1 /*
2 * Copyright (c) 2003, 2013, 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 // -28 [ argument word 1 ]
141 // -27 [ saved xmm15 ] <--- rsp_after_call
142 // [ saved xmm7-xmm14 ]
143 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
144 // -7 [ saved r15 ]
145 // -6 [ saved r14 ]
146 // -5 [ saved r13 ]
147 // -4 [ saved r12 ]
148 // -3 [ saved rdi ]
149 // -2 [ saved rsi ]
150 // -1 [ saved rbx ]
151 // 0 [ saved rbp ] <--- rbp
152 // 1 [ return address ]
153 // 2 [ call wrapper ]
154 // 3 [ result ]
155 // 4 [ result type ]
156 // 5 [ method ]
157 // 6 [ entry point ]
158 // 7 [ parameters ]
159 // 8 [ parameter size ]
160 // 9 [ thread ]
161 //
162 // Windows reserves the callers stack space for arguments 1-4.
163 // We spill c_rarg0-c_rarg3 to this space.
164
165 // Call stub stack layout word offsets from rbp
166 enum call_stub_layout {
167 #ifdef _WIN64
168 xmm_save_first = 6, // save from xmm6
169 xmm_save_last = 15, // to xmm15
170 xmm_save_base = -9,
171 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
172 r15_off = -7,
173 r14_off = -6,
174 r13_off = -5,
175 r12_off = -4,
176 rdi_off = -3,
177 rsi_off = -2,
178 rbx_off = -1,
179 rbp_off = 0,
180 retaddr_off = 1,
181 call_wrapper_off = 2,
182 result_off = 3,
183 result_type_off = 4,
184 method_off = 5,
185 entry_point_off = 6,
186 parameters_off = 7,
187 parameter_size_off = 8,
188 thread_off = 9
189 #else
190 rsp_after_call_off = -12,
191 mxcsr_off = rsp_after_call_off,
192 r15_off = -11,
193 r14_off = -10,
194 r13_off = -9,
195 r12_off = -8,
196 rbx_off = -7,
197 call_wrapper_off = -6,
198 result_off = -5,
199 result_type_off = -4,
200 method_off = -3,
201 entry_point_off = -2,
202 parameters_off = -1,
203 rbp_off = 0,
204 retaddr_off = 1,
205 parameter_size_off = 2,
206 thread_off = 3
207 #endif
208 };
209
210 #ifdef _WIN64
211 Address xmm_save(int reg) {
212 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
213 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
214 }
215 #endif
216
217 address generate_call_stub(address& return_address) {
218 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
219 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
220 "adjust this code");
221 StubCodeMark mark(this, "StubRoutines", "call_stub");
222 address start = __ pc();
223
224 // same as in generate_catch_exception()!
225 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
226
227 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
228 const Address result (rbp, result_off * wordSize);
229 const Address result_type (rbp, result_type_off * wordSize);
230 const Address method (rbp, method_off * wordSize);
231 const Address entry_point (rbp, entry_point_off * wordSize);
232 const Address parameters (rbp, parameters_off * wordSize);
233 const Address parameter_size(rbp, parameter_size_off * wordSize);
234
235 // same as in generate_catch_exception()!
236 const Address thread (rbp, thread_off * wordSize);
237
238 const Address r15_save(rbp, r15_off * wordSize);
239 const Address r14_save(rbp, r14_off * wordSize);
240 const Address r13_save(rbp, r13_off * wordSize);
241 const Address r12_save(rbp, r12_off * wordSize);
242 const Address rbx_save(rbp, rbx_off * wordSize);
243
244 // stub code
245 __ enter();
246 __ subptr(rsp, -rsp_after_call_off * wordSize);
247
248 // save register parameters
249 #ifndef _WIN64
250 __ movptr(parameters, c_rarg5); // parameters
251 __ movptr(entry_point, c_rarg4); // entry_point
252 #endif
253
254 __ movptr(method, c_rarg3); // method
255 __ movl(result_type, c_rarg2); // result type
256 __ movptr(result, c_rarg1); // result
257 __ movptr(call_wrapper, c_rarg0); // call wrapper
258
259 // save regs belonging to calling function
260 __ movptr(rbx_save, rbx);
261 __ movptr(r12_save, r12);
262 __ movptr(r13_save, r13);
263 __ movptr(r14_save, r14);
264 __ movptr(r15_save, r15);
265 #ifdef _WIN64
266 for (int i = 6; i <= 15; i++) {
267 __ movdqu(xmm_save(i), as_XMMRegister(i));
268 }
269
270 const Address rdi_save(rbp, rdi_off * wordSize);
271 const Address rsi_save(rbp, rsi_off * wordSize);
272
273 __ movptr(rsi_save, rsi);
274 __ movptr(rdi_save, rdi);
275 #else
276 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
277 {
278 Label skip_ldmx;
279 __ stmxcsr(mxcsr_save);
280 __ movl(rax, mxcsr_save);
281 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
282 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
283 __ cmp32(rax, mxcsr_std);
284 __ jcc(Assembler::equal, skip_ldmx);
285 __ ldmxcsr(mxcsr_std);
286 __ bind(skip_ldmx);
287 }
288 #endif
289
290 // Load up thread register
291 __ movptr(r15_thread, thread);
292 __ reinit_heapbase();
293
294 #ifdef ASSERT
295 // make sure we have no pending exceptions
296 {
297 Label L;
298 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
299 __ jcc(Assembler::equal, L);
300 __ stop("StubRoutines::call_stub: entered with pending exception");
301 __ bind(L);
302 }
303 #endif
304
305 // pass parameters if any
306 BLOCK_COMMENT("pass parameters if any");
307 Label parameters_done;
308 __ movl(c_rarg3, parameter_size);
309 __ testl(c_rarg3, c_rarg3);
310 __ jcc(Assembler::zero, parameters_done);
311
312 Label loop;
313 __ movptr(c_rarg2, parameters); // parameter pointer
314 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
315 __ BIND(loop);
316 __ movptr(rax, Address(c_rarg2, 0));// get parameter
317 __ addptr(c_rarg2, wordSize); // advance to next parameter
318 __ decrementl(c_rarg1); // decrement counter
319 __ push(rax); // pass parameter
320 __ jcc(Assembler::notZero, loop);
321
322 // call Java function
323 __ BIND(parameters_done);
324 __ movptr(rbx, method); // get Method*
325 __ movptr(c_rarg1, entry_point); // get entry_point
326 __ mov(r13, rsp); // set sender sp
327 BLOCK_COMMENT("call Java function");
328 __ call(c_rarg1);
329
330 BLOCK_COMMENT("call_stub_return_address:");
331 return_address = __ pc();
332
333 // store result depending on type (everything that is not
334 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
335 __ movptr(c_rarg0, result);
336 Label is_long, is_float, is_double, exit;
337 __ movl(c_rarg1, result_type);
338 __ cmpl(c_rarg1, T_OBJECT);
339 __ jcc(Assembler::equal, is_long);
340 __ cmpl(c_rarg1, T_LONG);
341 __ jcc(Assembler::equal, is_long);
342 __ cmpl(c_rarg1, T_FLOAT);
343 __ jcc(Assembler::equal, is_float);
344 __ cmpl(c_rarg1, T_DOUBLE);
345 __ jcc(Assembler::equal, is_double);
346
347 // handle T_INT case
348 __ movl(Address(c_rarg0, 0), rax);
349
350 __ BIND(exit);
351
352 // pop parameters
353 __ lea(rsp, rsp_after_call);
354
355 #ifdef ASSERT
356 // verify that threads correspond
357 {
358 Label L, S;
359 __ cmpptr(r15_thread, thread);
360 __ jcc(Assembler::notEqual, S);
361 __ get_thread(rbx);
362 __ cmpptr(r15_thread, rbx);
363 __ jcc(Assembler::equal, L);
364 __ bind(S);
365 __ jcc(Assembler::equal, L);
366 __ stop("StubRoutines::call_stub: threads must correspond");
367 __ bind(L);
368 }
369 #endif
370
371 // restore regs belonging to calling function
372 #ifdef _WIN64
373 for (int i = 15; i >= 6; i--) {
374 __ movdqu(as_XMMRegister(i), xmm_save(i));
375 }
376 #endif
377 __ movptr(r15, r15_save);
378 __ movptr(r14, r14_save);
379 __ movptr(r13, r13_save);
380 __ movptr(r12, r12_save);
381 __ movptr(rbx, rbx_save);
382
383 #ifdef _WIN64
384 __ movptr(rdi, rdi_save);
385 __ movptr(rsi, rsi_save);
386 #else
387 __ ldmxcsr(mxcsr_save);
388 #endif
389
390 // restore rsp
391 __ addptr(rsp, -rsp_after_call_off * wordSize);
392
393 // return
394 __ pop(rbp);
395 __ ret(0);
396
397 // handle return types different from T_INT
398 __ BIND(is_long);
399 __ movq(Address(c_rarg0, 0), rax);
400 __ jmp(exit);
401
402 __ BIND(is_float);
403 __ movflt(Address(c_rarg0, 0), xmm0);
404 __ jmp(exit);
405
406 __ BIND(is_double);
407 __ movdbl(Address(c_rarg0, 0), xmm0);
408 __ jmp(exit);
409
410 return start;
411 }
412
413 // Return point for a Java call if there's an exception thrown in
414 // Java code. The exception is caught and transformed into a
415 // pending exception stored in JavaThread that can be tested from
416 // within the VM.
417 //
418 // Note: Usually the parameters are removed by the callee. In case
419 // of an exception crossing an activation frame boundary, that is
420 // not the case if the callee is compiled code => need to setup the
421 // rsp.
422 //
423 // rax: exception oop
424
425 address generate_catch_exception() {
426 StubCodeMark mark(this, "StubRoutines", "catch_exception");
427 address start = __ pc();
428
429 // same as in generate_call_stub():
430 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
431 const Address thread (rbp, thread_off * wordSize);
432
433 #ifdef ASSERT
434 // verify that threads correspond
435 {
436 Label L, S;
437 __ cmpptr(r15_thread, thread);
438 __ jcc(Assembler::notEqual, S);
439 __ get_thread(rbx);
440 __ cmpptr(r15_thread, rbx);
441 __ jcc(Assembler::equal, L);
442 __ bind(S);
443 __ stop("StubRoutines::catch_exception: threads must correspond");
444 __ bind(L);
445 }
446 #endif
447
448 // set pending exception
449 __ verify_oop(rax);
450
451 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
452 __ lea(rscratch1, ExternalAddress((address)__FILE__));
453 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
454 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
455
456 // complete return to VM
457 assert(StubRoutines::_call_stub_return_address != NULL,
458 "_call_stub_return_address must have been generated before");
459 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
460
461 return start;
462 }
463
464 // Continuation point for runtime calls returning with a pending
465 // exception. The pending exception check happened in the runtime
466 // or native call stub. The pending exception in Thread is
467 // converted into a Java-level exception.
468 //
469 // Contract with Java-level exception handlers:
470 // rax: exception
471 // rdx: throwing pc
472 //
473 // NOTE: At entry of this stub, exception-pc must be on stack !!
474
475 address generate_forward_exception() {
476 StubCodeMark mark(this, "StubRoutines", "forward exception");
477 address start = __ pc();
478
479 // Upon entry, the sp points to the return address returning into
480 // Java (interpreted or compiled) code; i.e., the return address
481 // becomes the throwing pc.
482 //
483 // Arguments pushed before the runtime call are still on the stack
484 // but the exception handler will reset the stack pointer ->
485 // ignore them. A potential result in registers can be ignored as
486 // well.
487
488 #ifdef ASSERT
489 // make sure this code is only executed if there is a pending exception
490 {
491 Label L;
492 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
493 __ jcc(Assembler::notEqual, L);
494 __ stop("StubRoutines::forward exception: no pending exception (1)");
495 __ bind(L);
496 }
497 #endif
498
499 // compute exception handler into rbx
500 __ movptr(c_rarg0, Address(rsp, 0));
501 BLOCK_COMMENT("call exception_handler_for_return_address");
502 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
503 SharedRuntime::exception_handler_for_return_address),
504 r15_thread, c_rarg0);
505 __ mov(rbx, rax);
506
507 // setup rax & rdx, remove return address & clear pending exception
508 __ pop(rdx);
509 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
510 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
511
512 #ifdef ASSERT
513 // make sure exception is set
514 {
515 Label L;
516 __ testptr(rax, rax);
517 __ jcc(Assembler::notEqual, L);
518 __ stop("StubRoutines::forward exception: no pending exception (2)");
519 __ bind(L);
520 }
521 #endif
522
523 // continue at exception handler (return address removed)
524 // rax: exception
525 // rbx: exception handler
526 // rdx: throwing pc
527 __ verify_oop(rax);
528 __ jmp(rbx);
529
530 return start;
531 }
532
533 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
534 //
535 // Arguments :
536 // c_rarg0: exchange_value
537 // c_rarg0: dest
538 //
539 // Result:
540 // *dest <- ex, return (orig *dest)
541 address generate_atomic_xchg() {
542 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
543 address start = __ pc();
544
545 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
546 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
547 __ ret(0);
548
549 return start;
550 }
551
552 // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
553 //
554 // Arguments :
555 // c_rarg0: exchange_value
556 // c_rarg1: dest
557 //
558 // Result:
559 // *dest <- ex, return (orig *dest)
560 address generate_atomic_xchg_ptr() {
561 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
562 address start = __ pc();
563
564 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
565 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
566 __ ret(0);
567
568 return start;
569 }
570
571 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
572 // jint compare_value)
573 //
574 // Arguments :
575 // c_rarg0: exchange_value
576 // c_rarg1: dest
577 // c_rarg2: compare_value
578 //
579 // Result:
580 // if ( compare_value == *dest ) {
581 // *dest = exchange_value
582 // return compare_value;
583 // else
584 // return *dest;
585 address generate_atomic_cmpxchg() {
586 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
587 address start = __ pc();
588
589 __ movl(rax, c_rarg2);
590 if ( os::is_MP() ) __ lock();
591 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
592 __ ret(0);
593
594 return start;
595 }
596
597 // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
598 // volatile jlong* dest,
599 // jlong compare_value)
600 // Arguments :
601 // c_rarg0: exchange_value
602 // c_rarg1: dest
603 // c_rarg2: compare_value
604 //
605 // Result:
606 // if ( compare_value == *dest ) {
607 // *dest = exchange_value
608 // return compare_value;
609 // else
610 // return *dest;
611 address generate_atomic_cmpxchg_long() {
612 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
613 address start = __ pc();
614
615 __ movq(rax, c_rarg2);
616 if ( os::is_MP() ) __ lock();
617 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
618 __ ret(0);
619
620 return start;
621 }
622
623 // Support for jint atomic::add(jint add_value, volatile jint* dest)
624 //
625 // Arguments :
626 // c_rarg0: add_value
627 // c_rarg1: dest
628 //
629 // Result:
630 // *dest += add_value
631 // return *dest;
632 address generate_atomic_add() {
633 StubCodeMark mark(this, "StubRoutines", "atomic_add");
634 address start = __ pc();
635
636 __ movl(rax, c_rarg0);
637 if ( os::is_MP() ) __ lock();
638 __ xaddl(Address(c_rarg1, 0), c_rarg0);
639 __ addl(rax, c_rarg0);
640 __ ret(0);
641
642 return start;
643 }
644
645 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
646 //
647 // Arguments :
648 // c_rarg0: add_value
649 // c_rarg1: dest
650 //
651 // Result:
652 // *dest += add_value
653 // return *dest;
654 address generate_atomic_add_ptr() {
655 StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
656 address start = __ pc();
657
658 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
659 if ( os::is_MP() ) __ lock();
660 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
661 __ addptr(rax, c_rarg0);
662 __ ret(0);
663
664 return start;
665 }
666
667 // Support for intptr_t OrderAccess::fence()
668 //
669 // Arguments :
670 //
671 // Result:
672 address generate_orderaccess_fence() {
673 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
674 address start = __ pc();
675 __ membar(Assembler::StoreLoad);
676 __ ret(0);
677
678 return start;
679 }
680
681 // Support for intptr_t get_previous_fp()
682 //
683 // This routine is used to find the previous frame pointer for the
684 // caller (current_frame_guess). This is used as part of debugging
685 // ps() is seemingly lost trying to find frames.
686 // This code assumes that caller current_frame_guess) has a frame.
687 address generate_get_previous_fp() {
688 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
689 const Address old_fp(rbp, 0);
690 const Address older_fp(rax, 0);
691 address start = __ pc();
692
693 __ enter();
694 __ movptr(rax, old_fp); // callers fp
695 __ movptr(rax, older_fp); // the frame for ps()
696 __ pop(rbp);
697 __ ret(0);
698
699 return start;
700 }
701
702 // Support for intptr_t get_previous_sp()
703 //
704 // This routine is used to find the previous stack pointer for the
705 // caller.
706 address generate_get_previous_sp() {
707 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
708 address start = __ pc();
709
710 __ movptr(rax, rsp);
711 __ addptr(rax, 8); // return address is at the top of the stack.
712 __ ret(0);
713
714 return start;
715 }
716
717 //----------------------------------------------------------------------------------------------------
718 // Support for void verify_mxcsr()
719 //
720 // This routine is used with -Xcheck:jni to verify that native
721 // JNI code does not return to Java code without restoring the
722 // MXCSR register to our expected state.
723
724 address generate_verify_mxcsr() {
725 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
726 address start = __ pc();
727
728 const Address mxcsr_save(rsp, 0);
729
730 if (CheckJNICalls) {
731 Label ok_ret;
732 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
733 __ push(rax);
734 __ subptr(rsp, wordSize); // allocate a temp location
735 __ stmxcsr(mxcsr_save);
736 __ movl(rax, mxcsr_save);
737 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
738 __ cmp32(rax, mxcsr_std);
739 __ jcc(Assembler::equal, ok_ret);
740
741 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
742
743 __ ldmxcsr(mxcsr_std);
744
745 __ bind(ok_ret);
746 __ addptr(rsp, wordSize);
747 __ pop(rax);
748 }
749
750 __ ret(0);
751
752 return start;
753 }
754
755 address generate_f2i_fixup() {
756 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
757 Address inout(rsp, 5 * wordSize); // return address + 4 saves
758
759 address start = __ pc();
760
761 Label L;
762
763 __ push(rax);
764 __ push(c_rarg3);
765 __ push(c_rarg2);
766 __ push(c_rarg1);
767
768 __ movl(rax, 0x7f800000);
769 __ xorl(c_rarg3, c_rarg3);
770 __ movl(c_rarg2, inout);
771 __ movl(c_rarg1, c_rarg2);
772 __ andl(c_rarg1, 0x7fffffff);
773 __ cmpl(rax, c_rarg1); // NaN? -> 0
774 __ jcc(Assembler::negative, L);
775 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
776 __ movl(c_rarg3, 0x80000000);
777 __ movl(rax, 0x7fffffff);
778 __ cmovl(Assembler::positive, c_rarg3, rax);
779
780 __ bind(L);
781 __ movptr(inout, c_rarg3);
782
783 __ pop(c_rarg1);
784 __ pop(c_rarg2);
785 __ pop(c_rarg3);
786 __ pop(rax);
787
788 __ ret(0);
789
790 return start;
791 }
792
793 address generate_f2l_fixup() {
794 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
795 Address inout(rsp, 5 * wordSize); // return address + 4 saves
796 address start = __ pc();
797
798 Label L;
799
800 __ push(rax);
801 __ push(c_rarg3);
802 __ push(c_rarg2);
803 __ push(c_rarg1);
804
805 __ movl(rax, 0x7f800000);
806 __ xorl(c_rarg3, c_rarg3);
807 __ movl(c_rarg2, inout);
808 __ movl(c_rarg1, c_rarg2);
809 __ andl(c_rarg1, 0x7fffffff);
810 __ cmpl(rax, c_rarg1); // NaN? -> 0
811 __ jcc(Assembler::negative, L);
812 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
813 __ mov64(c_rarg3, 0x8000000000000000);
814 __ mov64(rax, 0x7fffffffffffffff);
815 __ cmov(Assembler::positive, c_rarg3, rax);
816
817 __ bind(L);
818 __ movptr(inout, c_rarg3);
819
820 __ pop(c_rarg1);
821 __ pop(c_rarg2);
822 __ pop(c_rarg3);
823 __ pop(rax);
824
825 __ ret(0);
826
827 return start;
828 }
829
830 address generate_d2i_fixup() {
831 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
832 Address inout(rsp, 6 * wordSize); // return address + 5 saves
833
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 __ push(c_rarg0);
843
844 __ movl(rax, 0x7ff00000);
845 __ movq(c_rarg2, inout);
846 __ movl(c_rarg3, c_rarg2);
847 __ mov(c_rarg1, c_rarg2);
848 __ mov(c_rarg0, c_rarg2);
849 __ negl(c_rarg3);
850 __ shrptr(c_rarg1, 0x20);
851 __ orl(c_rarg3, c_rarg2);
852 __ andl(c_rarg1, 0x7fffffff);
853 __ xorl(c_rarg2, c_rarg2);
854 __ shrl(c_rarg3, 0x1f);
855 __ orl(c_rarg1, c_rarg3);
856 __ cmpl(rax, c_rarg1);
857 __ jcc(Assembler::negative, L); // NaN -> 0
858 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
859 __ movl(c_rarg2, 0x80000000);
860 __ movl(rax, 0x7fffffff);
861 __ cmov(Assembler::positive, c_rarg2, rax);
862
863 __ bind(L);
864 __ movptr(inout, c_rarg2);
865
866 __ pop(c_rarg0);
867 __ pop(c_rarg1);
868 __ pop(c_rarg2);
869 __ pop(c_rarg3);
870 __ pop(rax);
871
872 __ ret(0);
873
874 return start;
875 }
876
877 address generate_d2l_fixup() {
878 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
879 Address inout(rsp, 6 * wordSize); // return address + 5 saves
880
881 address start = __ pc();
882
883 Label L;
884
885 __ push(rax);
886 __ push(c_rarg3);
887 __ push(c_rarg2);
888 __ push(c_rarg1);
889 __ push(c_rarg0);
890
891 __ movl(rax, 0x7ff00000);
892 __ movq(c_rarg2, inout);
893 __ movl(c_rarg3, c_rarg2);
894 __ mov(c_rarg1, c_rarg2);
895 __ mov(c_rarg0, c_rarg2);
896 __ negl(c_rarg3);
897 __ shrptr(c_rarg1, 0x20);
898 __ orl(c_rarg3, c_rarg2);
899 __ andl(c_rarg1, 0x7fffffff);
900 __ xorl(c_rarg2, c_rarg2);
901 __ shrl(c_rarg3, 0x1f);
902 __ orl(c_rarg1, c_rarg3);
903 __ cmpl(rax, c_rarg1);
904 __ jcc(Assembler::negative, L); // NaN -> 0
905 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
906 __ mov64(c_rarg2, 0x8000000000000000);
907 __ mov64(rax, 0x7fffffffffffffff);
908 __ cmovq(Assembler::positive, c_rarg2, rax);
909
910 __ bind(L);
911 __ movq(inout, c_rarg2);
912
913 __ pop(c_rarg0);
914 __ pop(c_rarg1);
915 __ pop(c_rarg2);
916 __ pop(c_rarg3);
917 __ pop(rax);
918
919 __ ret(0);
920
921 return start;
922 }
923
924 address generate_fp_mask(const char *stub_name, int64_t mask) {
925 __ align(CodeEntryAlignment);
926 StubCodeMark mark(this, "StubRoutines", stub_name);
927 address start = __ pc();
928
929 __ emit_data64( mask, relocInfo::none );
930 __ emit_data64( mask, relocInfo::none );
931
932 return start;
933 }
934
935 // The following routine generates a subroutine to throw an
936 // asynchronous UnknownError when an unsafe access gets a fault that
937 // could not be reasonably prevented by the programmer. (Example:
938 // SIGBUS/OBJERR.)
939 address generate_handler_for_unsafe_access() {
940 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
941 address start = __ pc();
942
943 __ push(0); // hole for return address-to-be
944 __ pusha(); // push registers
945 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
946
947 // FIXME: this probably needs alignment logic
948
949 __ subptr(rsp, frame::arg_reg_save_area_bytes);
950 BLOCK_COMMENT("call handle_unsafe_access");
951 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
952 __ addptr(rsp, frame::arg_reg_save_area_bytes);
953
954 __ movptr(next_pc, rax); // stuff next address
955 __ popa();
956 __ ret(0); // jump to next address
957
958 return start;
959 }
960
961 // Non-destructive plausibility checks for oops
962 //
963 // Arguments:
964 // all args on stack!
965 //
966 // Stack after saving c_rarg3:
967 // [tos + 0]: saved c_rarg3
968 // [tos + 1]: saved c_rarg2
969 // [tos + 2]: saved r12 (several TemplateTable methods use it)
970 // [tos + 3]: saved flags
971 // [tos + 4]: return address
972 // * [tos + 5]: error message (char*)
973 // * [tos + 6]: object to verify (oop)
974 // * [tos + 7]: saved rax - saved by caller and bashed
975 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
976 // * = popped on exit
977 address generate_verify_oop() {
978 StubCodeMark mark(this, "StubRoutines", "verify_oop");
979 address start = __ pc();
980
981 Label exit, error;
982
983 __ pushf();
984 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
985
986 __ push(r12);
987
988 // save c_rarg2 and c_rarg3
989 __ push(c_rarg2);
990 __ push(c_rarg3);
991
992 enum {
993 // After previous pushes.
994 oop_to_verify = 6 * wordSize,
995 saved_rax = 7 * wordSize,
996 saved_r10 = 8 * wordSize,
997
998 // Before the call to MacroAssembler::debug(), see below.
999 return_addr = 16 * wordSize,
1000 error_msg = 17 * wordSize
1001 };
1002
1003 // get object
1004 __ movptr(rax, Address(rsp, oop_to_verify));
1005
1006 // make sure object is 'reasonable'
1007 __ testptr(rax, rax);
1008 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1009 // Check if the oop is in the right area of memory
1010 __ movptr(c_rarg2, rax);
1011 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1012 __ andptr(c_rarg2, c_rarg3);
1013 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1014 __ cmpptr(c_rarg2, c_rarg3);
1015 __ jcc(Assembler::notZero, error);
1016
1017 // set r12 to heapbase for load_klass()
1018 __ reinit_heapbase();
1019
1020 // make sure klass is 'reasonable', which is not zero.
1021 __ load_klass(rax, rax); // get klass
1022 __ testptr(rax, rax);
1023 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1024
1025 // return if everything seems ok
1026 __ bind(exit);
1027 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1028 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1029 __ pop(c_rarg3); // restore c_rarg3
1030 __ pop(c_rarg2); // restore c_rarg2
1031 __ pop(r12); // restore r12
1032 __ popf(); // restore flags
1033 __ ret(4 * wordSize); // pop caller saved stuff
1034
1035 // handle errors
1036 __ bind(error);
1037 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1038 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1039 __ pop(c_rarg3); // get saved c_rarg3 back
1040 __ pop(c_rarg2); // get saved c_rarg2 back
1041 __ pop(r12); // get saved r12 back
1042 __ popf(); // get saved flags off stack --
1043 // will be ignored
1044
1045 __ pusha(); // push registers
1046 // (rip is already
1047 // already pushed)
1048 // debug(char* msg, int64_t pc, int64_t regs[])
1049 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1050 // pushed all the registers, so now the stack looks like:
1051 // [tos + 0] 16 saved registers
1052 // [tos + 16] return address
1053 // * [tos + 17] error message (char*)
1054 // * [tos + 18] object to verify (oop)
1055 // * [tos + 19] saved rax - saved by caller and bashed
1056 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1057 // * = popped on exit
1058
1059 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1060 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1061 __ movq(c_rarg2, rsp); // pass address of regs on stack
1062 __ mov(r12, rsp); // remember rsp
1063 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1064 __ andptr(rsp, -16); // align stack as required by ABI
1065 BLOCK_COMMENT("call MacroAssembler::debug");
1066 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1067 __ mov(rsp, r12); // restore rsp
1068 __ popa(); // pop registers (includes r12)
1069 __ ret(4 * wordSize); // pop caller saved stuff
1070
1071 return start;
1072 }
1073
1074 //
1075 // Verify that a register contains clean 32-bits positive value
1076 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1077 //
1078 // Input:
1079 // Rint - 32-bits value
1080 // Rtmp - scratch
1081 //
1082 void assert_clean_int(Register Rint, Register Rtmp) {
1083 #ifdef ASSERT
1084 Label L;
1085 assert_different_registers(Rtmp, Rint);
1086 __ movslq(Rtmp, Rint);
1087 __ cmpq(Rtmp, Rint);
1088 __ jcc(Assembler::equal, L);
1089 __ stop("high 32-bits of int value are not 0");
1090 __ bind(L);
1091 #endif
1092 }
1093
1094 // Generate overlap test for array copy stubs
1095 //
1096 // Input:
1097 // c_rarg0 - from
1098 // c_rarg1 - to
1099 // c_rarg2 - element count
1100 //
1101 // Output:
1102 // rax - &from[element count - 1]
1103 //
1104 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1105 assert(no_overlap_target != NULL, "must be generated");
1106 array_overlap_test(no_overlap_target, NULL, sf);
1107 }
1108 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1109 array_overlap_test(NULL, &L_no_overlap, sf);
1110 }
1111 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1112 const Register from = c_rarg0;
1113 const Register to = c_rarg1;
1114 const Register count = c_rarg2;
1115 const Register end_from = rax;
1116
1117 __ cmpptr(to, from);
1118 __ lea(end_from, Address(from, count, sf, 0));
1119 if (NOLp == NULL) {
1120 ExternalAddress no_overlap(no_overlap_target);
1121 __ jump_cc(Assembler::belowEqual, no_overlap);
1122 __ cmpptr(to, end_from);
1123 __ jump_cc(Assembler::aboveEqual, no_overlap);
1124 } else {
1125 __ jcc(Assembler::belowEqual, (*NOLp));
1126 __ cmpptr(to, end_from);
1127 __ jcc(Assembler::aboveEqual, (*NOLp));
1128 }
1129 }
1130
1131 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1132 //
1133 // Outputs:
1134 // rdi - rcx
1135 // rsi - rdx
1136 // rdx - r8
1137 // rcx - r9
1138 //
1139 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1140 // are non-volatile. r9 and r10 should not be used by the caller.
1141 //
1142 void setup_arg_regs(int nargs = 3) {
1143 const Register saved_rdi = r9;
1144 const Register saved_rsi = r10;
1145 assert(nargs == 3 || nargs == 4, "else fix");
1146 #ifdef _WIN64
1147 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1148 "unexpected argument registers");
1149 if (nargs >= 4)
1150 __ mov(rax, r9); // r9 is also saved_rdi
1151 __ movptr(saved_rdi, rdi);
1152 __ movptr(saved_rsi, rsi);
1153 __ mov(rdi, rcx); // c_rarg0
1154 __ mov(rsi, rdx); // c_rarg1
1155 __ mov(rdx, r8); // c_rarg2
1156 if (nargs >= 4)
1157 __ mov(rcx, rax); // c_rarg3 (via rax)
1158 #else
1159 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1160 "unexpected argument registers");
1161 #endif
1162 }
1163
1164 void restore_arg_regs() {
1165 const Register saved_rdi = r9;
1166 const Register saved_rsi = r10;
1167 #ifdef _WIN64
1168 __ movptr(rdi, saved_rdi);
1169 __ movptr(rsi, saved_rsi);
1170 #endif
1171 }
1172
1173 // Generate code for an array write pre barrier
1174 //
1175 // addr - starting address
1176 // count - element count
1177 // tmp - scratch register
1178 //
1179 // Destroy no registers!
1180 //
1181 void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) {
1182 BarrierSet* bs = Universe::heap()->barrier_set();
1183 switch (bs->kind()) {
1184 case BarrierSet::G1SATBCT:
1185 case BarrierSet::G1SATBCTLogging:
1186 // With G1, don't generate the call if we statically know that the target in uninitialized
1187 if (!dest_uninitialized) {
1188 __ pusha(); // push registers
1189 if (count == c_rarg0) {
1190 if (addr == c_rarg1) {
1191 // exactly backwards!!
1192 __ xchgptr(c_rarg1, c_rarg0);
1193 } else {
1194 __ movptr(c_rarg1, count);
1195 __ movptr(c_rarg0, addr);
1196 }
1197 } else {
1198 __ movptr(c_rarg0, addr);
1199 __ movptr(c_rarg1, count);
1200 }
1201 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1202 __ popa();
1203 }
1204 break;
1205 case BarrierSet::CardTableModRef:
1206 case BarrierSet::CardTableExtension:
1207 case BarrierSet::ModRef:
1208 case BarrierSet::Epsilon:
1209 break;
1210 default:
1211 ShouldNotReachHere();
1212
1213 }
1214 }
1215
1216 //
1217 // Generate code for an array write post barrier
1218 //
1219 // Input:
1220 // start - register containing starting address of destination array
1221 // count - elements count
1222 // scratch - scratch register
1223 //
1224 // The input registers are overwritten.
1225 //
1226 void gen_write_ref_array_post_barrier(Register start, Register count, Register scratch) {
1227 assert_different_registers(start, count, scratch);
1228 BarrierSet* bs = Universe::heap()->barrier_set();
1229 switch (bs->kind()) {
1230 case BarrierSet::G1SATBCT:
1231 case BarrierSet::G1SATBCTLogging:
1232 {
1233 __ pusha(); // push registers (overkill)
1234 if (c_rarg0 == count) { // On win64 c_rarg0 == rcx
1235 assert_different_registers(c_rarg1, start);
1236 __ mov(c_rarg1, count);
1237 __ mov(c_rarg0, start);
1238 } else {
1239 assert_different_registers(c_rarg0, count);
1240 __ mov(c_rarg0, start);
1241 __ mov(c_rarg1, count);
1242 }
1243 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1244 __ popa();
1245 }
1246 break;
1247 case BarrierSet::CardTableModRef:
1248 case BarrierSet::CardTableExtension:
1249 {
1250 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1251 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1252
1253 Label L_loop;
1254 const Register end = count;
1255
1256 __ leaq(end, Address(start, count, TIMES_OOP, 0)); // end == start+count*oop_size
1257 __ subptr(end, BytesPerHeapOop); // end - 1 to make inclusive
1258 __ shrptr(start, CardTableModRefBS::card_shift);
1259 __ shrptr(end, CardTableModRefBS::card_shift);
1260 __ subptr(end, start); // end --> cards count
1261
1262 int64_t disp = (int64_t) ct->byte_map_base;
1263 __ mov64(scratch, disp);
1264 __ addptr(start, scratch);
1265 __ BIND(L_loop);
1266 __ movb(Address(start, count, Address::times_1), 0);
1267 __ decrement(count);
1268 __ jcc(Assembler::greaterEqual, L_loop);
1269 }
1270 break;
1271 case BarrierSet::Epsilon:
1272 // No barriers
1273 break;
1274 default:
1275 ShouldNotReachHere();
1276
1277 }
1278 }
1279
1280
1281 // Copy big chunks forward
1282 //
1283 // Inputs:
1284 // end_from - source arrays end address
1285 // end_to - destination array end address
1286 // qword_count - 64-bits element count, negative
1287 // to - scratch
1288 // L_copy_bytes - entry label
1289 // L_copy_8_bytes - exit label
1290 //
1291 void copy_bytes_forward(Register end_from, Register end_to,
1292 Register qword_count, Register to,
1293 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1294 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1295 Label L_loop;
1296 __ align(OptoLoopAlignment);
1297 if (UseUnalignedLoadStores) {
1298 Label L_end;
1299 // Copy 64-bytes per iteration
1300 __ BIND(L_loop);
1301 if (UseAVX >= 2) {
1302 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1303 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1304 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1305 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1306 } else {
1307 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1308 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1309 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1310 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1311 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1312 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1313 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1314 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1315 }
1316 __ BIND(L_copy_bytes);
1317 __ addptr(qword_count, 8);
1318 __ jcc(Assembler::lessEqual, L_loop);
1319 __ subptr(qword_count, 4); // sub(8) and add(4)
1320 __ jccb(Assembler::greater, L_end);
1321 // Copy trailing 32 bytes
1322 if (UseAVX >= 2) {
1323 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1324 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1325 } else {
1326 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1327 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1328 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1329 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1330 }
1331 __ addptr(qword_count, 4);
1332 __ BIND(L_end);
1333 if (UseAVX >= 2) {
1334 // clean upper bits of YMM registers
1335 __ vpxor(xmm0, xmm0);
1336 __ vpxor(xmm1, xmm1);
1337 }
1338 } else {
1339 // Copy 32-bytes per iteration
1340 __ BIND(L_loop);
1341 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1342 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1343 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1344 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1345 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1346 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1347 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1348 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1349
1350 __ BIND(L_copy_bytes);
1351 __ addptr(qword_count, 4);
1352 __ jcc(Assembler::lessEqual, L_loop);
1353 }
1354 __ subptr(qword_count, 4);
1355 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1356 }
1357
1358 // Copy big chunks backward
1359 //
1360 // Inputs:
1361 // from - source arrays address
1362 // dest - destination array address
1363 // qword_count - 64-bits element count
1364 // to - scratch
1365 // L_copy_bytes - entry label
1366 // L_copy_8_bytes - exit label
1367 //
1368 void copy_bytes_backward(Register from, Register dest,
1369 Register qword_count, Register to,
1370 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1371 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1372 Label L_loop;
1373 __ align(OptoLoopAlignment);
1374 if (UseUnalignedLoadStores) {
1375 Label L_end;
1376 // Copy 64-bytes per iteration
1377 __ BIND(L_loop);
1378 if (UseAVX >= 2) {
1379 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1380 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1381 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1382 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1383 } else {
1384 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1385 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1386 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1387 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1388 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1389 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1390 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1391 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1392 }
1393 __ BIND(L_copy_bytes);
1394 __ subptr(qword_count, 8);
1395 __ jcc(Assembler::greaterEqual, L_loop);
1396
1397 __ addptr(qword_count, 4); // add(8) and sub(4)
1398 __ jccb(Assembler::less, L_end);
1399 // Copy trailing 32 bytes
1400 if (UseAVX >= 2) {
1401 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1402 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1403 } else {
1404 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1405 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1406 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1407 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1408 }
1409 __ subptr(qword_count, 4);
1410 __ BIND(L_end);
1411 if (UseAVX >= 2) {
1412 // clean upper bits of YMM registers
1413 __ vpxor(xmm0, xmm0);
1414 __ vpxor(xmm1, xmm1);
1415 }
1416 } else {
1417 // Copy 32-bytes per iteration
1418 __ BIND(L_loop);
1419 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1420 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1421 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1422 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1423 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1424 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1425 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1426 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1427
1428 __ BIND(L_copy_bytes);
1429 __ subptr(qword_count, 4);
1430 __ jcc(Assembler::greaterEqual, L_loop);
1431 }
1432 __ addptr(qword_count, 4);
1433 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1434 }
1435
1436
1437 // Arguments:
1438 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1439 // ignored
1440 // name - stub name string
1441 //
1442 // Inputs:
1443 // c_rarg0 - source array address
1444 // c_rarg1 - destination array address
1445 // c_rarg2 - element count, treated as ssize_t, can be zero
1446 //
1447 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1448 // we let the hardware handle it. The one to eight bytes within words,
1449 // dwords or qwords that span cache line boundaries will still be loaded
1450 // and stored atomically.
1451 //
1452 // Side Effects:
1453 // disjoint_byte_copy_entry is set to the no-overlap entry point
1454 // used by generate_conjoint_byte_copy().
1455 //
1456 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1457 __ align(CodeEntryAlignment);
1458 StubCodeMark mark(this, "StubRoutines", name);
1459 address start = __ pc();
1460
1461 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1462 Label L_copy_byte, L_exit;
1463 const Register from = rdi; // source array address
1464 const Register to = rsi; // destination array address
1465 const Register count = rdx; // elements count
1466 const Register byte_count = rcx;
1467 const Register qword_count = count;
1468 const Register end_from = from; // source array end address
1469 const Register end_to = to; // destination array end address
1470 // End pointers are inclusive, and if count is not zero they point
1471 // to the last unit copied: end_to[0] := end_from[0]
1472
1473 __ enter(); // required for proper stackwalking of RuntimeStub frame
1474 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1475
1476 if (entry != NULL) {
1477 *entry = __ pc();
1478 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1479 BLOCK_COMMENT("Entry:");
1480 }
1481
1482 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1483 // r9 and r10 may be used to save non-volatile registers
1484
1485 // 'from', 'to' and 'count' are now valid
1486 __ movptr(byte_count, count);
1487 __ shrptr(count, 3); // count => qword_count
1488
1489 // Copy from low to high addresses. Use 'to' as scratch.
1490 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1491 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1492 __ negptr(qword_count); // make the count negative
1493 __ jmp(L_copy_bytes);
1494
1495 // Copy trailing qwords
1496 __ BIND(L_copy_8_bytes);
1497 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1498 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1499 __ increment(qword_count);
1500 __ jcc(Assembler::notZero, L_copy_8_bytes);
1501
1502 // Check for and copy trailing dword
1503 __ BIND(L_copy_4_bytes);
1504 __ testl(byte_count, 4);
1505 __ jccb(Assembler::zero, L_copy_2_bytes);
1506 __ movl(rax, Address(end_from, 8));
1507 __ movl(Address(end_to, 8), rax);
1508
1509 __ addptr(end_from, 4);
1510 __ addptr(end_to, 4);
1511
1512 // Check for and copy trailing word
1513 __ BIND(L_copy_2_bytes);
1514 __ testl(byte_count, 2);
1515 __ jccb(Assembler::zero, L_copy_byte);
1516 __ movw(rax, Address(end_from, 8));
1517 __ movw(Address(end_to, 8), rax);
1518
1519 __ addptr(end_from, 2);
1520 __ addptr(end_to, 2);
1521
1522 // Check for and copy trailing byte
1523 __ BIND(L_copy_byte);
1524 __ testl(byte_count, 1);
1525 __ jccb(Assembler::zero, L_exit);
1526 __ movb(rax, Address(end_from, 8));
1527 __ movb(Address(end_to, 8), rax);
1528
1529 __ BIND(L_exit);
1530 restore_arg_regs();
1531 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1532 __ xorptr(rax, rax); // return 0
1533 __ leave(); // required for proper stackwalking of RuntimeStub frame
1534 __ ret(0);
1535
1536 // Copy in multi-bytes chunks
1537 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1538 __ jmp(L_copy_4_bytes);
1539
1540 return start;
1541 }
1542
1543 // Arguments:
1544 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1545 // ignored
1546 // name - stub name string
1547 //
1548 // Inputs:
1549 // c_rarg0 - source array address
1550 // c_rarg1 - destination array address
1551 // c_rarg2 - element count, treated as ssize_t, can be zero
1552 //
1553 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1554 // we let the hardware handle it. The one to eight bytes within words,
1555 // dwords or qwords that span cache line boundaries will still be loaded
1556 // and stored atomically.
1557 //
1558 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1559 address* entry, const char *name) {
1560 __ align(CodeEntryAlignment);
1561 StubCodeMark mark(this, "StubRoutines", name);
1562 address start = __ pc();
1563
1564 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1565 const Register from = rdi; // source array address
1566 const Register to = rsi; // destination array address
1567 const Register count = rdx; // elements count
1568 const Register byte_count = rcx;
1569 const Register qword_count = count;
1570
1571 __ enter(); // required for proper stackwalking of RuntimeStub frame
1572 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1573
1574 if (entry != NULL) {
1575 *entry = __ pc();
1576 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1577 BLOCK_COMMENT("Entry:");
1578 }
1579
1580 array_overlap_test(nooverlap_target, Address::times_1);
1581 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1582 // r9 and r10 may be used to save non-volatile registers
1583
1584 // 'from', 'to' and 'count' are now valid
1585 __ movptr(byte_count, count);
1586 __ shrptr(count, 3); // count => qword_count
1587
1588 // Copy from high to low addresses.
1589
1590 // Check for and copy trailing byte
1591 __ testl(byte_count, 1);
1592 __ jcc(Assembler::zero, L_copy_2_bytes);
1593 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1594 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1595 __ decrement(byte_count); // Adjust for possible trailing word
1596
1597 // Check for and copy trailing word
1598 __ BIND(L_copy_2_bytes);
1599 __ testl(byte_count, 2);
1600 __ jcc(Assembler::zero, L_copy_4_bytes);
1601 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1602 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1603
1604 // Check for and copy trailing dword
1605 __ BIND(L_copy_4_bytes);
1606 __ testl(byte_count, 4);
1607 __ jcc(Assembler::zero, L_copy_bytes);
1608 __ movl(rax, Address(from, qword_count, Address::times_8));
1609 __ movl(Address(to, qword_count, Address::times_8), rax);
1610 __ jmp(L_copy_bytes);
1611
1612 // Copy trailing qwords
1613 __ BIND(L_copy_8_bytes);
1614 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1615 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1616 __ decrement(qword_count);
1617 __ jcc(Assembler::notZero, L_copy_8_bytes);
1618
1619 restore_arg_regs();
1620 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1621 __ xorptr(rax, rax); // return 0
1622 __ leave(); // required for proper stackwalking of RuntimeStub frame
1623 __ ret(0);
1624
1625 // Copy in multi-bytes chunks
1626 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1627
1628 restore_arg_regs();
1629 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1630 __ xorptr(rax, rax); // return 0
1631 __ leave(); // required for proper stackwalking of RuntimeStub frame
1632 __ ret(0);
1633
1634 return start;
1635 }
1636
1637 // Arguments:
1638 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1639 // ignored
1640 // name - stub name string
1641 //
1642 // Inputs:
1643 // c_rarg0 - source array address
1644 // c_rarg1 - destination array address
1645 // c_rarg2 - element count, treated as ssize_t, can be zero
1646 //
1647 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1648 // let the hardware handle it. The two or four words within dwords
1649 // or qwords that span cache line boundaries will still be loaded
1650 // and stored atomically.
1651 //
1652 // Side Effects:
1653 // disjoint_short_copy_entry is set to the no-overlap entry point
1654 // used by generate_conjoint_short_copy().
1655 //
1656 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1657 __ align(CodeEntryAlignment);
1658 StubCodeMark mark(this, "StubRoutines", name);
1659 address start = __ pc();
1660
1661 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1662 const Register from = rdi; // source array address
1663 const Register to = rsi; // destination array address
1664 const Register count = rdx; // elements count
1665 const Register word_count = rcx;
1666 const Register qword_count = count;
1667 const Register end_from = from; // source array end address
1668 const Register end_to = to; // destination array end address
1669 // End pointers are inclusive, and if count is not zero they point
1670 // to the last unit copied: end_to[0] := end_from[0]
1671
1672 __ enter(); // required for proper stackwalking of RuntimeStub frame
1673 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1674
1675 if (entry != NULL) {
1676 *entry = __ pc();
1677 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1678 BLOCK_COMMENT("Entry:");
1679 }
1680
1681 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1682 // r9 and r10 may be used to save non-volatile registers
1683
1684 // 'from', 'to' and 'count' are now valid
1685 __ movptr(word_count, count);
1686 __ shrptr(count, 2); // count => qword_count
1687
1688 // Copy from low to high addresses. Use 'to' as scratch.
1689 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1690 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1691 __ negptr(qword_count);
1692 __ jmp(L_copy_bytes);
1693
1694 // Copy trailing qwords
1695 __ BIND(L_copy_8_bytes);
1696 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1697 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1698 __ increment(qword_count);
1699 __ jcc(Assembler::notZero, L_copy_8_bytes);
1700
1701 // Original 'dest' is trashed, so we can't use it as a
1702 // base register for a possible trailing word copy
1703
1704 // Check for and copy trailing dword
1705 __ BIND(L_copy_4_bytes);
1706 __ testl(word_count, 2);
1707 __ jccb(Assembler::zero, L_copy_2_bytes);
1708 __ movl(rax, Address(end_from, 8));
1709 __ movl(Address(end_to, 8), rax);
1710
1711 __ addptr(end_from, 4);
1712 __ addptr(end_to, 4);
1713
1714 // Check for and copy trailing word
1715 __ BIND(L_copy_2_bytes);
1716 __ testl(word_count, 1);
1717 __ jccb(Assembler::zero, L_exit);
1718 __ movw(rax, Address(end_from, 8));
1719 __ movw(Address(end_to, 8), rax);
1720
1721 __ BIND(L_exit);
1722 restore_arg_regs();
1723 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1724 __ xorptr(rax, rax); // return 0
1725 __ leave(); // required for proper stackwalking of RuntimeStub frame
1726 __ ret(0);
1727
1728 // Copy in multi-bytes chunks
1729 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1730 __ jmp(L_copy_4_bytes);
1731
1732 return start;
1733 }
1734
1735 address generate_fill(BasicType t, bool aligned, const char *name) {
1736 __ align(CodeEntryAlignment);
1737 StubCodeMark mark(this, "StubRoutines", name);
1738 address start = __ pc();
1739
1740 BLOCK_COMMENT("Entry:");
1741
1742 const Register to = c_rarg0; // source array address
1743 const Register value = c_rarg1; // value
1744 const Register count = c_rarg2; // elements count
1745
1746 __ enter(); // required for proper stackwalking of RuntimeStub frame
1747
1748 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1749
1750 __ leave(); // required for proper stackwalking of RuntimeStub frame
1751 __ ret(0);
1752 return start;
1753 }
1754
1755 // Arguments:
1756 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1757 // ignored
1758 // name - stub name string
1759 //
1760 // Inputs:
1761 // c_rarg0 - source array address
1762 // c_rarg1 - destination array address
1763 // c_rarg2 - element count, treated as ssize_t, can be zero
1764 //
1765 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1766 // let the hardware handle it. The two or four words within dwords
1767 // or qwords that span cache line boundaries will still be loaded
1768 // and stored atomically.
1769 //
1770 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1771 address *entry, const char *name) {
1772 __ align(CodeEntryAlignment);
1773 StubCodeMark mark(this, "StubRoutines", name);
1774 address start = __ pc();
1775
1776 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1777 const Register from = rdi; // source array address
1778 const Register to = rsi; // destination array address
1779 const Register count = rdx; // elements count
1780 const Register word_count = rcx;
1781 const Register qword_count = count;
1782
1783 __ enter(); // required for proper stackwalking of RuntimeStub frame
1784 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1785
1786 if (entry != NULL) {
1787 *entry = __ pc();
1788 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1789 BLOCK_COMMENT("Entry:");
1790 }
1791
1792 array_overlap_test(nooverlap_target, Address::times_2);
1793 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1794 // r9 and r10 may be used to save non-volatile registers
1795
1796 // 'from', 'to' and 'count' are now valid
1797 __ movptr(word_count, count);
1798 __ shrptr(count, 2); // count => qword_count
1799
1800 // Copy from high to low addresses. Use 'to' as scratch.
1801
1802 // Check for and copy trailing word
1803 __ testl(word_count, 1);
1804 __ jccb(Assembler::zero, L_copy_4_bytes);
1805 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1806 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1807
1808 // Check for and copy trailing dword
1809 __ BIND(L_copy_4_bytes);
1810 __ testl(word_count, 2);
1811 __ jcc(Assembler::zero, L_copy_bytes);
1812 __ movl(rax, Address(from, qword_count, Address::times_8));
1813 __ movl(Address(to, qword_count, Address::times_8), rax);
1814 __ jmp(L_copy_bytes);
1815
1816 // Copy trailing qwords
1817 __ BIND(L_copy_8_bytes);
1818 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1819 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1820 __ decrement(qword_count);
1821 __ jcc(Assembler::notZero, L_copy_8_bytes);
1822
1823 restore_arg_regs();
1824 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1825 __ xorptr(rax, rax); // return 0
1826 __ leave(); // required for proper stackwalking of RuntimeStub frame
1827 __ ret(0);
1828
1829 // Copy in multi-bytes chunks
1830 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1831
1832 restore_arg_regs();
1833 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1834 __ xorptr(rax, rax); // return 0
1835 __ leave(); // required for proper stackwalking of RuntimeStub frame
1836 __ ret(0);
1837
1838 return start;
1839 }
1840
1841 // Arguments:
1842 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1843 // ignored
1844 // is_oop - true => oop array, so generate store check code
1845 // name - stub name string
1846 //
1847 // Inputs:
1848 // c_rarg0 - source array address
1849 // c_rarg1 - destination array address
1850 // c_rarg2 - element count, treated as ssize_t, can be zero
1851 //
1852 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1853 // the hardware handle it. The two dwords within qwords that span
1854 // cache line boundaries will still be loaded and stored atomicly.
1855 //
1856 // Side Effects:
1857 // disjoint_int_copy_entry is set to the no-overlap entry point
1858 // used by generate_conjoint_int_oop_copy().
1859 //
1860 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1861 const char *name, bool dest_uninitialized = false) {
1862 __ align(CodeEntryAlignment);
1863 StubCodeMark mark(this, "StubRoutines", name);
1864 address start = __ pc();
1865
1866 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1867 const Register from = rdi; // source array address
1868 const Register to = rsi; // destination array address
1869 const Register count = rdx; // elements count
1870 const Register dword_count = rcx;
1871 const Register qword_count = count;
1872 const Register end_from = from; // source array end address
1873 const Register end_to = to; // destination array end address
1874 const Register saved_to = r11; // saved destination array address
1875 // End pointers are inclusive, and if count is not zero they point
1876 // to the last unit copied: end_to[0] := end_from[0]
1877
1878 __ enter(); // required for proper stackwalking of RuntimeStub frame
1879 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1880
1881 if (entry != NULL) {
1882 *entry = __ pc();
1883 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1884 BLOCK_COMMENT("Entry:");
1885 }
1886
1887 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1888 // r9 and r10 may be used to save non-volatile registers
1889 if (is_oop) {
1890 __ movq(saved_to, to);
1891 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1892 }
1893
1894 // 'from', 'to' and 'count' are now valid
1895 __ movptr(dword_count, count);
1896 __ shrptr(count, 1); // count => qword_count
1897
1898 // Copy from low to high addresses. Use 'to' as scratch.
1899 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1900 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1901 __ negptr(qword_count);
1902 __ jmp(L_copy_bytes);
1903
1904 // Copy trailing qwords
1905 __ BIND(L_copy_8_bytes);
1906 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1907 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1908 __ increment(qword_count);
1909 __ jcc(Assembler::notZero, L_copy_8_bytes);
1910
1911 // Check for and copy trailing dword
1912 __ BIND(L_copy_4_bytes);
1913 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1914 __ jccb(Assembler::zero, L_exit);
1915 __ movl(rax, Address(end_from, 8));
1916 __ movl(Address(end_to, 8), rax);
1917
1918 __ BIND(L_exit);
1919 if (is_oop) {
1920 gen_write_ref_array_post_barrier(saved_to, dword_count, rax);
1921 }
1922 restore_arg_regs();
1923 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1924 __ xorptr(rax, rax); // return 0
1925 __ leave(); // required for proper stackwalking of RuntimeStub frame
1926 __ ret(0);
1927
1928 // Copy in multi-bytes chunks
1929 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1930 __ jmp(L_copy_4_bytes);
1931
1932 return start;
1933 }
1934
1935 // Arguments:
1936 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1937 // ignored
1938 // is_oop - true => oop array, so generate store check code
1939 // name - stub name string
1940 //
1941 // Inputs:
1942 // c_rarg0 - source array address
1943 // c_rarg1 - destination array address
1944 // c_rarg2 - element count, treated as ssize_t, can be zero
1945 //
1946 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1947 // the hardware handle it. The two dwords within qwords that span
1948 // cache line boundaries will still be loaded and stored atomicly.
1949 //
1950 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1951 address *entry, const char *name,
1952 bool dest_uninitialized = false) {
1953 __ align(CodeEntryAlignment);
1954 StubCodeMark mark(this, "StubRoutines", name);
1955 address start = __ pc();
1956
1957 Label L_copy_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1958 const Register from = rdi; // source array address
1959 const Register to = rsi; // destination array address
1960 const Register count = rdx; // elements count
1961 const Register dword_count = rcx;
1962 const Register qword_count = count;
1963
1964 __ enter(); // required for proper stackwalking of RuntimeStub frame
1965 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1966
1967 if (entry != NULL) {
1968 *entry = __ pc();
1969 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1970 BLOCK_COMMENT("Entry:");
1971 }
1972
1973 array_overlap_test(nooverlap_target, Address::times_4);
1974 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1975 // r9 and r10 may be used to save non-volatile registers
1976
1977 if (is_oop) {
1978 // no registers are destroyed by this call
1979 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1980 }
1981
1982 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1983 // 'from', 'to' and 'count' are now valid
1984 __ movptr(dword_count, count);
1985 __ shrptr(count, 1); // count => qword_count
1986
1987 // Copy from high to low addresses. Use 'to' as scratch.
1988
1989 // Check for and copy trailing dword
1990 __ testl(dword_count, 1);
1991 __ jcc(Assembler::zero, L_copy_bytes);
1992 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1993 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1994 __ jmp(L_copy_bytes);
1995
1996 // Copy trailing qwords
1997 __ BIND(L_copy_8_bytes);
1998 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1999 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2000 __ decrement(qword_count);
2001 __ jcc(Assembler::notZero, L_copy_8_bytes);
2002
2003 if (is_oop) {
2004 __ jmp(L_exit);
2005 }
2006 restore_arg_regs();
2007 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2008 __ xorptr(rax, rax); // return 0
2009 __ leave(); // required for proper stackwalking of RuntimeStub frame
2010 __ ret(0);
2011
2012 // Copy in multi-bytes chunks
2013 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2014
2015 __ BIND(L_exit);
2016 if (is_oop) {
2017 gen_write_ref_array_post_barrier(to, dword_count, rax);
2018 }
2019 restore_arg_regs();
2020 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2021 __ xorptr(rax, rax); // return 0
2022 __ leave(); // required for proper stackwalking of RuntimeStub frame
2023 __ ret(0);
2024
2025 return start;
2026 }
2027
2028 // Arguments:
2029 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2030 // ignored
2031 // is_oop - true => oop array, so generate store check code
2032 // name - stub name string
2033 //
2034 // Inputs:
2035 // c_rarg0 - source array address
2036 // c_rarg1 - destination array address
2037 // c_rarg2 - element count, treated as ssize_t, can be zero
2038 //
2039 // Side Effects:
2040 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2041 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2042 //
2043 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2044 const char *name, bool dest_uninitialized = false) {
2045 __ align(CodeEntryAlignment);
2046 StubCodeMark mark(this, "StubRoutines", name);
2047 address start = __ pc();
2048
2049 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2050 const Register from = rdi; // source array address
2051 const Register to = rsi; // destination array address
2052 const Register qword_count = rdx; // elements count
2053 const Register end_from = from; // source array end address
2054 const Register end_to = rcx; // destination array end address
2055 const Register saved_to = to;
2056 const Register saved_count = r11;
2057 // End pointers are inclusive, and if count is not zero they point
2058 // to the last unit copied: end_to[0] := end_from[0]
2059
2060 __ enter(); // required for proper stackwalking of RuntimeStub frame
2061 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2062 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2063
2064 if (entry != NULL) {
2065 *entry = __ pc();
2066 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2067 BLOCK_COMMENT("Entry:");
2068 }
2069
2070 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2071 // r9 and r10 may be used to save non-volatile registers
2072 // 'from', 'to' and 'qword_count' are now valid
2073 if (is_oop) {
2074 // Save to and count for store barrier
2075 __ movptr(saved_count, qword_count);
2076 // no registers are destroyed by this call
2077 gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized);
2078 }
2079
2080 // Copy from low to high addresses. Use 'to' as scratch.
2081 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2082 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2083 __ negptr(qword_count);
2084 __ jmp(L_copy_bytes);
2085
2086 // Copy trailing qwords
2087 __ BIND(L_copy_8_bytes);
2088 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2089 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2090 __ increment(qword_count);
2091 __ jcc(Assembler::notZero, L_copy_8_bytes);
2092
2093 if (is_oop) {
2094 __ jmp(L_exit);
2095 } else {
2096 restore_arg_regs();
2097 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2098 __ xorptr(rax, rax); // return 0
2099 __ leave(); // required for proper stackwalking of RuntimeStub frame
2100 __ ret(0);
2101 }
2102
2103 // Copy in multi-bytes chunks
2104 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2105
2106 if (is_oop) {
2107 __ BIND(L_exit);
2108 gen_write_ref_array_post_barrier(saved_to, saved_count, rax);
2109 }
2110 restore_arg_regs();
2111 if (is_oop) {
2112 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2113 } else {
2114 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2115 }
2116 __ xorptr(rax, rax); // return 0
2117 __ leave(); // required for proper stackwalking of RuntimeStub frame
2118 __ ret(0);
2119
2120 return start;
2121 }
2122
2123 // Arguments:
2124 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2125 // ignored
2126 // is_oop - true => oop array, so generate store check code
2127 // name - stub name string
2128 //
2129 // Inputs:
2130 // c_rarg0 - source array address
2131 // c_rarg1 - destination array address
2132 // c_rarg2 - element count, treated as ssize_t, can be zero
2133 //
2134 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2135 address nooverlap_target, address *entry,
2136 const char *name, bool dest_uninitialized = false) {
2137 __ align(CodeEntryAlignment);
2138 StubCodeMark mark(this, "StubRoutines", name);
2139 address start = __ pc();
2140
2141 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2142 const Register from = rdi; // source array address
2143 const Register to = rsi; // destination array address
2144 const Register qword_count = rdx; // elements count
2145 const Register saved_count = rcx;
2146
2147 __ enter(); // required for proper stackwalking of RuntimeStub frame
2148 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2149
2150 if (entry != NULL) {
2151 *entry = __ pc();
2152 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2153 BLOCK_COMMENT("Entry:");
2154 }
2155
2156 array_overlap_test(nooverlap_target, Address::times_8);
2157 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2158 // r9 and r10 may be used to save non-volatile registers
2159 // 'from', 'to' and 'qword_count' are now valid
2160 if (is_oop) {
2161 // Save to and count for store barrier
2162 __ movptr(saved_count, qword_count);
2163 // No registers are destroyed by this call
2164 gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized);
2165 }
2166
2167 __ jmp(L_copy_bytes);
2168
2169 // Copy trailing qwords
2170 __ BIND(L_copy_8_bytes);
2171 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2172 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2173 __ decrement(qword_count);
2174 __ jcc(Assembler::notZero, L_copy_8_bytes);
2175
2176 if (is_oop) {
2177 __ jmp(L_exit);
2178 } else {
2179 restore_arg_regs();
2180 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2181 __ xorptr(rax, rax); // return 0
2182 __ leave(); // required for proper stackwalking of RuntimeStub frame
2183 __ ret(0);
2184 }
2185
2186 // Copy in multi-bytes chunks
2187 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2188
2189 if (is_oop) {
2190 __ BIND(L_exit);
2191 gen_write_ref_array_post_barrier(to, saved_count, rax);
2192 }
2193 restore_arg_regs();
2194 if (is_oop) {
2195 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2196 } else {
2197 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2198 }
2199 __ xorptr(rax, rax); // return 0
2200 __ leave(); // required for proper stackwalking of RuntimeStub frame
2201 __ ret(0);
2202
2203 return start;
2204 }
2205
2206
2207 // Helper for generating a dynamic type check.
2208 // Smashes no registers.
2209 void generate_type_check(Register sub_klass,
2210 Register super_check_offset,
2211 Register super_klass,
2212 Label& L_success) {
2213 assert_different_registers(sub_klass, super_check_offset, super_klass);
2214
2215 BLOCK_COMMENT("type_check:");
2216
2217 Label L_miss;
2218
2219 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2220 super_check_offset);
2221 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2222
2223 // Fall through on failure!
2224 __ BIND(L_miss);
2225 }
2226
2227 //
2228 // Generate checkcasting array copy stub
2229 //
2230 // Input:
2231 // c_rarg0 - source array address
2232 // c_rarg1 - destination array address
2233 // c_rarg2 - element count, treated as ssize_t, can be zero
2234 // c_rarg3 - size_t ckoff (super_check_offset)
2235 // not Win64
2236 // c_rarg4 - oop ckval (super_klass)
2237 // Win64
2238 // rsp+40 - oop ckval (super_klass)
2239 //
2240 // Output:
2241 // rax == 0 - success
2242 // rax == -1^K - failure, where K is partial transfer count
2243 //
2244 address generate_checkcast_copy(const char *name, address *entry,
2245 bool dest_uninitialized = false) {
2246
2247 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2248
2249 // Input registers (after setup_arg_regs)
2250 const Register from = rdi; // source array address
2251 const Register to = rsi; // destination array address
2252 const Register length = rdx; // elements count
2253 const Register ckoff = rcx; // super_check_offset
2254 const Register ckval = r8; // super_klass
2255
2256 // Registers used as temps (r13, r14 are save-on-entry)
2257 const Register end_from = from; // source array end address
2258 const Register end_to = r13; // destination array end address
2259 const Register count = rdx; // -(count_remaining)
2260 const Register r14_length = r14; // saved copy of length
2261 // End pointers are inclusive, and if length is not zero they point
2262 // to the last unit copied: end_to[0] := end_from[0]
2263
2264 const Register rax_oop = rax; // actual oop copied
2265 const Register r11_klass = r11; // oop._klass
2266
2267 //---------------------------------------------------------------
2268 // Assembler stub will be used for this call to arraycopy
2269 // if the two arrays are subtypes of Object[] but the
2270 // destination array type is not equal to or a supertype
2271 // of the source type. Each element must be separately
2272 // checked.
2273
2274 __ align(CodeEntryAlignment);
2275 StubCodeMark mark(this, "StubRoutines", name);
2276 address start = __ pc();
2277
2278 __ enter(); // required for proper stackwalking of RuntimeStub frame
2279
2280 #ifdef ASSERT
2281 // caller guarantees that the arrays really are different
2282 // otherwise, we would have to make conjoint checks
2283 { Label L;
2284 array_overlap_test(L, TIMES_OOP);
2285 __ stop("checkcast_copy within a single array");
2286 __ bind(L);
2287 }
2288 #endif //ASSERT
2289
2290 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2291 // ckoff => rcx, ckval => r8
2292 // r9 and r10 may be used to save non-volatile registers
2293 #ifdef _WIN64
2294 // last argument (#4) is on stack on Win64
2295 __ movptr(ckval, Address(rsp, 6 * wordSize));
2296 #endif
2297
2298 // Caller of this entry point must set up the argument registers.
2299 if (entry != NULL) {
2300 *entry = __ pc();
2301 BLOCK_COMMENT("Entry:");
2302 }
2303
2304 // allocate spill slots for r13, r14
2305 enum {
2306 saved_r13_offset,
2307 saved_r14_offset,
2308 saved_rbp_offset
2309 };
2310 __ subptr(rsp, saved_rbp_offset * wordSize);
2311 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2312 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2313
2314 // check that int operands are properly extended to size_t
2315 assert_clean_int(length, rax);
2316 assert_clean_int(ckoff, rax);
2317
2318 #ifdef ASSERT
2319 BLOCK_COMMENT("assert consistent ckoff/ckval");
2320 // The ckoff and ckval must be mutually consistent,
2321 // even though caller generates both.
2322 { Label L;
2323 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2324 __ cmpl(ckoff, Address(ckval, sco_offset));
2325 __ jcc(Assembler::equal, L);
2326 __ stop("super_check_offset inconsistent");
2327 __ bind(L);
2328 }
2329 #endif //ASSERT
2330
2331 // Loop-invariant addresses. They are exclusive end pointers.
2332 Address end_from_addr(from, length, TIMES_OOP, 0);
2333 Address end_to_addr(to, length, TIMES_OOP, 0);
2334 // Loop-variant addresses. They assume post-incremented count < 0.
2335 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2336 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2337
2338 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2339
2340 // Copy from low to high addresses, indexed from the end of each array.
2341 __ lea(end_from, end_from_addr);
2342 __ lea(end_to, end_to_addr);
2343 __ movptr(r14_length, length); // save a copy of the length
2344 assert(length == count, ""); // else fix next line:
2345 __ negptr(count); // negate and test the length
2346 __ jcc(Assembler::notZero, L_load_element);
2347
2348 // Empty array: Nothing to do.
2349 __ xorptr(rax, rax); // return 0 on (trivial) success
2350 __ jmp(L_done);
2351
2352 // ======== begin loop ========
2353 // (Loop is rotated; its entry is L_load_element.)
2354 // Loop control:
2355 // for (count = -count; count != 0; count++)
2356 // Base pointers src, dst are biased by 8*(count-1),to last element.
2357 __ align(OptoLoopAlignment);
2358
2359 __ BIND(L_store_element);
2360 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2361 __ increment(count); // increment the count toward zero
2362 __ jcc(Assembler::zero, L_do_card_marks);
2363
2364 // ======== loop entry is here ========
2365 __ BIND(L_load_element);
2366 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2367 __ testptr(rax_oop, rax_oop);
2368 __ jcc(Assembler::zero, L_store_element);
2369
2370 __ load_klass(r11_klass, rax_oop);// query the object klass
2371 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2372 // ======== end loop ========
2373
2374 // It was a real error; we must depend on the caller to finish the job.
2375 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2376 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2377 // and report their number to the caller.
2378 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2379 Label L_post_barrier;
2380 __ addptr(r14_length, count); // K = (original - remaining) oops
2381 __ movptr(rax, r14_length); // save the value
2382 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2383 __ jccb(Assembler::notZero, L_post_barrier);
2384 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2385
2386 // Come here on success only.
2387 __ BIND(L_do_card_marks);
2388 __ xorptr(rax, rax); // return 0 on success
2389
2390 __ BIND(L_post_barrier);
2391 gen_write_ref_array_post_barrier(to, r14_length, rscratch1);
2392
2393 // Common exit point (success or failure).
2394 __ BIND(L_done);
2395 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2396 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2397 restore_arg_regs();
2398 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2399 __ leave(); // required for proper stackwalking of RuntimeStub frame
2400 __ ret(0);
2401
2402 return start;
2403 }
2404
2405 //
2406 // Generate 'unsafe' array copy stub
2407 // Though just as safe as the other stubs, it takes an unscaled
2408 // size_t argument instead of an element count.
2409 //
2410 // Input:
2411 // c_rarg0 - source array address
2412 // c_rarg1 - destination array address
2413 // c_rarg2 - byte count, treated as ssize_t, can be zero
2414 //
2415 // Examines the alignment of the operands and dispatches
2416 // to a long, int, short, or byte copy loop.
2417 //
2418 address generate_unsafe_copy(const char *name,
2419 address byte_copy_entry, address short_copy_entry,
2420 address int_copy_entry, address long_copy_entry) {
2421
2422 Label L_long_aligned, L_int_aligned, L_short_aligned;
2423
2424 // Input registers (before setup_arg_regs)
2425 const Register from = c_rarg0; // source array address
2426 const Register to = c_rarg1; // destination array address
2427 const Register size = c_rarg2; // byte count (size_t)
2428
2429 // Register used as a temp
2430 const Register bits = rax; // test copy of low bits
2431
2432 __ align(CodeEntryAlignment);
2433 StubCodeMark mark(this, "StubRoutines", name);
2434 address start = __ pc();
2435
2436 __ enter(); // required for proper stackwalking of RuntimeStub frame
2437
2438 // bump this on entry, not on exit:
2439 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2440
2441 __ mov(bits, from);
2442 __ orptr(bits, to);
2443 __ orptr(bits, size);
2444
2445 __ testb(bits, BytesPerLong-1);
2446 __ jccb(Assembler::zero, L_long_aligned);
2447
2448 __ testb(bits, BytesPerInt-1);
2449 __ jccb(Assembler::zero, L_int_aligned);
2450
2451 __ testb(bits, BytesPerShort-1);
2452 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2453
2454 __ BIND(L_short_aligned);
2455 __ shrptr(size, LogBytesPerShort); // size => short_count
2456 __ jump(RuntimeAddress(short_copy_entry));
2457
2458 __ BIND(L_int_aligned);
2459 __ shrptr(size, LogBytesPerInt); // size => int_count
2460 __ jump(RuntimeAddress(int_copy_entry));
2461
2462 __ BIND(L_long_aligned);
2463 __ shrptr(size, LogBytesPerLong); // size => qword_count
2464 __ jump(RuntimeAddress(long_copy_entry));
2465
2466 return start;
2467 }
2468
2469 // Perform range checks on the proposed arraycopy.
2470 // Kills temp, but nothing else.
2471 // Also, clean the sign bits of src_pos and dst_pos.
2472 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2473 Register src_pos, // source position (c_rarg1)
2474 Register dst, // destination array oo (c_rarg2)
2475 Register dst_pos, // destination position (c_rarg3)
2476 Register length,
2477 Register temp,
2478 Label& L_failed) {
2479 BLOCK_COMMENT("arraycopy_range_checks:");
2480
2481 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2482 __ movl(temp, length);
2483 __ addl(temp, src_pos); // src_pos + length
2484 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2485 __ jcc(Assembler::above, L_failed);
2486
2487 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2488 __ movl(temp, length);
2489 __ addl(temp, dst_pos); // dst_pos + length
2490 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2491 __ jcc(Assembler::above, L_failed);
2492
2493 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2494 // Move with sign extension can be used since they are positive.
2495 __ movslq(src_pos, src_pos);
2496 __ movslq(dst_pos, dst_pos);
2497
2498 BLOCK_COMMENT("arraycopy_range_checks done");
2499 }
2500
2501 //
2502 // Generate generic array copy stubs
2503 //
2504 // Input:
2505 // c_rarg0 - src oop
2506 // c_rarg1 - src_pos (32-bits)
2507 // c_rarg2 - dst oop
2508 // c_rarg3 - dst_pos (32-bits)
2509 // not Win64
2510 // c_rarg4 - element count (32-bits)
2511 // Win64
2512 // rsp+40 - element count (32-bits)
2513 //
2514 // Output:
2515 // rax == 0 - success
2516 // rax == -1^K - failure, where K is partial transfer count
2517 //
2518 address generate_generic_copy(const char *name,
2519 address byte_copy_entry, address short_copy_entry,
2520 address int_copy_entry, address oop_copy_entry,
2521 address long_copy_entry, address checkcast_copy_entry) {
2522
2523 Label L_failed, L_failed_0, L_objArray;
2524 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2525
2526 // Input registers
2527 const Register src = c_rarg0; // source array oop
2528 const Register src_pos = c_rarg1; // source position
2529 const Register dst = c_rarg2; // destination array oop
2530 const Register dst_pos = c_rarg3; // destination position
2531 #ifndef _WIN64
2532 const Register length = c_rarg4;
2533 #else
2534 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2535 #endif
2536
2537 { int modulus = CodeEntryAlignment;
2538 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2539 int advance = target - (__ offset() % modulus);
2540 if (advance < 0) advance += modulus;
2541 if (advance > 0) __ nop(advance);
2542 }
2543 StubCodeMark mark(this, "StubRoutines", name);
2544
2545 // Short-hop target to L_failed. Makes for denser prologue code.
2546 __ BIND(L_failed_0);
2547 __ jmp(L_failed);
2548 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2549
2550 __ align(CodeEntryAlignment);
2551 address start = __ pc();
2552
2553 __ enter(); // required for proper stackwalking of RuntimeStub frame
2554
2555 // bump this on entry, not on exit:
2556 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2557
2558 //-----------------------------------------------------------------------
2559 // Assembler stub will be used for this call to arraycopy
2560 // if the following conditions are met:
2561 //
2562 // (1) src and dst must not be null.
2563 // (2) src_pos must not be negative.
2564 // (3) dst_pos must not be negative.
2565 // (4) length must not be negative.
2566 // (5) src klass and dst klass should be the same and not NULL.
2567 // (6) src and dst should be arrays.
2568 // (7) src_pos + length must not exceed length of src.
2569 // (8) dst_pos + length must not exceed length of dst.
2570 //
2571
2572 // if (src == NULL) return -1;
2573 __ testptr(src, src); // src oop
2574 size_t j1off = __ offset();
2575 __ jccb(Assembler::zero, L_failed_0);
2576
2577 // if (src_pos < 0) return -1;
2578 __ testl(src_pos, src_pos); // src_pos (32-bits)
2579 __ jccb(Assembler::negative, L_failed_0);
2580
2581 // if (dst == NULL) return -1;
2582 __ testptr(dst, dst); // dst oop
2583 __ jccb(Assembler::zero, L_failed_0);
2584
2585 // if (dst_pos < 0) return -1;
2586 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2587 size_t j4off = __ offset();
2588 __ jccb(Assembler::negative, L_failed_0);
2589
2590 // The first four tests are very dense code,
2591 // but not quite dense enough to put four
2592 // jumps in a 16-byte instruction fetch buffer.
2593 // That's good, because some branch predicters
2594 // do not like jumps so close together.
2595 // Make sure of this.
2596 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2597
2598 // registers used as temp
2599 const Register r11_length = r11; // elements count to copy
2600 const Register r10_src_klass = r10; // array klass
2601
2602 // if (length < 0) return -1;
2603 __ movl(r11_length, length); // length (elements count, 32-bits value)
2604 __ testl(r11_length, r11_length);
2605 __ jccb(Assembler::negative, L_failed_0);
2606
2607 __ load_klass(r10_src_klass, src);
2608 #ifdef ASSERT
2609 // assert(src->klass() != NULL);
2610 {
2611 BLOCK_COMMENT("assert klasses not null {");
2612 Label L1, L2;
2613 __ testptr(r10_src_klass, r10_src_klass);
2614 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2615 __ bind(L1);
2616 __ stop("broken null klass");
2617 __ bind(L2);
2618 __ load_klass(rax, dst);
2619 __ cmpq(rax, 0);
2620 __ jcc(Assembler::equal, L1); // this would be broken also
2621 BLOCK_COMMENT("} assert klasses not null done");
2622 }
2623 #endif
2624
2625 // Load layout helper (32-bits)
2626 //
2627 // |array_tag| | header_size | element_type | |log2_element_size|
2628 // 32 30 24 16 8 2 0
2629 //
2630 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2631 //
2632
2633 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2634
2635 // Handle objArrays completely differently...
2636 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2637 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2638 __ jcc(Assembler::equal, L_objArray);
2639
2640 // if (src->klass() != dst->klass()) return -1;
2641 __ load_klass(rax, dst);
2642 __ cmpq(r10_src_klass, rax);
2643 __ jcc(Assembler::notEqual, L_failed);
2644
2645 const Register rax_lh = rax; // layout helper
2646 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2647
2648 // if (!src->is_Array()) return -1;
2649 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2650 __ jcc(Assembler::greaterEqual, L_failed);
2651
2652 // At this point, it is known to be a typeArray (array_tag 0x3).
2653 #ifdef ASSERT
2654 {
2655 BLOCK_COMMENT("assert primitive array {");
2656 Label L;
2657 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2658 __ jcc(Assembler::greaterEqual, L);
2659 __ stop("must be a primitive array");
2660 __ bind(L);
2661 BLOCK_COMMENT("} assert primitive array done");
2662 }
2663 #endif
2664
2665 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2666 r10, L_failed);
2667
2668 // TypeArrayKlass
2669 //
2670 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2671 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2672 //
2673
2674 const Register r10_offset = r10; // array offset
2675 const Register rax_elsize = rax_lh; // element size
2676
2677 __ movl(r10_offset, rax_lh);
2678 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2679 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2680 __ addptr(src, r10_offset); // src array offset
2681 __ addptr(dst, r10_offset); // dst array offset
2682 BLOCK_COMMENT("choose copy loop based on element size");
2683 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2684
2685 // next registers should be set before the jump to corresponding stub
2686 const Register from = c_rarg0; // source array address
2687 const Register to = c_rarg1; // destination array address
2688 const Register count = c_rarg2; // elements count
2689
2690 // 'from', 'to', 'count' registers should be set in such order
2691 // since they are the same as 'src', 'src_pos', 'dst'.
2692
2693 __ BIND(L_copy_bytes);
2694 __ cmpl(rax_elsize, 0);
2695 __ jccb(Assembler::notEqual, L_copy_shorts);
2696 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2697 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2698 __ movl2ptr(count, r11_length); // length
2699 __ jump(RuntimeAddress(byte_copy_entry));
2700
2701 __ BIND(L_copy_shorts);
2702 __ cmpl(rax_elsize, LogBytesPerShort);
2703 __ jccb(Assembler::notEqual, L_copy_ints);
2704 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2705 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2706 __ movl2ptr(count, r11_length); // length
2707 __ jump(RuntimeAddress(short_copy_entry));
2708
2709 __ BIND(L_copy_ints);
2710 __ cmpl(rax_elsize, LogBytesPerInt);
2711 __ jccb(Assembler::notEqual, L_copy_longs);
2712 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2713 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2714 __ movl2ptr(count, r11_length); // length
2715 __ jump(RuntimeAddress(int_copy_entry));
2716
2717 __ BIND(L_copy_longs);
2718 #ifdef ASSERT
2719 {
2720 BLOCK_COMMENT("assert long copy {");
2721 Label L;
2722 __ cmpl(rax_elsize, LogBytesPerLong);
2723 __ jcc(Assembler::equal, L);
2724 __ stop("must be long copy, but elsize is wrong");
2725 __ bind(L);
2726 BLOCK_COMMENT("} assert long copy done");
2727 }
2728 #endif
2729 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2730 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2731 __ movl2ptr(count, r11_length); // length
2732 __ jump(RuntimeAddress(long_copy_entry));
2733
2734 // ObjArrayKlass
2735 __ BIND(L_objArray);
2736 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2737
2738 Label L_plain_copy, L_checkcast_copy;
2739 // test array classes for subtyping
2740 __ load_klass(rax, dst);
2741 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2742 __ jcc(Assembler::notEqual, L_checkcast_copy);
2743
2744 // Identically typed arrays can be copied without element-wise checks.
2745 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2746 r10, L_failed);
2747
2748 __ lea(from, Address(src, src_pos, TIMES_OOP,
2749 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2750 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2751 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2752 __ movl2ptr(count, r11_length); // length
2753 __ BIND(L_plain_copy);
2754 __ jump(RuntimeAddress(oop_copy_entry));
2755
2756 __ BIND(L_checkcast_copy);
2757 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2758 {
2759 // Before looking at dst.length, make sure dst is also an objArray.
2760 __ cmpl(Address(rax, lh_offset), objArray_lh);
2761 __ jcc(Assembler::notEqual, L_failed);
2762
2763 // It is safe to examine both src.length and dst.length.
2764 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2765 rax, L_failed);
2766
2767 const Register r11_dst_klass = r11;
2768 __ load_klass(r11_dst_klass, dst); // reload
2769
2770 // Marshal the base address arguments now, freeing registers.
2771 __ lea(from, Address(src, src_pos, TIMES_OOP,
2772 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2773 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2774 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2775 __ movl(count, length); // length (reloaded)
2776 Register sco_temp = c_rarg3; // this register is free now
2777 assert_different_registers(from, to, count, sco_temp,
2778 r11_dst_klass, r10_src_klass);
2779 assert_clean_int(count, sco_temp);
2780
2781 // Generate the type check.
2782 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2783 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2784 assert_clean_int(sco_temp, rax);
2785 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2786
2787 // Fetch destination element klass from the ObjArrayKlass header.
2788 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2789 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2790 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2791 assert_clean_int(sco_temp, rax);
2792
2793 // the checkcast_copy loop needs two extra arguments:
2794 assert(c_rarg3 == sco_temp, "#3 already in place");
2795 // Set up arguments for checkcast_copy_entry.
2796 setup_arg_regs(4);
2797 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2798 __ jump(RuntimeAddress(checkcast_copy_entry));
2799 }
2800
2801 __ BIND(L_failed);
2802 __ xorptr(rax, rax);
2803 __ notptr(rax); // return -1
2804 __ leave(); // required for proper stackwalking of RuntimeStub frame
2805 __ ret(0);
2806
2807 return start;
2808 }
2809
2810 void generate_arraycopy_stubs() {
2811 address entry;
2812 address entry_jbyte_arraycopy;
2813 address entry_jshort_arraycopy;
2814 address entry_jint_arraycopy;
2815 address entry_oop_arraycopy;
2816 address entry_jlong_arraycopy;
2817 address entry_checkcast_arraycopy;
2818
2819 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2820 "jbyte_disjoint_arraycopy");
2821 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2822 "jbyte_arraycopy");
2823
2824 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2825 "jshort_disjoint_arraycopy");
2826 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2827 "jshort_arraycopy");
2828
2829 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2830 "jint_disjoint_arraycopy");
2831 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2832 &entry_jint_arraycopy, "jint_arraycopy");
2833
2834 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2835 "jlong_disjoint_arraycopy");
2836 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2837 &entry_jlong_arraycopy, "jlong_arraycopy");
2838
2839
2840 if (UseCompressedOops) {
2841 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2842 "oop_disjoint_arraycopy");
2843 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2844 &entry_oop_arraycopy, "oop_arraycopy");
2845 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2846 "oop_disjoint_arraycopy_uninit",
2847 /*dest_uninitialized*/true);
2848 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2849 NULL, "oop_arraycopy_uninit",
2850 /*dest_uninitialized*/true);
2851 } else {
2852 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2853 "oop_disjoint_arraycopy");
2854 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2855 &entry_oop_arraycopy, "oop_arraycopy");
2856 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2857 "oop_disjoint_arraycopy_uninit",
2858 /*dest_uninitialized*/true);
2859 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2860 NULL, "oop_arraycopy_uninit",
2861 /*dest_uninitialized*/true);
2862 }
2863
2864 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2865 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2866 /*dest_uninitialized*/true);
2867
2868 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2869 entry_jbyte_arraycopy,
2870 entry_jshort_arraycopy,
2871 entry_jint_arraycopy,
2872 entry_jlong_arraycopy);
2873 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2874 entry_jbyte_arraycopy,
2875 entry_jshort_arraycopy,
2876 entry_jint_arraycopy,
2877 entry_oop_arraycopy,
2878 entry_jlong_arraycopy,
2879 entry_checkcast_arraycopy);
2880
2881 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2882 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2883 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2884 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2885 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2886 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2887
2888 // We don't generate specialized code for HeapWord-aligned source
2889 // arrays, so just use the code we've already generated
2890 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2891 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2892
2893 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2894 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2895
2896 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2897 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2898
2899 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2900 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2901
2902 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2903 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2904
2905 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2906 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2907 }
2908
2909 void generate_math_stubs() {
2910 {
2911 StubCodeMark mark(this, "StubRoutines", "log");
2912 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2913
2914 __ subq(rsp, 8);
2915 __ movdbl(Address(rsp, 0), xmm0);
2916 __ fld_d(Address(rsp, 0));
2917 __ flog();
2918 __ fstp_d(Address(rsp, 0));
2919 __ movdbl(xmm0, Address(rsp, 0));
2920 __ addq(rsp, 8);
2921 __ ret(0);
2922 }
2923 {
2924 StubCodeMark mark(this, "StubRoutines", "log10");
2925 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2926
2927 __ subq(rsp, 8);
2928 __ movdbl(Address(rsp, 0), xmm0);
2929 __ fld_d(Address(rsp, 0));
2930 __ flog10();
2931 __ fstp_d(Address(rsp, 0));
2932 __ movdbl(xmm0, Address(rsp, 0));
2933 __ addq(rsp, 8);
2934 __ ret(0);
2935 }
2936 {
2937 StubCodeMark mark(this, "StubRoutines", "sin");
2938 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2939
2940 __ subq(rsp, 8);
2941 __ movdbl(Address(rsp, 0), xmm0);
2942 __ fld_d(Address(rsp, 0));
2943 __ trigfunc('s');
2944 __ fstp_d(Address(rsp, 0));
2945 __ movdbl(xmm0, Address(rsp, 0));
2946 __ addq(rsp, 8);
2947 __ ret(0);
2948 }
2949 {
2950 StubCodeMark mark(this, "StubRoutines", "cos");
2951 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2952
2953 __ subq(rsp, 8);
2954 __ movdbl(Address(rsp, 0), xmm0);
2955 __ fld_d(Address(rsp, 0));
2956 __ trigfunc('c');
2957 __ fstp_d(Address(rsp, 0));
2958 __ movdbl(xmm0, Address(rsp, 0));
2959 __ addq(rsp, 8);
2960 __ ret(0);
2961 }
2962 {
2963 StubCodeMark mark(this, "StubRoutines", "tan");
2964 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2965
2966 __ subq(rsp, 8);
2967 __ movdbl(Address(rsp, 0), xmm0);
2968 __ fld_d(Address(rsp, 0));
2969 __ trigfunc('t');
2970 __ fstp_d(Address(rsp, 0));
2971 __ movdbl(xmm0, Address(rsp, 0));
2972 __ addq(rsp, 8);
2973 __ ret(0);
2974 }
2975 {
2976 StubCodeMark mark(this, "StubRoutines", "exp");
2977 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
2978
2979 __ subq(rsp, 8);
2980 __ movdbl(Address(rsp, 0), xmm0);
2981 __ fld_d(Address(rsp, 0));
2982 __ exp_with_fallback(0);
2983 __ fstp_d(Address(rsp, 0));
2984 __ movdbl(xmm0, Address(rsp, 0));
2985 __ addq(rsp, 8);
2986 __ ret(0);
2987 }
2988 {
2989 StubCodeMark mark(this, "StubRoutines", "pow");
2990 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
2991
2992 __ subq(rsp, 8);
2993 __ movdbl(Address(rsp, 0), xmm1);
2994 __ fld_d(Address(rsp, 0));
2995 __ movdbl(Address(rsp, 0), xmm0);
2996 __ fld_d(Address(rsp, 0));
2997 __ pow_with_fallback(0);
2998 __ fstp_d(Address(rsp, 0));
2999 __ movdbl(xmm0, Address(rsp, 0));
3000 __ addq(rsp, 8);
3001 __ ret(0);
3002 }
3003 }
3004
3005 // AES intrinsic stubs
3006 enum {AESBlockSize = 16};
3007
3008 address generate_key_shuffle_mask() {
3009 __ align(16);
3010 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
3011 address start = __ pc();
3012 __ emit_data64( 0x0405060700010203, relocInfo::none );
3013 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
3014 return start;
3015 }
3016
3017 // Utility routine for loading a 128-bit key word in little endian format
3018 // can optionally specify that the shuffle mask is already in an xmmregister
3019 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
3020 __ movdqu(xmmdst, Address(key, offset));
3021 if (xmm_shuf_mask != NULL) {
3022 __ pshufb(xmmdst, xmm_shuf_mask);
3023 } else {
3024 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3025 }
3026 }
3027
3028 // Arguments:
3029 //
3030 // Inputs:
3031 // c_rarg0 - source byte array address
3032 // c_rarg1 - destination byte array address
3033 // c_rarg2 - K (key) in little endian int array
3034 //
3035 address generate_aescrypt_encryptBlock() {
3036 assert(UseAES, "need AES instructions and misaligned SSE support");
3037 __ align(CodeEntryAlignment);
3038 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
3039 Label L_doLast;
3040 address start = __ pc();
3041
3042 const Register from = c_rarg0; // source array address
3043 const Register to = c_rarg1; // destination array address
3044 const Register key = c_rarg2; // key array address
3045 const Register keylen = rax;
3046
3047 const XMMRegister xmm_result = xmm0;
3048 const XMMRegister xmm_key_shuf_mask = xmm1;
3049 // On win64 xmm6-xmm15 must be preserved so don't use them.
3050 const XMMRegister xmm_temp1 = xmm2;
3051 const XMMRegister xmm_temp2 = xmm3;
3052 const XMMRegister xmm_temp3 = xmm4;
3053 const XMMRegister xmm_temp4 = xmm5;
3054
3055 __ enter(); // required for proper stackwalking of RuntimeStub frame
3056
3057 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3058 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3059
3060 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3061 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
3062
3063 // For encryption, the java expanded key ordering is just what we need
3064 // we don't know if the key is aligned, hence not using load-execute form
3065
3066 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3067 __ pxor(xmm_result, xmm_temp1);
3068
3069 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3070 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3071 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3072 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3073
3074 __ aesenc(xmm_result, xmm_temp1);
3075 __ aesenc(xmm_result, xmm_temp2);
3076 __ aesenc(xmm_result, xmm_temp3);
3077 __ aesenc(xmm_result, xmm_temp4);
3078
3079 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3080 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3081 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3082 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3083
3084 __ aesenc(xmm_result, xmm_temp1);
3085 __ aesenc(xmm_result, xmm_temp2);
3086 __ aesenc(xmm_result, xmm_temp3);
3087 __ aesenc(xmm_result, xmm_temp4);
3088
3089 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3090 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3091
3092 __ cmpl(keylen, 44);
3093 __ jccb(Assembler::equal, L_doLast);
3094
3095 __ aesenc(xmm_result, xmm_temp1);
3096 __ aesenc(xmm_result, xmm_temp2);
3097
3098 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3099 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3100
3101 __ cmpl(keylen, 52);
3102 __ jccb(Assembler::equal, L_doLast);
3103
3104 __ aesenc(xmm_result, xmm_temp1);
3105 __ aesenc(xmm_result, xmm_temp2);
3106
3107 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3108 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3109
3110 __ BIND(L_doLast);
3111 __ aesenc(xmm_result, xmm_temp1);
3112 __ aesenclast(xmm_result, xmm_temp2);
3113 __ movdqu(Address(to, 0), xmm_result); // store the result
3114 __ xorptr(rax, rax); // return 0
3115 __ leave(); // required for proper stackwalking of RuntimeStub frame
3116 __ ret(0);
3117
3118 return start;
3119 }
3120
3121
3122 // Arguments:
3123 //
3124 // Inputs:
3125 // c_rarg0 - source byte array address
3126 // c_rarg1 - destination byte array address
3127 // c_rarg2 - K (key) in little endian int array
3128 //
3129 address generate_aescrypt_decryptBlock() {
3130 assert(UseAES, "need AES instructions and misaligned SSE support");
3131 __ align(CodeEntryAlignment);
3132 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3133 Label L_doLast;
3134 address start = __ pc();
3135
3136 const Register from = c_rarg0; // source array address
3137 const Register to = c_rarg1; // destination array address
3138 const Register key = c_rarg2; // key array address
3139 const Register keylen = rax;
3140
3141 const XMMRegister xmm_result = xmm0;
3142 const XMMRegister xmm_key_shuf_mask = xmm1;
3143 // On win64 xmm6-xmm15 must be preserved so don't use them.
3144 const XMMRegister xmm_temp1 = xmm2;
3145 const XMMRegister xmm_temp2 = xmm3;
3146 const XMMRegister xmm_temp3 = xmm4;
3147 const XMMRegister xmm_temp4 = xmm5;
3148
3149 __ enter(); // required for proper stackwalking of RuntimeStub frame
3150
3151 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3152 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3153
3154 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3155 __ movdqu(xmm_result, Address(from, 0));
3156
3157 // for decryption java expanded key ordering is rotated one position from what we want
3158 // so we start from 0x10 here and hit 0x00 last
3159 // we don't know if the key is aligned, hence not using load-execute form
3160 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3161 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3162 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3163 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3164
3165 __ pxor (xmm_result, xmm_temp1);
3166 __ aesdec(xmm_result, xmm_temp2);
3167 __ aesdec(xmm_result, xmm_temp3);
3168 __ aesdec(xmm_result, xmm_temp4);
3169
3170 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3171 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3172 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3173 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3174
3175 __ aesdec(xmm_result, xmm_temp1);
3176 __ aesdec(xmm_result, xmm_temp2);
3177 __ aesdec(xmm_result, xmm_temp3);
3178 __ aesdec(xmm_result, xmm_temp4);
3179
3180 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3181 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3182 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3183
3184 __ cmpl(keylen, 44);
3185 __ jccb(Assembler::equal, L_doLast);
3186
3187 __ aesdec(xmm_result, xmm_temp1);
3188 __ aesdec(xmm_result, xmm_temp2);
3189
3190 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3191 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3192
3193 __ cmpl(keylen, 52);
3194 __ jccb(Assembler::equal, L_doLast);
3195
3196 __ aesdec(xmm_result, xmm_temp1);
3197 __ aesdec(xmm_result, xmm_temp2);
3198
3199 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3200 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3201
3202 __ BIND(L_doLast);
3203 __ aesdec(xmm_result, xmm_temp1);
3204 __ aesdec(xmm_result, xmm_temp2);
3205
3206 // for decryption the aesdeclast operation is always on key+0x00
3207 __ aesdeclast(xmm_result, xmm_temp3);
3208 __ movdqu(Address(to, 0), xmm_result); // store the result
3209 __ xorptr(rax, rax); // return 0
3210 __ leave(); // required for proper stackwalking of RuntimeStub frame
3211 __ ret(0);
3212
3213 return start;
3214 }
3215
3216
3217 // Arguments:
3218 //
3219 // Inputs:
3220 // c_rarg0 - source byte array address
3221 // c_rarg1 - destination byte array address
3222 // c_rarg2 - K (key) in little endian int array
3223 // c_rarg3 - r vector byte array address
3224 // c_rarg4 - input length
3225 //
3226 // Output:
3227 // rax - input length
3228 //
3229 address generate_cipherBlockChaining_encryptAESCrypt() {
3230 assert(UseAES, "need AES instructions and misaligned SSE support");
3231 __ align(CodeEntryAlignment);
3232 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3233 address start = __ pc();
3234
3235 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3236 const Register from = c_rarg0; // source array address
3237 const Register to = c_rarg1; // destination array address
3238 const Register key = c_rarg2; // key array address
3239 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3240 // and left with the results of the last encryption block
3241 #ifndef _WIN64
3242 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3243 #else
3244 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3245 const Register len_reg = r10; // pick the first volatile windows register
3246 #endif
3247 const Register pos = rax;
3248
3249 // xmm register assignments for the loops below
3250 const XMMRegister xmm_result = xmm0;
3251 const XMMRegister xmm_temp = xmm1;
3252 // keys 0-10 preloaded into xmm2-xmm12
3253 const int XMM_REG_NUM_KEY_FIRST = 2;
3254 const int XMM_REG_NUM_KEY_LAST = 15;
3255 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3256 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3257 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3258 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3259 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3260
3261 __ enter(); // required for proper stackwalking of RuntimeStub frame
3262
3263 #ifdef _WIN64
3264 // on win64, fill len_reg from stack position
3265 __ movl(len_reg, len_mem);
3266 // save the xmm registers which must be preserved 6-15
3267 __ subptr(rsp, -rsp_after_call_off * wordSize);
3268 for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3269 __ movdqu(xmm_save(i), as_XMMRegister(i));
3270 }
3271 #else
3272 __ push(len_reg); // Save
3273 #endif
3274
3275 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3276 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3277 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3278 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3279 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3280 offset += 0x10;
3281 }
3282 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3283
3284 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3285 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3286 __ cmpl(rax, 44);
3287 __ jcc(Assembler::notEqual, L_key_192_256);
3288
3289 // 128 bit code follows here
3290 __ movptr(pos, 0);
3291 __ align(OptoLoopAlignment);
3292
3293 __ BIND(L_loopTop_128);
3294 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3295 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3296 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3297 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3298 __ aesenc(xmm_result, as_XMMRegister(rnum));
3299 }
3300 __ aesenclast(xmm_result, xmm_key10);
3301 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3302 // no need to store r to memory until we exit
3303 __ addptr(pos, AESBlockSize);
3304 __ subptr(len_reg, AESBlockSize);
3305 __ jcc(Assembler::notEqual, L_loopTop_128);
3306
3307 __ BIND(L_exit);
3308 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3309
3310 #ifdef _WIN64
3311 // restore xmm regs belonging to calling function
3312 for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3313 __ movdqu(as_XMMRegister(i), xmm_save(i));
3314 }
3315 __ movl(rax, len_mem);
3316 #else
3317 __ pop(rax); // return length
3318 #endif
3319 __ leave(); // required for proper stackwalking of RuntimeStub frame
3320 __ ret(0);
3321
3322 __ BIND(L_key_192_256);
3323 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3324 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3325 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3326 __ cmpl(rax, 52);
3327 __ jcc(Assembler::notEqual, L_key_256);
3328
3329 // 192-bit code follows here (could be changed to use more xmm registers)
3330 __ movptr(pos, 0);
3331 __ align(OptoLoopAlignment);
3332
3333 __ BIND(L_loopTop_192);
3334 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3335 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3336 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3337 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3338 __ aesenc(xmm_result, as_XMMRegister(rnum));
3339 }
3340 __ aesenclast(xmm_result, xmm_key12);
3341 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3342 // no need to store r to memory until we exit
3343 __ addptr(pos, AESBlockSize);
3344 __ subptr(len_reg, AESBlockSize);
3345 __ jcc(Assembler::notEqual, L_loopTop_192);
3346 __ jmp(L_exit);
3347
3348 __ BIND(L_key_256);
3349 // 256-bit code follows here (could be changed to use more xmm registers)
3350 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3351 __ movptr(pos, 0);
3352 __ align(OptoLoopAlignment);
3353
3354 __ BIND(L_loopTop_256);
3355 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3356 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3357 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3358 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3359 __ aesenc(xmm_result, as_XMMRegister(rnum));
3360 }
3361 load_key(xmm_temp, key, 0xe0);
3362 __ aesenclast(xmm_result, xmm_temp);
3363 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3364 // no need to store r to memory until we exit
3365 __ addptr(pos, AESBlockSize);
3366 __ subptr(len_reg, AESBlockSize);
3367 __ jcc(Assembler::notEqual, L_loopTop_256);
3368 __ jmp(L_exit);
3369
3370 return start;
3371 }
3372
3373 // Safefetch stubs.
3374 void generate_safefetch(const char* name, int size, address* entry,
3375 address* fault_pc, address* continuation_pc) {
3376 // safefetch signatures:
3377 // int SafeFetch32(int* adr, int errValue);
3378 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3379 //
3380 // arguments:
3381 // c_rarg0 = adr
3382 // c_rarg1 = errValue
3383 //
3384 // result:
3385 // PPC_RET = *adr or errValue
3386
3387 StubCodeMark mark(this, "StubRoutines", name);
3388
3389 // Entry point, pc or function descriptor.
3390 *entry = __ pc();
3391
3392 // Load *adr into c_rarg1, may fault.
3393 *fault_pc = __ pc();
3394 switch (size) {
3395 case 4:
3396 // int32_t
3397 __ movl(c_rarg1, Address(c_rarg0, 0));
3398 break;
3399 case 8:
3400 // int64_t
3401 __ movq(c_rarg1, Address(c_rarg0, 0));
3402 break;
3403 default:
3404 ShouldNotReachHere();
3405 }
3406
3407 // return errValue or *adr
3408 *continuation_pc = __ pc();
3409 __ movq(rax, c_rarg1);
3410 __ ret(0);
3411 }
3412
3413 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3414 // to hide instruction latency
3415 //
3416 // Arguments:
3417 //
3418 // Inputs:
3419 // c_rarg0 - source byte array address
3420 // c_rarg1 - destination byte array address
3421 // c_rarg2 - K (key) in little endian int array
3422 // c_rarg3 - r vector byte array address
3423 // c_rarg4 - input length
3424 //
3425 // Output:
3426 // rax - input length
3427 //
3428
3429 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3430 assert(UseAES, "need AES instructions and misaligned SSE support");
3431 __ align(CodeEntryAlignment);
3432 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3433 address start = __ pc();
3434
3435 Label L_exit, L_key_192_256, L_key_256;
3436 Label L_singleBlock_loopTop_128, L_multiBlock_loopTop_128;
3437 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
3438 const Register from = c_rarg0; // source array address
3439 const Register to = c_rarg1; // destination array address
3440 const Register key = c_rarg2; // key array address
3441 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3442 // and left with the results of the last encryption block
3443 #ifndef _WIN64
3444 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3445 #else
3446 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3447 const Register len_reg = r10; // pick the first volatile windows register
3448 #endif
3449 const Register pos = rax;
3450
3451 // keys 0-10 preloaded into xmm2-xmm12
3452 const int XMM_REG_NUM_KEY_FIRST = 5;
3453 const int XMM_REG_NUM_KEY_LAST = 15;
3454 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3455 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3456
3457 __ enter(); // required for proper stackwalking of RuntimeStub frame
3458
3459 #ifdef _WIN64
3460 // on win64, fill len_reg from stack position
3461 __ movl(len_reg, len_mem);
3462 // save the xmm registers which must be preserved 6-15
3463 __ subptr(rsp, -rsp_after_call_off * wordSize);
3464 for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3465 __ movdqu(xmm_save(i), as_XMMRegister(i));
3466 }
3467 #else
3468 __ push(len_reg); // Save
3469 #endif
3470
3471 // the java expanded key ordering is rotated one position from what we want
3472 // so we start from 0x10 here and hit 0x00 last
3473 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3474 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3475 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3476 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3477 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3478 offset += 0x10;
3479 }
3480 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3481
3482 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3483
3484 // registers holding the four results in the parallelized loop
3485 const XMMRegister xmm_result0 = xmm0;
3486 const XMMRegister xmm_result1 = xmm2;
3487 const XMMRegister xmm_result2 = xmm3;
3488 const XMMRegister xmm_result3 = xmm4;
3489
3490 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3491
3492 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3493 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3494 __ cmpl(rax, 44);
3495 __ jcc(Assembler::notEqual, L_key_192_256);
3496
3497
3498 // 128-bit code follows here, parallelized
3499 __ movptr(pos, 0);
3500 __ align(OptoLoopAlignment);
3501 __ BIND(L_multiBlock_loopTop_128);
3502 __ cmpptr(len_reg, 4*AESBlockSize); // see if at least 4 blocks left
3503 __ jcc(Assembler::less, L_singleBlock_loopTop_128);
3504
3505 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0*AESBlockSize)); // get next 4 blocks into xmmresult registers
3506 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1*AESBlockSize));
3507 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2*AESBlockSize));
3508 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3*AESBlockSize));
3509
3510 #define DoFour(opc, src_reg) \
3511 __ opc(xmm_result0, src_reg); \
3512 __ opc(xmm_result1, src_reg); \
3513 __ opc(xmm_result2, src_reg); \
3514 __ opc(xmm_result3, src_reg);
3515
3516 DoFour(pxor, xmm_key_first);
3517 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3518 DoFour(aesdec, as_XMMRegister(rnum));
3519 }
3520 DoFour(aesdeclast, xmm_key_last);
3521 // for each result, xor with the r vector of previous cipher block
3522 __ pxor(xmm_result0, xmm_prev_block_cipher);
3523 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0*AESBlockSize));
3524 __ pxor(xmm_result1, xmm_prev_block_cipher);
3525 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1*AESBlockSize));
3526 __ pxor(xmm_result2, xmm_prev_block_cipher);
3527 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2*AESBlockSize));
3528 __ pxor(xmm_result3, xmm_prev_block_cipher);
3529 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3*AESBlockSize)); // this will carry over to next set of blocks
3530
3531 __ movdqu(Address(to, pos, Address::times_1, 0*AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3532 __ movdqu(Address(to, pos, Address::times_1, 1*AESBlockSize), xmm_result1);
3533 __ movdqu(Address(to, pos, Address::times_1, 2*AESBlockSize), xmm_result2);
3534 __ movdqu(Address(to, pos, Address::times_1, 3*AESBlockSize), xmm_result3);
3535
3536 __ addptr(pos, 4*AESBlockSize);
3537 __ subptr(len_reg, 4*AESBlockSize);
3538 __ jmp(L_multiBlock_loopTop_128);
3539
3540 // registers used in the non-parallelized loops
3541 // xmm register assignments for the loops below
3542 const XMMRegister xmm_result = xmm0;
3543 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3544 const XMMRegister xmm_key11 = xmm3;
3545 const XMMRegister xmm_key12 = xmm4;
3546 const XMMRegister xmm_temp = xmm4;
3547
3548 __ align(OptoLoopAlignment);
3549 __ BIND(L_singleBlock_loopTop_128);
3550 __ cmpptr(len_reg, 0); // any blocks left??
3551 __ jcc(Assembler::equal, L_exit);
3552 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3553 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3554 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
3555 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3556 __ aesdec(xmm_result, as_XMMRegister(rnum));
3557 }
3558 __ aesdeclast(xmm_result, xmm_key_last);
3559 __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3560 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3561 // no need to store r to memory until we exit
3562 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3563
3564 __ addptr(pos, AESBlockSize);
3565 __ subptr(len_reg, AESBlockSize);
3566 __ jmp(L_singleBlock_loopTop_128);
3567
3568
3569 __ BIND(L_exit);
3570 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3571 #ifdef _WIN64
3572 // restore regs belonging to calling function
3573 for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
3574 __ movdqu(as_XMMRegister(i), xmm_save(i));
3575 }
3576 __ movl(rax, len_mem);
3577 #else
3578 __ pop(rax); // return length
3579 #endif
3580 __ leave(); // required for proper stackwalking of RuntimeStub frame
3581 __ ret(0);
3582
3583
3584 __ BIND(L_key_192_256);
3585 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3586 load_key(xmm_key11, key, 0xb0);
3587 __ cmpl(rax, 52);
3588 __ jcc(Assembler::notEqual, L_key_256);
3589
3590 // 192-bit code follows here (could be optimized to use parallelism)
3591 load_key(xmm_key12, key, 0xc0); // 192-bit key goes up to c0
3592 __ movptr(pos, 0);
3593 __ align(OptoLoopAlignment);
3594
3595 __ BIND(L_singleBlock_loopTop_192);
3596 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3597 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3598 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
3599 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3600 __ aesdec(xmm_result, as_XMMRegister(rnum));
3601 }
3602 __ aesdec(xmm_result, xmm_key11);
3603 __ aesdec(xmm_result, xmm_key12);
3604 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3605 __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3606 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3607 // no need to store r to memory until we exit
3608 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3609 __ addptr(pos, AESBlockSize);
3610 __ subptr(len_reg, AESBlockSize);
3611 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
3612 __ jmp(L_exit);
3613
3614 __ BIND(L_key_256);
3615 // 256-bit code follows here (could be optimized to use parallelism)
3616 __ movptr(pos, 0);
3617 __ align(OptoLoopAlignment);
3618
3619 __ BIND(L_singleBlock_loopTop_256);
3620 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3621 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3622 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
3623 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
3624 __ aesdec(xmm_result, as_XMMRegister(rnum));
3625 }
3626 __ aesdec(xmm_result, xmm_key11);
3627 load_key(xmm_temp, key, 0xc0);
3628 __ aesdec(xmm_result, xmm_temp);
3629 load_key(xmm_temp, key, 0xd0);
3630 __ aesdec(xmm_result, xmm_temp);
3631 load_key(xmm_temp, key, 0xe0); // 256-bit key goes up to e0
3632 __ aesdec(xmm_result, xmm_temp);
3633 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 came from key+0
3634 __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3635 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3636 // no need to store r to memory until we exit
3637 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3638 __ addptr(pos, AESBlockSize);
3639 __ subptr(len_reg, AESBlockSize);
3640 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
3641 __ jmp(L_exit);
3642
3643 return start;
3644 }
3645
3646 /**
3647 * Arguments:
3648 *
3649 * Inputs:
3650 * c_rarg0 - int crc
3651 * c_rarg1 - byte* buf
3652 * c_rarg2 - int length
3653 *
3654 * Ouput:
3655 * rax - int crc result
3656 */
3657 address generate_updateBytesCRC32() {
3658 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3659
3660 __ align(CodeEntryAlignment);
3661 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
3662
3663 address start = __ pc();
3664 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3665 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3666 // rscratch1: r10
3667 const Register crc = c_rarg0; // crc
3668 const Register buf = c_rarg1; // source java byte array address
3669 const Register len = c_rarg2; // length
3670 const Register table = c_rarg3; // crc_table address (reuse register)
3671 const Register tmp = r11;
3672 assert_different_registers(crc, buf, len, table, tmp, rax);
3673
3674 BLOCK_COMMENT("Entry:");
3675 __ enter(); // required for proper stackwalking of RuntimeStub frame
3676
3677 __ kernel_crc32(crc, buf, len, table, tmp);
3678
3679 __ movl(rax, crc);
3680 __ leave(); // required for proper stackwalking of RuntimeStub frame
3681 __ ret(0);
3682
3683 return start;
3684 }
3685
3686
3687 /**
3688 * Arguments:
3689 *
3690 * Input:
3691 * c_rarg0 - x address
3692 * c_rarg1 - x length
3693 * c_rarg2 - y address
3694 * c_rarg3 - y lenth
3695 * not Win64
3696 * c_rarg4 - z address
3697 * c_rarg5 - z length
3698 * Win64
3699 * rsp+40 - z address
3700 * rsp+48 - z length
3701 */
3702 address generate_multiplyToLen() {
3703 __ align(CodeEntryAlignment);
3704 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
3705
3706 address start = __ pc();
3707 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3708 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3709 const Register x = rdi;
3710 const Register xlen = rax;
3711 const Register y = rsi;
3712 const Register ylen = rcx;
3713 const Register z = r8;
3714 const Register zlen = r11;
3715
3716 // Next registers will be saved on stack in multiply_to_len().
3717 const Register tmp1 = r12;
3718 const Register tmp2 = r13;
3719 const Register tmp3 = r14;
3720 const Register tmp4 = r15;
3721 const Register tmp5 = rbx;
3722
3723 BLOCK_COMMENT("Entry:");
3724 __ enter(); // required for proper stackwalking of RuntimeStub frame
3725
3726 #ifndef _WIN64
3727 __ movptr(zlen, r9); // Save r9 in r11 - zlen
3728 #endif
3729 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
3730 // ylen => rcx, z => r8, zlen => r11
3731 // r9 and r10 may be used to save non-volatile registers
3732 #ifdef _WIN64
3733 // last 2 arguments (#4, #5) are on stack on Win64
3734 __ movptr(z, Address(rsp, 6 * wordSize));
3735 __ movptr(zlen, Address(rsp, 7 * wordSize));
3736 #endif
3737
3738 __ movptr(xlen, rsi);
3739 __ movptr(y, rdx);
3740 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
3741
3742 restore_arg_regs();
3743
3744 __ leave(); // required for proper stackwalking of RuntimeStub frame
3745 __ ret(0);
3746
3747 return start;
3748 }
3749
3750 /**
3751 * Arguments:
3752 *
3753 // Input:
3754 // c_rarg0 - x address
3755 // c_rarg1 - x length
3756 // c_rarg2 - z address
3757 // c_rarg3 - z lenth
3758 *
3759 */
3760 address generate_squareToLen() {
3761
3762 __ align(CodeEntryAlignment);
3763 StubCodeMark mark(this, "StubRoutines", "squareToLen");
3764
3765 address start = __ pc();
3766 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3767 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
3768 const Register x = rdi;
3769 const Register len = rsi;
3770 const Register z = r8;
3771 const Register zlen = rcx;
3772
3773 const Register tmp1 = r12;
3774 const Register tmp2 = r13;
3775 const Register tmp3 = r14;
3776 const Register tmp4 = r15;
3777 const Register tmp5 = rbx;
3778
3779 BLOCK_COMMENT("Entry:");
3780 __ enter(); // required for proper stackwalking of RuntimeStub frame
3781
3782 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
3783 // zlen => rcx
3784 // r9 and r10 may be used to save non-volatile registers
3785 __ movptr(r8, rdx);
3786 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3787
3788 restore_arg_regs();
3789
3790 __ leave(); // required for proper stackwalking of RuntimeStub frame
3791 __ ret(0);
3792
3793 return start;
3794 }
3795
3796 /**
3797 * Arguments:
3798 *
3799 * Input:
3800 * c_rarg0 - out address
3801 * c_rarg1 - in address
3802 * c_rarg2 - offset
3803 * c_rarg3 - len
3804 * not Win64
3805 * c_rarg4 - k
3806 * Win64
3807 * rsp+40 - k
3808 */
3809 address generate_mulAdd() {
3810 __ align(CodeEntryAlignment);
3811 StubCodeMark mark(this, "StubRoutines", "mulAdd");
3812
3813 address start = __ pc();
3814 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3815 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3816 const Register out = rdi;
3817 const Register in = rsi;
3818 const Register offset = r11;
3819 const Register len = rcx;
3820 const Register k = r8;
3821
3822 // Next registers will be saved on stack in mul_add().
3823 const Register tmp1 = r12;
3824 const Register tmp2 = r13;
3825 const Register tmp3 = r14;
3826 const Register tmp4 = r15;
3827 const Register tmp5 = rbx;
3828
3829 BLOCK_COMMENT("Entry:");
3830 __ enter(); // required for proper stackwalking of RuntimeStub frame
3831
3832 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
3833 // len => rcx, k => r8
3834 // r9 and r10 may be used to save non-volatile registers
3835 #ifdef _WIN64
3836 // last argument is on stack on Win64
3837 __ movl(k, Address(rsp, 6 * wordSize));
3838 #endif
3839 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
3840 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3841
3842 restore_arg_regs();
3843
3844 __ leave(); // required for proper stackwalking of RuntimeStub frame
3845 __ ret(0);
3846
3847 return start;
3848 }
3849
3850
3851 #undef __
3852 #define __ masm->
3853
3854 // Continuation point for throwing of implicit exceptions that are
3855 // not handled in the current activation. Fabricates an exception
3856 // oop and initiates normal exception dispatching in this
3857 // frame. Since we need to preserve callee-saved values (currently
3858 // only for C2, but done for C1 as well) we need a callee-saved oop
3859 // map and therefore have to make these stubs into RuntimeStubs
3860 // rather than BufferBlobs. If the compiler needs all registers to
3861 // be preserved between the fault point and the exception handler
3862 // then it must assume responsibility for that in
3863 // AbstractCompiler::continuation_for_implicit_null_exception or
3864 // continuation_for_implicit_division_by_zero_exception. All other
3865 // implicit exceptions (e.g., NullPointerException or
3866 // AbstractMethodError on entry) are either at call sites or
3867 // otherwise assume that stack unwinding will be initiated, so
3868 // caller saved registers were assumed volatile in the compiler.
3869 address generate_throw_exception(const char* name,
3870 address runtime_entry,
3871 Register arg1 = noreg,
3872 Register arg2 = noreg) {
3873 // Information about frame layout at time of blocking runtime call.
3874 // Note that we only have to preserve callee-saved registers since
3875 // the compilers are responsible for supplying a continuation point
3876 // if they expect all registers to be preserved.
3877 enum layout {
3878 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
3879 rbp_off2,
3880 return_off,
3881 return_off2,
3882 framesize // inclusive of return address
3883 };
3884
3885 int insts_size = 512;
3886 int locs_size = 64;
3887
3888 CodeBuffer code(name, insts_size, locs_size);
3889 OopMapSet* oop_maps = new OopMapSet();
3890 MacroAssembler* masm = new MacroAssembler(&code);
3891
3892 address start = __ pc();
3893
3894 // This is an inlined and slightly modified version of call_VM
3895 // which has the ability to fetch the return PC out of
3896 // thread-local storage and also sets up last_Java_sp slightly
3897 // differently than the real call_VM
3898
3899 __ enter(); // required for proper stackwalking of RuntimeStub frame
3900
3901 assert(is_even(framesize/2), "sp not 16-byte aligned");
3902
3903 // return address and rbp are already in place
3904 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
3905
3906 int frame_complete = __ pc() - start;
3907
3908 // Set up last_Java_sp and last_Java_fp
3909 address the_pc = __ pc();
3910 __ set_last_Java_frame(rsp, rbp, the_pc);
3911 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
3912
3913 // Call runtime
3914 if (arg1 != noreg) {
3915 assert(arg2 != c_rarg1, "clobbered");
3916 __ movptr(c_rarg1, arg1);
3917 }
3918 if (arg2 != noreg) {
3919 __ movptr(c_rarg2, arg2);
3920 }
3921 __ movptr(c_rarg0, r15_thread);
3922 BLOCK_COMMENT("call runtime_entry");
3923 __ call(RuntimeAddress(runtime_entry));
3924
3925 // Generate oop map
3926 OopMap* map = new OopMap(framesize, 0);
3927
3928 oop_maps->add_gc_map(the_pc - start, map);
3929
3930 __ reset_last_Java_frame(true);
3931
3932 __ leave(); // required for proper stackwalking of RuntimeStub frame
3933
3934 // check for pending exceptions
3935 #ifdef ASSERT
3936 Label L;
3937 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
3938 (int32_t) NULL_WORD);
3939 __ jcc(Assembler::notEqual, L);
3940 __ should_not_reach_here();
3941 __ bind(L);
3942 #endif // ASSERT
3943 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3944
3945
3946 // codeBlob framesize is in words (not VMRegImpl::slot_size)
3947 RuntimeStub* stub =
3948 RuntimeStub::new_runtime_stub(name,
3949 &code,
3950 frame_complete,
3951 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3952 oop_maps, false);
3953 return stub->entry_point();
3954 }
3955
3956 void create_control_words() {
3957 // Round to nearest, 53-bit mode, exceptions masked
3958 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
3959 // Round to zero, 53-bit mode, exception mased
3960 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
3961 // Round to nearest, 24-bit mode, exceptions masked
3962 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
3963 // Round to nearest, 64-bit mode, exceptions masked
3964 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
3965 // Round to nearest, 64-bit mode, exceptions masked
3966 StubRoutines::_mxcsr_std = 0x1F80;
3967 // Note: the following two constants are 80-bit values
3968 // layout is critical for correct loading by FPU.
3969 // Bias for strict fp multiply/divide
3970 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
3971 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
3972 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
3973 // Un-Bias for strict fp multiply/divide
3974 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
3975 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
3976 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
3977 }
3978
3979 // Initialization
3980 void generate_initial() {
3981 // Generates all stubs and initializes the entry points
3982
3983 // This platform-specific settings are needed by generate_call_stub()
3984 create_control_words();
3985
3986 // entry points that exist in all platforms Note: This is code
3987 // that could be shared among different platforms - however the
3988 // benefit seems to be smaller than the disadvantage of having a
3989 // much more complicated generator structure. See also comment in
3990 // stubRoutines.hpp.
3991
3992 StubRoutines::_forward_exception_entry = generate_forward_exception();
3993
3994 StubRoutines::_call_stub_entry =
3995 generate_call_stub(StubRoutines::_call_stub_return_address);
3996
3997 // is referenced by megamorphic call
3998 StubRoutines::_catch_exception_entry = generate_catch_exception();
3999
4000 // atomic calls
4001 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
4002 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
4003 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
4004 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
4005 StubRoutines::_atomic_add_entry = generate_atomic_add();
4006 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
4007 StubRoutines::_fence_entry = generate_orderaccess_fence();
4008
4009 StubRoutines::_handler_for_unsafe_access_entry =
4010 generate_handler_for_unsafe_access();
4011
4012 // platform dependent
4013 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
4014 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
4015
4016 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
4017
4018 // Build this early so it's available for the interpreter.
4019 StubRoutines::_throw_StackOverflowError_entry =
4020 generate_throw_exception("StackOverflowError throw_exception",
4021 CAST_FROM_FN_PTR(address,
4022 SharedRuntime::
4023 throw_StackOverflowError));
4024 if (UseCRC32Intrinsics) {
4025 // set table address before stub generation which use it
4026 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
4027 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
4028 }
4029 }
4030
4031 void generate_all() {
4032 // Generates all stubs and initializes the entry points
4033
4034 // These entry points require SharedInfo::stack0 to be set up in
4035 // non-core builds and need to be relocatable, so they each
4036 // fabricate a RuntimeStub internally.
4037 StubRoutines::_throw_AbstractMethodError_entry =
4038 generate_throw_exception("AbstractMethodError throw_exception",
4039 CAST_FROM_FN_PTR(address,
4040 SharedRuntime::
4041 throw_AbstractMethodError));
4042
4043 StubRoutines::_throw_IncompatibleClassChangeError_entry =
4044 generate_throw_exception("IncompatibleClassChangeError throw_exception",
4045 CAST_FROM_FN_PTR(address,
4046 SharedRuntime::
4047 throw_IncompatibleClassChangeError));
4048
4049 StubRoutines::_throw_NullPointerException_at_call_entry =
4050 generate_throw_exception("NullPointerException at call throw_exception",
4051 CAST_FROM_FN_PTR(address,
4052 SharedRuntime::
4053 throw_NullPointerException_at_call));
4054
4055 // entry points that are platform specific
4056 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
4057 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
4058 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
4059 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
4060
4061 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
4062 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
4063 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4064 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
4065
4066 // support for verify_oop (must happen after universe_init)
4067 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
4068
4069 // arraycopy stubs used by compilers
4070 generate_arraycopy_stubs();
4071
4072 generate_math_stubs();
4073
4074 // don't bother generating these AES intrinsic stubs unless global flag is set
4075 if (UseAESIntrinsics) {
4076 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
4077
4078 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
4079 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
4080 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
4081 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
4082 }
4083
4084 // Safefetch stubs.
4085 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
4086 &StubRoutines::_safefetch32_fault_pc,
4087 &StubRoutines::_safefetch32_continuation_pc);
4088 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
4089 &StubRoutines::_safefetchN_fault_pc,
4090 &StubRoutines::_safefetchN_continuation_pc);
4091 #ifdef COMPILER2
4092 if (UseMultiplyToLenIntrinsic) {
4093 StubRoutines::_multiplyToLen = generate_multiplyToLen();
4094 }
4095 if (UseSquareToLenIntrinsic) {
4096 StubRoutines::_squareToLen = generate_squareToLen();
4097 }
4098 if (UseMulAddIntrinsic) {
4099 StubRoutines::_mulAdd = generate_mulAdd();
4100 }
4101
4102 #ifndef _WINDOWS
4103 if (UseMontgomeryMultiplyIntrinsic) {
4104 StubRoutines::_montgomeryMultiply
4105 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
4106 }
4107 if (UseMontgomerySquareIntrinsic) {
4108 StubRoutines::_montgomerySquare
4109 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
4110 }
4111 #endif // WINDOWS
4112 #endif // COMPILER2
4113 }
4114
4115 public:
4116 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
4117 if (all) {
4118 generate_all();
4119 } else {
4120 generate_initial();
4121 }
4122 }
4123 }; // end class declaration
4124
4125 void StubGenerator_generate(CodeBuffer* code, bool all) {
4126 StubGenerator g(code, all);
4127 }