1 /*
2 * Copyright (c) 2003, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "nativeInst_x86.hpp"
30 #include "oops/instanceOop.hpp"
31 #include "oops/method.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "oops/oop.inline.hpp"
34 #include "prims/methodHandles.hpp"
35 #include "runtime/frame.inline.hpp"
36 #include "runtime/handles.inline.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubCodeGenerator.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "runtime/thread.inline.hpp"
41 #include "utilities/top.hpp"
42 #ifdef COMPILER2
43 #include "opto/runtime.hpp"
44 #endif
45
46 // Declaration and definition of StubGenerator (no .hpp file).
47 // For a more detailed description of the stub routine structure
48 // see the comment in stubRoutines.hpp
49
50 #define __ _masm->
51 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
52 #define a__ ((Assembler*)_masm)->
53
54 #ifdef PRODUCT
55 #define BLOCK_COMMENT(str) /* nothing */
56 #else
57 #define BLOCK_COMMENT(str) __ block_comment(str)
58 #endif
59
60 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
61 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
62
63 // Stub Code definitions
64
65 static address handle_unsafe_access() {
66 JavaThread* thread = JavaThread::current();
67 address pc = thread->saved_exception_pc();
68 // pc is the instruction which we must emulate
69 // doing a no-op is fine: return garbage from the load
70 // therefore, compute npc
71 address npc = Assembler::locate_next_instruction(pc);
72
73 // request an async exception
74 thread->set_pending_unsafe_access_error();
75
76 // return address of next instruction to execute
77 return npc;
78 }
79
80 class StubGenerator: public StubCodeGenerator {
81 private:
82
83 #ifdef PRODUCT
84 #define inc_counter_np(counter) ((void)0)
85 #else
86 void inc_counter_np_(int& counter) {
87 // This can destroy rscratch1 if counter is far from the code cache
88 __ incrementl(ExternalAddress((address)&counter));
89 }
90 #define inc_counter_np(counter) \
91 BLOCK_COMMENT("inc_counter " #counter); \
92 inc_counter_np_(counter);
93 #endif
94
95 // Call stubs are used to call Java from C
96 //
97 // Linux Arguments:
98 // c_rarg0: call wrapper address address
99 // c_rarg1: result address
100 // c_rarg2: result type BasicType
101 // c_rarg3: method Method*
102 // c_rarg4: (interpreter) entry point address
103 // c_rarg5: parameters intptr_t*
104 // 16(rbp): parameter size (in words) int
105 // 24(rbp): thread Thread*
106 //
107 // [ return_from_Java ] <--- rsp
108 // [ argument word n ]
109 // ...
110 // -12 [ argument word 1 ]
111 // -11 [ saved r15 ] <--- rsp_after_call
112 // -10 [ saved r14 ]
113 // -9 [ saved r13 ]
114 // -8 [ saved r12 ]
115 // -7 [ saved rbx ]
116 // -6 [ call wrapper ]
117 // -5 [ result ]
118 // -4 [ result type ]
119 // -3 [ method ]
120 // -2 [ entry point ]
121 // -1 [ parameters ]
122 // 0 [ saved rbp ] <--- rbp
123 // 1 [ return address ]
124 // 2 [ parameter size ]
125 // 3 [ thread ]
126 //
127 // Windows Arguments:
128 // c_rarg0: call wrapper address address
129 // c_rarg1: result address
130 // c_rarg2: result type BasicType
131 // c_rarg3: method Method*
132 // 48(rbp): (interpreter) entry point address
133 // 56(rbp): parameters intptr_t*
134 // 64(rbp): parameter size (in words) int
135 // 72(rbp): thread Thread*
136 //
137 // [ return_from_Java ] <--- rsp
138 // [ argument word n ]
139 // ...
140 // -60 [ argument word 1 ]
141 // -59 [ saved xmm31 ] <--- rsp after_call
142 // [ saved xmm16-xmm30 ] (EVEX enabled, else the space is blank)
143 // -27 [ saved xmm15 ]
144 // [ saved xmm7-xmm14 ]
145 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
146 // -7 [ saved r15 ]
147 // -6 [ saved r14 ]
148 // -5 [ saved r13 ]
149 // -4 [ saved r12 ]
150 // -3 [ saved rdi ]
151 // -2 [ saved rsi ]
152 // -1 [ saved rbx ]
153 // 0 [ saved rbp ] <--- rbp
154 // 1 [ return address ]
155 // 2 [ call wrapper ]
156 // 3 [ result ]
157 // 4 [ result type ]
158 // 5 [ method ]
159 // 6 [ entry point ]
160 // 7 [ parameters ]
161 // 8 [ parameter size ]
162 // 9 [ thread ]
163 //
164 // Windows reserves the callers stack space for arguments 1-4.
165 // We spill c_rarg0-c_rarg3 to this space.
166
167 // Call stub stack layout word offsets from rbp
168 enum call_stub_layout {
169 #ifdef _WIN64
170 xmm_save_first = 6, // save from xmm6
171 xmm_save_last = 31, // to xmm31
172 xmm_save_base = -9,
173 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
174 r15_off = -7,
175 r14_off = -6,
176 r13_off = -5,
177 r12_off = -4,
178 rdi_off = -3,
179 rsi_off = -2,
180 rbx_off = -1,
181 rbp_off = 0,
182 retaddr_off = 1,
183 call_wrapper_off = 2,
184 result_off = 3,
185 result_type_off = 4,
186 method_off = 5,
187 entry_point_off = 6,
188 parameters_off = 7,
189 parameter_size_off = 8,
190 thread_off = 9
191 #else
192 rsp_after_call_off = -12,
193 mxcsr_off = rsp_after_call_off,
194 r15_off = -11,
195 r14_off = -10,
196 r13_off = -9,
197 r12_off = -8,
198 rbx_off = -7,
199 call_wrapper_off = -6,
200 result_off = -5,
201 result_type_off = -4,
202 method_off = -3,
203 entry_point_off = -2,
204 parameters_off = -1,
205 rbp_off = 0,
206 retaddr_off = 1,
207 parameter_size_off = 2,
208 thread_off = 3
209 #endif
210 };
211
212 #ifdef _WIN64
213 Address xmm_save(int reg) {
214 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
215 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
216 }
217 #endif
218
219 address generate_call_stub(address& return_address) {
220 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
221 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
222 "adjust this code");
223 StubCodeMark mark(this, "StubRoutines", "call_stub");
224 address start = __ pc();
225
226 // same as in generate_catch_exception()!
227 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
228
229 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
230 const Address result (rbp, result_off * wordSize);
231 const Address result_type (rbp, result_type_off * wordSize);
232 const Address method (rbp, method_off * wordSize);
233 const Address entry_point (rbp, entry_point_off * wordSize);
234 const Address parameters (rbp, parameters_off * wordSize);
235 const Address parameter_size(rbp, parameter_size_off * wordSize);
236
237 // same as in generate_catch_exception()!
238 const Address thread (rbp, thread_off * wordSize);
239
240 const Address r15_save(rbp, r15_off * wordSize);
241 const Address r14_save(rbp, r14_off * wordSize);
242 const Address r13_save(rbp, r13_off * wordSize);
243 const Address r12_save(rbp, r12_off * wordSize);
244 const Address rbx_save(rbp, rbx_off * wordSize);
245
246 // stub code
247 __ enter();
248 __ subptr(rsp, -rsp_after_call_off * wordSize);
249
250 // save register parameters
251 #ifndef _WIN64
252 __ movptr(parameters, c_rarg5); // parameters
253 __ movptr(entry_point, c_rarg4); // entry_point
254 #endif
255
256 __ movptr(method, c_rarg3); // method
257 __ movl(result_type, c_rarg2); // result type
258 __ movptr(result, c_rarg1); // result
259 __ movptr(call_wrapper, c_rarg0); // call wrapper
260
261 // save regs belonging to calling function
262 __ movptr(rbx_save, rbx);
263 __ movptr(r12_save, r12);
264 __ movptr(r13_save, r13);
265 __ movptr(r14_save, r14);
266 __ movptr(r15_save, r15);
267 if (UseAVX > 2) {
268 __ movl(rbx, 0xffff);
269 __ kmovql(k1, rbx);
270 }
271 #ifdef _WIN64
272 if (UseAVX > 2) {
273 for (int i = 6; i <= 31; i++) {
274 __ movdqu(xmm_save(i), as_XMMRegister(i));
275 }
276 } else {
277 for (int i = 6; i <= 15; i++) {
278 __ movdqu(xmm_save(i), as_XMMRegister(i));
279 }
280 }
281
282 const Address rdi_save(rbp, rdi_off * wordSize);
283 const Address rsi_save(rbp, rsi_off * wordSize);
284
285 __ movptr(rsi_save, rsi);
286 __ movptr(rdi_save, rdi);
287 #else
288 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
289 {
290 Label skip_ldmx;
291 __ stmxcsr(mxcsr_save);
292 __ movl(rax, mxcsr_save);
293 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
294 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
295 __ cmp32(rax, mxcsr_std);
296 __ jcc(Assembler::equal, skip_ldmx);
297 __ ldmxcsr(mxcsr_std);
298 __ bind(skip_ldmx);
299 }
300 #endif
301
302 // Load up thread register
303 __ movptr(r15_thread, thread);
304 __ reinit_heapbase();
305
306 #ifdef ASSERT
307 // make sure we have no pending exceptions
308 {
309 Label L;
310 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
311 __ jcc(Assembler::equal, L);
312 __ stop("StubRoutines::call_stub: entered with pending exception");
313 __ bind(L);
314 }
315 #endif
316
317 // pass parameters if any
318 BLOCK_COMMENT("pass parameters if any");
319 Label parameters_done;
320 __ movl(c_rarg3, parameter_size);
321 __ testl(c_rarg3, c_rarg3);
322 __ jcc(Assembler::zero, parameters_done);
323
324 Label loop;
325 __ movptr(c_rarg2, parameters); // parameter pointer
326 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
327 __ BIND(loop);
328 __ movptr(rax, Address(c_rarg2, 0));// get parameter
329 __ addptr(c_rarg2, wordSize); // advance to next parameter
330 __ decrementl(c_rarg1); // decrement counter
331 __ push(rax); // pass parameter
332 __ jcc(Assembler::notZero, loop);
333
334 // call Java function
335 __ BIND(parameters_done);
336 __ movptr(rbx, method); // get Method*
337 __ movptr(c_rarg1, entry_point); // get entry_point
338 __ mov(r13, rsp); // set sender sp
339 BLOCK_COMMENT("call Java function");
340 __ call(c_rarg1);
341
342 BLOCK_COMMENT("call_stub_return_address:");
343 return_address = __ pc();
344
345 // store result depending on type (everything that is not
346 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
347 __ movptr(c_rarg0, result);
348 Label is_long, is_float, is_double, exit;
349 __ movl(c_rarg1, result_type);
350 __ cmpl(c_rarg1, T_OBJECT);
351 __ jcc(Assembler::equal, is_long);
352 __ cmpl(c_rarg1, T_LONG);
353 __ jcc(Assembler::equal, is_long);
354 __ cmpl(c_rarg1, T_FLOAT);
355 __ jcc(Assembler::equal, is_float);
356 __ cmpl(c_rarg1, T_DOUBLE);
357 __ jcc(Assembler::equal, is_double);
358
359 // handle T_INT case
360 __ movl(Address(c_rarg0, 0), rax);
361
362 __ BIND(exit);
363
364 // pop parameters
365 __ lea(rsp, rsp_after_call);
366
367 #ifdef ASSERT
368 // verify that threads correspond
369 {
370 Label L, S;
371 __ cmpptr(r15_thread, thread);
372 __ jcc(Assembler::notEqual, S);
373 __ get_thread(rbx);
374 __ cmpptr(r15_thread, rbx);
375 __ jcc(Assembler::equal, L);
376 __ bind(S);
377 __ jcc(Assembler::equal, L);
378 __ stop("StubRoutines::call_stub: threads must correspond");
379 __ bind(L);
380 }
381 #endif
382
383 // restore regs belonging to calling function
384 #ifdef _WIN64
385 for (int i = 15; i >= 6; i--) {
386 __ movdqu(as_XMMRegister(i), xmm_save(i));
387 }
388 #endif
389 __ movptr(r15, r15_save);
390 __ movptr(r14, r14_save);
391 __ movptr(r13, r13_save);
392 __ movptr(r12, r12_save);
393 __ movptr(rbx, rbx_save);
394
395 #ifdef _WIN64
396 __ movptr(rdi, rdi_save);
397 __ movptr(rsi, rsi_save);
398 #else
399 __ ldmxcsr(mxcsr_save);
400 #endif
401
402 // restore rsp
403 __ addptr(rsp, -rsp_after_call_off * wordSize);
404
405 // return
406 __ pop(rbp);
407 __ ret(0);
408
409 // handle return types different from T_INT
410 __ BIND(is_long);
411 __ movq(Address(c_rarg0, 0), rax);
412 __ jmp(exit);
413
414 __ BIND(is_float);
415 __ movflt(Address(c_rarg0, 0), xmm0);
416 __ jmp(exit);
417
418 __ BIND(is_double);
419 __ movdbl(Address(c_rarg0, 0), xmm0);
420 __ jmp(exit);
421
422 return start;
423 }
424
425 // Return point for a Java call if there's an exception thrown in
426 // Java code. The exception is caught and transformed into a
427 // pending exception stored in JavaThread that can be tested from
428 // within the VM.
429 //
430 // Note: Usually the parameters are removed by the callee. In case
431 // of an exception crossing an activation frame boundary, that is
432 // not the case if the callee is compiled code => need to setup the
433 // rsp.
434 //
435 // rax: exception oop
436
437 address generate_catch_exception() {
438 StubCodeMark mark(this, "StubRoutines", "catch_exception");
439 address start = __ pc();
440
441 // same as in generate_call_stub():
442 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
443 const Address thread (rbp, thread_off * wordSize);
444
445 #ifdef ASSERT
446 // verify that threads correspond
447 {
448 Label L, S;
449 __ cmpptr(r15_thread, thread);
450 __ jcc(Assembler::notEqual, S);
451 __ get_thread(rbx);
452 __ cmpptr(r15_thread, rbx);
453 __ jcc(Assembler::equal, L);
454 __ bind(S);
455 __ stop("StubRoutines::catch_exception: threads must correspond");
456 __ bind(L);
457 }
458 #endif
459
460 // set pending exception
461 __ verify_oop(rax);
462
463 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
464 __ lea(rscratch1, ExternalAddress((address)__FILE__));
465 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
466 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
467
468 // complete return to VM
469 assert(StubRoutines::_call_stub_return_address != NULL,
470 "_call_stub_return_address must have been generated before");
471 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
472
473 return start;
474 }
475
476 // Continuation point for runtime calls returning with a pending
477 // exception. The pending exception check happened in the runtime
478 // or native call stub. The pending exception in Thread is
479 // converted into a Java-level exception.
480 //
481 // Contract with Java-level exception handlers:
482 // rax: exception
483 // rdx: throwing pc
484 //
485 // NOTE: At entry of this stub, exception-pc must be on stack !!
486
487 address generate_forward_exception() {
488 StubCodeMark mark(this, "StubRoutines", "forward exception");
489 address start = __ pc();
490
491 // Upon entry, the sp points to the return address returning into
492 // Java (interpreted or compiled) code; i.e., the return address
493 // becomes the throwing pc.
494 //
495 // Arguments pushed before the runtime call are still on the stack
496 // but the exception handler will reset the stack pointer ->
497 // ignore them. A potential result in registers can be ignored as
498 // well.
499
500 #ifdef ASSERT
501 // make sure this code is only executed if there is a pending exception
502 {
503 Label L;
504 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
505 __ jcc(Assembler::notEqual, L);
506 __ stop("StubRoutines::forward exception: no pending exception (1)");
507 __ bind(L);
508 }
509 #endif
510
511 // compute exception handler into rbx
512 __ movptr(c_rarg0, Address(rsp, 0));
513 BLOCK_COMMENT("call exception_handler_for_return_address");
514 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
515 SharedRuntime::exception_handler_for_return_address),
516 r15_thread, c_rarg0);
517 __ mov(rbx, rax);
518
519 // setup rax & rdx, remove return address & clear pending exception
520 __ pop(rdx);
521 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
522 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
523
524 #ifdef ASSERT
525 // make sure exception is set
526 {
527 Label L;
528 __ testptr(rax, rax);
529 __ jcc(Assembler::notEqual, L);
530 __ stop("StubRoutines::forward exception: no pending exception (2)");
531 __ bind(L);
532 }
533 #endif
534
535 // continue at exception handler (return address removed)
536 // rax: exception
537 // rbx: exception handler
538 // rdx: throwing pc
539 __ verify_oop(rax);
540 __ jmp(rbx);
541
542 return start;
543 }
544
545 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
546 //
547 // Arguments :
548 // c_rarg0: exchange_value
549 // c_rarg0: dest
550 //
551 // Result:
552 // *dest <- ex, return (orig *dest)
553 address generate_atomic_xchg() {
554 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
555 address start = __ pc();
556
557 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
558 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
559 __ ret(0);
560
561 return start;
562 }
563
564 // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
565 //
566 // Arguments :
567 // c_rarg0: exchange_value
568 // c_rarg1: dest
569 //
570 // Result:
571 // *dest <- ex, return (orig *dest)
572 address generate_atomic_xchg_ptr() {
573 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
574 address start = __ pc();
575
576 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
577 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
578 __ ret(0);
579
580 return start;
581 }
582
583 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
584 // jint compare_value)
585 //
586 // Arguments :
587 // c_rarg0: exchange_value
588 // c_rarg1: dest
589 // c_rarg2: compare_value
590 //
591 // Result:
592 // if ( compare_value == *dest ) {
593 // *dest = exchange_value
594 // return compare_value;
595 // else
596 // return *dest;
597 address generate_atomic_cmpxchg() {
598 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
599 address start = __ pc();
600
601 __ movl(rax, c_rarg2);
602 if ( os::is_MP() ) __ lock();
603 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
604 __ ret(0);
605
606 return start;
607 }
608
609 // Support for jbyte atomic::atomic_cmpxchg(jbyte exchange_value, volatile jbyte* dest,
610 // jbyte compare_value)
611 //
612 // Arguments :
613 // c_rarg0: exchange_value
614 // c_rarg1: dest
615 // c_rarg2: compare_value
616 //
617 // Result:
618 // if ( compare_value == *dest ) {
619 // *dest = exchange_value
620 // return compare_value;
621 // else
622 // return *dest;
623 address generate_atomic_cmpxchg_byte() {
624 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_byte");
625 address start = __ pc();
626
627 __ movsbq(rax, c_rarg2);
628 if ( os::is_MP() ) __ lock();
629 __ cmpxchgb(c_rarg0, Address(c_rarg1, 0));
630 __ ret(0);
631
632 return start;
633 }
634
635 // Support for jlong atomic::atomic_cmpxchg(jlong exchange_value,
636 // volatile jlong* dest,
637 // jlong compare_value)
638 // Arguments :
639 // c_rarg0: exchange_value
640 // c_rarg1: dest
641 // c_rarg2: compare_value
642 //
643 // Result:
644 // if ( compare_value == *dest ) {
645 // *dest = exchange_value
646 // return compare_value;
647 // else
648 // return *dest;
649 address generate_atomic_cmpxchg_long() {
650 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
651 address start = __ pc();
652
653 __ movq(rax, c_rarg2);
654 if ( os::is_MP() ) __ lock();
655 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
656 __ ret(0);
657
658 return start;
659 }
660
661 // Support for jint atomic::add(jint add_value, volatile jint* dest)
662 //
663 // Arguments :
664 // c_rarg0: add_value
665 // c_rarg1: dest
666 //
667 // Result:
668 // *dest += add_value
669 // return *dest;
670 address generate_atomic_add() {
671 StubCodeMark mark(this, "StubRoutines", "atomic_add");
672 address start = __ pc();
673
674 __ movl(rax, c_rarg0);
675 if ( os::is_MP() ) __ lock();
676 __ xaddl(Address(c_rarg1, 0), c_rarg0);
677 __ addl(rax, c_rarg0);
678 __ ret(0);
679
680 return start;
681 }
682
683 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
684 //
685 // Arguments :
686 // c_rarg0: add_value
687 // c_rarg1: dest
688 //
689 // Result:
690 // *dest += add_value
691 // return *dest;
692 address generate_atomic_add_ptr() {
693 StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
694 address start = __ pc();
695
696 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
697 if ( os::is_MP() ) __ lock();
698 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
699 __ addptr(rax, c_rarg0);
700 __ ret(0);
701
702 return start;
703 }
704
705 // Support for intptr_t OrderAccess::fence()
706 //
707 // Arguments :
708 //
709 // Result:
710 address generate_orderaccess_fence() {
711 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
712 address start = __ pc();
713 __ membar(Assembler::StoreLoad);
714 __ ret(0);
715
716 return start;
717 }
718
719 // Support for intptr_t get_previous_fp()
720 //
721 // This routine is used to find the previous frame pointer for the
722 // caller (current_frame_guess). This is used as part of debugging
723 // ps() is seemingly lost trying to find frames.
724 // This code assumes that caller current_frame_guess) has a frame.
725 address generate_get_previous_fp() {
726 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
727 const Address old_fp(rbp, 0);
728 const Address older_fp(rax, 0);
729 address start = __ pc();
730
731 __ enter();
732 __ movptr(rax, old_fp); // callers fp
733 __ movptr(rax, older_fp); // the frame for ps()
734 __ pop(rbp);
735 __ ret(0);
736
737 return start;
738 }
739
740 // Support for intptr_t get_previous_sp()
741 //
742 // This routine is used to find the previous stack pointer for the
743 // caller.
744 address generate_get_previous_sp() {
745 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
746 address start = __ pc();
747
748 __ movptr(rax, rsp);
749 __ addptr(rax, 8); // return address is at the top of the stack.
750 __ ret(0);
751
752 return start;
753 }
754
755 //----------------------------------------------------------------------------------------------------
756 // Support for void verify_mxcsr()
757 //
758 // This routine is used with -Xcheck:jni to verify that native
759 // JNI code does not return to Java code without restoring the
760 // MXCSR register to our expected state.
761
762 address generate_verify_mxcsr() {
763 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
764 address start = __ pc();
765
766 const Address mxcsr_save(rsp, 0);
767
768 if (CheckJNICalls) {
769 Label ok_ret;
770 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
771 __ push(rax);
772 __ subptr(rsp, wordSize); // allocate a temp location
773 __ stmxcsr(mxcsr_save);
774 __ movl(rax, mxcsr_save);
775 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
776 __ cmp32(rax, mxcsr_std);
777 __ jcc(Assembler::equal, ok_ret);
778
779 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
780
781 __ ldmxcsr(mxcsr_std);
782
783 __ bind(ok_ret);
784 __ addptr(rsp, wordSize);
785 __ pop(rax);
786 }
787
788 __ ret(0);
789
790 return start;
791 }
792
793 address generate_f2i_fixup() {
794 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
795 Address inout(rsp, 5 * wordSize); // return address + 4 saves
796
797 address start = __ pc();
798
799 Label L;
800
801 __ push(rax);
802 __ push(c_rarg3);
803 __ push(c_rarg2);
804 __ push(c_rarg1);
805
806 __ movl(rax, 0x7f800000);
807 __ xorl(c_rarg3, c_rarg3);
808 __ movl(c_rarg2, inout);
809 __ movl(c_rarg1, c_rarg2);
810 __ andl(c_rarg1, 0x7fffffff);
811 __ cmpl(rax, c_rarg1); // NaN? -> 0
812 __ jcc(Assembler::negative, L);
813 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
814 __ movl(c_rarg3, 0x80000000);
815 __ movl(rax, 0x7fffffff);
816 __ cmovl(Assembler::positive, c_rarg3, rax);
817
818 __ bind(L);
819 __ movptr(inout, c_rarg3);
820
821 __ pop(c_rarg1);
822 __ pop(c_rarg2);
823 __ pop(c_rarg3);
824 __ pop(rax);
825
826 __ ret(0);
827
828 return start;
829 }
830
831 address generate_f2l_fixup() {
832 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
833 Address inout(rsp, 5 * wordSize); // return address + 4 saves
834 address start = __ pc();
835
836 Label L;
837
838 __ push(rax);
839 __ push(c_rarg3);
840 __ push(c_rarg2);
841 __ push(c_rarg1);
842
843 __ movl(rax, 0x7f800000);
844 __ xorl(c_rarg3, c_rarg3);
845 __ movl(c_rarg2, inout);
846 __ movl(c_rarg1, c_rarg2);
847 __ andl(c_rarg1, 0x7fffffff);
848 __ cmpl(rax, c_rarg1); // NaN? -> 0
849 __ jcc(Assembler::negative, L);
850 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
851 __ mov64(c_rarg3, 0x8000000000000000);
852 __ mov64(rax, 0x7fffffffffffffff);
853 __ cmov(Assembler::positive, c_rarg3, rax);
854
855 __ bind(L);
856 __ movptr(inout, c_rarg3);
857
858 __ pop(c_rarg1);
859 __ pop(c_rarg2);
860 __ pop(c_rarg3);
861 __ pop(rax);
862
863 __ ret(0);
864
865 return start;
866 }
867
868 address generate_d2i_fixup() {
869 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
870 Address inout(rsp, 6 * wordSize); // return address + 5 saves
871
872 address start = __ pc();
873
874 Label L;
875
876 __ push(rax);
877 __ push(c_rarg3);
878 __ push(c_rarg2);
879 __ push(c_rarg1);
880 __ push(c_rarg0);
881
882 __ movl(rax, 0x7ff00000);
883 __ movq(c_rarg2, inout);
884 __ movl(c_rarg3, c_rarg2);
885 __ mov(c_rarg1, c_rarg2);
886 __ mov(c_rarg0, c_rarg2);
887 __ negl(c_rarg3);
888 __ shrptr(c_rarg1, 0x20);
889 __ orl(c_rarg3, c_rarg2);
890 __ andl(c_rarg1, 0x7fffffff);
891 __ xorl(c_rarg2, c_rarg2);
892 __ shrl(c_rarg3, 0x1f);
893 __ orl(c_rarg1, c_rarg3);
894 __ cmpl(rax, c_rarg1);
895 __ jcc(Assembler::negative, L); // NaN -> 0
896 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
897 __ movl(c_rarg2, 0x80000000);
898 __ movl(rax, 0x7fffffff);
899 __ cmov(Assembler::positive, c_rarg2, rax);
900
901 __ bind(L);
902 __ movptr(inout, c_rarg2);
903
904 __ pop(c_rarg0);
905 __ pop(c_rarg1);
906 __ pop(c_rarg2);
907 __ pop(c_rarg3);
908 __ pop(rax);
909
910 __ ret(0);
911
912 return start;
913 }
914
915 address generate_d2l_fixup() {
916 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
917 Address inout(rsp, 6 * wordSize); // return address + 5 saves
918
919 address start = __ pc();
920
921 Label L;
922
923 __ push(rax);
924 __ push(c_rarg3);
925 __ push(c_rarg2);
926 __ push(c_rarg1);
927 __ push(c_rarg0);
928
929 __ movl(rax, 0x7ff00000);
930 __ movq(c_rarg2, inout);
931 __ movl(c_rarg3, c_rarg2);
932 __ mov(c_rarg1, c_rarg2);
933 __ mov(c_rarg0, c_rarg2);
934 __ negl(c_rarg3);
935 __ shrptr(c_rarg1, 0x20);
936 __ orl(c_rarg3, c_rarg2);
937 __ andl(c_rarg1, 0x7fffffff);
938 __ xorl(c_rarg2, c_rarg2);
939 __ shrl(c_rarg3, 0x1f);
940 __ orl(c_rarg1, c_rarg3);
941 __ cmpl(rax, c_rarg1);
942 __ jcc(Assembler::negative, L); // NaN -> 0
943 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
944 __ mov64(c_rarg2, 0x8000000000000000);
945 __ mov64(rax, 0x7fffffffffffffff);
946 __ cmovq(Assembler::positive, c_rarg2, rax);
947
948 __ bind(L);
949 __ movq(inout, c_rarg2);
950
951 __ pop(c_rarg0);
952 __ pop(c_rarg1);
953 __ pop(c_rarg2);
954 __ pop(c_rarg3);
955 __ pop(rax);
956
957 __ ret(0);
958
959 return start;
960 }
961
962 address generate_fp_mask(const char *stub_name, int64_t mask) {
963 __ align(CodeEntryAlignment);
964 StubCodeMark mark(this, "StubRoutines", stub_name);
965 address start = __ pc();
966
967 __ emit_data64( mask, relocInfo::none );
968 __ emit_data64( mask, relocInfo::none );
969
970 return start;
971 }
972
973 // The following routine generates a subroutine to throw an
974 // asynchronous UnknownError when an unsafe access gets a fault that
975 // could not be reasonably prevented by the programmer. (Example:
976 // SIGBUS/OBJERR.)
977 address generate_handler_for_unsafe_access() {
978 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
979 address start = __ pc();
980
981 __ push(0); // hole for return address-to-be
982 __ pusha(); // push registers
983 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
984
985 // FIXME: this probably needs alignment logic
986
987 __ subptr(rsp, frame::arg_reg_save_area_bytes);
988 BLOCK_COMMENT("call handle_unsafe_access");
989 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
990 __ addptr(rsp, frame::arg_reg_save_area_bytes);
991
992 __ movptr(next_pc, rax); // stuff next address
993 __ popa();
994 __ ret(0); // jump to next address
995
996 return start;
997 }
998
999 // Non-destructive plausibility checks for oops
1000 //
1001 // Arguments:
1002 // all args on stack!
1003 //
1004 // Stack after saving c_rarg3:
1005 // [tos + 0]: saved c_rarg3
1006 // [tos + 1]: saved c_rarg2
1007 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1008 // [tos + 3]: saved flags
1009 // [tos + 4]: return address
1010 // * [tos + 5]: error message (char*)
1011 // * [tos + 6]: object to verify (oop)
1012 // * [tos + 7]: saved rax - saved by caller and bashed
1013 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1014 // * = popped on exit
1015 address generate_verify_oop() {
1016 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1017 address start = __ pc();
1018
1019 Label exit, error;
1020
1021 __ pushf();
1022 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1023
1024 __ push(r12);
1025
1026 // save c_rarg2 and c_rarg3
1027 __ push(c_rarg2);
1028 __ push(c_rarg3);
1029
1030 enum {
1031 // After previous pushes.
1032 oop_to_verify = 6 * wordSize,
1033 saved_rax = 7 * wordSize,
1034 saved_r10 = 8 * wordSize,
1035
1036 // Before the call to MacroAssembler::debug(), see below.
1037 return_addr = 16 * wordSize,
1038 error_msg = 17 * wordSize
1039 };
1040
1041 // get object
1042 __ movptr(rax, Address(rsp, oop_to_verify));
1043
1044 // make sure object is 'reasonable'
1045 __ testptr(rax, rax);
1046 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1047 // Check if the oop is in the right area of memory
1048 __ movptr(c_rarg2, rax);
1049 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1050 __ andptr(c_rarg2, c_rarg3);
1051 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1052 __ cmpptr(c_rarg2, c_rarg3);
1053 __ jcc(Assembler::notZero, error);
1054
1055 // set r12 to heapbase for load_klass()
1056 __ reinit_heapbase();
1057
1058 // make sure klass is 'reasonable', which is not zero.
1059 __ load_klass(rax, rax); // get klass
1060 __ testptr(rax, rax);
1061 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1062
1063 // return if everything seems ok
1064 __ bind(exit);
1065 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1066 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1067 __ pop(c_rarg3); // restore c_rarg3
1068 __ pop(c_rarg2); // restore c_rarg2
1069 __ pop(r12); // restore r12
1070 __ popf(); // restore flags
1071 __ ret(4 * wordSize); // pop caller saved stuff
1072
1073 // handle errors
1074 __ bind(error);
1075 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1076 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1077 __ pop(c_rarg3); // get saved c_rarg3 back
1078 __ pop(c_rarg2); // get saved c_rarg2 back
1079 __ pop(r12); // get saved r12 back
1080 __ popf(); // get saved flags off stack --
1081 // will be ignored
1082
1083 __ pusha(); // push registers
1084 // (rip is already
1085 // already pushed)
1086 // debug(char* msg, int64_t pc, int64_t regs[])
1087 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1088 // pushed all the registers, so now the stack looks like:
1089 // [tos + 0] 16 saved registers
1090 // [tos + 16] return address
1091 // * [tos + 17] error message (char*)
1092 // * [tos + 18] object to verify (oop)
1093 // * [tos + 19] saved rax - saved by caller and bashed
1094 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1095 // * = popped on exit
1096
1097 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1098 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1099 __ movq(c_rarg2, rsp); // pass address of regs on stack
1100 __ mov(r12, rsp); // remember rsp
1101 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1102 __ andptr(rsp, -16); // align stack as required by ABI
1103 BLOCK_COMMENT("call MacroAssembler::debug");
1104 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1105 __ mov(rsp, r12); // restore rsp
1106 __ popa(); // pop registers (includes r12)
1107 __ ret(4 * wordSize); // pop caller saved stuff
1108
1109 return start;
1110 }
1111
1112 //
1113 // Verify that a register contains clean 32-bits positive value
1114 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1115 //
1116 // Input:
1117 // Rint - 32-bits value
1118 // Rtmp - scratch
1119 //
1120 void assert_clean_int(Register Rint, Register Rtmp) {
1121 #ifdef ASSERT
1122 Label L;
1123 assert_different_registers(Rtmp, Rint);
1124 __ movslq(Rtmp, Rint);
1125 __ cmpq(Rtmp, Rint);
1126 __ jcc(Assembler::equal, L);
1127 __ stop("high 32-bits of int value are not 0");
1128 __ bind(L);
1129 #endif
1130 }
1131
1132 // Generate overlap test for array copy stubs
1133 //
1134 // Input:
1135 // c_rarg0 - from
1136 // c_rarg1 - to
1137 // c_rarg2 - element count
1138 //
1139 // Output:
1140 // rax - &from[element count - 1]
1141 //
1142 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1143 assert(no_overlap_target != NULL, "must be generated");
1144 array_overlap_test(no_overlap_target, NULL, sf);
1145 }
1146 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1147 array_overlap_test(NULL, &L_no_overlap, sf);
1148 }
1149 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1150 const Register from = c_rarg0;
1151 const Register to = c_rarg1;
1152 const Register count = c_rarg2;
1153 const Register end_from = rax;
1154
1155 __ cmpptr(to, from);
1156 __ lea(end_from, Address(from, count, sf, 0));
1157 if (NOLp == NULL) {
1158 ExternalAddress no_overlap(no_overlap_target);
1159 __ jump_cc(Assembler::belowEqual, no_overlap);
1160 __ cmpptr(to, end_from);
1161 __ jump_cc(Assembler::aboveEqual, no_overlap);
1162 } else {
1163 __ jcc(Assembler::belowEqual, (*NOLp));
1164 __ cmpptr(to, end_from);
1165 __ jcc(Assembler::aboveEqual, (*NOLp));
1166 }
1167 }
1168
1169 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1170 //
1171 // Outputs:
1172 // rdi - rcx
1173 // rsi - rdx
1174 // rdx - r8
1175 // rcx - r9
1176 //
1177 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1178 // are non-volatile. r9 and r10 should not be used by the caller.
1179 //
1180 void setup_arg_regs(int nargs = 3) {
1181 const Register saved_rdi = r9;
1182 const Register saved_rsi = r10;
1183 assert(nargs == 3 || nargs == 4, "else fix");
1184 #ifdef _WIN64
1185 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1186 "unexpected argument registers");
1187 if (nargs >= 4)
1188 __ mov(rax, r9); // r9 is also saved_rdi
1189 __ movptr(saved_rdi, rdi);
1190 __ movptr(saved_rsi, rsi);
1191 __ mov(rdi, rcx); // c_rarg0
1192 __ mov(rsi, rdx); // c_rarg1
1193 __ mov(rdx, r8); // c_rarg2
1194 if (nargs >= 4)
1195 __ mov(rcx, rax); // c_rarg3 (via rax)
1196 #else
1197 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1198 "unexpected argument registers");
1199 #endif
1200 }
1201
1202 void restore_arg_regs() {
1203 const Register saved_rdi = r9;
1204 const Register saved_rsi = r10;
1205 #ifdef _WIN64
1206 __ movptr(rdi, saved_rdi);
1207 __ movptr(rsi, saved_rsi);
1208 #endif
1209 }
1210
1211 // Generate code for an array write pre barrier
1212 //
1213 // addr - starting address
1214 // count - element count
1215 // tmp - scratch register
1216 //
1217 // Destroy no registers!
1218 //
1219 void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) {
1220 BarrierSet* bs = Universe::heap()->barrier_set();
1221 switch (bs->kind()) {
1222 case BarrierSet::G1SATBCTLogging:
1223 // With G1, don't generate the call if we statically know that the target in uninitialized
1224 if (!dest_uninitialized) {
1225 __ pusha(); // push registers
1226 if (count == c_rarg0) {
1227 if (addr == c_rarg1) {
1228 // exactly backwards!!
1229 __ xchgptr(c_rarg1, c_rarg0);
1230 } else {
1231 __ movptr(c_rarg1, count);
1232 __ movptr(c_rarg0, addr);
1233 }
1234 } else {
1235 __ movptr(c_rarg0, addr);
1236 __ movptr(c_rarg1, count);
1237 }
1238 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1239 __ popa();
1240 }
1241 break;
1242 case BarrierSet::CardTableModRef:
1243 case BarrierSet::CardTableExtension:
1244 case BarrierSet::ModRef:
1245 break;
1246 default:
1247 ShouldNotReachHere();
1248
1249 }
1250 }
1251
1252 //
1253 // Generate code for an array write post barrier
1254 //
1255 // Input:
1256 // start - register containing starting address of destination array
1257 // count - elements count
1258 // scratch - scratch register
1259 //
1260 // The input registers are overwritten.
1261 //
1262 void gen_write_ref_array_post_barrier(Register start, Register count, Register scratch) {
1263 assert_different_registers(start, count, scratch);
1264 BarrierSet* bs = Universe::heap()->barrier_set();
1265 switch (bs->kind()) {
1266 case BarrierSet::G1SATBCTLogging:
1267 {
1268 __ pusha(); // push registers (overkill)
1269 if (c_rarg0 == count) { // On win64 c_rarg0 == rcx
1270 assert_different_registers(c_rarg1, start);
1271 __ mov(c_rarg1, count);
1272 __ mov(c_rarg0, start);
1273 } else {
1274 assert_different_registers(c_rarg0, count);
1275 __ mov(c_rarg0, start);
1276 __ mov(c_rarg1, count);
1277 }
1278 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1279 __ popa();
1280 }
1281 break;
1282 case BarrierSet::CardTableModRef:
1283 case BarrierSet::CardTableExtension:
1284 {
1285 CardTableModRefBS* ct = barrier_set_cast<CardTableModRefBS>(bs);
1286 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1287
1288 Label L_loop;
1289 const Register end = count;
1290
1291 __ leaq(end, Address(start, count, TIMES_OOP, 0)); // end == start+count*oop_size
1292 __ subptr(end, BytesPerHeapOop); // end - 1 to make inclusive
1293 __ shrptr(start, CardTableModRefBS::card_shift);
1294 __ shrptr(end, CardTableModRefBS::card_shift);
1295 __ subptr(end, start); // end --> cards count
1296
1297 int64_t disp = (int64_t) ct->byte_map_base;
1298 __ mov64(scratch, disp);
1299 __ addptr(start, scratch);
1300 __ BIND(L_loop);
1301 __ movb(Address(start, count, Address::times_1), 0);
1302 __ decrement(count);
1303 __ jcc(Assembler::greaterEqual, L_loop);
1304 }
1305 break;
1306 default:
1307 ShouldNotReachHere();
1308
1309 }
1310 }
1311
1312
1313 // Copy big chunks forward
1314 //
1315 // Inputs:
1316 // end_from - source arrays end address
1317 // end_to - destination array end address
1318 // qword_count - 64-bits element count, negative
1319 // to - scratch
1320 // L_copy_bytes - entry label
1321 // L_copy_8_bytes - exit label
1322 //
1323 void copy_bytes_forward(Register end_from, Register end_to,
1324 Register qword_count, Register to,
1325 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1326 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1327 Label L_loop;
1328 __ align(OptoLoopAlignment);
1329 if (UseUnalignedLoadStores) {
1330 Label L_end;
1331 // Copy 64-bytes per iteration
1332 __ BIND(L_loop);
1333 if (UseAVX > 2) {
1334 __ evmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1335 __ evmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1336 } else if (UseAVX == 2) {
1337 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1338 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1339 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1340 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1341 } else {
1342 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1343 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1344 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1345 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1346 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1347 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1348 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1349 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1350 }
1351 __ BIND(L_copy_bytes);
1352 __ addptr(qword_count, 8);
1353 __ jcc(Assembler::lessEqual, L_loop);
1354 __ subptr(qword_count, 4); // sub(8) and add(4)
1355 __ jccb(Assembler::greater, L_end);
1356 // Copy trailing 32 bytes
1357 if (UseAVX >= 2) {
1358 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1359 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1360 } else {
1361 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1362 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1363 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1364 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1365 }
1366 __ addptr(qword_count, 4);
1367 __ BIND(L_end);
1368 if (UseAVX >= 2) {
1369 // clean upper bits of YMM registers
1370 __ vzeroupper();
1371 }
1372 } else {
1373 // Copy 32-bytes per iteration
1374 __ BIND(L_loop);
1375 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1376 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1377 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1378 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1379 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1380 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1381 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1382 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1383
1384 __ BIND(L_copy_bytes);
1385 __ addptr(qword_count, 4);
1386 __ jcc(Assembler::lessEqual, L_loop);
1387 }
1388 __ subptr(qword_count, 4);
1389 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1390 }
1391
1392 // Copy big chunks backward
1393 //
1394 // Inputs:
1395 // from - source arrays address
1396 // dest - destination array address
1397 // qword_count - 64-bits element count
1398 // to - scratch
1399 // L_copy_bytes - entry label
1400 // L_copy_8_bytes - exit label
1401 //
1402 void copy_bytes_backward(Register from, Register dest,
1403 Register qword_count, Register to,
1404 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1405 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1406 Label L_loop;
1407 __ align(OptoLoopAlignment);
1408 if (UseUnalignedLoadStores) {
1409 Label L_end;
1410 // Copy 64-bytes per iteration
1411 __ BIND(L_loop);
1412 if (UseAVX > 2) {
1413 __ evmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32), Assembler::AVX_512bit);
1414 __ evmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0, Assembler::AVX_512bit);
1415 } else if (UseAVX == 2) {
1416 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1417 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1418 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1419 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1420 } else {
1421 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1422 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1423 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1424 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1425 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1426 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1427 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1428 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1429 }
1430 __ BIND(L_copy_bytes);
1431 __ subptr(qword_count, 8);
1432 __ jcc(Assembler::greaterEqual, L_loop);
1433
1434 __ addptr(qword_count, 4); // add(8) and sub(4)
1435 __ jccb(Assembler::less, L_end);
1436 // Copy trailing 32 bytes
1437 if (UseAVX >= 2) {
1438 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1439 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1440 } else {
1441 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1442 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1443 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1444 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1445 }
1446 __ subptr(qword_count, 4);
1447 __ BIND(L_end);
1448 if (UseAVX >= 2) {
1449 // clean upper bits of YMM registers
1450 __ vzeroupper();
1451 }
1452 } else {
1453 // Copy 32-bytes per iteration
1454 __ BIND(L_loop);
1455 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1456 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1457 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1458 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1459 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1460 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1461 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1462 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1463
1464 __ BIND(L_copy_bytes);
1465 __ subptr(qword_count, 4);
1466 __ jcc(Assembler::greaterEqual, L_loop);
1467 }
1468 __ addptr(qword_count, 4);
1469 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1470 }
1471
1472
1473 // Arguments:
1474 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1475 // ignored
1476 // name - stub name string
1477 //
1478 // Inputs:
1479 // c_rarg0 - source array address
1480 // c_rarg1 - destination array address
1481 // c_rarg2 - element count, treated as ssize_t, can be zero
1482 //
1483 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1484 // we let the hardware handle it. The one to eight bytes within words,
1485 // dwords or qwords that span cache line boundaries will still be loaded
1486 // and stored atomically.
1487 //
1488 // Side Effects:
1489 // disjoint_byte_copy_entry is set to the no-overlap entry point
1490 // used by generate_conjoint_byte_copy().
1491 //
1492 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1493 __ align(CodeEntryAlignment);
1494 StubCodeMark mark(this, "StubRoutines", name);
1495 address start = __ pc();
1496
1497 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1498 Label L_copy_byte, L_exit;
1499 const Register from = rdi; // source array address
1500 const Register to = rsi; // destination array address
1501 const Register count = rdx; // elements count
1502 const Register byte_count = rcx;
1503 const Register qword_count = count;
1504 const Register end_from = from; // source array end address
1505 const Register end_to = to; // destination array end address
1506 // End pointers are inclusive, and if count is not zero they point
1507 // to the last unit copied: end_to[0] := end_from[0]
1508
1509 __ enter(); // required for proper stackwalking of RuntimeStub frame
1510 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1511
1512 if (entry != NULL) {
1513 *entry = __ pc();
1514 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1515 BLOCK_COMMENT("Entry:");
1516 }
1517
1518 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1519 // r9 and r10 may be used to save non-volatile registers
1520
1521 // 'from', 'to' and 'count' are now valid
1522 __ movptr(byte_count, count);
1523 __ shrptr(count, 3); // count => qword_count
1524
1525 // Copy from low to high addresses. Use 'to' as scratch.
1526 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1527 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1528 __ negptr(qword_count); // make the count negative
1529 __ jmp(L_copy_bytes);
1530
1531 // Copy trailing qwords
1532 __ BIND(L_copy_8_bytes);
1533 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1534 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1535 __ increment(qword_count);
1536 __ jcc(Assembler::notZero, L_copy_8_bytes);
1537
1538 // Check for and copy trailing dword
1539 __ BIND(L_copy_4_bytes);
1540 __ testl(byte_count, 4);
1541 __ jccb(Assembler::zero, L_copy_2_bytes);
1542 __ movl(rax, Address(end_from, 8));
1543 __ movl(Address(end_to, 8), rax);
1544
1545 __ addptr(end_from, 4);
1546 __ addptr(end_to, 4);
1547
1548 // Check for and copy trailing word
1549 __ BIND(L_copy_2_bytes);
1550 __ testl(byte_count, 2);
1551 __ jccb(Assembler::zero, L_copy_byte);
1552 __ movw(rax, Address(end_from, 8));
1553 __ movw(Address(end_to, 8), rax);
1554
1555 __ addptr(end_from, 2);
1556 __ addptr(end_to, 2);
1557
1558 // Check for and copy trailing byte
1559 __ BIND(L_copy_byte);
1560 __ testl(byte_count, 1);
1561 __ jccb(Assembler::zero, L_exit);
1562 __ movb(rax, Address(end_from, 8));
1563 __ movb(Address(end_to, 8), rax);
1564
1565 __ BIND(L_exit);
1566 restore_arg_regs();
1567 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1568 __ xorptr(rax, rax); // return 0
1569 __ leave(); // required for proper stackwalking of RuntimeStub frame
1570 __ ret(0);
1571
1572 // Copy in multi-bytes chunks
1573 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1574 __ jmp(L_copy_4_bytes);
1575
1576 return start;
1577 }
1578
1579 // Arguments:
1580 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1581 // ignored
1582 // name - stub name string
1583 //
1584 // Inputs:
1585 // c_rarg0 - source array address
1586 // c_rarg1 - destination array address
1587 // c_rarg2 - element count, treated as ssize_t, can be zero
1588 //
1589 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1590 // we let the hardware handle it. The one to eight bytes within words,
1591 // dwords or qwords that span cache line boundaries will still be loaded
1592 // and stored atomically.
1593 //
1594 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1595 address* entry, const char *name) {
1596 __ align(CodeEntryAlignment);
1597 StubCodeMark mark(this, "StubRoutines", name);
1598 address start = __ pc();
1599
1600 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1601 const Register from = rdi; // source array address
1602 const Register to = rsi; // destination array address
1603 const Register count = rdx; // elements count
1604 const Register byte_count = rcx;
1605 const Register qword_count = count;
1606
1607 __ enter(); // required for proper stackwalking of RuntimeStub frame
1608 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1609
1610 if (entry != NULL) {
1611 *entry = __ pc();
1612 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1613 BLOCK_COMMENT("Entry:");
1614 }
1615
1616 array_overlap_test(nooverlap_target, Address::times_1);
1617 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1618 // r9 and r10 may be used to save non-volatile registers
1619
1620 // 'from', 'to' and 'count' are now valid
1621 __ movptr(byte_count, count);
1622 __ shrptr(count, 3); // count => qword_count
1623
1624 // Copy from high to low addresses.
1625
1626 // Check for and copy trailing byte
1627 __ testl(byte_count, 1);
1628 __ jcc(Assembler::zero, L_copy_2_bytes);
1629 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1630 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1631 __ decrement(byte_count); // Adjust for possible trailing word
1632
1633 // Check for and copy trailing word
1634 __ BIND(L_copy_2_bytes);
1635 __ testl(byte_count, 2);
1636 __ jcc(Assembler::zero, L_copy_4_bytes);
1637 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1638 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1639
1640 // Check for and copy trailing dword
1641 __ BIND(L_copy_4_bytes);
1642 __ testl(byte_count, 4);
1643 __ jcc(Assembler::zero, L_copy_bytes);
1644 __ movl(rax, Address(from, qword_count, Address::times_8));
1645 __ movl(Address(to, qword_count, Address::times_8), rax);
1646 __ jmp(L_copy_bytes);
1647
1648 // Copy trailing qwords
1649 __ BIND(L_copy_8_bytes);
1650 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1651 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1652 __ decrement(qword_count);
1653 __ jcc(Assembler::notZero, L_copy_8_bytes);
1654
1655 restore_arg_regs();
1656 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1657 __ xorptr(rax, rax); // return 0
1658 __ leave(); // required for proper stackwalking of RuntimeStub frame
1659 __ ret(0);
1660
1661 // Copy in multi-bytes chunks
1662 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1663
1664 restore_arg_regs();
1665 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1666 __ xorptr(rax, rax); // return 0
1667 __ leave(); // required for proper stackwalking of RuntimeStub frame
1668 __ ret(0);
1669
1670 return start;
1671 }
1672
1673 // Arguments:
1674 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1675 // ignored
1676 // name - stub name string
1677 //
1678 // Inputs:
1679 // c_rarg0 - source array address
1680 // c_rarg1 - destination array address
1681 // c_rarg2 - element count, treated as ssize_t, can be zero
1682 //
1683 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1684 // let the hardware handle it. The two or four words within dwords
1685 // or qwords that span cache line boundaries will still be loaded
1686 // and stored atomically.
1687 //
1688 // Side Effects:
1689 // disjoint_short_copy_entry is set to the no-overlap entry point
1690 // used by generate_conjoint_short_copy().
1691 //
1692 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1693 __ align(CodeEntryAlignment);
1694 StubCodeMark mark(this, "StubRoutines", name);
1695 address start = __ pc();
1696
1697 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1698 const Register from = rdi; // source array address
1699 const Register to = rsi; // destination array address
1700 const Register count = rdx; // elements count
1701 const Register word_count = rcx;
1702 const Register qword_count = count;
1703 const Register end_from = from; // source array end address
1704 const Register end_to = to; // destination array end address
1705 // End pointers are inclusive, and if count is not zero they point
1706 // to the last unit copied: end_to[0] := end_from[0]
1707
1708 __ enter(); // required for proper stackwalking of RuntimeStub frame
1709 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1710
1711 if (entry != NULL) {
1712 *entry = __ pc();
1713 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1714 BLOCK_COMMENT("Entry:");
1715 }
1716
1717 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1718 // r9 and r10 may be used to save non-volatile registers
1719
1720 // 'from', 'to' and 'count' are now valid
1721 __ movptr(word_count, count);
1722 __ shrptr(count, 2); // count => qword_count
1723
1724 // Copy from low to high addresses. Use 'to' as scratch.
1725 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1726 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1727 __ negptr(qword_count);
1728 __ jmp(L_copy_bytes);
1729
1730 // Copy trailing qwords
1731 __ BIND(L_copy_8_bytes);
1732 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1733 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1734 __ increment(qword_count);
1735 __ jcc(Assembler::notZero, L_copy_8_bytes);
1736
1737 // Original 'dest' is trashed, so we can't use it as a
1738 // base register for a possible trailing word copy
1739
1740 // Check for and copy trailing dword
1741 __ BIND(L_copy_4_bytes);
1742 __ testl(word_count, 2);
1743 __ jccb(Assembler::zero, L_copy_2_bytes);
1744 __ movl(rax, Address(end_from, 8));
1745 __ movl(Address(end_to, 8), rax);
1746
1747 __ addptr(end_from, 4);
1748 __ addptr(end_to, 4);
1749
1750 // Check for and copy trailing word
1751 __ BIND(L_copy_2_bytes);
1752 __ testl(word_count, 1);
1753 __ jccb(Assembler::zero, L_exit);
1754 __ movw(rax, Address(end_from, 8));
1755 __ movw(Address(end_to, 8), rax);
1756
1757 __ BIND(L_exit);
1758 restore_arg_regs();
1759 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1760 __ xorptr(rax, rax); // return 0
1761 __ leave(); // required for proper stackwalking of RuntimeStub frame
1762 __ ret(0);
1763
1764 // Copy in multi-bytes chunks
1765 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1766 __ jmp(L_copy_4_bytes);
1767
1768 return start;
1769 }
1770
1771 address generate_fill(BasicType t, bool aligned, const char *name) {
1772 __ align(CodeEntryAlignment);
1773 StubCodeMark mark(this, "StubRoutines", name);
1774 address start = __ pc();
1775
1776 BLOCK_COMMENT("Entry:");
1777
1778 const Register to = c_rarg0; // source array address
1779 const Register value = c_rarg1; // value
1780 const Register count = c_rarg2; // elements count
1781
1782 __ enter(); // required for proper stackwalking of RuntimeStub frame
1783
1784 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1785
1786 __ leave(); // required for proper stackwalking of RuntimeStub frame
1787 __ ret(0);
1788 return start;
1789 }
1790
1791 // Arguments:
1792 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1793 // ignored
1794 // name - stub name string
1795 //
1796 // Inputs:
1797 // c_rarg0 - source array address
1798 // c_rarg1 - destination array address
1799 // c_rarg2 - element count, treated as ssize_t, can be zero
1800 //
1801 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1802 // let the hardware handle it. The two or four words within dwords
1803 // or qwords that span cache line boundaries will still be loaded
1804 // and stored atomically.
1805 //
1806 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1807 address *entry, const char *name) {
1808 __ align(CodeEntryAlignment);
1809 StubCodeMark mark(this, "StubRoutines", name);
1810 address start = __ pc();
1811
1812 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1813 const Register from = rdi; // source array address
1814 const Register to = rsi; // destination array address
1815 const Register count = rdx; // elements count
1816 const Register word_count = rcx;
1817 const Register qword_count = count;
1818
1819 __ enter(); // required for proper stackwalking of RuntimeStub frame
1820 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1821
1822 if (entry != NULL) {
1823 *entry = __ pc();
1824 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1825 BLOCK_COMMENT("Entry:");
1826 }
1827
1828 array_overlap_test(nooverlap_target, Address::times_2);
1829 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1830 // r9 and r10 may be used to save non-volatile registers
1831
1832 // 'from', 'to' and 'count' are now valid
1833 __ movptr(word_count, count);
1834 __ shrptr(count, 2); // count => qword_count
1835
1836 // Copy from high to low addresses. Use 'to' as scratch.
1837
1838 // Check for and copy trailing word
1839 __ testl(word_count, 1);
1840 __ jccb(Assembler::zero, L_copy_4_bytes);
1841 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1842 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1843
1844 // Check for and copy trailing dword
1845 __ BIND(L_copy_4_bytes);
1846 __ testl(word_count, 2);
1847 __ jcc(Assembler::zero, L_copy_bytes);
1848 __ movl(rax, Address(from, qword_count, Address::times_8));
1849 __ movl(Address(to, qword_count, Address::times_8), rax);
1850 __ jmp(L_copy_bytes);
1851
1852 // Copy trailing qwords
1853 __ BIND(L_copy_8_bytes);
1854 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1855 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1856 __ decrement(qword_count);
1857 __ jcc(Assembler::notZero, L_copy_8_bytes);
1858
1859 restore_arg_regs();
1860 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1861 __ xorptr(rax, rax); // return 0
1862 __ leave(); // required for proper stackwalking of RuntimeStub frame
1863 __ ret(0);
1864
1865 // Copy in multi-bytes chunks
1866 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1867
1868 restore_arg_regs();
1869 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1870 __ xorptr(rax, rax); // return 0
1871 __ leave(); // required for proper stackwalking of RuntimeStub frame
1872 __ ret(0);
1873
1874 return start;
1875 }
1876
1877 // Arguments:
1878 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1879 // ignored
1880 // is_oop - true => oop array, so generate store check code
1881 // name - stub name string
1882 //
1883 // Inputs:
1884 // c_rarg0 - source array address
1885 // c_rarg1 - destination array address
1886 // c_rarg2 - element count, treated as ssize_t, can be zero
1887 //
1888 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1889 // the hardware handle it. The two dwords within qwords that span
1890 // cache line boundaries will still be loaded and stored atomicly.
1891 //
1892 // Side Effects:
1893 // disjoint_int_copy_entry is set to the no-overlap entry point
1894 // used by generate_conjoint_int_oop_copy().
1895 //
1896 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1897 const char *name, bool dest_uninitialized = false) {
1898 __ align(CodeEntryAlignment);
1899 StubCodeMark mark(this, "StubRoutines", name);
1900 address start = __ pc();
1901
1902 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1903 const Register from = rdi; // source array address
1904 const Register to = rsi; // destination array address
1905 const Register count = rdx; // elements count
1906 const Register dword_count = rcx;
1907 const Register qword_count = count;
1908 const Register end_from = from; // source array end address
1909 const Register end_to = to; // destination array end address
1910 const Register saved_to = r11; // saved destination array address
1911 // End pointers are inclusive, and if count is not zero they point
1912 // to the last unit copied: end_to[0] := end_from[0]
1913
1914 __ enter(); // required for proper stackwalking of RuntimeStub frame
1915 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1916
1917 if (entry != NULL) {
1918 *entry = __ pc();
1919 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1920 BLOCK_COMMENT("Entry:");
1921 }
1922
1923 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1924 // r9 and r10 may be used to save non-volatile registers
1925 if (is_oop) {
1926 __ movq(saved_to, to);
1927 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1928 }
1929
1930 // 'from', 'to' and 'count' are now valid
1931 __ movptr(dword_count, count);
1932 __ shrptr(count, 1); // count => qword_count
1933
1934 // Copy from low to high addresses. Use 'to' as scratch.
1935 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1936 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1937 __ negptr(qword_count);
1938 __ jmp(L_copy_bytes);
1939
1940 // Copy trailing qwords
1941 __ BIND(L_copy_8_bytes);
1942 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1943 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1944 __ increment(qword_count);
1945 __ jcc(Assembler::notZero, L_copy_8_bytes);
1946
1947 // Check for and copy trailing dword
1948 __ BIND(L_copy_4_bytes);
1949 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1950 __ jccb(Assembler::zero, L_exit);
1951 __ movl(rax, Address(end_from, 8));
1952 __ movl(Address(end_to, 8), rax);
1953
1954 __ BIND(L_exit);
1955 if (is_oop) {
1956 gen_write_ref_array_post_barrier(saved_to, dword_count, rax);
1957 }
1958 restore_arg_regs();
1959 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1960 __ xorptr(rax, rax); // return 0
1961 __ leave(); // required for proper stackwalking of RuntimeStub frame
1962 __ ret(0);
1963
1964 // Copy in multi-bytes chunks
1965 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1966 __ jmp(L_copy_4_bytes);
1967
1968 return start;
1969 }
1970
1971 // Arguments:
1972 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1973 // ignored
1974 // is_oop - true => oop array, so generate store check code
1975 // name - stub name string
1976 //
1977 // Inputs:
1978 // c_rarg0 - source array address
1979 // c_rarg1 - destination array address
1980 // c_rarg2 - element count, treated as ssize_t, can be zero
1981 //
1982 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1983 // the hardware handle it. The two dwords within qwords that span
1984 // cache line boundaries will still be loaded and stored atomicly.
1985 //
1986 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1987 address *entry, const char *name,
1988 bool dest_uninitialized = false) {
1989 __ align(CodeEntryAlignment);
1990 StubCodeMark mark(this, "StubRoutines", name);
1991 address start = __ pc();
1992
1993 Label L_copy_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1994 const Register from = rdi; // source array address
1995 const Register to = rsi; // destination array address
1996 const Register count = rdx; // elements count
1997 const Register dword_count = rcx;
1998 const Register qword_count = count;
1999
2000 __ enter(); // required for proper stackwalking of RuntimeStub frame
2001 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2002
2003 if (entry != NULL) {
2004 *entry = __ pc();
2005 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2006 BLOCK_COMMENT("Entry:");
2007 }
2008
2009 array_overlap_test(nooverlap_target, Address::times_4);
2010 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2011 // r9 and r10 may be used to save non-volatile registers
2012
2013 if (is_oop) {
2014 // no registers are destroyed by this call
2015 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2016 }
2017
2018 assert_clean_int(count, rax); // Make sure 'count' is clean int.
2019 // 'from', 'to' and 'count' are now valid
2020 __ movptr(dword_count, count);
2021 __ shrptr(count, 1); // count => qword_count
2022
2023 // Copy from high to low addresses. Use 'to' as scratch.
2024
2025 // Check for and copy trailing dword
2026 __ testl(dword_count, 1);
2027 __ jcc(Assembler::zero, L_copy_bytes);
2028 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
2029 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
2030 __ jmp(L_copy_bytes);
2031
2032 // Copy trailing qwords
2033 __ BIND(L_copy_8_bytes);
2034 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2035 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2036 __ decrement(qword_count);
2037 __ jcc(Assembler::notZero, L_copy_8_bytes);
2038
2039 if (is_oop) {
2040 __ jmp(L_exit);
2041 }
2042 restore_arg_regs();
2043 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2044 __ xorptr(rax, rax); // return 0
2045 __ leave(); // required for proper stackwalking of RuntimeStub frame
2046 __ ret(0);
2047
2048 // Copy in multi-bytes chunks
2049 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2050
2051 __ BIND(L_exit);
2052 if (is_oop) {
2053 gen_write_ref_array_post_barrier(to, dword_count, rax);
2054 }
2055 restore_arg_regs();
2056 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2057 __ xorptr(rax, rax); // return 0
2058 __ leave(); // required for proper stackwalking of RuntimeStub frame
2059 __ ret(0);
2060
2061 return start;
2062 }
2063
2064 // Arguments:
2065 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2066 // ignored
2067 // is_oop - true => oop array, so generate store check code
2068 // name - stub name string
2069 //
2070 // Inputs:
2071 // c_rarg0 - source array address
2072 // c_rarg1 - destination array address
2073 // c_rarg2 - element count, treated as ssize_t, can be zero
2074 //
2075 // Side Effects:
2076 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2077 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2078 //
2079 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2080 const char *name, bool dest_uninitialized = false) {
2081 __ align(CodeEntryAlignment);
2082 StubCodeMark mark(this, "StubRoutines", name);
2083 address start = __ pc();
2084
2085 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2086 const Register from = rdi; // source array address
2087 const Register to = rsi; // destination array address
2088 const Register qword_count = rdx; // elements count
2089 const Register end_from = from; // source array end address
2090 const Register end_to = rcx; // destination array end address
2091 const Register saved_to = to;
2092 const Register saved_count = r11;
2093 // End pointers are inclusive, and if count is not zero they point
2094 // to the last unit copied: end_to[0] := end_from[0]
2095
2096 __ enter(); // required for proper stackwalking of RuntimeStub frame
2097 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2098 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2099
2100 if (entry != NULL) {
2101 *entry = __ pc();
2102 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2103 BLOCK_COMMENT("Entry:");
2104 }
2105
2106 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2107 // r9 and r10 may be used to save non-volatile registers
2108 // 'from', 'to' and 'qword_count' are now valid
2109 if (is_oop) {
2110 // Save to and count for store barrier
2111 __ movptr(saved_count, qword_count);
2112 // no registers are destroyed by this call
2113 gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized);
2114 }
2115
2116 // Copy from low to high addresses. Use 'to' as scratch.
2117 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2118 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2119 __ negptr(qword_count);
2120 __ jmp(L_copy_bytes);
2121
2122 // Copy trailing qwords
2123 __ BIND(L_copy_8_bytes);
2124 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2125 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2126 __ increment(qword_count);
2127 __ jcc(Assembler::notZero, L_copy_8_bytes);
2128
2129 if (is_oop) {
2130 __ jmp(L_exit);
2131 } else {
2132 restore_arg_regs();
2133 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2134 __ xorptr(rax, rax); // return 0
2135 __ leave(); // required for proper stackwalking of RuntimeStub frame
2136 __ ret(0);
2137 }
2138
2139 // Copy in multi-bytes chunks
2140 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2141
2142 if (is_oop) {
2143 __ BIND(L_exit);
2144 gen_write_ref_array_post_barrier(saved_to, saved_count, rax);
2145 }
2146 restore_arg_regs();
2147 if (is_oop) {
2148 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2149 } else {
2150 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2151 }
2152 __ xorptr(rax, rax); // return 0
2153 __ leave(); // required for proper stackwalking of RuntimeStub frame
2154 __ ret(0);
2155
2156 return start;
2157 }
2158
2159 // Arguments:
2160 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2161 // ignored
2162 // is_oop - true => oop array, so generate store check code
2163 // name - stub name string
2164 //
2165 // Inputs:
2166 // c_rarg0 - source array address
2167 // c_rarg1 - destination array address
2168 // c_rarg2 - element count, treated as ssize_t, can be zero
2169 //
2170 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2171 address nooverlap_target, address *entry,
2172 const char *name, bool dest_uninitialized = false) {
2173 __ align(CodeEntryAlignment);
2174 StubCodeMark mark(this, "StubRoutines", name);
2175 address start = __ pc();
2176
2177 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2178 const Register from = rdi; // source array address
2179 const Register to = rsi; // destination array address
2180 const Register qword_count = rdx; // elements count
2181 const Register saved_count = rcx;
2182
2183 __ enter(); // required for proper stackwalking of RuntimeStub frame
2184 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2185
2186 if (entry != NULL) {
2187 *entry = __ pc();
2188 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2189 BLOCK_COMMENT("Entry:");
2190 }
2191
2192 array_overlap_test(nooverlap_target, Address::times_8);
2193 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2194 // r9 and r10 may be used to save non-volatile registers
2195 // 'from', 'to' and 'qword_count' are now valid
2196 if (is_oop) {
2197 // Save to and count for store barrier
2198 __ movptr(saved_count, qword_count);
2199 // No registers are destroyed by this call
2200 gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized);
2201 }
2202
2203 __ jmp(L_copy_bytes);
2204
2205 // Copy trailing qwords
2206 __ BIND(L_copy_8_bytes);
2207 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2208 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2209 __ decrement(qword_count);
2210 __ jcc(Assembler::notZero, L_copy_8_bytes);
2211
2212 if (is_oop) {
2213 __ jmp(L_exit);
2214 } else {
2215 restore_arg_regs();
2216 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2217 __ xorptr(rax, rax); // return 0
2218 __ leave(); // required for proper stackwalking of RuntimeStub frame
2219 __ ret(0);
2220 }
2221
2222 // Copy in multi-bytes chunks
2223 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2224
2225 if (is_oop) {
2226 __ BIND(L_exit);
2227 gen_write_ref_array_post_barrier(to, saved_count, rax);
2228 }
2229 restore_arg_regs();
2230 if (is_oop) {
2231 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2232 } else {
2233 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2234 }
2235 __ xorptr(rax, rax); // return 0
2236 __ leave(); // required for proper stackwalking of RuntimeStub frame
2237 __ ret(0);
2238
2239 return start;
2240 }
2241
2242
2243 // Helper for generating a dynamic type check.
2244 // Smashes no registers.
2245 void generate_type_check(Register sub_klass,
2246 Register super_check_offset,
2247 Register super_klass,
2248 Label& L_success) {
2249 assert_different_registers(sub_klass, super_check_offset, super_klass);
2250
2251 BLOCK_COMMENT("type_check:");
2252
2253 Label L_miss;
2254
2255 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2256 super_check_offset);
2257 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2258
2259 // Fall through on failure!
2260 __ BIND(L_miss);
2261 }
2262
2263 //
2264 // Generate checkcasting array copy stub
2265 //
2266 // Input:
2267 // c_rarg0 - source array address
2268 // c_rarg1 - destination array address
2269 // c_rarg2 - element count, treated as ssize_t, can be zero
2270 // c_rarg3 - size_t ckoff (super_check_offset)
2271 // not Win64
2272 // c_rarg4 - oop ckval (super_klass)
2273 // Win64
2274 // rsp+40 - oop ckval (super_klass)
2275 //
2276 // Output:
2277 // rax == 0 - success
2278 // rax == -1^K - failure, where K is partial transfer count
2279 //
2280 address generate_checkcast_copy(const char *name, address *entry,
2281 bool dest_uninitialized = false) {
2282
2283 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2284
2285 // Input registers (after setup_arg_regs)
2286 const Register from = rdi; // source array address
2287 const Register to = rsi; // destination array address
2288 const Register length = rdx; // elements count
2289 const Register ckoff = rcx; // super_check_offset
2290 const Register ckval = r8; // super_klass
2291
2292 // Registers used as temps (r13, r14 are save-on-entry)
2293 const Register end_from = from; // source array end address
2294 const Register end_to = r13; // destination array end address
2295 const Register count = rdx; // -(count_remaining)
2296 const Register r14_length = r14; // saved copy of length
2297 // End pointers are inclusive, and if length is not zero they point
2298 // to the last unit copied: end_to[0] := end_from[0]
2299
2300 const Register rax_oop = rax; // actual oop copied
2301 const Register r11_klass = r11; // oop._klass
2302
2303 //---------------------------------------------------------------
2304 // Assembler stub will be used for this call to arraycopy
2305 // if the two arrays are subtypes of Object[] but the
2306 // destination array type is not equal to or a supertype
2307 // of the source type. Each element must be separately
2308 // checked.
2309
2310 __ align(CodeEntryAlignment);
2311 StubCodeMark mark(this, "StubRoutines", name);
2312 address start = __ pc();
2313
2314 __ enter(); // required for proper stackwalking of RuntimeStub frame
2315
2316 #ifdef ASSERT
2317 // caller guarantees that the arrays really are different
2318 // otherwise, we would have to make conjoint checks
2319 { Label L;
2320 array_overlap_test(L, TIMES_OOP);
2321 __ stop("checkcast_copy within a single array");
2322 __ bind(L);
2323 }
2324 #endif //ASSERT
2325
2326 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2327 // ckoff => rcx, ckval => r8
2328 // r9 and r10 may be used to save non-volatile registers
2329 #ifdef _WIN64
2330 // last argument (#4) is on stack on Win64
2331 __ movptr(ckval, Address(rsp, 6 * wordSize));
2332 #endif
2333
2334 // Caller of this entry point must set up the argument registers.
2335 if (entry != NULL) {
2336 *entry = __ pc();
2337 BLOCK_COMMENT("Entry:");
2338 }
2339
2340 // allocate spill slots for r13, r14
2341 enum {
2342 saved_r13_offset,
2343 saved_r14_offset,
2344 saved_rbp_offset
2345 };
2346 __ subptr(rsp, saved_rbp_offset * wordSize);
2347 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2348 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2349
2350 // check that int operands are properly extended to size_t
2351 assert_clean_int(length, rax);
2352 assert_clean_int(ckoff, rax);
2353
2354 #ifdef ASSERT
2355 BLOCK_COMMENT("assert consistent ckoff/ckval");
2356 // The ckoff and ckval must be mutually consistent,
2357 // even though caller generates both.
2358 { Label L;
2359 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2360 __ cmpl(ckoff, Address(ckval, sco_offset));
2361 __ jcc(Assembler::equal, L);
2362 __ stop("super_check_offset inconsistent");
2363 __ bind(L);
2364 }
2365 #endif //ASSERT
2366
2367 // Loop-invariant addresses. They are exclusive end pointers.
2368 Address end_from_addr(from, length, TIMES_OOP, 0);
2369 Address end_to_addr(to, length, TIMES_OOP, 0);
2370 // Loop-variant addresses. They assume post-incremented count < 0.
2371 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2372 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2373
2374 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2375
2376 // Copy from low to high addresses, indexed from the end of each array.
2377 __ lea(end_from, end_from_addr);
2378 __ lea(end_to, end_to_addr);
2379 __ movptr(r14_length, length); // save a copy of the length
2380 assert(length == count, ""); // else fix next line:
2381 __ negptr(count); // negate and test the length
2382 __ jcc(Assembler::notZero, L_load_element);
2383
2384 // Empty array: Nothing to do.
2385 __ xorptr(rax, rax); // return 0 on (trivial) success
2386 __ jmp(L_done);
2387
2388 // ======== begin loop ========
2389 // (Loop is rotated; its entry is L_load_element.)
2390 // Loop control:
2391 // for (count = -count; count != 0; count++)
2392 // Base pointers src, dst are biased by 8*(count-1),to last element.
2393 __ align(OptoLoopAlignment);
2394
2395 __ BIND(L_store_element);
2396 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2397 __ increment(count); // increment the count toward zero
2398 __ jcc(Assembler::zero, L_do_card_marks);
2399
2400 // ======== loop entry is here ========
2401 __ BIND(L_load_element);
2402 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2403 __ testptr(rax_oop, rax_oop);
2404 __ jcc(Assembler::zero, L_store_element);
2405
2406 __ load_klass(r11_klass, rax_oop);// query the object klass
2407 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2408 // ======== end loop ========
2409
2410 // It was a real error; we must depend on the caller to finish the job.
2411 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2412 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2413 // and report their number to the caller.
2414 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2415 Label L_post_barrier;
2416 __ addptr(r14_length, count); // K = (original - remaining) oops
2417 __ movptr(rax, r14_length); // save the value
2418 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2419 __ jccb(Assembler::notZero, L_post_barrier);
2420 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2421
2422 // Come here on success only.
2423 __ BIND(L_do_card_marks);
2424 __ xorptr(rax, rax); // return 0 on success
2425
2426 __ BIND(L_post_barrier);
2427 gen_write_ref_array_post_barrier(to, r14_length, rscratch1);
2428
2429 // Common exit point (success or failure).
2430 __ BIND(L_done);
2431 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2432 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2433 restore_arg_regs();
2434 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2435 __ leave(); // required for proper stackwalking of RuntimeStub frame
2436 __ ret(0);
2437
2438 return start;
2439 }
2440
2441 //
2442 // Generate 'unsafe' array copy stub
2443 // Though just as safe as the other stubs, it takes an unscaled
2444 // size_t argument instead of an element count.
2445 //
2446 // Input:
2447 // c_rarg0 - source array address
2448 // c_rarg1 - destination array address
2449 // c_rarg2 - byte count, treated as ssize_t, can be zero
2450 //
2451 // Examines the alignment of the operands and dispatches
2452 // to a long, int, short, or byte copy loop.
2453 //
2454 address generate_unsafe_copy(const char *name,
2455 address byte_copy_entry, address short_copy_entry,
2456 address int_copy_entry, address long_copy_entry) {
2457
2458 Label L_long_aligned, L_int_aligned, L_short_aligned;
2459
2460 // Input registers (before setup_arg_regs)
2461 const Register from = c_rarg0; // source array address
2462 const Register to = c_rarg1; // destination array address
2463 const Register size = c_rarg2; // byte count (size_t)
2464
2465 // Register used as a temp
2466 const Register bits = rax; // test copy of low bits
2467
2468 __ align(CodeEntryAlignment);
2469 StubCodeMark mark(this, "StubRoutines", name);
2470 address start = __ pc();
2471
2472 __ enter(); // required for proper stackwalking of RuntimeStub frame
2473
2474 // bump this on entry, not on exit:
2475 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2476
2477 __ mov(bits, from);
2478 __ orptr(bits, to);
2479 __ orptr(bits, size);
2480
2481 __ testb(bits, BytesPerLong-1);
2482 __ jccb(Assembler::zero, L_long_aligned);
2483
2484 __ testb(bits, BytesPerInt-1);
2485 __ jccb(Assembler::zero, L_int_aligned);
2486
2487 __ testb(bits, BytesPerShort-1);
2488 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2489
2490 __ BIND(L_short_aligned);
2491 __ shrptr(size, LogBytesPerShort); // size => short_count
2492 __ jump(RuntimeAddress(short_copy_entry));
2493
2494 __ BIND(L_int_aligned);
2495 __ shrptr(size, LogBytesPerInt); // size => int_count
2496 __ jump(RuntimeAddress(int_copy_entry));
2497
2498 __ BIND(L_long_aligned);
2499 __ shrptr(size, LogBytesPerLong); // size => qword_count
2500 __ jump(RuntimeAddress(long_copy_entry));
2501
2502 return start;
2503 }
2504
2505 // Perform range checks on the proposed arraycopy.
2506 // Kills temp, but nothing else.
2507 // Also, clean the sign bits of src_pos and dst_pos.
2508 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2509 Register src_pos, // source position (c_rarg1)
2510 Register dst, // destination array oo (c_rarg2)
2511 Register dst_pos, // destination position (c_rarg3)
2512 Register length,
2513 Register temp,
2514 Label& L_failed) {
2515 BLOCK_COMMENT("arraycopy_range_checks:");
2516
2517 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2518 __ movl(temp, length);
2519 __ addl(temp, src_pos); // src_pos + length
2520 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2521 __ jcc(Assembler::above, L_failed);
2522
2523 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2524 __ movl(temp, length);
2525 __ addl(temp, dst_pos); // dst_pos + length
2526 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2527 __ jcc(Assembler::above, L_failed);
2528
2529 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2530 // Move with sign extension can be used since they are positive.
2531 __ movslq(src_pos, src_pos);
2532 __ movslq(dst_pos, dst_pos);
2533
2534 BLOCK_COMMENT("arraycopy_range_checks done");
2535 }
2536
2537 //
2538 // Generate generic array copy stubs
2539 //
2540 // Input:
2541 // c_rarg0 - src oop
2542 // c_rarg1 - src_pos (32-bits)
2543 // c_rarg2 - dst oop
2544 // c_rarg3 - dst_pos (32-bits)
2545 // not Win64
2546 // c_rarg4 - element count (32-bits)
2547 // Win64
2548 // rsp+40 - element count (32-bits)
2549 //
2550 // Output:
2551 // rax == 0 - success
2552 // rax == -1^K - failure, where K is partial transfer count
2553 //
2554 address generate_generic_copy(const char *name,
2555 address byte_copy_entry, address short_copy_entry,
2556 address int_copy_entry, address oop_copy_entry,
2557 address long_copy_entry, address checkcast_copy_entry) {
2558
2559 Label L_failed, L_failed_0, L_objArray;
2560 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2561
2562 // Input registers
2563 const Register src = c_rarg0; // source array oop
2564 const Register src_pos = c_rarg1; // source position
2565 const Register dst = c_rarg2; // destination array oop
2566 const Register dst_pos = c_rarg3; // destination position
2567 #ifndef _WIN64
2568 const Register length = c_rarg4;
2569 #else
2570 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2571 #endif
2572
2573 { int modulus = CodeEntryAlignment;
2574 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2575 int advance = target - (__ offset() % modulus);
2576 if (advance < 0) advance += modulus;
2577 if (advance > 0) __ nop(advance);
2578 }
2579 StubCodeMark mark(this, "StubRoutines", name);
2580
2581 // Short-hop target to L_failed. Makes for denser prologue code.
2582 __ BIND(L_failed_0);
2583 __ jmp(L_failed);
2584 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2585
2586 __ align(CodeEntryAlignment);
2587 address start = __ pc();
2588
2589 __ enter(); // required for proper stackwalking of RuntimeStub frame
2590
2591 // bump this on entry, not on exit:
2592 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2593
2594 //-----------------------------------------------------------------------
2595 // Assembler stub will be used for this call to arraycopy
2596 // if the following conditions are met:
2597 //
2598 // (1) src and dst must not be null.
2599 // (2) src_pos must not be negative.
2600 // (3) dst_pos must not be negative.
2601 // (4) length must not be negative.
2602 // (5) src klass and dst klass should be the same and not NULL.
2603 // (6) src and dst should be arrays.
2604 // (7) src_pos + length must not exceed length of src.
2605 // (8) dst_pos + length must not exceed length of dst.
2606 //
2607
2608 // if (src == NULL) return -1;
2609 __ testptr(src, src); // src oop
2610 size_t j1off = __ offset();
2611 __ jccb(Assembler::zero, L_failed_0);
2612
2613 // if (src_pos < 0) return -1;
2614 __ testl(src_pos, src_pos); // src_pos (32-bits)
2615 __ jccb(Assembler::negative, L_failed_0);
2616
2617 // if (dst == NULL) return -1;
2618 __ testptr(dst, dst); // dst oop
2619 __ jccb(Assembler::zero, L_failed_0);
2620
2621 // if (dst_pos < 0) return -1;
2622 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2623 size_t j4off = __ offset();
2624 __ jccb(Assembler::negative, L_failed_0);
2625
2626 // The first four tests are very dense code,
2627 // but not quite dense enough to put four
2628 // jumps in a 16-byte instruction fetch buffer.
2629 // That's good, because some branch predicters
2630 // do not like jumps so close together.
2631 // Make sure of this.
2632 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2633
2634 // registers used as temp
2635 const Register r11_length = r11; // elements count to copy
2636 const Register r10_src_klass = r10; // array klass
2637
2638 // if (length < 0) return -1;
2639 __ movl(r11_length, length); // length (elements count, 32-bits value)
2640 __ testl(r11_length, r11_length);
2641 __ jccb(Assembler::negative, L_failed_0);
2642
2643 __ load_klass(r10_src_klass, src);
2644 #ifdef ASSERT
2645 // assert(src->klass() != NULL);
2646 {
2647 BLOCK_COMMENT("assert klasses not null {");
2648 Label L1, L2;
2649 __ testptr(r10_src_klass, r10_src_klass);
2650 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2651 __ bind(L1);
2652 __ stop("broken null klass");
2653 __ bind(L2);
2654 __ load_klass(rax, dst);
2655 __ cmpq(rax, 0);
2656 __ jcc(Assembler::equal, L1); // this would be broken also
2657 BLOCK_COMMENT("} assert klasses not null done");
2658 }
2659 #endif
2660
2661 // Load layout helper (32-bits)
2662 //
2663 // |array_tag| | header_size | element_type | |log2_element_size|
2664 // 32 30 24 16 8 2 0
2665 //
2666 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2667 //
2668
2669 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2670
2671 // Handle objArrays completely differently...
2672 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2673 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2674 __ jcc(Assembler::equal, L_objArray);
2675
2676 // if (src->klass() != dst->klass()) return -1;
2677 __ load_klass(rax, dst);
2678 __ cmpq(r10_src_klass, rax);
2679 __ jcc(Assembler::notEqual, L_failed);
2680
2681 const Register rax_lh = rax; // layout helper
2682 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2683
2684 // if (!src->is_Array()) return -1;
2685 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2686 __ jcc(Assembler::greaterEqual, L_failed);
2687
2688 // At this point, it is known to be a typeArray (array_tag 0x3).
2689 #ifdef ASSERT
2690 {
2691 BLOCK_COMMENT("assert primitive array {");
2692 Label L;
2693 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2694 __ jcc(Assembler::greaterEqual, L);
2695 __ stop("must be a primitive array");
2696 __ bind(L);
2697 BLOCK_COMMENT("} assert primitive array done");
2698 }
2699 #endif
2700
2701 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2702 r10, L_failed);
2703
2704 // TypeArrayKlass
2705 //
2706 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2707 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2708 //
2709
2710 const Register r10_offset = r10; // array offset
2711 const Register rax_elsize = rax_lh; // element size
2712
2713 __ movl(r10_offset, rax_lh);
2714 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2715 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2716 __ addptr(src, r10_offset); // src array offset
2717 __ addptr(dst, r10_offset); // dst array offset
2718 BLOCK_COMMENT("choose copy loop based on element size");
2719 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2720
2721 // next registers should be set before the jump to corresponding stub
2722 const Register from = c_rarg0; // source array address
2723 const Register to = c_rarg1; // destination array address
2724 const Register count = c_rarg2; // elements count
2725
2726 // 'from', 'to', 'count' registers should be set in such order
2727 // since they are the same as 'src', 'src_pos', 'dst'.
2728
2729 __ BIND(L_copy_bytes);
2730 __ cmpl(rax_elsize, 0);
2731 __ jccb(Assembler::notEqual, L_copy_shorts);
2732 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2733 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2734 __ movl2ptr(count, r11_length); // length
2735 __ jump(RuntimeAddress(byte_copy_entry));
2736
2737 __ BIND(L_copy_shorts);
2738 __ cmpl(rax_elsize, LogBytesPerShort);
2739 __ jccb(Assembler::notEqual, L_copy_ints);
2740 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2741 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2742 __ movl2ptr(count, r11_length); // length
2743 __ jump(RuntimeAddress(short_copy_entry));
2744
2745 __ BIND(L_copy_ints);
2746 __ cmpl(rax_elsize, LogBytesPerInt);
2747 __ jccb(Assembler::notEqual, L_copy_longs);
2748 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2749 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2750 __ movl2ptr(count, r11_length); // length
2751 __ jump(RuntimeAddress(int_copy_entry));
2752
2753 __ BIND(L_copy_longs);
2754 #ifdef ASSERT
2755 {
2756 BLOCK_COMMENT("assert long copy {");
2757 Label L;
2758 __ cmpl(rax_elsize, LogBytesPerLong);
2759 __ jcc(Assembler::equal, L);
2760 __ stop("must be long copy, but elsize is wrong");
2761 __ bind(L);
2762 BLOCK_COMMENT("} assert long copy done");
2763 }
2764 #endif
2765 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2766 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2767 __ movl2ptr(count, r11_length); // length
2768 __ jump(RuntimeAddress(long_copy_entry));
2769
2770 // ObjArrayKlass
2771 __ BIND(L_objArray);
2772 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2773
2774 Label L_plain_copy, L_checkcast_copy;
2775 // test array classes for subtyping
2776 __ load_klass(rax, dst);
2777 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2778 __ jcc(Assembler::notEqual, L_checkcast_copy);
2779
2780 // Identically typed arrays can be copied without element-wise checks.
2781 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2782 r10, L_failed);
2783
2784 __ lea(from, Address(src, src_pos, TIMES_OOP,
2785 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2786 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2787 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2788 __ movl2ptr(count, r11_length); // length
2789 __ BIND(L_plain_copy);
2790 __ jump(RuntimeAddress(oop_copy_entry));
2791
2792 __ BIND(L_checkcast_copy);
2793 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2794 {
2795 // Before looking at dst.length, make sure dst is also an objArray.
2796 __ cmpl(Address(rax, lh_offset), objArray_lh);
2797 __ jcc(Assembler::notEqual, L_failed);
2798
2799 // It is safe to examine both src.length and dst.length.
2800 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2801 rax, L_failed);
2802
2803 const Register r11_dst_klass = r11;
2804 __ load_klass(r11_dst_klass, dst); // reload
2805
2806 // Marshal the base address arguments now, freeing registers.
2807 __ lea(from, Address(src, src_pos, TIMES_OOP,
2808 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2809 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2810 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2811 __ movl(count, length); // length (reloaded)
2812 Register sco_temp = c_rarg3; // this register is free now
2813 assert_different_registers(from, to, count, sco_temp,
2814 r11_dst_klass, r10_src_klass);
2815 assert_clean_int(count, sco_temp);
2816
2817 // Generate the type check.
2818 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2819 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2820 assert_clean_int(sco_temp, rax);
2821 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2822
2823 // Fetch destination element klass from the ObjArrayKlass header.
2824 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2825 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2826 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2827 assert_clean_int(sco_temp, rax);
2828
2829 // the checkcast_copy loop needs two extra arguments:
2830 assert(c_rarg3 == sco_temp, "#3 already in place");
2831 // Set up arguments for checkcast_copy_entry.
2832 setup_arg_regs(4);
2833 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2834 __ jump(RuntimeAddress(checkcast_copy_entry));
2835 }
2836
2837 __ BIND(L_failed);
2838 __ xorptr(rax, rax);
2839 __ notptr(rax); // return -1
2840 __ leave(); // required for proper stackwalking of RuntimeStub frame
2841 __ ret(0);
2842
2843 return start;
2844 }
2845
2846 void generate_arraycopy_stubs() {
2847 address entry;
2848 address entry_jbyte_arraycopy;
2849 address entry_jshort_arraycopy;
2850 address entry_jint_arraycopy;
2851 address entry_oop_arraycopy;
2852 address entry_jlong_arraycopy;
2853 address entry_checkcast_arraycopy;
2854
2855 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2856 "jbyte_disjoint_arraycopy");
2857 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2858 "jbyte_arraycopy");
2859
2860 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2861 "jshort_disjoint_arraycopy");
2862 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2863 "jshort_arraycopy");
2864
2865 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2866 "jint_disjoint_arraycopy");
2867 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2868 &entry_jint_arraycopy, "jint_arraycopy");
2869
2870 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2871 "jlong_disjoint_arraycopy");
2872 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2873 &entry_jlong_arraycopy, "jlong_arraycopy");
2874
2875
2876 if (UseCompressedOops) {
2877 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2878 "oop_disjoint_arraycopy");
2879 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2880 &entry_oop_arraycopy, "oop_arraycopy");
2881 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2882 "oop_disjoint_arraycopy_uninit",
2883 /*dest_uninitialized*/true);
2884 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2885 NULL, "oop_arraycopy_uninit",
2886 /*dest_uninitialized*/true);
2887 } else {
2888 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2889 "oop_disjoint_arraycopy");
2890 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2891 &entry_oop_arraycopy, "oop_arraycopy");
2892 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2893 "oop_disjoint_arraycopy_uninit",
2894 /*dest_uninitialized*/true);
2895 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2896 NULL, "oop_arraycopy_uninit",
2897 /*dest_uninitialized*/true);
2898 }
2899
2900 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2901 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2902 /*dest_uninitialized*/true);
2903
2904 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2905 entry_jbyte_arraycopy,
2906 entry_jshort_arraycopy,
2907 entry_jint_arraycopy,
2908 entry_jlong_arraycopy);
2909 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2910 entry_jbyte_arraycopy,
2911 entry_jshort_arraycopy,
2912 entry_jint_arraycopy,
2913 entry_oop_arraycopy,
2914 entry_jlong_arraycopy,
2915 entry_checkcast_arraycopy);
2916
2917 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2918 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2919 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2920 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2921 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2922 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2923
2924 // We don't generate specialized code for HeapWord-aligned source
2925 // arrays, so just use the code we've already generated
2926 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2927 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2928
2929 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2930 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2931
2932 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2933 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2934
2935 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2936 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2937
2938 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2939 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2940
2941 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2942 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2943 }
2944
2945 void generate_math_stubs() {
2946 {
2947 StubCodeMark mark(this, "StubRoutines", "log");
2948 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2949
2950 __ subq(rsp, 8);
2951 __ movdbl(Address(rsp, 0), xmm0);
2952 __ fld_d(Address(rsp, 0));
2953 __ flog();
2954 __ fstp_d(Address(rsp, 0));
2955 __ movdbl(xmm0, Address(rsp, 0));
2956 __ addq(rsp, 8);
2957 __ ret(0);
2958 }
2959 {
2960 StubCodeMark mark(this, "StubRoutines", "log10");
2961 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2962
2963 __ subq(rsp, 8);
2964 __ movdbl(Address(rsp, 0), xmm0);
2965 __ fld_d(Address(rsp, 0));
2966 __ flog10();
2967 __ fstp_d(Address(rsp, 0));
2968 __ movdbl(xmm0, Address(rsp, 0));
2969 __ addq(rsp, 8);
2970 __ ret(0);
2971 }
2972 {
2973 StubCodeMark mark(this, "StubRoutines", "sin");
2974 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2975
2976 __ subq(rsp, 8);
2977 __ movdbl(Address(rsp, 0), xmm0);
2978 __ fld_d(Address(rsp, 0));
2979 __ trigfunc('s');
2980 __ fstp_d(Address(rsp, 0));
2981 __ movdbl(xmm0, Address(rsp, 0));
2982 __ addq(rsp, 8);
2983 __ ret(0);
2984 }
2985 {
2986 StubCodeMark mark(this, "StubRoutines", "cos");
2987 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2988
2989 __ subq(rsp, 8);
2990 __ movdbl(Address(rsp, 0), xmm0);
2991 __ fld_d(Address(rsp, 0));
2992 __ trigfunc('c');
2993 __ fstp_d(Address(rsp, 0));
2994 __ movdbl(xmm0, Address(rsp, 0));
2995 __ addq(rsp, 8);
2996 __ ret(0);
2997 }
2998 {
2999 StubCodeMark mark(this, "StubRoutines", "tan");
3000 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
3001
3002 __ subq(rsp, 8);
3003 __ movdbl(Address(rsp, 0), xmm0);
3004 __ fld_d(Address(rsp, 0));
3005 __ trigfunc('t');
3006 __ fstp_d(Address(rsp, 0));
3007 __ movdbl(xmm0, Address(rsp, 0));
3008 __ addq(rsp, 8);
3009 __ ret(0);
3010 }
3011 {
3012 StubCodeMark mark(this, "StubRoutines", "exp");
3013 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
3014
3015 __ subq(rsp, 8);
3016 __ movdbl(Address(rsp, 0), xmm0);
3017 __ fld_d(Address(rsp, 0));
3018 __ exp_with_fallback(0);
3019 __ fstp_d(Address(rsp, 0));
3020 __ movdbl(xmm0, Address(rsp, 0));
3021 __ addq(rsp, 8);
3022 __ ret(0);
3023 }
3024 {
3025 StubCodeMark mark(this, "StubRoutines", "pow");
3026 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
3027
3028 __ subq(rsp, 8);
3029 __ movdbl(Address(rsp, 0), xmm1);
3030 __ fld_d(Address(rsp, 0));
3031 __ movdbl(Address(rsp, 0), xmm0);
3032 __ fld_d(Address(rsp, 0));
3033 __ pow_with_fallback(0);
3034 __ fstp_d(Address(rsp, 0));
3035 __ movdbl(xmm0, Address(rsp, 0));
3036 __ addq(rsp, 8);
3037 __ ret(0);
3038 }
3039 }
3040
3041 // AES intrinsic stubs
3042 enum {AESBlockSize = 16};
3043
3044 address generate_key_shuffle_mask() {
3045 __ align(16);
3046 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
3047 address start = __ pc();
3048 __ emit_data64( 0x0405060700010203, relocInfo::none );
3049 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
3050 return start;
3051 }
3052
3053 // Utility routine for loading a 128-bit key word in little endian format
3054 // can optionally specify that the shuffle mask is already in an xmmregister
3055 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
3056 __ movdqu(xmmdst, Address(key, offset));
3057 if (xmm_shuf_mask != NULL) {
3058 __ pshufb(xmmdst, xmm_shuf_mask);
3059 } else {
3060 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3061 }
3062 }
3063
3064 // Arguments:
3065 //
3066 // Inputs:
3067 // c_rarg0 - source byte array address
3068 // c_rarg1 - destination byte array address
3069 // c_rarg2 - K (key) in little endian int array
3070 //
3071 address generate_aescrypt_encryptBlock() {
3072 assert(UseAES, "need AES instructions and misaligned SSE support");
3073 __ align(CodeEntryAlignment);
3074 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
3075 Label L_doLast;
3076 address start = __ pc();
3077
3078 const Register from = c_rarg0; // source array address
3079 const Register to = c_rarg1; // destination array address
3080 const Register key = c_rarg2; // key array address
3081 const Register keylen = rax;
3082
3083 const XMMRegister xmm_result = xmm0;
3084 const XMMRegister xmm_key_shuf_mask = xmm1;
3085 // On win64 xmm6-xmm15 must be preserved so don't use them.
3086 const XMMRegister xmm_temp1 = xmm2;
3087 const XMMRegister xmm_temp2 = xmm3;
3088 const XMMRegister xmm_temp3 = xmm4;
3089 const XMMRegister xmm_temp4 = xmm5;
3090
3091 __ enter(); // required for proper stackwalking of RuntimeStub frame
3092
3093 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3094 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3095
3096 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3097 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
3098
3099 // For encryption, the java expanded key ordering is just what we need
3100 // we don't know if the key is aligned, hence not using load-execute form
3101
3102 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3103 __ pxor(xmm_result, xmm_temp1);
3104
3105 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3106 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3107 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3108 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3109
3110 __ aesenc(xmm_result, xmm_temp1);
3111 __ aesenc(xmm_result, xmm_temp2);
3112 __ aesenc(xmm_result, xmm_temp3);
3113 __ aesenc(xmm_result, xmm_temp4);
3114
3115 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3116 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3117 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3118 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3119
3120 __ aesenc(xmm_result, xmm_temp1);
3121 __ aesenc(xmm_result, xmm_temp2);
3122 __ aesenc(xmm_result, xmm_temp3);
3123 __ aesenc(xmm_result, xmm_temp4);
3124
3125 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3126 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3127
3128 __ cmpl(keylen, 44);
3129 __ jccb(Assembler::equal, L_doLast);
3130
3131 __ aesenc(xmm_result, xmm_temp1);
3132 __ aesenc(xmm_result, xmm_temp2);
3133
3134 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3135 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3136
3137 __ cmpl(keylen, 52);
3138 __ jccb(Assembler::equal, L_doLast);
3139
3140 __ aesenc(xmm_result, xmm_temp1);
3141 __ aesenc(xmm_result, xmm_temp2);
3142
3143 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3144 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3145
3146 __ BIND(L_doLast);
3147 __ aesenc(xmm_result, xmm_temp1);
3148 __ aesenclast(xmm_result, xmm_temp2);
3149 __ movdqu(Address(to, 0), xmm_result); // store the result
3150 __ xorptr(rax, rax); // return 0
3151 __ leave(); // required for proper stackwalking of RuntimeStub frame
3152 __ ret(0);
3153
3154 return start;
3155 }
3156
3157
3158 // Arguments:
3159 //
3160 // Inputs:
3161 // c_rarg0 - source byte array address
3162 // c_rarg1 - destination byte array address
3163 // c_rarg2 - K (key) in little endian int array
3164 //
3165 address generate_aescrypt_decryptBlock() {
3166 assert(UseAES, "need AES instructions and misaligned SSE support");
3167 __ align(CodeEntryAlignment);
3168 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3169 Label L_doLast;
3170 address start = __ pc();
3171
3172 const Register from = c_rarg0; // source array address
3173 const Register to = c_rarg1; // destination array address
3174 const Register key = c_rarg2; // key array address
3175 const Register keylen = rax;
3176
3177 const XMMRegister xmm_result = xmm0;
3178 const XMMRegister xmm_key_shuf_mask = xmm1;
3179 // On win64 xmm6-xmm15 must be preserved so don't use them.
3180 const XMMRegister xmm_temp1 = xmm2;
3181 const XMMRegister xmm_temp2 = xmm3;
3182 const XMMRegister xmm_temp3 = xmm4;
3183 const XMMRegister xmm_temp4 = xmm5;
3184
3185 __ enter(); // required for proper stackwalking of RuntimeStub frame
3186
3187 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3188 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3189
3190 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3191 __ movdqu(xmm_result, Address(from, 0));
3192
3193 // for decryption java expanded key ordering is rotated one position from what we want
3194 // so we start from 0x10 here and hit 0x00 last
3195 // we don't know if the key is aligned, hence not using load-execute form
3196 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3197 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3198 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3199 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3200
3201 __ pxor (xmm_result, xmm_temp1);
3202 __ aesdec(xmm_result, xmm_temp2);
3203 __ aesdec(xmm_result, xmm_temp3);
3204 __ aesdec(xmm_result, xmm_temp4);
3205
3206 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3207 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3208 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3209 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3210
3211 __ aesdec(xmm_result, xmm_temp1);
3212 __ aesdec(xmm_result, xmm_temp2);
3213 __ aesdec(xmm_result, xmm_temp3);
3214 __ aesdec(xmm_result, xmm_temp4);
3215
3216 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3217 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3218 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3219
3220 __ cmpl(keylen, 44);
3221 __ jccb(Assembler::equal, L_doLast);
3222
3223 __ aesdec(xmm_result, xmm_temp1);
3224 __ aesdec(xmm_result, xmm_temp2);
3225
3226 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3227 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3228
3229 __ cmpl(keylen, 52);
3230 __ jccb(Assembler::equal, L_doLast);
3231
3232 __ aesdec(xmm_result, xmm_temp1);
3233 __ aesdec(xmm_result, xmm_temp2);
3234
3235 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3236 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3237
3238 __ BIND(L_doLast);
3239 __ aesdec(xmm_result, xmm_temp1);
3240 __ aesdec(xmm_result, xmm_temp2);
3241
3242 // for decryption the aesdeclast operation is always on key+0x00
3243 __ aesdeclast(xmm_result, xmm_temp3);
3244 __ movdqu(Address(to, 0), xmm_result); // store the result
3245 __ xorptr(rax, rax); // return 0
3246 __ leave(); // required for proper stackwalking of RuntimeStub frame
3247 __ ret(0);
3248
3249 return start;
3250 }
3251
3252
3253 // Arguments:
3254 //
3255 // Inputs:
3256 // c_rarg0 - source byte array address
3257 // c_rarg1 - destination byte array address
3258 // c_rarg2 - K (key) in little endian int array
3259 // c_rarg3 - r vector byte array address
3260 // c_rarg4 - input length
3261 //
3262 // Output:
3263 // rax - input length
3264 //
3265 address generate_cipherBlockChaining_encryptAESCrypt() {
3266 assert(UseAES, "need AES instructions and misaligned SSE support");
3267 __ align(CodeEntryAlignment);
3268 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3269 address start = __ pc();
3270
3271 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3272 const Register from = c_rarg0; // source array address
3273 const Register to = c_rarg1; // destination array address
3274 const Register key = c_rarg2; // key array address
3275 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3276 // and left with the results of the last encryption block
3277 #ifndef _WIN64
3278 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3279 #else
3280 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3281 const Register len_reg = r10; // pick the first volatile windows register
3282 #endif
3283 const Register pos = rax;
3284
3285 // xmm register assignments for the loops below
3286 const XMMRegister xmm_result = xmm0;
3287 const XMMRegister xmm_temp = xmm1;
3288 // keys 0-10 preloaded into xmm2-xmm12
3289 const int XMM_REG_NUM_KEY_FIRST = 2;
3290 const int XMM_REG_NUM_KEY_LAST = 15;
3291 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3292 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3293 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3294 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3295 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3296
3297 __ enter(); // required for proper stackwalking of RuntimeStub frame
3298
3299 #ifdef _WIN64
3300 // on win64, fill len_reg from stack position
3301 __ movl(len_reg, len_mem);
3302 // save the xmm registers which must be preserved 6-15
3303 __ subptr(rsp, -rsp_after_call_off * wordSize);
3304 for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3305 __ movdqu(xmm_save(i), as_XMMRegister(i));
3306 }
3307 #else
3308 __ push(len_reg); // Save
3309 #endif
3310
3311 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3312 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3313 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3314 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3315 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3316 offset += 0x10;
3317 }
3318 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3319
3320 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3321 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3322 __ cmpl(rax, 44);
3323 __ jcc(Assembler::notEqual, L_key_192_256);
3324
3325 // 128 bit code follows here
3326 __ movptr(pos, 0);
3327 __ align(OptoLoopAlignment);
3328
3329 __ BIND(L_loopTop_128);
3330 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3331 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3332 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3333 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3334 __ aesenc(xmm_result, as_XMMRegister(rnum));
3335 }
3336 __ aesenclast(xmm_result, xmm_key10);
3337 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3338 // no need to store r to memory until we exit
3339 __ addptr(pos, AESBlockSize);
3340 __ subptr(len_reg, AESBlockSize);
3341 __ jcc(Assembler::notEqual, L_loopTop_128);
3342
3343 __ BIND(L_exit);
3344 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3345
3346 #ifdef _WIN64
3347 // restore xmm regs belonging to calling function
3348 for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3349 __ movdqu(as_XMMRegister(i), xmm_save(i));
3350 }
3351 __ movl(rax, len_mem);
3352 #else
3353 __ pop(rax); // return length
3354 #endif
3355 __ leave(); // required for proper stackwalking of RuntimeStub frame
3356 __ ret(0);
3357
3358 __ BIND(L_key_192_256);
3359 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3360 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3361 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3362 __ cmpl(rax, 52);
3363 __ jcc(Assembler::notEqual, L_key_256);
3364
3365 // 192-bit code follows here (could be changed to use more xmm registers)
3366 __ movptr(pos, 0);
3367 __ align(OptoLoopAlignment);
3368
3369 __ BIND(L_loopTop_192);
3370 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3371 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3372 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3373 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3374 __ aesenc(xmm_result, as_XMMRegister(rnum));
3375 }
3376 __ aesenclast(xmm_result, xmm_key12);
3377 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3378 // no need to store r to memory until we exit
3379 __ addptr(pos, AESBlockSize);
3380 __ subptr(len_reg, AESBlockSize);
3381 __ jcc(Assembler::notEqual, L_loopTop_192);
3382 __ jmp(L_exit);
3383
3384 __ BIND(L_key_256);
3385 // 256-bit code follows here (could be changed to use more xmm registers)
3386 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3387 __ movptr(pos, 0);
3388 __ align(OptoLoopAlignment);
3389
3390 __ BIND(L_loopTop_256);
3391 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3392 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3393 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3394 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3395 __ aesenc(xmm_result, as_XMMRegister(rnum));
3396 }
3397 load_key(xmm_temp, key, 0xe0);
3398 __ aesenclast(xmm_result, xmm_temp);
3399 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3400 // no need to store r to memory until we exit
3401 __ addptr(pos, AESBlockSize);
3402 __ subptr(len_reg, AESBlockSize);
3403 __ jcc(Assembler::notEqual, L_loopTop_256);
3404 __ jmp(L_exit);
3405
3406 return start;
3407 }
3408
3409 // Safefetch stubs.
3410 void generate_safefetch(const char* name, int size, address* entry,
3411 address* fault_pc, address* continuation_pc) {
3412 // safefetch signatures:
3413 // int SafeFetch32(int* adr, int errValue);
3414 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3415 //
3416 // arguments:
3417 // c_rarg0 = adr
3418 // c_rarg1 = errValue
3419 //
3420 // result:
3421 // PPC_RET = *adr or errValue
3422
3423 StubCodeMark mark(this, "StubRoutines", name);
3424
3425 // Entry point, pc or function descriptor.
3426 *entry = __ pc();
3427
3428 // Load *adr into c_rarg1, may fault.
3429 *fault_pc = __ pc();
3430 switch (size) {
3431 case 4:
3432 // int32_t
3433 __ movl(c_rarg1, Address(c_rarg0, 0));
3434 break;
3435 case 8:
3436 // int64_t
3437 __ movq(c_rarg1, Address(c_rarg0, 0));
3438 break;
3439 default:
3440 ShouldNotReachHere();
3441 }
3442
3443 // return errValue or *adr
3444 *continuation_pc = __ pc();
3445 __ movq(rax, c_rarg1);
3446 __ ret(0);
3447 }
3448
3449 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3450 // to hide instruction latency
3451 //
3452 // Arguments:
3453 //
3454 // Inputs:
3455 // c_rarg0 - source byte array address
3456 // c_rarg1 - destination byte array address
3457 // c_rarg2 - K (key) in little endian int array
3458 // c_rarg3 - r vector byte array address
3459 // c_rarg4 - input length
3460 //
3461 // Output:
3462 // rax - input length
3463 //
3464
3465 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3466 assert(UseAES, "need AES instructions and misaligned SSE support");
3467 __ align(CodeEntryAlignment);
3468 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3469 address start = __ pc();
3470
3471 Label L_exit, L_key_192_256, L_key_256;
3472 Label L_singleBlock_loopTop_128, L_multiBlock_loopTop_128;
3473 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
3474 const Register from = c_rarg0; // source array address
3475 const Register to = c_rarg1; // destination array address
3476 const Register key = c_rarg2; // key array address
3477 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3478 // and left with the results of the last encryption block
3479 #ifndef _WIN64
3480 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3481 #else
3482 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3483 const Register len_reg = r10; // pick the first volatile windows register
3484 #endif
3485 const Register pos = rax;
3486
3487 // keys 0-10 preloaded into xmm2-xmm12
3488 const int XMM_REG_NUM_KEY_FIRST = 5;
3489 const int XMM_REG_NUM_KEY_LAST = 15;
3490 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3491 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3492
3493 __ enter(); // required for proper stackwalking of RuntimeStub frame
3494
3495 #ifdef _WIN64
3496 // on win64, fill len_reg from stack position
3497 __ movl(len_reg, len_mem);
3498 // save the xmm registers which must be preserved 6-15
3499 __ subptr(rsp, -rsp_after_call_off * wordSize);
3500 for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3501 __ movdqu(xmm_save(i), as_XMMRegister(i));
3502 }
3503 #else
3504 __ push(len_reg); // Save
3505 #endif
3506
3507 // the java expanded key ordering is rotated one position from what we want
3508 // so we start from 0x10 here and hit 0x00 last
3509 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3510 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3511 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3512 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3513 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3514 offset += 0x10;
3515 }
3516 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3517
3518 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3519
3520 // registers holding the four results in the parallelized loop
3521 const XMMRegister xmm_result0 = xmm0;
3522 const XMMRegister xmm_result1 = xmm2;
3523 const XMMRegister xmm_result2 = xmm3;
3524 const XMMRegister xmm_result3 = xmm4;
3525
3526 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3527
3528 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3529 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3530 __ cmpl(rax, 44);
3531 __ jcc(Assembler::notEqual, L_key_192_256);
3532
3533
3534 // 128-bit code follows here, parallelized
3535 __ movptr(pos, 0);
3536 __ align(OptoLoopAlignment);
3537 __ BIND(L_multiBlock_loopTop_128);
3538 __ cmpptr(len_reg, 4*AESBlockSize); // see if at least 4 blocks left
3539 __ jcc(Assembler::less, L_singleBlock_loopTop_128);
3540
3541 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0*AESBlockSize)); // get next 4 blocks into xmmresult registers
3542 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1*AESBlockSize));
3543 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2*AESBlockSize));
3544 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3*AESBlockSize));
3545
3546 #define DoFour(opc, src_reg) \
3547 __ opc(xmm_result0, src_reg); \
3548 __ opc(xmm_result1, src_reg); \
3549 __ opc(xmm_result2, src_reg); \
3550 __ opc(xmm_result3, src_reg);
3551
3552 DoFour(pxor, xmm_key_first);
3553 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3554 DoFour(aesdec, as_XMMRegister(rnum));
3555 }
3556 DoFour(aesdeclast, xmm_key_last);
3557 // for each result, xor with the r vector of previous cipher block
3558 __ pxor(xmm_result0, xmm_prev_block_cipher);
3559 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0*AESBlockSize));
3560 __ pxor(xmm_result1, xmm_prev_block_cipher);
3561 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1*AESBlockSize));
3562 __ pxor(xmm_result2, xmm_prev_block_cipher);
3563 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2*AESBlockSize));
3564 __ pxor(xmm_result3, xmm_prev_block_cipher);
3565 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3*AESBlockSize)); // this will carry over to next set of blocks
3566
3567 __ movdqu(Address(to, pos, Address::times_1, 0*AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3568 __ movdqu(Address(to, pos, Address::times_1, 1*AESBlockSize), xmm_result1);
3569 __ movdqu(Address(to, pos, Address::times_1, 2*AESBlockSize), xmm_result2);
3570 __ movdqu(Address(to, pos, Address::times_1, 3*AESBlockSize), xmm_result3);
3571
3572 __ addptr(pos, 4*AESBlockSize);
3573 __ subptr(len_reg, 4*AESBlockSize);
3574 __ jmp(L_multiBlock_loopTop_128);
3575
3576 // registers used in the non-parallelized loops
3577 // xmm register assignments for the loops below
3578 const XMMRegister xmm_result = xmm0;
3579 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3580 const XMMRegister xmm_key11 = xmm3;
3581 const XMMRegister xmm_key12 = xmm4;
3582 const XMMRegister xmm_temp = xmm4;
3583
3584 __ align(OptoLoopAlignment);
3585 __ BIND(L_singleBlock_loopTop_128);
3586 __ cmpptr(len_reg, 0); // any blocks left??
3587 __ jcc(Assembler::equal, L_exit);
3588 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3589 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3590 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
3591 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3592 __ aesdec(xmm_result, as_XMMRegister(rnum));
3593 }
3594 __ aesdeclast(xmm_result, xmm_key_last);
3595 __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3596 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3597 // no need to store r to memory until we exit
3598 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3599
3600 __ addptr(pos, AESBlockSize);
3601 __ subptr(len_reg, AESBlockSize);
3602 __ jmp(L_singleBlock_loopTop_128);
3603
3604
3605 __ BIND(L_exit);
3606 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3607 #ifdef _WIN64
3608 // restore regs belonging to calling function
3609 for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3610 __ movdqu(as_XMMRegister(i), xmm_save(i));
3611 }
3612 __ movl(rax, len_mem);
3613 #else
3614 __ pop(rax); // return length
3615 #endif
3616 __ leave(); // required for proper stackwalking of RuntimeStub frame
3617 __ ret(0);
3618
3619
3620 __ BIND(L_key_192_256);
3621 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3622 load_key(xmm_key11, key, 0xb0);
3623 __ cmpl(rax, 52);
3624 __ jcc(Assembler::notEqual, L_key_256);
3625
3626 // 192-bit code follows here (could be optimized to use parallelism)
3627 load_key(xmm_key12, key, 0xc0); // 192-bit key goes up to c0
3628 __ movptr(pos, 0);
3629 __ align(OptoLoopAlignment);
3630
3631 __ BIND(L_singleBlock_loopTop_192);
3632 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3633 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3634 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
3635 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3636 __ aesdec(xmm_result, as_XMMRegister(rnum));
3637 }
3638 __ aesdec(xmm_result, xmm_key11);
3639 __ aesdec(xmm_result, xmm_key12);
3640 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3641 __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3642 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3643 // no need to store r to memory until we exit
3644 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3645 __ addptr(pos, AESBlockSize);
3646 __ subptr(len_reg, AESBlockSize);
3647 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
3648 __ jmp(L_exit);
3649
3650 __ BIND(L_key_256);
3651 // 256-bit code follows here (could be optimized to use parallelism)
3652 __ movptr(pos, 0);
3653 __ align(OptoLoopAlignment);
3654
3655 __ BIND(L_singleBlock_loopTop_256);
3656 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3657 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3658 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
3659 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3660 __ aesdec(xmm_result, as_XMMRegister(rnum));
3661 }
3662 __ aesdec(xmm_result, xmm_key11);
3663 load_key(xmm_temp, key, 0xc0);
3664 __ aesdec(xmm_result, xmm_temp);
3665 load_key(xmm_temp, key, 0xd0);
3666 __ aesdec(xmm_result, xmm_temp);
3667 load_key(xmm_temp, key, 0xe0); // 256-bit key goes up to e0
3668 __ aesdec(xmm_result, xmm_temp);
3669 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 came from key+0
3670 __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3671 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3672 // no need to store r to memory until we exit
3673 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3674 __ addptr(pos, AESBlockSize);
3675 __ subptr(len_reg, AESBlockSize);
3676 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
3677 __ jmp(L_exit);
3678
3679 return start;
3680 }
3681
3682 /**
3683 * Arguments:
3684 *
3685 * Inputs:
3686 * c_rarg0 - int crc
3687 * c_rarg1 - byte* buf
3688 * c_rarg2 - int length
3689 *
3690 * Ouput:
3691 * rax - int crc result
3692 */
3693 address generate_updateBytesCRC32() {
3694 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3695
3696 __ align(CodeEntryAlignment);
3697 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
3698
3699 address start = __ pc();
3700 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3701 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3702 // rscratch1: r10
3703 const Register crc = c_rarg0; // crc
3704 const Register buf = c_rarg1; // source java byte array address
3705 const Register len = c_rarg2; // length
3706 const Register table = c_rarg3; // crc_table address (reuse register)
3707 const Register tmp = r11;
3708 assert_different_registers(crc, buf, len, table, tmp, rax);
3709
3710 BLOCK_COMMENT("Entry:");
3711 __ enter(); // required for proper stackwalking of RuntimeStub frame
3712
3713 __ kernel_crc32(crc, buf, len, table, tmp);
3714
3715 __ movl(rax, crc);
3716 __ leave(); // required for proper stackwalking of RuntimeStub frame
3717 __ ret(0);
3718
3719 return start;
3720 }
3721
3722
3723 /**
3724 * Arguments:
3725 *
3726 * Input:
3727 * c_rarg0 - x address
3728 * c_rarg1 - x length
3729 * c_rarg2 - y address
3730 * c_rarg3 - y lenth
3731 * not Win64
3732 * c_rarg4 - z address
3733 * c_rarg5 - z length
3734 * Win64
3735 * rsp+40 - z address
3736 * rsp+48 - z length
3737 */
3738 address generate_multiplyToLen() {
3739 __ align(CodeEntryAlignment);
3740 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
3741
3742 address start = __ pc();
3743 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3744 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3745 const Register x = rdi;
3746 const Register xlen = rax;
3747 const Register y = rsi;
3748 const Register ylen = rcx;
3749 const Register z = r8;
3750 const Register zlen = r11;
3751
3752 // Next registers will be saved on stack in multiply_to_len().
3753 const Register tmp1 = r12;
3754 const Register tmp2 = r13;
3755 const Register tmp3 = r14;
3756 const Register tmp4 = r15;
3757 const Register tmp5 = rbx;
3758
3759 BLOCK_COMMENT("Entry:");
3760 __ enter(); // required for proper stackwalking of RuntimeStub frame
3761
3762 #ifndef _WIN64
3763 __ movptr(zlen, r9); // Save r9 in r11 - zlen
3764 #endif
3765 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
3766 // ylen => rcx, z => r8, zlen => r11
3767 // r9 and r10 may be used to save non-volatile registers
3768 #ifdef _WIN64
3769 // last 2 arguments (#4, #5) are on stack on Win64
3770 __ movptr(z, Address(rsp, 6 * wordSize));
3771 __ movptr(zlen, Address(rsp, 7 * wordSize));
3772 #endif
3773
3774 __ movptr(xlen, rsi);
3775 __ movptr(y, rdx);
3776 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
3777
3778 restore_arg_regs();
3779
3780 __ leave(); // required for proper stackwalking of RuntimeStub frame
3781 __ ret(0);
3782
3783 return start;
3784 }
3785
3786 #undef __
3787 #define __ masm->
3788
3789 // Continuation point for throwing of implicit exceptions that are
3790 // not handled in the current activation. Fabricates an exception
3791 // oop and initiates normal exception dispatching in this
3792 // frame. Since we need to preserve callee-saved values (currently
3793 // only for C2, but done for C1 as well) we need a callee-saved oop
3794 // map and therefore have to make these stubs into RuntimeStubs
3795 // rather than BufferBlobs. If the compiler needs all registers to
3796 // be preserved between the fault point and the exception handler
3797 // then it must assume responsibility for that in
3798 // AbstractCompiler::continuation_for_implicit_null_exception or
3799 // continuation_for_implicit_division_by_zero_exception. All other
3800 // implicit exceptions (e.g., NullPointerException or
3801 // AbstractMethodError on entry) are either at call sites or
3802 // otherwise assume that stack unwinding will be initiated, so
3803 // caller saved registers were assumed volatile in the compiler.
3804 address generate_throw_exception(const char* name,
3805 address runtime_entry,
3806 Register arg1 = noreg,
3807 Register arg2 = noreg) {
3808 // Information about frame layout at time of blocking runtime call.
3809 // Note that we only have to preserve callee-saved registers since
3810 // the compilers are responsible for supplying a continuation point
3811 // if they expect all registers to be preserved.
3812 enum layout {
3813 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
3814 rbp_off2,
3815 return_off,
3816 return_off2,
3817 framesize // inclusive of return address
3818 };
3819
3820 int insts_size = 512;
3821 int locs_size = 64;
3822
3823 CodeBuffer code(name, insts_size, locs_size);
3824 OopMapSet* oop_maps = new OopMapSet();
3825 MacroAssembler* masm = new MacroAssembler(&code);
3826
3827 address start = __ pc();
3828
3829 // This is an inlined and slightly modified version of call_VM
3830 // which has the ability to fetch the return PC out of
3831 // thread-local storage and also sets up last_Java_sp slightly
3832 // differently than the real call_VM
3833
3834 __ enter(); // required for proper stackwalking of RuntimeStub frame
3835
3836 assert(is_even(framesize/2), "sp not 16-byte aligned");
3837
3838 // return address and rbp are already in place
3839 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
3840
3841 int frame_complete = __ pc() - start;
3842
3843 // Set up last_Java_sp and last_Java_fp
3844 address the_pc = __ pc();
3845 __ set_last_Java_frame(rsp, rbp, the_pc);
3846 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
3847
3848 // Call runtime
3849 if (arg1 != noreg) {
3850 assert(arg2 != c_rarg1, "clobbered");
3851 __ movptr(c_rarg1, arg1);
3852 }
3853 if (arg2 != noreg) {
3854 __ movptr(c_rarg2, arg2);
3855 }
3856 __ movptr(c_rarg0, r15_thread);
3857 BLOCK_COMMENT("call runtime_entry");
3858 __ call(RuntimeAddress(runtime_entry));
3859
3860 // Generate oop map
3861 OopMap* map = new OopMap(framesize, 0);
3862
3863 oop_maps->add_gc_map(the_pc - start, map);
3864
3865 __ reset_last_Java_frame(true, true);
3866
3867 __ leave(); // required for proper stackwalking of RuntimeStub frame
3868
3869 // check for pending exceptions
3870 #ifdef ASSERT
3871 Label L;
3872 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
3873 (int32_t) NULL_WORD);
3874 __ jcc(Assembler::notEqual, L);
3875 __ should_not_reach_here();
3876 __ bind(L);
3877 #endif // ASSERT
3878 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3879
3880
3881 // codeBlob framesize is in words (not VMRegImpl::slot_size)
3882 RuntimeStub* stub =
3883 RuntimeStub::new_runtime_stub(name,
3884 &code,
3885 frame_complete,
3886 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3887 oop_maps, false);
3888 return stub->entry_point();
3889 }
3890
3891 void create_control_words() {
3892 // Round to nearest, 53-bit mode, exceptions masked
3893 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
3894 // Round to zero, 53-bit mode, exception mased
3895 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
3896 // Round to nearest, 24-bit mode, exceptions masked
3897 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
3898 // Round to nearest, 64-bit mode, exceptions masked
3899 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
3900 // Round to nearest, 64-bit mode, exceptions masked
3901 StubRoutines::_mxcsr_std = 0x1F80;
3902 // Note: the following two constants are 80-bit values
3903 // layout is critical for correct loading by FPU.
3904 // Bias for strict fp multiply/divide
3905 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
3906 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
3907 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
3908 // Un-Bias for strict fp multiply/divide
3909 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
3910 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
3911 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
3912 }
3913
3914 // Initialization
3915 void generate_initial() {
3916 // Generates all stubs and initializes the entry points
3917
3918 // This platform-specific settings are needed by generate_call_stub()
3919 create_control_words();
3920
3921 // entry points that exist in all platforms Note: This is code
3922 // that could be shared among different platforms - however the
3923 // benefit seems to be smaller than the disadvantage of having a
3924 // much more complicated generator structure. See also comment in
3925 // stubRoutines.hpp.
3926
3927 StubRoutines::_forward_exception_entry = generate_forward_exception();
3928
3929 StubRoutines::_call_stub_entry =
3930 generate_call_stub(StubRoutines::_call_stub_return_address);
3931
3932 // is referenced by megamorphic call
3933 StubRoutines::_catch_exception_entry = generate_catch_exception();
3934
3935 // atomic calls
3936 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3937 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
3938 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
3939 StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
3940 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
3941 StubRoutines::_atomic_add_entry = generate_atomic_add();
3942 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
3943 StubRoutines::_fence_entry = generate_orderaccess_fence();
3944
3945 StubRoutines::_handler_for_unsafe_access_entry =
3946 generate_handler_for_unsafe_access();
3947
3948 // platform dependent
3949 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
3950 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
3951
3952 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3953
3954 // Build this early so it's available for the interpreter.
3955 StubRoutines::_throw_StackOverflowError_entry =
3956 generate_throw_exception("StackOverflowError throw_exception",
3957 CAST_FROM_FN_PTR(address,
3958 SharedRuntime::
3959 throw_StackOverflowError));
3960 if (UseCRC32Intrinsics) {
3961 // set table address before stub generation which use it
3962 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3963 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3964 }
3965 }
3966
3967 void generate_all() {
3968 // Generates all stubs and initializes the entry points
3969
3970 // These entry points require SharedInfo::stack0 to be set up in
3971 // non-core builds and need to be relocatable, so they each
3972 // fabricate a RuntimeStub internally.
3973 StubRoutines::_throw_AbstractMethodError_entry =
3974 generate_throw_exception("AbstractMethodError throw_exception",
3975 CAST_FROM_FN_PTR(address,
3976 SharedRuntime::
3977 throw_AbstractMethodError));
3978
3979 StubRoutines::_throw_IncompatibleClassChangeError_entry =
3980 generate_throw_exception("IncompatibleClassChangeError throw_exception",
3981 CAST_FROM_FN_PTR(address,
3982 SharedRuntime::
3983 throw_IncompatibleClassChangeError));
3984
3985 StubRoutines::_throw_NullPointerException_at_call_entry =
3986 generate_throw_exception("NullPointerException at call throw_exception",
3987 CAST_FROM_FN_PTR(address,
3988 SharedRuntime::
3989 throw_NullPointerException_at_call));
3990
3991 // entry points that are platform specific
3992 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
3993 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
3994 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
3995 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
3996
3997 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
3998 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
3999 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4000 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
4001
4002 // support for verify_oop (must happen after universe_init)
4003 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
4004
4005 // arraycopy stubs used by compilers
4006 generate_arraycopy_stubs();
4007
4008 generate_math_stubs();
4009
4010 // don't bother generating these AES intrinsic stubs unless global flag is set
4011 if (UseAESIntrinsics) {
4012 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
4013
4014 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
4015 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
4016 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
4017 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
4018 }
4019
4020 // Safefetch stubs.
4021 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
4022 &StubRoutines::_safefetch32_fault_pc,
4023 &StubRoutines::_safefetch32_continuation_pc);
4024 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
4025 &StubRoutines::_safefetchN_fault_pc,
4026 &StubRoutines::_safefetchN_continuation_pc);
4027 #ifdef COMPILER2
4028 if (UseMultiplyToLenIntrinsic) {
4029 StubRoutines::_multiplyToLen = generate_multiplyToLen();
4030 }
4031 #endif
4032 }
4033
4034 public:
4035 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
4036 if (all) {
4037 generate_all();
4038 } else {
4039 generate_initial();
4040 }
4041 }
4042 }; // end class declaration
4043
4044 void StubGenerator_generate(CodeBuffer* code, bool all) {
4045 StubGenerator g(code, all);
4046 }