 const Register count = rdx; // elements count
1366 const Register byte_count = rcx;
1367 const Register qword_count = count;
1368 const Register end_from = from; // source array end address
1369 const Register end_to = to; // destination array end address
1370 // End pointers are inclusive, and if count is not zero they point
1371 // to the last unit copied: end_to[0] := end_from[0]
1373 __ enter(); // required for proper stackwalking of RuntimeStub frame
1374 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1376 disjoint_byte_copy_entry = __ pc();
1377 BLOCK_COMMENT("Entry:");
1378 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1380 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1381 // r9 and r10 may be used to save non-volatile registers
1383 // 'from', 'to' and 'count' are now valid
1384 __ movptr(byte_count, count);
1385 __ shrptr(count, 3); // count => qword_count
1387 // Copy from low to high addresses. Use 'to' as scratch.
1388 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1389 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1390 __ negptr(qword_count); // make the count negative
1391 __ jmp(L_copy_32_bytes);
1393 // Copy trailing qwords
1394 __ BIND(L_copy_8_bytes);
1395 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1396 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1397 __ increment(qword_count);
1398 __ jcc(Assembler::notZero, L_copy_8_bytes);
1400 // Check for and copy trailing dword
1401 __ BIND(L_copy_4_bytes);
1402 __ testl(byte_count, 4);
1403 __ jccb(Assembler::zero, L_copy_2_bytes);
1404 __ movl(rax, Address(end_from, 8));
1405 __ movl(Address(end_to, 8), rax);
1407 __ addptr(end_from, 4);
1408 __ addptr(end_to, 4);
1410 // Check for and copy trailing word
1411 __ BIND(L_copy_2_bytes);
1412 __ testl(byte_count, 2);
1413 __ jccb(Assembler::zero, L_copy_byte);
1414 __ movw(rax, Address(end_from, 8));
1415 __ movw(Address(end_to, 8), rax);
1417 __ addptr(end_from, 2);
1418 __ addptr(end_to, 2);
1420 // Check for and copy trailing byte
1421 __ BIND(L_copy_byte);
1422 __ testl(byte_count, 1);
1423 __ jccb(Assembler::zero, L_exit);
1424 __ movb(rax, Address(end_from, 8));
1425 __ movb(Address(end_to, 8), rax);
1427 __ BIND(L_exit);
1428 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1429 restore_arg_regs();
1430 __ xorptr(rax, rax); // return 0
1431 __ leave(); // required for proper stackwalking of RuntimeStub frame
1432 __ ret(0);
1434 // Copy in 32-bytes chunks
1435 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1436 __ jmp(L_copy_4_bytes);
1438 return start;
1439 }
1441 // Arguments:
1442 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1443 // ignored
1444 // name - stub name string
1445 //
1446 // Inputs:
1447 // c_rarg0 - source array address
1448 // c_rarg1 - destination array address
1449 // c_rarg2 - element count, treated as ssize_t, can be zero
1450 //
1451 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1452 // we let the hardware handle it. The one to eight bytes within words,
1453 // dwords or qwords that span cache line boundaries will still be loaded
1454 // and stored atomically.
1455 //
1456 address generate_conjoint_byte_copy(bool aligned, const char *name) {
1457 __ align(CodeEntryAlignment);
1458 StubCodeMark mark(this, "StubRoutines", name);
1459 address start = __ pc();
1461 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1462 const Register from = rdi; // source array address
1463 const Register to = rsi; // destination array address
1464 const Register count = rdx; // elements count
1465 const Register byte_count = rcx;
1466 const Register qword_count = count;
1468 __ enter(); // required for proper stackwalking of RuntimeStub frame
1469 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1471 byte_copy_entry = __ pc();
1472 BLOCK_COMMENT("Entry:");
1473 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1475 array_overlap_test(disjoint_byte_copy_entry, Address::times_1);
1476 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1477 // r9 and r10 may be used to save non-volatile registers
1479 // 'from', 'to' and 'count' are now valid
1480 __ movptr(byte_count, count);
1481 __ shrptr(count, 3); // count => qword_count
1483 // Copy from high to low addresses.
1485 // Check for and copy trailing byte
1486 __ testl(byte_count, 1);
1487 __ jcc(Assembler::zero, L_copy_2_bytes);
1488 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1489 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1490 __ decrement(byte_count); // Adjust for possible trailing word
1492 // Check for and copy trailing word
1493 __ BIND(L_copy_2_bytes);
1494 __ testl(byte_count, 2);
1495 __ jcc(Assembler::zero, L_copy_4_bytes);
1496 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1497 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1499 // Check for and copy trailing dword
1500 __ BIND(L_copy_4_bytes);
1501 __ testl(byte_count, 4);
1502 __ jcc(Assembler::zero, L_copy_32_bytes);
1503 __ movl(rax, Address(from, qword_count, Address::times_8));
1504 __ movl(Address(to, qword_count, Address::times_8), rax);
1505 __ jmp(L_copy_32_bytes);
1507 // Copy trailing qwords
1508 __ BIND(L_copy_8_bytes);
1509 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1510 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1511 __ decrement(qword_count);
1512 __ jcc(Assembler::notZero, L_copy_8_bytes);
1514 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1515 restore_arg_regs();
1516 __ xorptr(rax, rax); // return 0
1517 __ leave(); // required for proper stackwalking of RuntimeStub frame
1518 __ ret(0);
1520 // Copy in 32-bytes chunks
1521 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1523 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
1524 restore_arg_regs();
1525 __ xorptr(rax, rax); // return 0
1526 __ leave(); // required for proper stackwalking of RuntimeStub frame
1527 __ ret(0);
1529 return start;
1530 }
1532 // Arguments:
1533 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1534 // ignored
1535 // name - stub name string
1536 //
1537 // Inputs:
1538 // c_rarg0 - source array address
1539 // c_rarg1 - destination array address
1540 // c_rarg2 - element count, treated as ssize_t, can be zero
1541 //
1542 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1543 // let the hardware handle it. The two or four words within dwords
1544 // or qwords that span cache line boundaries will still be loaded
1545 // and stored atomically.
1546 //
1547 // Side Effects:
1548 // disjoint_short_copy_entry is set to the no-overlap entry point
1549 // used by generate_conjoint_short_copy().
1550 //
1551 address generate_disjoint_short_copy(bool aligned, const char *name) {
1552 __ align(CodeEntryAlignment);
1553 StubCodeMark mark(this, "StubRoutines", name);
1554 address start = __ pc();
1556 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1557 const Register from = rdi; // source array address
1558 const Register to = rsi; // destination array address
1559 const Register count = rdx; // elements count
1560 const Register word_count = rcx;
1561 const Register qword_count = count;
1562 const Register end_from = from; // source array end address
1563 const Register end_to = to; // destination array end address
1564 // End pointers are inclusive, and if count is not zero they point
1565 // to the last unit copied: end_to[0] := end_from[0]
1567 __ enter(); // required for proper stackwalking of RuntimeStub frame
1568 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1570 disjoint_short_copy_entry = __ pc();
1571 BLOCK_COMMENT("Entry:");
1572 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1574 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1575 // r9 and r10 may be used to save non-volatile registers
1577 // 'from', 'to' and 'count' are now valid
1578 __ movptr(word_count, count);
1579 __ shrptr(count, 2); // count => qword_count
1581 // Copy from low to high addresses. Use 'to' as scratch.
1582 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1583 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1584 __ negptr(qword_count);
1585 __ jmp(L_copy_32_bytes);
1587 // Copy trailing qwords
1588 __ BIND(L_copy_8_bytes);
1589 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1590 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1591 __ increment(qword_count);
1592 __ jcc(Assembler::notZero, L_copy_8_bytes);
1594 // Original 'dest' is trashed, so we can't use it as a
1595 // base register for a possible trailing word copy
1597 // Check for and copy trailing dword
1598 __ BIND(L_copy_4_bytes);
1599 __ testl(word_count, 2);
1600 __ jccb(Assembler::zero, L_copy_2_bytes);
1601 __ movl(rax, Address(end_from, 8));
1602 __ movl(Address(end_to, 8), rax);
1604 __ addptr(end_from, 4);
1605 __ addptr(end_to, 4);
1607 // Check for and copy trailing word
1608 __ BIND(L_copy_2_bytes);
1609 __ testl(word_count, 1);
1610 __ jccb(Assembler::zero, L_exit);
1611 __ movw(rax, Address(end_from, 8));
1612 __ movw(Address(end_to, 8), rax);
1614 __ BIND(L_exit);
1615 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1616 restore_arg_regs();
1617 __ xorptr(rax, rax); // return 0
1618 __ leave(); // required for proper stackwalking of RuntimeStub frame
1619 __ ret(0);
1621 // Copy in 32-bytes chunks
1622 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1623 __ jmp(L_copy_4_bytes);
1625 return start;
1626 }
1628 // Arguments:
1629 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1630 // ignored
1631 // name - stub name string
1632 //
1633 // Inputs:
1634 // c_rarg0 - source array address
1635 // c_rarg1 - destination array address
1636 // c_rarg2 - element count, treated as ssize_t, can be zero
1637 //
1638 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1639 // let the hardware handle it. The two or four words within dwords
1640 // or qwords that span cache line boundaries will still be loaded
1641 // and stored atomically.
1642 //
1643 address generate_conjoint_short_copy(bool aligned, const char *name) {
1644 __ align(CodeEntryAlignment);
1645 StubCodeMark mark(this, "StubRoutines", name);
1646 address start = __ pc();
1648 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1649 const Register from = rdi; // source array address
1650 const Register to = rsi; // destination array address
1651 const Register count = rdx; // elements count
1652 const Register word_count = rcx;
1653 const Register qword_count = count;
1655 __ enter(); // required for proper stackwalking of RuntimeStub frame
1656 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1658 short_copy_entry = __ pc();
1659 BLOCK_COMMENT("Entry:");
1660 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1662 array_overlap_test(disjoint_short_copy_entry, Address::times_2);
1663 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1664 // r9 and r10 may be used to save non-volatile registers
1666 // 'from', 'to' and 'count' are now valid
1667 __ movptr(word_count, count);
1668 __ shrptr(count, 2); // count => qword_count
1670 // Copy from high to low addresses. Use 'to' as scratch.
1672 // Check for and copy trailing word
1673 __ testl(word_count, 1);
1674 __ jccb(Assembler::zero, L_copy_4_bytes);
1675 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1676 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1678 // Check for and copy trailing dword
1679 __ BIND(L_copy_4_bytes);
1680 __ testl(word_count, 2);
1681 __ jcc(Assembler::zero, L_copy_32_bytes);
1682 __ movl(rax, Address(from, qword_count, Address::times_8));
1683 __ movl(Address(to, qword_count, Address::times_8), rax);
1684 __ jmp(L_copy_32_bytes);
1686 // Copy trailing qwords
1687 __ BIND(L_copy_8_bytes);
1688 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1689 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1690 __ decrement(qword_count);
1691 __ jcc(Assembler::notZero, L_copy_8_bytes);
1693 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1694 restore_arg_regs();
1695 __ xorptr(rax, rax); // return 0
1696 __ leave(); // required for proper stackwalking of RuntimeStub frame
1697 __ ret(0);
1699 // Copy in 32-bytes chunks
1700 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1702 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
1703 restore_arg_regs();
1704 __ xorptr(rax, rax); // return 0
1705 __ leave(); // required for proper stackwalking of RuntimeStub frame
1706 __ ret(0);
1708 return start;
1709 }
1711 // Arguments:
1712 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1713 // ignored
1714 // is_oop - true => oop array, so generate store check code
1715 // name - stub name string
1716 //
1717 // Inputs:
1718 // c_rarg0 - source array address
1719 // c_rarg1 - destination array address
1720 // c_rarg2 - element count, treated as ssize_t, can be zero
1721 //
1722 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1723 // the hardware handle it. The two dwords within qwords that span
1724 // cache line boundaries will still be loaded and stored atomicly.
1725 //
1726 // Side Effects:
1727 // disjoint_int_copy_entry is set to the no-overlap entry point
1728 // used by generate_conjoint_int_oop_copy().
1729 //
1730 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1731 __ align(CodeEntryAlignment);
1732 StubCodeMark mark(this, "StubRoutines", name);
1733 address start = __ pc();
1735 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1736 const Register from = rdi; // source array address
1737 const Register to = rsi; // destination array address
1738 const Register count = rdx; // elements count
1739 const Register dword_count = rcx;
1740 const Register qword_count = count;
1741 const Register end_from = from; // source array end address
1742 const Register end_to = to; // destination array end address
1743 const Register saved_to = r11; // saved destination array address
1744 // End pointers are inclusive, and if count is not zero they point
1745 // to the last unit copied: end_to[0] := end_from[0]
1747 __ enter(); // required for proper stackwalking of RuntimeStub frame
1748 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1750 (is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry) = __ pc();
1752 if (is_oop) {
1753 // no registers are destroyed by this call
1754 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1755 }
1757 BLOCK_COMMENT("Entry:");
1758 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1760 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1761 // r9 and r10 may be used to save non-volatile registers
1763 if (is_oop) {
1764 __ movq(saved_to, to);
1765 }
1767 // 'from', 'to' and 'count' are now valid
1768 __ movptr(dword_count, count);
1769 __ shrptr(count, 1); // count => qword_count
1771 // Copy from low to high addresses. Use 'to' as scratch.
1772 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1773 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1774 __ negptr(qword_count);
1775 __ jmp(L_copy_32_bytes);
1777 // Copy trailing qwords
1778 __ BIND(L_copy_8_bytes);
1779 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1780 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1781 __ increment(qword_count);
1782 __ jcc(Assembler::notZero, L_copy_8_bytes);
1784 // Check for and copy trailing dword
1785 __ BIND(L_copy_4_bytes);
1786 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1787 __ jccb(Assembler::zero, L_exit);
1788 __ movl(rax, Address(end_from, 8));
1789 __ movl(Address(end_to, 8), rax);
1791 __ BIND(L_exit);
1792 if (is_oop) {
1793 __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1794 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1795 }
1796 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1797 restore_arg_regs();
1798 __ xorptr(rax, rax); // return 0
1799 __ leave(); // required for proper stackwalking of RuntimeStub frame
1800 __ ret(0);
1802 // Copy 32-bytes chunks
1803 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1804 __ jmp(L_copy_4_bytes);
1806 return start;
1807 }
1809 // Arguments:
1810 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1811 // ignored
1812 // is_oop - true => oop array, so generate store check code
1813 // name - stub name string
1814 //
1815 // Inputs:
1816 // c_rarg0 - source array address
1817 // c_rarg1 - destination array address
1818 // c_rarg2 - element count, treated as ssize_t, can be zero
1819 //
1820 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1821 // the hardware handle it. The two dwords within qwords that span
1822 // cache line boundaries will still be loaded and stored atomicly.
1823 //
1824 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1825 __ align(CodeEntryAlignment);
1826 StubCodeMark mark(this, "StubRoutines", name);
1827 address start = __ pc();
1829 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1830 const Register from = rdi; // source array address
1831 const Register to = rsi; // destination array address
1832 const Register count = rdx; // elements count
1833 const Register dword_count = rcx;
1834 const Register qword_count = count;
1836 __ enter(); // required for proper stackwalking of RuntimeStub frame
1837 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1839 if (is_oop) {
1840 // no registers are destroyed by this call
1841 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1842 }
1844 (is_oop ? oop_copy_entry : int_copy_entry) = __ pc();
1845 BLOCK_COMMENT("Entry:");
1846 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1848 array_overlap_test(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry,
1849 Address::times_4);
1850 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1851 // r9 and r10 may be used to save non-volatile registers
1853 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1854 // 'from', 'to' and 'count' are now valid
1855 __ movptr(dword_count, count);
1856 __ shrptr(count, 1); // count => qword_count
1858 // Copy from high to low addresses. Use 'to' as scratch.
1860 // Check for and copy trailing dword
1861 __ testl(dword_count, 1);
1862 __ jcc(Assembler::zero, L_copy_32_bytes);
1863 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1864 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1865 __ jmp(L_copy_32_bytes);
1867 // Copy trailing qwords
1868 __ BIND(L_copy_8_bytes);
1869 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1870 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1871 __ decrement(qword_count);
1872 __ jcc(Assembler::notZero, L_copy_8_bytes);
1874 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1875 if (is_oop) {
1876 __ jmp(L_exit);
1877 }
1878 restore_arg_regs();
1879 __ xorptr(rax, rax); // return 0
1880 __ leave(); // required for proper stackwalking of RuntimeStub frame
1881 __ ret(0);
1883 // Copy in 32-bytes chunks
1884 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1886 inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
1887 __ bind(L_exit);
1888 if (is_oop) {
1889 Register end_to = rdx;
1890 __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1891 gen_write_ref_array_post_barrier(to, end_to, rax);
1892 }
1893 restore_arg_regs();
1894 __ xorptr(rax, rax); // return 0
1895 __ leave(); // required for proper stackwalking of RuntimeStub frame
1896 __ ret(0);
1898 return start;
1899 }
1901 // Arguments:
1902 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1903 // ignored
1904 // is_oop - true => oop array, so generate store check code
1905 // name - stub name string
1906 //
1907 // Inputs:
1908 // c_rarg0 - source array address
1909 // c_rarg1 - destination array address
1910 // c_rarg2 - element count, treated as ssize_t, can be zero
1911 //
1912 // Side Effects:
1913 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1914 // no-overlap entry point used by generate_conjoint_long_oop_copy().
1915 //
1916 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1917 __ align(CodeEntryAlignment);
1918 StubCodeMark mark(this, "StubRoutines", name);
1919 address start = __ pc();
1921 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1922 const Register from = rdi; // source array address
1923 const Register to = rsi; // destination array address
1924 const Register qword_count = rdx; // elements count
1925 const Register end_from = from; // source array end address
1926 const Register end_to = rcx; // destination array end address
1927 const Register saved_to = to;
1928 // End pointers are inclusive, and if count is not zero they point
1929 // to the last unit copied: end_to[0] := end_from[0]
1931 __ enter(); // required for proper stackwalking of RuntimeStub frame
1932 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
1933 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1935 if (is_oop) {
1936 disjoint_oop_copy_entry = __ pc();
1937 // no registers are destroyed by this call
1938 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
1939 } else {
1940 disjoint_long_copy_entry = __ pc();
1941 }
1942 BLOCK_COMMENT("Entry:");
1943 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1945 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1946 // r9 and r10 may be used to save non-volatile registers
1948 // 'from', 'to' and 'qword_count' are now valid
1950 // Copy from low to high addresses. Use 'to' as scratch.
1951 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1952 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1953 __ negptr(qword_count);
1954 __ jmp(L_copy_32_bytes);
1956 // Copy trailing qwords
1957 __ BIND(L_copy_8_bytes);
1958 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1959 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1960 __ increment(qword_count);
1961 __ jcc(Assembler::notZero, L_copy_8_bytes);
1963 if (is_oop) {
1964 __ jmp(L_exit);
1965 } else {
1966 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1967 restore_arg_regs();
1968 __ xorptr(rax, rax); // return 0
1969 __ leave(); // required for proper stackwalking of RuntimeStub frame
1970 __ ret(0);
1971 }
1973 // Copy 64-byte chunks
1974 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1976 if (is_oop) {
1977 __ BIND(L_exit);
1978 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1979 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
1980 } else {
1981 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1982 }
1983 restore_arg_regs();
1984 __ xorptr(rax, rax); // return 0
1985 __ leave(); // required for proper stackwalking of RuntimeStub frame
1986 __ ret(0);
1988 return start;
1989 }
1991 // Arguments:
1992 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1993 // ignored
1994 // is_oop - true => oop array, so generate store check code
1995 // name - stub name string
1996 //
1997 // Inputs:
1998 // c_rarg0 - source array address
1999 // c_rarg1 - destination array address
2000 // c_rarg2 - element count, treated as ssize_t, can be zero
2001 //
2002 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
2003 __ align(CodeEntryAlignment);
2004 StubCodeMark mark(this, "StubRoutines", name);
2005 address start = __ pc();
2007 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2008 const Register from = rdi; // source array address
2009 const Register to = rsi; // destination array address
2010 const Register qword_count = rdx; // elements count
2011 const Register saved_count = rcx;
2013 __ enter(); // required for proper stackwalking of RuntimeStub frame
2014 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2016 address disjoint_copy_entry = NULL;
2017 if (is_oop) {
2018 assert(!UseCompressedOops, "shouldn't be called for compressed oops");
2019 disjoint_copy_entry = disjoint_oop_copy_entry;
2020 oop_copy_entry = __ pc();
2021 array_overlap_test(disjoint_oop_copy_entry, Address::times_8);
2022 } else {
2023 disjoint_copy_entry = disjoint_long_copy_entry;
2024 long_copy_entry = __ pc();
2025 array_overlap_test(disjoint_long_copy_entry, Address::times_8);
2026 }
2027 BLOCK_COMMENT("Entry:");
2028 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2030 array_overlap_test(disjoint_copy_entry, Address::times_8);
2031 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2032 // r9 and r10 may be used to save non-volatile registers
2034 // 'from', 'to' and 'qword_count' are now valid
2036 if (is_oop) {
2037 // Save to and count for store barrier
2038 __ movptr(saved_count, qword_count);
2039 // No registers are destroyed by this call
2040 gen_write_ref_array_pre_barrier(to, saved_count);
2041 }
2043 __ jmp(L_copy_32_bytes);
2045 // Copy trailing qwords
2046 __ BIND(L_copy_8_bytes);
2047 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2048 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2049 __ decrement(qword_count);
2050 __ jcc(Assembler::notZero, L_copy_8_bytes);
2052 if (is_oop) {
2053 __ jmp(L_exit);
2054 } else {
2055 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2056 restore_arg_regs();
2057 __ xorptr(rax, rax); // return 0
2058 __ leave(); // required for proper stackwalking of RuntimeStub frame
2059 __ ret(0);
2060 }
2062 // Copy in 32-bytes chunks
2063 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2065 if (is_oop) {
2066 __ BIND(L_exit);
2067 __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2068 gen_write_ref_array_post_barrier(to, rcx, rax);
2069 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
2070 } else {
2071 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
2072 }
2073 restore_arg_regs();
2074 __ xorptr(rax, rax); // return 0
2075 __ leave(); // required for proper stackwalking of RuntimeStub frame
2076 __ ret(0);
2078 return start;
2079 }
2082 // Helper for generating a dynamic type check.
2083 // Smashes no registers.
2084 void generate_type_check(Register sub_klass,
2085 Register super_check_offset,
2086 Register super_klass,
2087 Label& L_success) {
2088 assert_different_registers(sub_klass, super_check_offset, super_klass);
2090 BLOCK_COMMENT("type_check:");
2092 Label L_miss;
2094 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2095 super_check_offset);
2096 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2098 // Fall through on failure!
2099 __ BIND(L_miss);
2100 }
2102 //
2103 // Generate checkcasting array copy stub
2104 //
2105 // Input:
2106 // c_rarg0 - source array address
2107 // c_rarg1 - destination array address
2108 // c_rarg2 - element count, treated as ssize_t, can be zero
2109 // c_rarg3 - size_t ckoff (super_check_offset)
2110 // not Win64
2111 // c_rarg4 - oop ckval (super_klass)
2112 // Win64
2113 // rsp+40 - oop ckval (super_klass)
2114 //
2115 // Output:
2116 // rax == 0 - success
2117 // rax == -1^K - failure, where K is partial transfer count
2118 //
2119 address generate_checkcast_copy(const char *name) {
2121 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2123 // Input registers (after setup_arg_regs)
2124 const Register from = rdi; // source array address
2125 const Register to = rsi; // destination array address
2126 const Register length = rdx; // elements count
2127 const Register ckoff = rcx; // super_check_offset
2128 const Register ckval = r8; // super_klass
2130 // Registers used as temps (r13, r14 are save-on-entry)
2131 const Register end_from = from; // source array end address
2132 const Register end_to = r13; // destination array end address
2133 const Register count = rdx; // -(count_remaining)
2134 const Register r14_length = r14; // saved copy of length
2135 // End pointers are inclusive, and if length is not zero they point
2136 // to the last unit copied: end_to[0] := end_from[0]
2138 const Register rax_oop = rax; // actual oop copied
2139 const Register r11_klass = r11; // oop._klass
2141 //---------------------------------------------------------------
2142 // Assembler stub will be used for this call to arraycopy
2143 // if the two arrays are subtypes of Object[] but the
2144 // destination array type is not equal to or a supertype
2145 // of the source type. Each element must be separately
2146 // checked.
2148 __ align(CodeEntryAlignment);
2149 StubCodeMark mark(this, "StubRoutines", name);
2150 address start = __ pc();
2152 __ enter(); // required for proper stackwalking of RuntimeStub frame
2154 checkcast_copy_entry = __ pc();
2155 BLOCK_COMMENT("Entry:");
2157 #ifdef ASSERT
2158 // caller guarantees that the arrays really are different
2159 // otherwise, we would have to make conjoint checks
2160 { Label L;
2161 array_overlap_test(L, TIMES_OOP);
2162 __ stop("checkcast_copy within a single array");
2163 __ bind(L);
2164 }
2165 #endif //ASSERT
2167 // allocate spill slots for r13, r14
2168 enum {
2169 saved_r13_offset,
2170 saved_r14_offset,
2171 saved_rbp_offset,
2172 saved_rip_offset,
2173 saved_rarg0_offset
2174 };
2175 __ subptr(rsp, saved_rbp_offset * wordSize);
2176 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2177 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2178 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2179 // ckoff => rcx, ckval => r8
2180 // r9 and r10 may be used to save non-volatile registers
2181 #ifdef _WIN64
2182 // last argument (#4) is on stack on Win64
2183 const int ckval_offset = saved_rarg0_offset + 4;
2184 __ movptr(ckval, Address(rsp, ckval_offset * wordSize));
2185 #endif
2187 // check that int operands are properly extended to size_t
2188 assert_clean_int(length, rax);
2189 assert_clean_int(ckoff, rax);
2191 #ifdef ASSERT
2192 BLOCK_COMMENT("assert consistent ckoff/ckval");
2193 // The ckoff and ckval must be mutually consistent,
2194 // even though caller generates both.
2195 { Label L;
2196 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2197 Klass::super_check_offset_offset_in_bytes());
2198 __ cmpl(ckoff, Address(ckval, sco_offset));
2199 __ jcc(Assembler::equal, L);
2200 __ stop("super_check_offset inconsistent");
2201 __ bind(L);
2202 }
2203 #endif //ASSERT
2205 // Loop-invariant addresses. They are exclusive end pointers.
2206 Address end_from_addr(from, length, TIMES_OOP, 0);
2207 Address end_to_addr(to, length, TIMES_OOP, 0);
2208 // Loop-variant addresses. They assume post-incremented count < 0.
2209 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2210 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2212 gen_write_ref_array_pre_barrier(to, count);
2214 // Copy from low to high addresses, indexed from the end of each array.
2215 __ lea(end_from, end_from_addr);
2216 __ lea(end_to, end_to_addr);
2217 __ movptr(r14_length, length); // save a copy of the length
2218 assert(length == count, ""); // else fix next line:
2219 __ negptr(count); // negate and test the length
2220 __ jcc(Assembler::notZero, L_load_element);
2222 // Empty array: Nothing to do.
2223 __ xorptr(rax, rax); // return 0 on (trivial) success
2224 __ jmp(L_done);
2226 // ======== begin loop ========
2227 // (Loop is rotated; its entry is L_load_element.)
2228 // Loop control:
2229 // for (count = -count; count != 0; count++)
2230 // Base pointers src, dst are biased by 8*(count-1),to last element.
2231 __ align(16);
2233 __ BIND(L_store_element);
2234 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2235 __ increment(count); // increment the count toward zero
2236 __ jcc(Assembler::zero, L_do_card_marks);
2238 // ======== loop entry is here ========
2239 __ BIND(L_load_element);
2240 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2241 __ testptr(rax_oop, rax_oop);
2242 __ jcc(Assembler::zero, L_store_element);
2244 __ load_klass(r11_klass, rax_oop);// query the object klass
2245 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2246 // ======== end loop ========
2248 // It was a real error; we must depend on the caller to finish the job.
2249 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2250 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2251 // and report their number to the caller.
2252 assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2253 __ lea(end_to, to_element_addr);
2254 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2255 __ movptr(rax, r14_length); // original oops
2256 __ addptr(rax, count); // K = (original - remaining) oops
2257 __ notptr(rax); // report (-1^K) to caller
2258 __ jmp(L_done);
2260 // Come here on success only.
2261 __ BIND(L_do_card_marks);
2262 __ addptr(end_to, -wordSize); // make an inclusive end pointer
2263 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2264 __ xorptr(rax, rax); // return 0 on success
2266 // Common exit point (success or failure).
2267 __ BIND(L_done);
2268 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2269 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2270 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
2271 restore_arg_regs();
2272 __ leave(); // required for proper stackwalking of RuntimeStub frame
2273 __ ret(0);
2275 return start;
2276 }
2278 //
2279 // Generate 'unsafe' array copy stub
2280 // Though just as safe as the other stubs, it takes an unscaled
2281 // size_t argument instead of an element count.
2282 //
2283 // Input:
2284 // c_rarg0 - source array address
2285 // c_rarg1 - destination array address
2286 // c_rarg2 - byte count, treated as ssize_t, can be zero
2287 //
2288 // Examines the alignment of the operands and dispatches
2289 // to a long, int, short, or byte copy loop.
2290 //
2291 address generate_unsafe_copy(const char *name) {
2293 Label L_long_aligned, L_int_aligned, L_short_aligned;
2295 // Input registers (before setup_arg_regs)
2296 const Register from = c_rarg0; // source array address
2297 const Register to = c_rarg1; // destination array address
2298 const Register size = c_rarg2; // byte count (size_t)
2300 // Register used as a temp
2301 const Register bits = rax; // test copy of low bits
2303 __ align(CodeEntryAlignment);
2304 StubCodeMark mark(this, "StubRoutines", name);
2305 address start = __ pc();
2307 __ enter(); // required for proper stackwalking of RuntimeStub frame
2309 // bump this on entry, not on exit:
2310 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2312 __ mov(bits, from);
2313 __ orptr(bits, to);
2314 __ orptr(bits, size);
2316 __ testb(bits, BytesPerLong-1);
2317 __ jccb(Assembler::zero, L_long_aligned);
2319 __ testb(bits, BytesPerInt-1);
2320 __ jccb(Assembler::zero, L_int_aligned);
2322 __ testb(bits, BytesPerShort-1);
2323 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2325 __ BIND(L_short_aligned);
2326 __ shrptr(size, LogBytesPerShort); // size => short_count
2327 __ jump(RuntimeAddress(short_copy_entry));
2329 __ BIND(L_int_aligned);
2330 __ shrptr(size, LogBytesPerInt); // size => int_count
2331 __ jump(RuntimeAddress(int_copy_entry));
2333 __ BIND(L_long_aligned);
2334 __ shrptr(size, LogBytesPerLong); // size => qword_count
2335 __ jump(RuntimeAddress(long_copy_entry));
2337 return start;
2338 }
2340 // Perform range checks on the proposed arraycopy.
2341 // Kills temp, but nothing else.
2342 // Also, clean the sign bits of src_pos and dst_pos.
2343 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2344 Register src_pos, // source position (c_rarg1)
2345 Register dst, // destination array oo (c_rarg2)
2346 Register dst_pos, // destination position (c_rarg3)
2347 Register length,
2348 Register temp,
2349 Label& L_failed) {
2350 BLOCK_COMMENT("arraycopy_range_checks:");
2352 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2353 __ movl(temp, length);
2354 __ addl(temp, src_pos); // src_pos + length
2355 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2356 __ jcc(Assembler::above, L_failed);
2358 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2359 __ movl(temp, length);
2360 __ addl(temp, dst_pos); // dst_pos + length
2361 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2362 __ jcc(Assembler::above, L_failed);
2364 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2365 // Move with sign extension can be used since they are positive.
2366 __ movslq(src_pos, src_pos);
2367 __ movslq(dst_pos, dst_pos);
2369 BLOCK_COMMENT("arraycopy_range_checks done");
2370 }
2372 //
2373 // Generate generic array copy stubs
2374 //
2375 // Input:
2376 // c_rarg0 - src oop
2377 // c_rarg1 - src_pos (32-bits)
2378 // c_rarg2 - dst oop
2379 // c_rarg3 - dst_pos (32-bits)
2380 // not Win64
2381 // c_rarg4 - element count (32-bits)
2382 // Win64
2383 // rsp+40 - element count (32-bits)
2384 //
2385 // Output:
2386 // rax == 0 - success
2387 // rax == -1^K - failure, where K is partial transfer count
2388 //
2389 address generate_generic_copy(const char *name) {
2391 Label L_failed, L_failed_0, L_objArray;
2392 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2394 // Input registers
2395 const Register src = c_rarg0; // source array oop
2396 const Register src_pos = c_rarg1; // source position
2397 const Register dst = c_rarg2; // destination array oop
2398 const Register dst_pos = c_rarg3; // destination position
2399 // elements count is on stack on Win64
2400 #ifdef _WIN64
2401 #define C_RARG4 Address(rsp, 6 * wordSize)
2402 #else
2403 #define C_RARG4 c_rarg4
2404 #endif
2406 { int modulus = CodeEntryAlignment;
2407 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2408 int advance = target - (__ offset() % modulus);
2409 if (advance < 0) advance += modulus;
2410 if (advance > 0) __ nop(advance);
2411 }
2412 StubCodeMark mark(this, "StubRoutines", name);
2414 // Short-hop target to L_failed. Makes for denser prologue code.
2415 __ BIND(L_failed_0);
2416 __ jmp(L_failed);
2417 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2419 __ align(CodeEntryAlignment);
2420 address start = __ pc();
2422 __ enter(); // required for proper stackwalking of RuntimeStub frame
2424 // bump this on entry, not on exit:
2425 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2427 //-----------------------------------------------------------------------
2428 // Assembler stub will be used for this call to arraycopy
2429 // if the following conditions are met:
2430 //
2431 // (1) src and dst must not be null.
2432 // (2) src_pos must not be negative.
2433 // (3) dst_pos must not be negative.
2434 // (4) length must not be negative.
2435 // (5) src klass and dst klass should be the same and not NULL.
2436 // (6) src and dst should be arrays.
2437 // (7) src_pos + length must not exceed length of src.
2438 // (8) dst_pos + length must not exceed length of dst.
2439 //
2441 // if (src == NULL) return -1;
2442 __ testptr(src, src); // src oop
2443 size_t j1off = __ offset();
2444 __ jccb(Assembler::zero, L_failed_0);
2446 // if (src_pos < 0) return -1;
2447 __ testl(src_pos, src_pos); // src_pos (32-bits)
2448 __ jccb(Assembler::negative, L_failed_0);
2450 // if (dst == NULL) return -1;
2451 __ testptr(dst, dst); // dst oop
2452 __ jccb(Assembler::zero, L_failed_0);
2454 // if (dst_pos < 0) return -1;
2455 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2456 size_t j4off = __ offset();
2457 __ jccb(Assembler::negative, L_failed_0);
2459 // The first four tests are very dense code,
2460 // but not quite dense enough to put four
2461 // jumps in a 16-byte instruction fetch buffer.
2462 // That's good, because some branch predicters
2463 // do not like jumps so close together.
2464 // Make sure of this.
2465 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2467 // registers used as temp
2468 const Register r11_length = r11; // elements count to copy
2469 const Register r10_src_klass = r10; // array klass
2470 const Register r9_dst_klass = r9; // dest array klass
2472 // if (length < 0) return -1;
2473 __ movl(r11_length, C_RARG4); // length (elements count, 32-bits value)
2474 __ testl(r11_length, r11_length);
2475 __ jccb(Assembler::negative, L_failed_0);
2477 __ load_klass(r10_src_klass, src);
2478 #ifdef ASSERT
2479 // assert(src->klass() != NULL);
2480 BLOCK_COMMENT("assert klasses not null");
2481 { Label L1, L2;
2482 __ testptr(r10_src_klass, r10_src_klass);
2483 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2484 __ bind(L1);
2485 __ stop("broken null klass");
2486 __ bind(L2);
2487 __ load_klass(r9_dst_klass, dst);
2488 __ cmpq(r9_dst_klass, 0);
2489 __ jcc(Assembler::equal, L1); // this would be broken also
2490 BLOCK_COMMENT("assert done");
2491 }
2492 #endif
2494 // Load layout helper (32-bits)
2495 //
2496 // |array_tag| | header_size | element_type | |log2_element_size|
2497 // 32 30 24 16 8 2 0
2498 //
2499 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2500 //
2502 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2503 Klass::layout_helper_offset_in_bytes();
2505 const Register rax_lh = rax; // layout helper
2507 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2509 // Handle objArrays completely differently...
2510 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2511 __ cmpl(rax_lh, objArray_lh);
2512 __ jcc(Assembler::equal, L_objArray);
2514 // if (src->klass() != dst->klass()) return -1;
2515 __ load_klass(r9_dst_klass, dst);
2516 __ cmpq(r10_src_klass, r9_dst_klass);
2517 __ jcc(Assembler::notEqual, L_failed);
2519 // if (!src->is_Array()) return -1;
2520 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2521 __ jcc(Assembler::greaterEqual, L_failed);
2523 // At this point, it is known to be a typeArray (array_tag 0x3).
2524 #ifdef ASSERT
2525 { Label L;
2526 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2527 __ jcc(Assembler::greaterEqual, L);
2528 __ stop("must be a primitive array");
2529 __ bind(L);
2530 }
2531 #endif
2533 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2534 r10, L_failed);
2536 // typeArrayKlass
2537 //
2538 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2539 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2540 //
2542 const Register r10_offset = r10; // array offset
2543 const Register rax_elsize = rax_lh; // element size
2545 __ movl(r10_offset, rax_lh);
2546 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2547 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2548 __ addptr(src, r10_offset); // src array offset
2549 __ addptr(dst, r10_offset); // dst array offset
2550 BLOCK_COMMENT("choose copy loop based on element size");
2551 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2553 // next registers should be set before the jump to corresponding stub
2554 const Register from = c_rarg0; // source array address
2555 const Register to = c_rarg1; // destination array address
2556 const Register count = c_rarg2; // elements count
2558 // 'from', 'to', 'count' registers should be set in such order
2559 // since they are the same as 'src', 'src_pos', 'dst'.
2561 __ BIND(L_copy_bytes);
2562 __ cmpl(rax_elsize, 0);
2563 __ jccb(Assembler::notEqual, L_copy_shorts);
2564 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2565 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2566 __ movl2ptr(count, r11_length); // length
2567 __ jump(RuntimeAddress(byte_copy_entry));
2569 __ BIND(L_copy_shorts);
2570 __ cmpl(rax_elsize, LogBytesPerShort);
2571 __ jccb(Assembler::notEqual, L_copy_ints);
2572 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2573 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2574 __ movl2ptr(count, r11_length); // length
2575 __ jump(RuntimeAddress(short_copy_entry));
2577 __ BIND(L_copy_ints);
2578 __ cmpl(rax_elsize, LogBytesPerInt);
2579 __ jccb(Assembler::notEqual, L_copy_longs);
2580 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2581 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2582 __ movl2ptr(count, r11_length); // length
2583 __ jump(RuntimeAddress(int_copy_entry));
2585 __ BIND(L_copy_longs);
2586 #ifdef ASSERT
2587 { Label L;
2588 __ cmpl(rax_elsize, LogBytesPerLong);
2589 __ jcc(Assembler::equal, L);
2590 __ stop("must be long copy, but elsize is wrong");
2591 __ bind(L);
2592 }
2593 #endif
2594 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2595 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2596 __ movl2ptr(count, r11_length); // length
2597 __ jump(RuntimeAddress(long_copy_entry));
2599 // objArrayKlass
2600 __ BIND(L_objArray);
2601 // live at this point: r10_src_klass, src[_pos], dst[_pos]
2603 Label L_plain_copy, L_checkcast_copy;
2604 // test array classes for subtyping
2605 __ load_klass(r9_dst_klass, dst);
2606 __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality
2607 __ jcc(Assembler::notEqual, L_checkcast_copy);
2609 // Identically typed arrays can be copied without element-wise checks.
2610 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2611 r10, L_failed);
2613 __ lea(from, Address(src, src_pos, TIMES_OOP,
2614 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2615 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2616 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2617 __ movl2ptr(count, r11_length); // length
2618 __ BIND(L_plain_copy);
2619 __ jump(RuntimeAddress(oop_copy_entry));
2621 __ BIND(L_checkcast_copy);
2622 // live at this point: r10_src_klass, !r11_length
2623 {
2624 // assert(r11_length == C_RARG4); // will reload from here
2625 Register r11_dst_klass = r11;
2626 __ load_klass(r11_dst_klass, dst);
2628 // Before looking at dst.length, make sure dst is also an objArray.
2629 __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh);
2630 __ jcc(Assembler::notEqual, L_failed);
2632 // It is safe to examine both src.length and dst.length.
2633 #ifndef _WIN64
2634 arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4,
2635 rax, L_failed);
2636 #else
2637 __ movl(r11_length, C_RARG4); // reload
2638 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2639 rax, L_failed);
2640 __ load_klass(r11_dst_klass, dst); // reload
2641 #endif
2643 // Marshal the base address arguments now, freeing registers.
2644 __ lea(from, Address(src, src_pos, TIMES_OOP,
2645 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2646 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2647 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2648 __ movl(count, C_RARG4); // length (reloaded)
2649 Register sco_temp = c_rarg3; // this register is free now
2650 assert_different_registers(from, to, count, sco_temp,
2651 r11_dst_klass, r10_src_klass);
2652 assert_clean_int(count, sco_temp);
2654 // Generate the type check.
2655 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2656 Klass::super_check_offset_offset_in_bytes());
2657 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2658 assert_clean_int(sco_temp, rax);
2659 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2661 // Fetch destination element klass from the objArrayKlass header.
2662 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2663 objArrayKlass::element_klass_offset_in_bytes());
2664 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2665 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2666 assert_clean_int(sco_temp, rax);
2668 // the checkcast_copy loop needs two extra arguments:
2669 assert(c_rarg3 == sco_temp, "#3 already in place");
2670 __ movptr(C_RARG4, r11_dst_klass); // dst.klass.element_klass
2671 __ jump(RuntimeAddress(checkcast_copy_entry));
2672 }
2674 __ BIND(L_failed);
2675 __ xorptr(rax, rax);
2676 __ notptr(rax); // return -1
2677 __ leave(); // required for proper stackwalking of RuntimeStub frame
2678 __ ret(0);
2680 return start;
2681 }
2683 #undef length_arg
2685 void generate_arraycopy_stubs() {
2686 // Call the conjoint generation methods immediately after
2687 // the disjoint ones so that short branches from the former
2688 // to the latter can be generated.
2689 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
2690 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
2692 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
2693 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
2695 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy");
2696 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy");
2698 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy");
2699 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy");
2702 if (UseCompressedOops) {
2703 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy");
2704 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy");
2705 } else {
2706 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy");
2707 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy");
2708 }
2710 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
2711 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy");
2712 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy");
2714 // We don't generate specialized code for HeapWord-aligned source
2715 // arrays, so just use the code we've already generated
2716 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2717 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2719 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2720 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2722 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2723 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2725 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2726 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2728 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2729 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2730 }
2732 #undef __
2733 #define __ masm->
2735 // Continuation point for throwing of implicit exceptions that are
2736 // not handled in the current activation. Fabricates an exception
2737 // oop and initiates normal exception dispatching in this
2738 // frame. Since we need to preserve callee-saved values (currently
2739 // only for C2, but done for C1 as well) we need a callee-saved oop
2740 // map and therefore have to make these stubs into RuntimeStubs
2741 // rather than BufferBlobs. If the compiler needs all registers to
2742 // be preserved between the fault point and the exception handler
2743 // then it must assume responsibility for that in
2744 // AbstractCompiler::continuation_for_implicit_null_exception or
2745 // continuation_for_implicit_division_by_zero_exception. All other
2746 // implicit exceptions (e.g., NullPointerException or
2747 // AbstractMethodError on entry) are either at call sites or
2748 // otherwise assume that stack unwinding will be initiated, so
2749 // caller saved registers were assumed volatile in the compiler.
2750 address generate_throw_exception(const char* name,
2751 address runtime_entry,
2752 bool restore_saved_exception_pc) {
2753 // Information about frame layout at time of blocking runtime call.
2754 // Note that we only have to preserve callee-saved registers since
2755 // the compilers are responsible for supplying a continuation point
2756 // if they expect all registers to be preserved.
2757 enum layout {
2758 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2759 rbp_off2,
2760 return_off,
2761 return_off2,
2762 framesize // inclusive of return address
2763 };
2765 int insts_size = 512;
2766 int locs_size = 64;
2768 CodeBuffer code(name, insts_size, locs_size);
2769 OopMapSet* oop_maps = new OopMapSet();
2770 MacroAssembler* masm = new MacroAssembler(&code);
2772 address start = __ pc();
2774 // This is an inlined and slightly modified version of call_VM
2775 // which has the ability to fetch the return PC out of
2776 // thread-local storage and also sets up last_Java_sp slightly
2777 // differently than the real call_VM
2778 if (restore_saved_exception_pc) {
2779 __ movptr(rax,
2780 Address(r15_thread,
2781 in_bytes(JavaThread::saved_exception_pc_offset())));
2782 __ push(rax);
2783 }
2785 __ enter(); // required for proper stackwalking of RuntimeStub frame
2787 assert(is_even(framesize/2), "sp not 16-byte aligned");
2789 // return address and rbp are already in place
2790 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
2792 int frame_complete = __ pc() - start;
2794 // Set up last_Java_sp and last_Java_fp
2795 __ set_last_Java_frame(rsp, rbp, NULL);
2797 // Call runtime
2798 __ movptr(c_rarg0, r15_thread);
2799 BLOCK_COMMENT("call runtime_entry");
2800 __ call(RuntimeAddress(runtime_entry));
2802 // Generate oop map
2803 OopMap* map = new OopMap(framesize, 0);
2805 oop_maps->add_gc_map(__ pc() - start, map);
2807 __ reset_last_Java_frame(true, false);
2809 __ leave(); // required for proper stackwalking of RuntimeStub frame
2811 // check for pending exceptions
2812 #ifdef ASSERT
2813 Label L;
2814 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
2815 (int32_t) NULL_WORD);
2816 __ jcc(Assembler::notEqual, L);
2817 __ should_not_reach_here();
2818 __ bind(L);
2819 #endif // ASSERT
2820 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2823 // codeBlob framesize is in words (not VMRegImpl::slot_size)
2824 RuntimeStub* stub =
2825 RuntimeStub::new_runtime_stub(name,
2826 &code,
2827 frame_complete,
2828 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2829 oop_maps, false);
2830 return stub->entry_point();
2831 }
2833 // Initialization
2834 void generate_initial() {
2835 // Generates all stubs and initializes the entry points
2837 // This platform-specific stub is needed by generate_call_stub()
2838 StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
2840 // entry points that exist in all platforms Note: This is code
2841 // that could be shared among different platforms - however the
2842 // benefit seems to be smaller than the disadvantage of having a
2843 // much more complicated generator structure. See also comment in
2844 // stubRoutines.hpp.
2846 StubRoutines::_forward_exception_entry = generate_forward_exception();
2848 StubRoutines::_call_stub_entry =
2849 generate_call_stub(StubRoutines::_call_stub_return_address);
2851 // is referenced by megamorphic call
2852 StubRoutines::_catch_exception_entry = generate_catch_exception();
2854 // atomic calls
2855 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2856 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
2857 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
2858 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2859 StubRoutines::_atomic_add_entry = generate_atomic_add();
2860 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
2861 StubRoutines::_fence_entry = generate_orderaccess_fence();
2863 StubRoutines::_handler_for_unsafe_access_entry =
2864 generate_handler_for_unsafe_access();
2866 // platform dependent
2867 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
2869 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2870 }
2872 void generate_all() {
2873 // Generates all stubs and initializes the entry points
2875 // These entry points require SharedInfo::stack0 to be set up in
2876 // non-core builds and need to be relocatable, so they each
2877 // fabricate a RuntimeStub internally.
2878 StubRoutines::_throw_AbstractMethodError_entry =
2879 generate_throw_exception("AbstractMethodError throw_exception",
2880 CAST_FROM_FN_PTR(address,
2881 SharedRuntime::
2882 throw_AbstractMethodError),
2883 false);
2885 StubRoutines::_throw_IncompatibleClassChangeError_entry =
2886 generate_throw_exception("IncompatibleClassChangeError throw_exception",
2887 CAST_FROM_FN_PTR(address,
2888 SharedRuntime::
2889 throw_IncompatibleClassChangeError),
2890 false);
2892 StubRoutines::_throw_ArithmeticException_entry =
2893 generate_throw_exception("ArithmeticException throw_exception",
2894 CAST_FROM_FN_PTR(address,
2895 SharedRuntime::
2896 throw_ArithmeticException),
2897 true);
2899 StubRoutines::_throw_NullPointerException_entry =
2900 generate_throw_exception("NullPointerException throw_exception",
2901 CAST_FROM_FN_PTR(address,
2902 SharedRuntime::
2903 throw_NullPointerException),
2904 true);
2906 StubRoutines::_throw_NullPointerException_at_call_entry =
2907 generate_throw_exception("NullPointerException at call throw_exception",
2908 CAST_FROM_FN_PTR(address,
2909 SharedRuntime::
2910 throw_NullPointerException_at_call),
2911 false);
2913 StubRoutines::_throw_StackOverflowError_entry =
2914 generate_throw_exception("StackOverflowError throw_exception",
2915 CAST_FROM_FN_PTR(address,
2916 SharedRuntime::
2917 throw_StackOverflowError),
2918 false);
2920 // entry points that are platform specific
2921 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
2922 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
2923 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
2924 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
2926 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
2927 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
2928 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
2929 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
2931 // support for verify_oop (must happen after universe_init)
2932 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2934 // arraycopy stubs used by compilers
2935 generate_arraycopy_stubs();
2936 }
2938 public:
2939 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2940 if (all) {
2941 generate_all();
2942 } else {
2943 generate_initial();
2944 }
2945 }
2946 }; // end class declaration
2948 address StubGenerator::disjoint_byte_copy_entry = NULL;
2949 address StubGenerator::disjoint_short_copy_entry = NULL;
2950 address StubGenerator::disjoint_int_copy_entry = NULL;
2951 address StubGenerator::disjoint_long_copy_entry = NULL;
2952 address StubGenerator::disjoint_oop_copy_entry = NULL;
2954 address StubGenerator::byte_copy_entry = NULL;
2955 address StubGenerator::short_copy_entry = NULL;
2956 address StubGenerator::int_copy_entry = NULL;
2957 address StubGenerator::long_copy_entry = NULL;
2958 address StubGenerator::oop_copy_entry = NULL;
2960 address StubGenerator::checkcast_copy_entry = NULL;
2962 void StubGenerator_generate(CodeBuffer* code, bool all) {
2963 StubGenerator g(code, all);
2964 }