1 /*
2 * Copyright (c) 2003, 2019, 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 "ci/ciUtilities.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/barrierSetNMethod.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "memory/universe.hpp"
34 #include "nativeInst_x86.hpp"
35 #include "oops/instanceOop.hpp"
36 #include "oops/method.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/methodHandles.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/handles.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubCodeGenerator.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/thread.inline.hpp"
46 #ifdef COMPILER2
47 #include "opto/runtime.hpp"
48 #endif
49 #if INCLUDE_ZGC
50 #include "gc/z/zThreadLocalData.hpp"
51 #endif
52
53 // Declaration and definition of StubGenerator (no .hpp file).
54 // For a more detailed description of the stub routine structure
55 // see the comment in stubRoutines.hpp
56
57 #define __ _masm->
58 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
59 #define a__ ((Assembler*)_masm)->
60
61 #ifdef PRODUCT
62 #define BLOCK_COMMENT(str) /* nothing */
63 #else
64 #define BLOCK_COMMENT(str) __ block_comment(str)
65 #endif
66
67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
68 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
69
70 // Stub Code definitions
71
72 class StubGenerator: public StubCodeGenerator {
73 private:
74
75 #ifdef PRODUCT
76 #define inc_counter_np(counter) ((void)0)
77 #else
78 void inc_counter_np_(int& counter) {
79 // This can destroy rscratch1 if counter is far from the code cache
80 __ incrementl(ExternalAddress((address)&counter));
81 }
82 #define inc_counter_np(counter) \
83 BLOCK_COMMENT("inc_counter " #counter); \
84 inc_counter_np_(counter);
85 #endif
86
87 // Call stubs are used to call Java from C
88 //
89 // Linux Arguments:
90 // c_rarg0: call wrapper address address
91 // c_rarg1: result address
92 // c_rarg2: result type BasicType
93 // c_rarg3: method Method*
94 // c_rarg4: (interpreter) entry point address
95 // c_rarg5: parameters intptr_t*
96 // 16(rbp): parameter size (in words) int
97 // 24(rbp): thread Thread*
98 //
99 // [ return_from_Java ] <--- rsp
100 // [ argument word n ]
101 // ...
102 // -12 [ argument word 1 ]
103 // -11 [ saved r15 ] <--- rsp_after_call
104 // -10 [ saved r14 ]
105 // -9 [ saved r13 ]
106 // -8 [ saved r12 ]
107 // -7 [ saved rbx ]
108 // -6 [ call wrapper ]
109 // -5 [ result ]
110 // -4 [ result type ]
111 // -3 [ method ]
112 // -2 [ entry point ]
113 // -1 [ parameters ]
114 // 0 [ saved rbp ] <--- rbp
115 // 1 [ return address ]
116 // 2 [ parameter size ]
117 // 3 [ thread ]
118 //
119 // Windows Arguments:
120 // c_rarg0: call wrapper address address
121 // c_rarg1: result address
122 // c_rarg2: result type BasicType
123 // c_rarg3: method Method*
124 // 48(rbp): (interpreter) entry point address
125 // 56(rbp): parameters intptr_t*
126 // 64(rbp): parameter size (in words) int
127 // 72(rbp): thread Thread*
128 //
129 // [ return_from_Java ] <--- rsp
130 // [ argument word n ]
131 // ...
132 // -60 [ argument word 1 ]
133 // -59 [ saved xmm31 ] <--- rsp after_call
134 // [ saved xmm16-xmm30 ] (EVEX enabled, else the space is blank)
135 // -27 [ saved xmm15 ]
136 // [ saved xmm7-xmm14 ]
137 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
138 // -7 [ saved r15 ]
139 // -6 [ saved r14 ]
140 // -5 [ saved r13 ]
141 // -4 [ saved r12 ]
142 // -3 [ saved rdi ]
143 // -2 [ saved rsi ]
144 // -1 [ saved rbx ]
145 // 0 [ saved rbp ] <--- rbp
146 // 1 [ return address ]
147 // 2 [ call wrapper ]
148 // 3 [ result ]
149 // 4 [ result type ]
150 // 5 [ method ]
151 // 6 [ entry point ]
152 // 7 [ parameters ]
153 // 8 [ parameter size ]
154 // 9 [ thread ]
155 //
156 // Windows reserves the callers stack space for arguments 1-4.
157 // We spill c_rarg0-c_rarg3 to this space.
158
159 // Call stub stack layout word offsets from rbp
160 enum call_stub_layout {
161 #ifdef _WIN64
162 xmm_save_first = 6, // save from xmm6
163 xmm_save_last = 31, // to xmm31
164 xmm_save_base = -9,
165 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
166 r15_off = -7,
167 r14_off = -6,
168 r13_off = -5,
169 r12_off = -4,
170 rdi_off = -3,
171 rsi_off = -2,
172 rbx_off = -1,
173 rbp_off = 0,
174 retaddr_off = 1,
175 call_wrapper_off = 2,
176 result_off = 3,
177 result_type_off = 4,
178 method_off = 5,
179 entry_point_off = 6,
180 parameters_off = 7,
181 parameter_size_off = 8,
182 thread_off = 9
183 #else
184 rsp_after_call_off = -12,
185 mxcsr_off = rsp_after_call_off,
186 r15_off = -11,
187 r14_off = -10,
188 r13_off = -9,
189 r12_off = -8,
190 rbx_off = -7,
191 call_wrapper_off = -6,
192 result_off = -5,
193 result_type_off = -4,
194 method_off = -3,
195 entry_point_off = -2,
196 parameters_off = -1,
197 rbp_off = 0,
198 retaddr_off = 1,
199 parameter_size_off = 2,
200 thread_off = 3
201 #endif
202 };
203
204 #ifdef _WIN64
205 Address xmm_save(int reg) {
206 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
207 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
208 }
209 #endif
210
211 address generate_call_stub(address& return_address) {
212 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
213 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
214 "adjust this code");
215 StubCodeMark mark(this, "StubRoutines", "call_stub");
216 address start = __ pc();
217
218 // same as in generate_catch_exception()!
219 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
220
221 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
222 const Address result (rbp, result_off * wordSize);
223 const Address result_type (rbp, result_type_off * wordSize);
224 const Address method (rbp, method_off * wordSize);
225 const Address entry_point (rbp, entry_point_off * wordSize);
226 const Address parameters (rbp, parameters_off * wordSize);
227 const Address parameter_size(rbp, parameter_size_off * wordSize);
228
229 // same as in generate_catch_exception()!
230 const Address thread (rbp, thread_off * wordSize);
231
232 const Address r15_save(rbp, r15_off * wordSize);
233 const Address r14_save(rbp, r14_off * wordSize);
234 const Address r13_save(rbp, r13_off * wordSize);
235 const Address r12_save(rbp, r12_off * wordSize);
236 const Address rbx_save(rbp, rbx_off * wordSize);
237
238 // stub code
239 __ enter();
240 __ subptr(rsp, -rsp_after_call_off * wordSize);
241
242 // save register parameters
243 #ifndef _WIN64
244 __ movptr(parameters, c_rarg5); // parameters
245 __ movptr(entry_point, c_rarg4); // entry_point
246 #endif
247
248 __ movptr(method, c_rarg3); // method
249 __ movl(result_type, c_rarg2); // result type
250 __ movptr(result, c_rarg1); // result
251 __ movptr(call_wrapper, c_rarg0); // call wrapper
252
253 // save regs belonging to calling function
254 __ movptr(rbx_save, rbx);
255 __ movptr(r12_save, r12);
256 __ movptr(r13_save, r13);
257 __ movptr(r14_save, r14);
258 __ movptr(r15_save, r15);
259
260 #ifdef _WIN64
261 int last_reg = 15;
262 if (UseAVX > 2) {
263 last_reg = 31;
264 }
265 if (VM_Version::supports_evex()) {
266 for (int i = xmm_save_first; i <= last_reg; i++) {
267 __ vextractf32x4(xmm_save(i), as_XMMRegister(i), 0);
268 }
269 } else {
270 for (int i = xmm_save_first; i <= last_reg; i++) {
271 __ movdqu(xmm_save(i), as_XMMRegister(i));
272 }
273 }
274
275 const Address rdi_save(rbp, rdi_off * wordSize);
276 const Address rsi_save(rbp, rsi_off * wordSize);
277
278 __ movptr(rsi_save, rsi);
279 __ movptr(rdi_save, rdi);
280 #else
281 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
282 {
283 Label skip_ldmx;
284 __ stmxcsr(mxcsr_save);
285 __ movl(rax, mxcsr_save);
286 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
287 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
288 __ cmp32(rax, mxcsr_std);
289 __ jcc(Assembler::equal, skip_ldmx);
290 __ ldmxcsr(mxcsr_std);
291 __ bind(skip_ldmx);
292 }
293 #endif
294
295 // Load up thread register
296 __ movptr(r15_thread, thread);
297 __ reinit_heapbase();
298
299 #ifdef ASSERT
300 // make sure we have no pending exceptions
301 {
302 Label L;
303 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
304 __ jcc(Assembler::equal, L);
305 __ stop("StubRoutines::call_stub: entered with pending exception");
306 __ bind(L);
307 }
308 #endif
309
310 // pass parameters if any
311 BLOCK_COMMENT("pass parameters if any");
312 Label parameters_done;
313 __ movl(c_rarg3, parameter_size);
314 __ testl(c_rarg3, c_rarg3);
315 __ jcc(Assembler::zero, parameters_done);
316
317 Label loop;
318 __ movptr(c_rarg2, parameters); // parameter pointer
319 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
320 __ BIND(loop);
321 __ movptr(rax, Address(c_rarg2, 0));// get parameter
322 __ addptr(c_rarg2, wordSize); // advance to next parameter
323 __ decrementl(c_rarg1); // decrement counter
324 __ push(rax); // pass parameter
325 __ jcc(Assembler::notZero, loop);
326
327 // call Java function
328 __ BIND(parameters_done);
329 __ movptr(rbx, method); // get Method*
330 __ movptr(c_rarg1, entry_point); // get entry_point
331 __ mov(r13, rsp); // set sender sp
332 BLOCK_COMMENT("call Java function");
333 __ call(c_rarg1);
334
335 BLOCK_COMMENT("call_stub_return_address:");
336 return_address = __ pc();
337
338 // store result depending on type (everything that is not
339 // T_OBJECT, T_VALUETYPE, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
340 __ movptr(r13, result);
341 Label is_long, is_float, is_double, is_value, exit;
342 __ movl(rbx, result_type);
343 __ cmpl(rbx, T_OBJECT);
344 __ jcc(Assembler::equal, is_long);
345 __ cmpl(rbx, T_VALUETYPE);
346 __ jcc(Assembler::equal, is_value);
347 __ cmpl(rbx, T_LONG);
348 __ jcc(Assembler::equal, is_long);
349 __ cmpl(rbx, T_FLOAT);
350 __ jcc(Assembler::equal, is_float);
351 __ cmpl(rbx, T_DOUBLE);
352 __ jcc(Assembler::equal, is_double);
353
354 // handle T_INT case
355 __ movl(Address(r13, 0), rax);
356
357 __ BIND(exit);
358
359 // pop parameters
360 __ lea(rsp, rsp_after_call);
361
362 #ifdef ASSERT
363 // verify that threads correspond
364 {
365 Label L1, L2, L3;
366 __ cmpptr(r15_thread, thread);
367 __ jcc(Assembler::equal, L1);
368 __ stop("StubRoutines::call_stub: r15_thread is corrupted");
369 __ bind(L1);
370 __ get_thread(rbx);
371 __ cmpptr(r15_thread, thread);
372 __ jcc(Assembler::equal, L2);
373 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
374 __ bind(L2);
375 __ cmpptr(r15_thread, rbx);
376 __ jcc(Assembler::equal, L3);
377 __ stop("StubRoutines::call_stub: threads must correspond");
378 __ bind(L3);
379 }
380 #endif
381
382 // restore regs belonging to calling function
383 #ifdef _WIN64
384 // emit the restores for xmm regs
385 if (VM_Version::supports_evex()) {
386 for (int i = xmm_save_first; i <= last_reg; i++) {
387 __ vinsertf32x4(as_XMMRegister(i), as_XMMRegister(i), xmm_save(i), 0);
388 }
389 } else {
390 for (int i = xmm_save_first; i <= last_reg; i++) {
391 __ movdqu(as_XMMRegister(i), xmm_save(i));
392 }
393 }
394 #endif
395 __ movptr(r15, r15_save);
396 __ movptr(r14, r14_save);
397 __ movptr(r13, r13_save);
398 __ movptr(r12, r12_save);
399 __ movptr(rbx, rbx_save);
400
401 #ifdef _WIN64
402 __ movptr(rdi, rdi_save);
403 __ movptr(rsi, rsi_save);
404 #else
405 __ ldmxcsr(mxcsr_save);
406 #endif
407
408 // restore rsp
409 __ addptr(rsp, -rsp_after_call_off * wordSize);
410
411 // return
412 __ vzeroupper();
413 __ pop(rbp);
414 __ ret(0);
415
416 // handle return types different from T_INT
417 __ BIND(is_value);
418 if (ValueTypeReturnedAsFields) {
419 // Handle value type returned as fields
420 __ store_value_type_fields_to_buf(NULL);
421 __ movptr(r13, result);
422 }
423 __ BIND(is_long);
424 __ movq(Address(r13, 0), rax);
425 __ jmp(exit);
426
427 __ BIND(is_float);
428 __ movflt(Address(r13, 0), xmm0);
429 __ jmp(exit);
430
431 __ BIND(is_double);
432 __ movdbl(Address(r13, 0), xmm0);
433 __ jmp(exit);
434
435 return start;
436 }
437
438 // Return point for a Java call if there's an exception thrown in
439 // Java code. The exception is caught and transformed into a
440 // pending exception stored in JavaThread that can be tested from
441 // within the VM.
442 //
443 // Note: Usually the parameters are removed by the callee. In case
444 // of an exception crossing an activation frame boundary, that is
445 // not the case if the callee is compiled code => need to setup the
446 // rsp.
447 //
448 // rax: exception oop
449
450 address generate_catch_exception() {
451 StubCodeMark mark(this, "StubRoutines", "catch_exception");
452 address start = __ pc();
453
454 // same as in generate_call_stub():
455 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
456 const Address thread (rbp, thread_off * wordSize);
457
458 #ifdef ASSERT
459 // verify that threads correspond
460 {
461 Label L1, L2, L3;
462 __ cmpptr(r15_thread, thread);
463 __ jcc(Assembler::equal, L1);
464 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
465 __ bind(L1);
466 __ get_thread(rbx);
467 __ cmpptr(r15_thread, thread);
468 __ jcc(Assembler::equal, L2);
469 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
470 __ bind(L2);
471 __ cmpptr(r15_thread, rbx);
472 __ jcc(Assembler::equal, L3);
473 __ stop("StubRoutines::catch_exception: threads must correspond");
474 __ bind(L3);
475 }
476 #endif
477
478 // set pending exception
479 __ verify_oop(rax);
480
481 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
482 __ lea(rscratch1, ExternalAddress((address)__FILE__));
483 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
484 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
485
486 // complete return to VM
487 assert(StubRoutines::_call_stub_return_address != NULL,
488 "_call_stub_return_address must have been generated before");
489 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
490
491 return start;
492 }
493
494 // Continuation point for runtime calls returning with a pending
495 // exception. The pending exception check happened in the runtime
496 // or native call stub. The pending exception in Thread is
497 // converted into a Java-level exception.
498 //
499 // Contract with Java-level exception handlers:
500 // rax: exception
501 // rdx: throwing pc
502 //
503 // NOTE: At entry of this stub, exception-pc must be on stack !!
504
505 address generate_forward_exception() {
506 StubCodeMark mark(this, "StubRoutines", "forward exception");
507 address start = __ pc();
508
509 // Upon entry, the sp points to the return address returning into
510 // Java (interpreted or compiled) code; i.e., the return address
511 // becomes the throwing pc.
512 //
513 // Arguments pushed before the runtime call are still on the stack
514 // but the exception handler will reset the stack pointer ->
515 // ignore them. A potential result in registers can be ignored as
516 // well.
517
518 #ifdef ASSERT
519 // make sure this code is only executed if there is a pending exception
520 {
521 Label L;
522 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
523 __ jcc(Assembler::notEqual, L);
524 __ stop("StubRoutines::forward exception: no pending exception (1)");
525 __ bind(L);
526 }
527 #endif
528
529 // compute exception handler into rbx
530 __ movptr(c_rarg0, Address(rsp, 0));
531 BLOCK_COMMENT("call exception_handler_for_return_address");
532 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
533 SharedRuntime::exception_handler_for_return_address),
534 r15_thread, c_rarg0);
535 __ mov(rbx, rax);
536
537 // setup rax & rdx, remove return address & clear pending exception
538 __ pop(rdx);
539 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
540 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
541
542 #ifdef ASSERT
543 // make sure exception is set
544 {
545 Label L;
546 __ testptr(rax, rax);
547 __ jcc(Assembler::notEqual, L);
548 __ stop("StubRoutines::forward exception: no pending exception (2)");
549 __ bind(L);
550 }
551 #endif
552
553 // continue at exception handler (return address removed)
554 // rax: exception
555 // rbx: exception handler
556 // rdx: throwing pc
557 __ verify_oop(rax);
558 __ jmp(rbx);
559
560 return start;
561 }
562
563 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
564 //
565 // Arguments :
566 // c_rarg0: exchange_value
567 // c_rarg0: dest
568 //
569 // Result:
570 // *dest <- ex, return (orig *dest)
571 address generate_atomic_xchg() {
572 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
573 address start = __ pc();
574
575 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
576 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
577 __ ret(0);
578
579 return start;
580 }
581
582 // Support for intptr_t atomic::xchg_long(jlong exchange_value, volatile jlong* dest)
583 //
584 // Arguments :
585 // c_rarg0: exchange_value
586 // c_rarg1: dest
587 //
588 // Result:
589 // *dest <- ex, return (orig *dest)
590 address generate_atomic_xchg_long() {
591 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_long");
592 address start = __ pc();
593
594 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
595 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
596 __ ret(0);
597
598 return start;
599 }
600
601 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
602 // jint compare_value)
603 //
604 // Arguments :
605 // c_rarg0: exchange_value
606 // c_rarg1: dest
607 // c_rarg2: compare_value
608 //
609 // Result:
610 // if ( compare_value == *dest ) {
611 // *dest = exchange_value
612 // return compare_value;
613 // else
614 // return *dest;
615 address generate_atomic_cmpxchg() {
616 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
617 address start = __ pc();
618
619 __ movl(rax, c_rarg2);
620 __ lock();
621 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
622 __ ret(0);
623
624 return start;
625 }
626
627 // Support for int8_t atomic::atomic_cmpxchg(int8_t exchange_value, volatile int8_t* dest,
628 // int8_t compare_value)
629 //
630 // Arguments :
631 // c_rarg0: exchange_value
632 // c_rarg1: dest
633 // c_rarg2: compare_value
634 //
635 // Result:
636 // if ( compare_value == *dest ) {
637 // *dest = exchange_value
638 // return compare_value;
639 // else
640 // return *dest;
641 address generate_atomic_cmpxchg_byte() {
642 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_byte");
643 address start = __ pc();
644
645 __ movsbq(rax, c_rarg2);
646 __ lock();
647 __ cmpxchgb(c_rarg0, Address(c_rarg1, 0));
648 __ ret(0);
649
650 return start;
651 }
652
653 // Support for int64_t atomic::atomic_cmpxchg(int64_t exchange_value,
654 // volatile int64_t* dest,
655 // int64_t compare_value)
656 // Arguments :
657 // c_rarg0: exchange_value
658 // c_rarg1: dest
659 // c_rarg2: compare_value
660 //
661 // Result:
662 // if ( compare_value == *dest ) {
663 // *dest = exchange_value
664 // return compare_value;
665 // else
666 // return *dest;
667 address generate_atomic_cmpxchg_long() {
668 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
669 address start = __ pc();
670
671 __ movq(rax, c_rarg2);
672 __ lock();
673 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
674 __ ret(0);
675
676 return start;
677 }
678
679 // Support for jint atomic::add(jint add_value, volatile jint* dest)
680 //
681 // Arguments :
682 // c_rarg0: add_value
683 // c_rarg1: dest
684 //
685 // Result:
686 // *dest += add_value
687 // return *dest;
688 address generate_atomic_add() {
689 StubCodeMark mark(this, "StubRoutines", "atomic_add");
690 address start = __ pc();
691
692 __ movl(rax, c_rarg0);
693 __ lock();
694 __ xaddl(Address(c_rarg1, 0), c_rarg0);
695 __ addl(rax, c_rarg0);
696 __ ret(0);
697
698 return start;
699 }
700
701 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
702 //
703 // Arguments :
704 // c_rarg0: add_value
705 // c_rarg1: dest
706 //
707 // Result:
708 // *dest += add_value
709 // return *dest;
710 address generate_atomic_add_long() {
711 StubCodeMark mark(this, "StubRoutines", "atomic_add_long");
712 address start = __ pc();
713
714 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
715 __ lock();
716 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
717 __ addptr(rax, c_rarg0);
718 __ ret(0);
719
720 return start;
721 }
722
723 // Support for intptr_t OrderAccess::fence()
724 //
725 // Arguments :
726 //
727 // Result:
728 address generate_orderaccess_fence() {
729 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
730 address start = __ pc();
731 __ membar(Assembler::StoreLoad);
732 __ ret(0);
733
734 return start;
735 }
736
737 // Support for intptr_t get_previous_fp()
738 //
739 // This routine is used to find the previous frame pointer for the
740 // caller (current_frame_guess). This is used as part of debugging
741 // ps() is seemingly lost trying to find frames.
742 // This code assumes that caller current_frame_guess) has a frame.
743 address generate_get_previous_fp() {
744 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
745 const Address old_fp(rbp, 0);
746 const Address older_fp(rax, 0);
747 address start = __ pc();
748
749 __ enter();
750 __ movptr(rax, old_fp); // callers fp
751 __ movptr(rax, older_fp); // the frame for ps()
752 __ pop(rbp);
753 __ ret(0);
754
755 return start;
756 }
757
758 // Support for intptr_t get_previous_sp()
759 //
760 // This routine is used to find the previous stack pointer for the
761 // caller.
762 address generate_get_previous_sp() {
763 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
764 address start = __ pc();
765
766 __ movptr(rax, rsp);
767 __ addptr(rax, 8); // return address is at the top of the stack.
768 __ ret(0);
769
770 return start;
771 }
772
773 //----------------------------------------------------------------------------------------------------
774 // Support for void verify_mxcsr()
775 //
776 // This routine is used with -Xcheck:jni to verify that native
777 // JNI code does not return to Java code without restoring the
778 // MXCSR register to our expected state.
779
780 address generate_verify_mxcsr() {
781 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
782 address start = __ pc();
783
784 const Address mxcsr_save(rsp, 0);
785
786 if (CheckJNICalls) {
787 Label ok_ret;
788 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
789 __ push(rax);
790 __ subptr(rsp, wordSize); // allocate a temp location
791 __ stmxcsr(mxcsr_save);
792 __ movl(rax, mxcsr_save);
793 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
794 __ cmp32(rax, mxcsr_std);
795 __ jcc(Assembler::equal, ok_ret);
796
797 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
798
799 __ ldmxcsr(mxcsr_std);
800
801 __ bind(ok_ret);
802 __ addptr(rsp, wordSize);
803 __ pop(rax);
804 }
805
806 __ ret(0);
807
808 return start;
809 }
810
811 address generate_f2i_fixup() {
812 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
813 Address inout(rsp, 5 * wordSize); // return address + 4 saves
814
815 address start = __ pc();
816
817 Label L;
818
819 __ push(rax);
820 __ push(c_rarg3);
821 __ push(c_rarg2);
822 __ push(c_rarg1);
823
824 __ movl(rax, 0x7f800000);
825 __ xorl(c_rarg3, c_rarg3);
826 __ movl(c_rarg2, inout);
827 __ movl(c_rarg1, c_rarg2);
828 __ andl(c_rarg1, 0x7fffffff);
829 __ cmpl(rax, c_rarg1); // NaN? -> 0
830 __ jcc(Assembler::negative, L);
831 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
832 __ movl(c_rarg3, 0x80000000);
833 __ movl(rax, 0x7fffffff);
834 __ cmovl(Assembler::positive, c_rarg3, rax);
835
836 __ bind(L);
837 __ movptr(inout, c_rarg3);
838
839 __ pop(c_rarg1);
840 __ pop(c_rarg2);
841 __ pop(c_rarg3);
842 __ pop(rax);
843
844 __ ret(0);
845
846 return start;
847 }
848
849 address generate_f2l_fixup() {
850 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
851 Address inout(rsp, 5 * wordSize); // return address + 4 saves
852 address start = __ pc();
853
854 Label L;
855
856 __ push(rax);
857 __ push(c_rarg3);
858 __ push(c_rarg2);
859 __ push(c_rarg1);
860
861 __ movl(rax, 0x7f800000);
862 __ xorl(c_rarg3, c_rarg3);
863 __ movl(c_rarg2, inout);
864 __ movl(c_rarg1, c_rarg2);
865 __ andl(c_rarg1, 0x7fffffff);
866 __ cmpl(rax, c_rarg1); // NaN? -> 0
867 __ jcc(Assembler::negative, L);
868 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
869 __ mov64(c_rarg3, 0x8000000000000000);
870 __ mov64(rax, 0x7fffffffffffffff);
871 __ cmov(Assembler::positive, c_rarg3, rax);
872
873 __ bind(L);
874 __ movptr(inout, c_rarg3);
875
876 __ pop(c_rarg1);
877 __ pop(c_rarg2);
878 __ pop(c_rarg3);
879 __ pop(rax);
880
881 __ ret(0);
882
883 return start;
884 }
885
886 address generate_d2i_fixup() {
887 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
888 Address inout(rsp, 6 * wordSize); // return address + 5 saves
889
890 address start = __ pc();
891
892 Label L;
893
894 __ push(rax);
895 __ push(c_rarg3);
896 __ push(c_rarg2);
897 __ push(c_rarg1);
898 __ push(c_rarg0);
899
900 __ movl(rax, 0x7ff00000);
901 __ movq(c_rarg2, inout);
902 __ movl(c_rarg3, c_rarg2);
903 __ mov(c_rarg1, c_rarg2);
904 __ mov(c_rarg0, c_rarg2);
905 __ negl(c_rarg3);
906 __ shrptr(c_rarg1, 0x20);
907 __ orl(c_rarg3, c_rarg2);
908 __ andl(c_rarg1, 0x7fffffff);
909 __ xorl(c_rarg2, c_rarg2);
910 __ shrl(c_rarg3, 0x1f);
911 __ orl(c_rarg1, c_rarg3);
912 __ cmpl(rax, c_rarg1);
913 __ jcc(Assembler::negative, L); // NaN -> 0
914 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
915 __ movl(c_rarg2, 0x80000000);
916 __ movl(rax, 0x7fffffff);
917 __ cmov(Assembler::positive, c_rarg2, rax);
918
919 __ bind(L);
920 __ movptr(inout, c_rarg2);
921
922 __ pop(c_rarg0);
923 __ pop(c_rarg1);
924 __ pop(c_rarg2);
925 __ pop(c_rarg3);
926 __ pop(rax);
927
928 __ ret(0);
929
930 return start;
931 }
932
933 address generate_d2l_fixup() {
934 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
935 Address inout(rsp, 6 * wordSize); // return address + 5 saves
936
937 address start = __ pc();
938
939 Label L;
940
941 __ push(rax);
942 __ push(c_rarg3);
943 __ push(c_rarg2);
944 __ push(c_rarg1);
945 __ push(c_rarg0);
946
947 __ movl(rax, 0x7ff00000);
948 __ movq(c_rarg2, inout);
949 __ movl(c_rarg3, c_rarg2);
950 __ mov(c_rarg1, c_rarg2);
951 __ mov(c_rarg0, c_rarg2);
952 __ negl(c_rarg3);
953 __ shrptr(c_rarg1, 0x20);
954 __ orl(c_rarg3, c_rarg2);
955 __ andl(c_rarg1, 0x7fffffff);
956 __ xorl(c_rarg2, c_rarg2);
957 __ shrl(c_rarg3, 0x1f);
958 __ orl(c_rarg1, c_rarg3);
959 __ cmpl(rax, c_rarg1);
960 __ jcc(Assembler::negative, L); // NaN -> 0
961 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
962 __ mov64(c_rarg2, 0x8000000000000000);
963 __ mov64(rax, 0x7fffffffffffffff);
964 __ cmovq(Assembler::positive, c_rarg2, rax);
965
966 __ bind(L);
967 __ movq(inout, c_rarg2);
968
969 __ pop(c_rarg0);
970 __ pop(c_rarg1);
971 __ pop(c_rarg2);
972 __ pop(c_rarg3);
973 __ pop(rax);
974
975 __ ret(0);
976
977 return start;
978 }
979
980 address generate_fp_mask(const char *stub_name, int64_t mask) {
981 __ align(CodeEntryAlignment);
982 StubCodeMark mark(this, "StubRoutines", stub_name);
983 address start = __ pc();
984
985 __ emit_data64( mask, relocInfo::none );
986 __ emit_data64( mask, relocInfo::none );
987
988 return start;
989 }
990
991 address generate_vector_mask(const char *stub_name, int64_t mask) {
992 __ align(CodeEntryAlignment);
993 StubCodeMark mark(this, "StubRoutines", stub_name);
994 address start = __ pc();
995
996 __ emit_data64(mask, relocInfo::none);
997 __ emit_data64(mask, relocInfo::none);
998 __ emit_data64(mask, relocInfo::none);
999 __ emit_data64(mask, relocInfo::none);
1000 __ emit_data64(mask, relocInfo::none);
1001 __ emit_data64(mask, relocInfo::none);
1002 __ emit_data64(mask, relocInfo::none);
1003 __ emit_data64(mask, relocInfo::none);
1004
1005 return start;
1006 }
1007
1008 address generate_vector_byte_perm_mask(const char *stub_name) {
1009 __ align(CodeEntryAlignment);
1010 StubCodeMark mark(this, "StubRoutines", stub_name);
1011 address start = __ pc();
1012
1013 __ emit_data64(0x0000000000000001, relocInfo::none);
1014 __ emit_data64(0x0000000000000003, relocInfo::none);
1015 __ emit_data64(0x0000000000000005, relocInfo::none);
1016 __ emit_data64(0x0000000000000007, relocInfo::none);
1017 __ emit_data64(0x0000000000000000, relocInfo::none);
1018 __ emit_data64(0x0000000000000002, relocInfo::none);
1019 __ emit_data64(0x0000000000000004, relocInfo::none);
1020 __ emit_data64(0x0000000000000006, relocInfo::none);
1021
1022 return start;
1023 }
1024
1025 // Non-destructive plausibility checks for oops
1026 //
1027 // Arguments:
1028 // all args on stack!
1029 //
1030 // Stack after saving c_rarg3:
1031 // [tos + 0]: saved c_rarg3
1032 // [tos + 1]: saved c_rarg2
1033 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1034 // [tos + 3]: saved flags
1035 // [tos + 4]: return address
1036 // * [tos + 5]: error message (char*)
1037 // * [tos + 6]: object to verify (oop)
1038 // * [tos + 7]: saved rax - saved by caller and bashed
1039 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1040 // * = popped on exit
1041 address generate_verify_oop() {
1042 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1043 address start = __ pc();
1044
1045 Label exit, error, in_Java_heap;
1046
1047 __ pushf();
1048 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1049
1050 __ push(r12);
1051
1052 // save c_rarg2 and c_rarg3
1053 __ push(c_rarg2);
1054 __ push(c_rarg3);
1055
1056 enum {
1057 // After previous pushes.
1058 oop_to_verify = 6 * wordSize,
1059 saved_rax = 7 * wordSize,
1060 saved_r10 = 8 * wordSize,
1061
1062 // Before the call to MacroAssembler::debug(), see below.
1063 return_addr = 16 * wordSize,
1064 error_msg = 17 * wordSize
1065 };
1066
1067 // get object
1068 __ movptr(rax, Address(rsp, oop_to_verify));
1069
1070 // make sure object is 'reasonable'
1071 __ testptr(rax, rax);
1072 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1073
1074 #if INCLUDE_ZGC
1075 if (UseZGC) {
1076 // Check if metadata bits indicate a bad oop
1077 __ testptr(rax, Address(r15_thread, ZThreadLocalData::address_bad_mask_offset()));
1078 __ jcc(Assembler::notZero, error);
1079 }
1080 #endif
1081
1082 // Check if the oop is in the right area of memory
1083 __ movptr(c_rarg2, rax);
1084 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1085 __ andptr(c_rarg2, c_rarg3);
1086 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1087 __ cmpptr(c_rarg2, c_rarg3);
1088 __ jcc(Assembler::zero, in_Java_heap);
1089 // Not in Java heap, but could be valid if it's a bufferable value type
1090 __ load_klass(c_rarg2, rax);
1091 __ movbool(c_rarg2, Address(c_rarg2, InstanceKlass::extra_flags_offset()));
1092 __ andptr(c_rarg2, InstanceKlass::_extra_is_bufferable);
1093 __ testbool(c_rarg2);
1094 __ jcc(Assembler::zero, error);
1095 __ bind(in_Java_heap);
1096
1097 // set r12 to heapbase for load_klass()
1098 __ reinit_heapbase();
1099
1100 // make sure klass is 'reasonable', which is not zero.
1101 __ load_klass(rax, rax); // get klass
1102 __ testptr(rax, rax);
1103 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1104
1105 // return if everything seems ok
1106 __ bind(exit);
1107 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1108 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1109 __ pop(c_rarg3); // restore c_rarg3
1110 __ pop(c_rarg2); // restore c_rarg2
1111 __ pop(r12); // restore r12
1112 __ popf(); // restore flags
1113 __ ret(4 * wordSize); // pop caller saved stuff
1114
1115 // handle errors
1116 __ bind(error);
1117 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1118 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1119 __ pop(c_rarg3); // get saved c_rarg3 back
1120 __ pop(c_rarg2); // get saved c_rarg2 back
1121 __ pop(r12); // get saved r12 back
1122 __ popf(); // get saved flags off stack --
1123 // will be ignored
1124
1125 __ pusha(); // push registers
1126 // (rip is already
1127 // already pushed)
1128 // debug(char* msg, int64_t pc, int64_t regs[])
1129 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1130 // pushed all the registers, so now the stack looks like:
1131 // [tos + 0] 16 saved registers
1132 // [tos + 16] return address
1133 // * [tos + 17] error message (char*)
1134 // * [tos + 18] object to verify (oop)
1135 // * [tos + 19] saved rax - saved by caller and bashed
1136 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1137 // * = popped on exit
1138
1139 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1140 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1141 __ movq(c_rarg2, rsp); // pass address of regs on stack
1142 __ mov(r12, rsp); // remember rsp
1143 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1144 __ andptr(rsp, -16); // align stack as required by ABI
1145 BLOCK_COMMENT("call MacroAssembler::debug");
1146 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1147 __ mov(rsp, r12); // restore rsp
1148 __ popa(); // pop registers (includes r12)
1149 __ ret(4 * wordSize); // pop caller saved stuff
1150
1151 return start;
1152 }
1153
1154 //
1155 // Verify that a register contains clean 32-bits positive value
1156 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1157 //
1158 // Input:
1159 // Rint - 32-bits value
1160 // Rtmp - scratch
1161 //
1162 void assert_clean_int(Register Rint, Register Rtmp) {
1163 #ifdef ASSERT
1164 Label L;
1165 assert_different_registers(Rtmp, Rint);
1166 __ movslq(Rtmp, Rint);
1167 __ cmpq(Rtmp, Rint);
1168 __ jcc(Assembler::equal, L);
1169 __ stop("high 32-bits of int value are not 0");
1170 __ bind(L);
1171 #endif
1172 }
1173
1174 // Generate overlap test for array copy stubs
1175 //
1176 // Input:
1177 // c_rarg0 - from
1178 // c_rarg1 - to
1179 // c_rarg2 - element count
1180 //
1181 // Output:
1182 // rax - &from[element count - 1]
1183 //
1184 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1185 assert(no_overlap_target != NULL, "must be generated");
1186 array_overlap_test(no_overlap_target, NULL, sf);
1187 }
1188 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1189 array_overlap_test(NULL, &L_no_overlap, sf);
1190 }
1191 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1192 const Register from = c_rarg0;
1193 const Register to = c_rarg1;
1194 const Register count = c_rarg2;
1195 const Register end_from = rax;
1196
1197 __ cmpptr(to, from);
1198 __ lea(end_from, Address(from, count, sf, 0));
1199 if (NOLp == NULL) {
1200 ExternalAddress no_overlap(no_overlap_target);
1201 __ jump_cc(Assembler::belowEqual, no_overlap);
1202 __ cmpptr(to, end_from);
1203 __ jump_cc(Assembler::aboveEqual, no_overlap);
1204 } else {
1205 __ jcc(Assembler::belowEqual, (*NOLp));
1206 __ cmpptr(to, end_from);
1207 __ jcc(Assembler::aboveEqual, (*NOLp));
1208 }
1209 }
1210
1211 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1212 //
1213 // Outputs:
1214 // rdi - rcx
1215 // rsi - rdx
1216 // rdx - r8
1217 // rcx - r9
1218 //
1219 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1220 // are non-volatile. r9 and r10 should not be used by the caller.
1221 //
1222 DEBUG_ONLY(bool regs_in_thread;)
1223
1224 void setup_arg_regs(int nargs = 3) {
1225 const Register saved_rdi = r9;
1226 const Register saved_rsi = r10;
1227 assert(nargs == 3 || nargs == 4, "else fix");
1228 #ifdef _WIN64
1229 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1230 "unexpected argument registers");
1231 if (nargs >= 4)
1232 __ mov(rax, r9); // r9 is also saved_rdi
1233 __ movptr(saved_rdi, rdi);
1234 __ movptr(saved_rsi, rsi);
1235 __ mov(rdi, rcx); // c_rarg0
1236 __ mov(rsi, rdx); // c_rarg1
1237 __ mov(rdx, r8); // c_rarg2
1238 if (nargs >= 4)
1239 __ mov(rcx, rax); // c_rarg3 (via rax)
1240 #else
1241 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1242 "unexpected argument registers");
1243 #endif
1244 DEBUG_ONLY(regs_in_thread = false;)
1245 }
1246
1247 void restore_arg_regs() {
1248 assert(!regs_in_thread, "wrong call to restore_arg_regs");
1249 const Register saved_rdi = r9;
1250 const Register saved_rsi = r10;
1251 #ifdef _WIN64
1252 __ movptr(rdi, saved_rdi);
1253 __ movptr(rsi, saved_rsi);
1254 #endif
1255 }
1256
1257 // This is used in places where r10 is a scratch register, and can
1258 // be adapted if r9 is needed also.
1259 void setup_arg_regs_using_thread() {
1260 const Register saved_r15 = r9;
1261 #ifdef _WIN64
1262 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1263 __ get_thread(r15_thread);
1264 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1265 "unexpected argument registers");
1266 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1267 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1268
1269 __ mov(rdi, rcx); // c_rarg0
1270 __ mov(rsi, rdx); // c_rarg1
1271 __ mov(rdx, r8); // c_rarg2
1272 #else
1273 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1274 "unexpected argument registers");
1275 #endif
1276 DEBUG_ONLY(regs_in_thread = true;)
1277 }
1278
1279 void restore_arg_regs_using_thread() {
1280 assert(regs_in_thread, "wrong call to restore_arg_regs");
1281 const Register saved_r15 = r9;
1282 #ifdef _WIN64
1283 __ get_thread(r15_thread);
1284 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1285 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1286 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1287 #endif
1288 }
1289
1290 // Copy big chunks forward
1291 //
1292 // Inputs:
1293 // end_from - source arrays end address
1294 // end_to - destination array end address
1295 // qword_count - 64-bits element count, negative
1296 // to - scratch
1297 // L_copy_bytes - entry label
1298 // L_copy_8_bytes - exit label
1299 //
1300 void copy_bytes_forward(Register end_from, Register end_to,
1301 Register qword_count, Register to,
1302 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1303 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1304 Label L_loop;
1305 __ align(OptoLoopAlignment);
1306 if (UseUnalignedLoadStores) {
1307 Label L_end;
1308 // Copy 64-bytes per iteration
1309 __ BIND(L_loop);
1310 if (UseAVX > 2) {
1311 __ evmovdqul(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1312 __ evmovdqul(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1313 } else if (UseAVX == 2) {
1314 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1315 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1316 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1317 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1318 } else {
1319 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1320 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1321 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1322 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1323 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1324 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1325 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1326 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1327 }
1328 __ BIND(L_copy_bytes);
1329 __ addptr(qword_count, 8);
1330 __ jcc(Assembler::lessEqual, L_loop);
1331 __ subptr(qword_count, 4); // sub(8) and add(4)
1332 __ jccb(Assembler::greater, L_end);
1333 // Copy trailing 32 bytes
1334 if (UseAVX >= 2) {
1335 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1336 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1337 } else {
1338 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1339 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1340 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1341 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1342 }
1343 __ addptr(qword_count, 4);
1344 __ BIND(L_end);
1345 if (UseAVX >= 2) {
1346 // clean upper bits of YMM registers
1347 __ vpxor(xmm0, xmm0);
1348 __ vpxor(xmm1, xmm1);
1349 }
1350 } else {
1351 // Copy 32-bytes per iteration
1352 __ BIND(L_loop);
1353 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1354 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1355 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1356 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1357 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1358 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1359 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1360 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1361
1362 __ BIND(L_copy_bytes);
1363 __ addptr(qword_count, 4);
1364 __ jcc(Assembler::lessEqual, L_loop);
1365 }
1366 __ subptr(qword_count, 4);
1367 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1368 }
1369
1370 // Copy big chunks backward
1371 //
1372 // Inputs:
1373 // from - source arrays address
1374 // dest - destination array address
1375 // qword_count - 64-bits element count
1376 // to - scratch
1377 // L_copy_bytes - entry label
1378 // L_copy_8_bytes - exit label
1379 //
1380 void copy_bytes_backward(Register from, Register dest,
1381 Register qword_count, Register to,
1382 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1383 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1384 Label L_loop;
1385 __ align(OptoLoopAlignment);
1386 if (UseUnalignedLoadStores) {
1387 Label L_end;
1388 // Copy 64-bytes per iteration
1389 __ BIND(L_loop);
1390 if (UseAVX > 2) {
1391 __ evmovdqul(xmm0, Address(from, qword_count, Address::times_8, 0), Assembler::AVX_512bit);
1392 __ evmovdqul(Address(dest, qword_count, Address::times_8, 0), xmm0, Assembler::AVX_512bit);
1393 } else if (UseAVX == 2) {
1394 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1395 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1396 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1397 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1398 } else {
1399 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1400 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1401 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1402 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1403 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1404 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1405 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1406 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1407 }
1408 __ BIND(L_copy_bytes);
1409 __ subptr(qword_count, 8);
1410 __ jcc(Assembler::greaterEqual, L_loop);
1411
1412 __ addptr(qword_count, 4); // add(8) and sub(4)
1413 __ jccb(Assembler::less, L_end);
1414 // Copy trailing 32 bytes
1415 if (UseAVX >= 2) {
1416 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1417 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1418 } else {
1419 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1420 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1421 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1422 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1423 }
1424 __ subptr(qword_count, 4);
1425 __ BIND(L_end);
1426 if (UseAVX >= 2) {
1427 // clean upper bits of YMM registers
1428 __ vpxor(xmm0, xmm0);
1429 __ vpxor(xmm1, xmm1);
1430 }
1431 } else {
1432 // Copy 32-bytes per iteration
1433 __ BIND(L_loop);
1434 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1435 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1436 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1437 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1438 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1439 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1440 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1441 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1442
1443 __ BIND(L_copy_bytes);
1444 __ subptr(qword_count, 4);
1445 __ jcc(Assembler::greaterEqual, L_loop);
1446 }
1447 __ addptr(qword_count, 4);
1448 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1449 }
1450
1451
1452 // Arguments:
1453 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1454 // ignored
1455 // name - stub name string
1456 //
1457 // Inputs:
1458 // c_rarg0 - source array address
1459 // c_rarg1 - destination array address
1460 // c_rarg2 - element count, treated as ssize_t, can be zero
1461 //
1462 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1463 // we let the hardware handle it. The one to eight bytes within words,
1464 // dwords or qwords that span cache line boundaries will still be loaded
1465 // and stored atomically.
1466 //
1467 // Side Effects:
1468 // disjoint_byte_copy_entry is set to the no-overlap entry point
1469 // used by generate_conjoint_byte_copy().
1470 //
1471 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1472 __ align(CodeEntryAlignment);
1473 StubCodeMark mark(this, "StubRoutines", name);
1474 address start = __ pc();
1475
1476 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1477 Label L_copy_byte, L_exit;
1478 const Register from = rdi; // source array address
1479 const Register to = rsi; // destination array address
1480 const Register count = rdx; // elements count
1481 const Register byte_count = rcx;
1482 const Register qword_count = count;
1483 const Register end_from = from; // source array end address
1484 const Register end_to = to; // destination array end address
1485 // End pointers are inclusive, and if count is not zero they point
1486 // to the last unit copied: end_to[0] := end_from[0]
1487
1488 __ enter(); // required for proper stackwalking of RuntimeStub frame
1489 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1490
1491 if (entry != NULL) {
1492 *entry = __ pc();
1493 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1494 BLOCK_COMMENT("Entry:");
1495 }
1496
1497 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1498 // r9 and r10 may be used to save non-volatile registers
1499
1500 // 'from', 'to' and 'count' are now valid
1501 __ movptr(byte_count, count);
1502 __ shrptr(count, 3); // count => qword_count
1503
1504 // Copy from low to high addresses. Use 'to' as scratch.
1505 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1506 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1507 __ negptr(qword_count); // make the count negative
1508 __ jmp(L_copy_bytes);
1509
1510 // Copy trailing qwords
1511 __ BIND(L_copy_8_bytes);
1512 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1513 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1514 __ increment(qword_count);
1515 __ jcc(Assembler::notZero, L_copy_8_bytes);
1516
1517 // Check for and copy trailing dword
1518 __ BIND(L_copy_4_bytes);
1519 __ testl(byte_count, 4);
1520 __ jccb(Assembler::zero, L_copy_2_bytes);
1521 __ movl(rax, Address(end_from, 8));
1522 __ movl(Address(end_to, 8), rax);
1523
1524 __ addptr(end_from, 4);
1525 __ addptr(end_to, 4);
1526
1527 // Check for and copy trailing word
1528 __ BIND(L_copy_2_bytes);
1529 __ testl(byte_count, 2);
1530 __ jccb(Assembler::zero, L_copy_byte);
1531 __ movw(rax, Address(end_from, 8));
1532 __ movw(Address(end_to, 8), rax);
1533
1534 __ addptr(end_from, 2);
1535 __ addptr(end_to, 2);
1536
1537 // Check for and copy trailing byte
1538 __ BIND(L_copy_byte);
1539 __ testl(byte_count, 1);
1540 __ jccb(Assembler::zero, L_exit);
1541 __ movb(rax, Address(end_from, 8));
1542 __ movb(Address(end_to, 8), rax);
1543
1544 __ BIND(L_exit);
1545 restore_arg_regs();
1546 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1547 __ xorptr(rax, rax); // return 0
1548 __ vzeroupper();
1549 __ leave(); // required for proper stackwalking of RuntimeStub frame
1550 __ ret(0);
1551
1552 // Copy in multi-bytes chunks
1553 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1554 __ jmp(L_copy_4_bytes);
1555
1556 return start;
1557 }
1558
1559 // Arguments:
1560 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1561 // ignored
1562 // name - stub name string
1563 //
1564 // Inputs:
1565 // c_rarg0 - source array address
1566 // c_rarg1 - destination array address
1567 // c_rarg2 - element count, treated as ssize_t, can be zero
1568 //
1569 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1570 // we let the hardware handle it. The one to eight bytes within words,
1571 // dwords or qwords that span cache line boundaries will still be loaded
1572 // and stored atomically.
1573 //
1574 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1575 address* entry, const char *name) {
1576 __ align(CodeEntryAlignment);
1577 StubCodeMark mark(this, "StubRoutines", name);
1578 address start = __ pc();
1579
1580 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1581 const Register from = rdi; // source array address
1582 const Register to = rsi; // destination array address
1583 const Register count = rdx; // elements count
1584 const Register byte_count = rcx;
1585 const Register qword_count = count;
1586
1587 __ enter(); // required for proper stackwalking of RuntimeStub frame
1588 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1589
1590 if (entry != NULL) {
1591 *entry = __ pc();
1592 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1593 BLOCK_COMMENT("Entry:");
1594 }
1595
1596 array_overlap_test(nooverlap_target, Address::times_1);
1597 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1598 // r9 and r10 may be used to save non-volatile registers
1599
1600 // 'from', 'to' and 'count' are now valid
1601 __ movptr(byte_count, count);
1602 __ shrptr(count, 3); // count => qword_count
1603
1604 // Copy from high to low addresses.
1605
1606 // Check for and copy trailing byte
1607 __ testl(byte_count, 1);
1608 __ jcc(Assembler::zero, L_copy_2_bytes);
1609 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1610 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1611 __ decrement(byte_count); // Adjust for possible trailing word
1612
1613 // Check for and copy trailing word
1614 __ BIND(L_copy_2_bytes);
1615 __ testl(byte_count, 2);
1616 __ jcc(Assembler::zero, L_copy_4_bytes);
1617 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1618 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1619
1620 // Check for and copy trailing dword
1621 __ BIND(L_copy_4_bytes);
1622 __ testl(byte_count, 4);
1623 __ jcc(Assembler::zero, L_copy_bytes);
1624 __ movl(rax, Address(from, qword_count, Address::times_8));
1625 __ movl(Address(to, qword_count, Address::times_8), rax);
1626 __ jmp(L_copy_bytes);
1627
1628 // Copy trailing qwords
1629 __ BIND(L_copy_8_bytes);
1630 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1631 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1632 __ decrement(qword_count);
1633 __ jcc(Assembler::notZero, L_copy_8_bytes);
1634
1635 restore_arg_regs();
1636 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1637 __ xorptr(rax, rax); // return 0
1638 __ vzeroupper();
1639 __ leave(); // required for proper stackwalking of RuntimeStub frame
1640 __ ret(0);
1641
1642 // Copy in multi-bytes chunks
1643 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1644
1645 restore_arg_regs();
1646 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1647 __ xorptr(rax, rax); // return 0
1648 __ vzeroupper();
1649 __ leave(); // required for proper stackwalking of RuntimeStub frame
1650 __ ret(0);
1651
1652 return start;
1653 }
1654
1655 // Arguments:
1656 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1657 // ignored
1658 // name - stub name string
1659 //
1660 // Inputs:
1661 // c_rarg0 - source array address
1662 // c_rarg1 - destination array address
1663 // c_rarg2 - element count, treated as ssize_t, can be zero
1664 //
1665 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1666 // let the hardware handle it. The two or four words within dwords
1667 // or qwords that span cache line boundaries will still be loaded
1668 // and stored atomically.
1669 //
1670 // Side Effects:
1671 // disjoint_short_copy_entry is set to the no-overlap entry point
1672 // used by generate_conjoint_short_copy().
1673 //
1674 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1675 __ align(CodeEntryAlignment);
1676 StubCodeMark mark(this, "StubRoutines", name);
1677 address start = __ pc();
1678
1679 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1680 const Register from = rdi; // source array address
1681 const Register to = rsi; // destination array address
1682 const Register count = rdx; // elements count
1683 const Register word_count = rcx;
1684 const Register qword_count = count;
1685 const Register end_from = from; // source array end address
1686 const Register end_to = to; // destination array end address
1687 // End pointers are inclusive, and if count is not zero they point
1688 // to the last unit copied: end_to[0] := end_from[0]
1689
1690 __ enter(); // required for proper stackwalking of RuntimeStub frame
1691 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1692
1693 if (entry != NULL) {
1694 *entry = __ pc();
1695 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1696 BLOCK_COMMENT("Entry:");
1697 }
1698
1699 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1700 // r9 and r10 may be used to save non-volatile registers
1701
1702 // 'from', 'to' and 'count' are now valid
1703 __ movptr(word_count, count);
1704 __ shrptr(count, 2); // count => qword_count
1705
1706 // Copy from low to high addresses. Use 'to' as scratch.
1707 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1708 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1709 __ negptr(qword_count);
1710 __ jmp(L_copy_bytes);
1711
1712 // Copy trailing qwords
1713 __ BIND(L_copy_8_bytes);
1714 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1715 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1716 __ increment(qword_count);
1717 __ jcc(Assembler::notZero, L_copy_8_bytes);
1718
1719 // Original 'dest' is trashed, so we can't use it as a
1720 // base register for a possible trailing word copy
1721
1722 // Check for and copy trailing dword
1723 __ BIND(L_copy_4_bytes);
1724 __ testl(word_count, 2);
1725 __ jccb(Assembler::zero, L_copy_2_bytes);
1726 __ movl(rax, Address(end_from, 8));
1727 __ movl(Address(end_to, 8), rax);
1728
1729 __ addptr(end_from, 4);
1730 __ addptr(end_to, 4);
1731
1732 // Check for and copy trailing word
1733 __ BIND(L_copy_2_bytes);
1734 __ testl(word_count, 1);
1735 __ jccb(Assembler::zero, L_exit);
1736 __ movw(rax, Address(end_from, 8));
1737 __ movw(Address(end_to, 8), rax);
1738
1739 __ BIND(L_exit);
1740 restore_arg_regs();
1741 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1742 __ xorptr(rax, rax); // return 0
1743 __ vzeroupper();
1744 __ leave(); // required for proper stackwalking of RuntimeStub frame
1745 __ ret(0);
1746
1747 // Copy in multi-bytes chunks
1748 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1749 __ jmp(L_copy_4_bytes);
1750
1751 return start;
1752 }
1753
1754 address generate_fill(BasicType t, bool aligned, const char *name) {
1755 __ align(CodeEntryAlignment);
1756 StubCodeMark mark(this, "StubRoutines", name);
1757 address start = __ pc();
1758
1759 BLOCK_COMMENT("Entry:");
1760
1761 const Register to = c_rarg0; // source array address
1762 const Register value = c_rarg1; // value
1763 const Register count = c_rarg2; // elements count
1764
1765 __ enter(); // required for proper stackwalking of RuntimeStub frame
1766
1767 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1768
1769 __ vzeroupper();
1770 __ leave(); // required for proper stackwalking of RuntimeStub frame
1771 __ ret(0);
1772 return start;
1773 }
1774
1775 // Arguments:
1776 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1777 // ignored
1778 // name - stub name string
1779 //
1780 // Inputs:
1781 // c_rarg0 - source array address
1782 // c_rarg1 - destination array address
1783 // c_rarg2 - element count, treated as ssize_t, can be zero
1784 //
1785 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1786 // let the hardware handle it. The two or four words within dwords
1787 // or qwords that span cache line boundaries will still be loaded
1788 // and stored atomically.
1789 //
1790 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1791 address *entry, const char *name) {
1792 __ align(CodeEntryAlignment);
1793 StubCodeMark mark(this, "StubRoutines", name);
1794 address start = __ pc();
1795
1796 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1797 const Register from = rdi; // source array address
1798 const Register to = rsi; // destination array address
1799 const Register count = rdx; // elements count
1800 const Register word_count = rcx;
1801 const Register qword_count = count;
1802
1803 __ enter(); // required for proper stackwalking of RuntimeStub frame
1804 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1805
1806 if (entry != NULL) {
1807 *entry = __ pc();
1808 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1809 BLOCK_COMMENT("Entry:");
1810 }
1811
1812 array_overlap_test(nooverlap_target, Address::times_2);
1813 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1814 // r9 and r10 may be used to save non-volatile registers
1815
1816 // 'from', 'to' and 'count' are now valid
1817 __ movptr(word_count, count);
1818 __ shrptr(count, 2); // count => qword_count
1819
1820 // Copy from high to low addresses. Use 'to' as scratch.
1821
1822 // Check for and copy trailing word
1823 __ testl(word_count, 1);
1824 __ jccb(Assembler::zero, L_copy_4_bytes);
1825 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1826 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1827
1828 // Check for and copy trailing dword
1829 __ BIND(L_copy_4_bytes);
1830 __ testl(word_count, 2);
1831 __ jcc(Assembler::zero, L_copy_bytes);
1832 __ movl(rax, Address(from, qword_count, Address::times_8));
1833 __ movl(Address(to, qword_count, Address::times_8), rax);
1834 __ jmp(L_copy_bytes);
1835
1836 // Copy trailing qwords
1837 __ BIND(L_copy_8_bytes);
1838 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1839 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1840 __ decrement(qword_count);
1841 __ jcc(Assembler::notZero, L_copy_8_bytes);
1842
1843 restore_arg_regs();
1844 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1845 __ xorptr(rax, rax); // return 0
1846 __ vzeroupper();
1847 __ leave(); // required for proper stackwalking of RuntimeStub frame
1848 __ ret(0);
1849
1850 // Copy in multi-bytes chunks
1851 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1852
1853 restore_arg_regs();
1854 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1855 __ xorptr(rax, rax); // return 0
1856 __ vzeroupper();
1857 __ leave(); // required for proper stackwalking of RuntimeStub frame
1858 __ ret(0);
1859
1860 return start;
1861 }
1862
1863 // Arguments:
1864 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1865 // ignored
1866 // is_oop - true => oop array, so generate store check code
1867 // name - stub name string
1868 //
1869 // Inputs:
1870 // c_rarg0 - source array address
1871 // c_rarg1 - destination array address
1872 // c_rarg2 - element count, treated as ssize_t, can be zero
1873 //
1874 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1875 // the hardware handle it. The two dwords within qwords that span
1876 // cache line boundaries will still be loaded and stored atomicly.
1877 //
1878 // Side Effects:
1879 // disjoint_int_copy_entry is set to the no-overlap entry point
1880 // used by generate_conjoint_int_oop_copy().
1881 //
1882 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1883 const char *name, bool dest_uninitialized = false) {
1884 __ align(CodeEntryAlignment);
1885 StubCodeMark mark(this, "StubRoutines", name);
1886 address start = __ pc();
1887
1888 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1889 const Register from = rdi; // source array address
1890 const Register to = rsi; // destination array address
1891 const Register count = rdx; // elements count
1892 const Register dword_count = rcx;
1893 const Register qword_count = count;
1894 const Register end_from = from; // source array end address
1895 const Register end_to = to; // destination array end address
1896 // End pointers are inclusive, and if count is not zero they point
1897 // to the last unit copied: end_to[0] := end_from[0]
1898
1899 __ enter(); // required for proper stackwalking of RuntimeStub frame
1900 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1901
1902 if (entry != NULL) {
1903 *entry = __ pc();
1904 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1905 BLOCK_COMMENT("Entry:");
1906 }
1907
1908 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1909 // r9 is used to save r15_thread
1910
1911 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
1912 if (dest_uninitialized) {
1913 decorators |= IS_DEST_UNINITIALIZED;
1914 }
1915 if (aligned) {
1916 decorators |= ARRAYCOPY_ALIGNED;
1917 }
1918
1919 BasicType type = is_oop ? T_OBJECT : T_INT;
1920 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1921 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1922
1923 // 'from', 'to' and 'count' are now valid
1924 __ movptr(dword_count, count);
1925 __ shrptr(count, 1); // count => qword_count
1926
1927 // Copy from low to high addresses. Use 'to' as scratch.
1928 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1929 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1930 __ negptr(qword_count);
1931 __ jmp(L_copy_bytes);
1932
1933 // Copy trailing qwords
1934 __ BIND(L_copy_8_bytes);
1935 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1936 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1937 __ increment(qword_count);
1938 __ jcc(Assembler::notZero, L_copy_8_bytes);
1939
1940 // Check for and copy trailing dword
1941 __ BIND(L_copy_4_bytes);
1942 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1943 __ jccb(Assembler::zero, L_exit);
1944 __ movl(rax, Address(end_from, 8));
1945 __ movl(Address(end_to, 8), rax);
1946
1947 __ BIND(L_exit);
1948 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
1949 restore_arg_regs_using_thread();
1950 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1951 __ vzeroupper();
1952 __ xorptr(rax, rax); // return 0
1953 __ leave(); // required for proper stackwalking of RuntimeStub frame
1954 __ ret(0);
1955
1956 // Copy in multi-bytes chunks
1957 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1958 __ jmp(L_copy_4_bytes);
1959
1960 return start;
1961 }
1962
1963 // Arguments:
1964 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1965 // ignored
1966 // is_oop - true => oop array, so generate store check code
1967 // name - stub name string
1968 //
1969 // Inputs:
1970 // c_rarg0 - source array address
1971 // c_rarg1 - destination array address
1972 // c_rarg2 - element count, treated as ssize_t, can be zero
1973 //
1974 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1975 // the hardware handle it. The two dwords within qwords that span
1976 // cache line boundaries will still be loaded and stored atomicly.
1977 //
1978 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1979 address *entry, const char *name,
1980 bool dest_uninitialized = false) {
1981 __ align(CodeEntryAlignment);
1982 StubCodeMark mark(this, "StubRoutines", name);
1983 address start = __ pc();
1984
1985 Label L_copy_bytes, L_copy_8_bytes, L_exit;
1986 const Register from = rdi; // source array address
1987 const Register to = rsi; // destination array address
1988 const Register count = rdx; // elements count
1989 const Register dword_count = rcx;
1990 const Register qword_count = count;
1991
1992 __ enter(); // required for proper stackwalking of RuntimeStub frame
1993 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1994
1995 if (entry != NULL) {
1996 *entry = __ pc();
1997 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1998 BLOCK_COMMENT("Entry:");
1999 }
2000
2001 array_overlap_test(nooverlap_target, Address::times_4);
2002 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2003 // r9 is used to save r15_thread
2004
2005 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2006 if (dest_uninitialized) {
2007 decorators |= IS_DEST_UNINITIALIZED;
2008 }
2009 if (aligned) {
2010 decorators |= ARRAYCOPY_ALIGNED;
2011 }
2012
2013 BasicType type = is_oop ? T_OBJECT : T_INT;
2014 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2015 // no registers are destroyed by this call
2016 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2017
2018 assert_clean_int(count, rax); // Make sure 'count' is clean int.
2019 // 'from', 'to' and 'count' are now valid
2020 __ movptr(dword_count, count);
2021 __ shrptr(count, 1); // count => qword_count
2022
2023 // Copy from high to low addresses. Use 'to' as scratch.
2024
2025 // Check for and copy trailing dword
2026 __ testl(dword_count, 1);
2027 __ jcc(Assembler::zero, L_copy_bytes);
2028 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
2029 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
2030 __ jmp(L_copy_bytes);
2031
2032 // Copy trailing qwords
2033 __ BIND(L_copy_8_bytes);
2034 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2035 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2036 __ decrement(qword_count);
2037 __ jcc(Assembler::notZero, L_copy_8_bytes);
2038
2039 if (is_oop) {
2040 __ jmp(L_exit);
2041 }
2042 restore_arg_regs_using_thread();
2043 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2044 __ xorptr(rax, rax); // return 0
2045 __ vzeroupper();
2046 __ leave(); // required for proper stackwalking of RuntimeStub frame
2047 __ ret(0);
2048
2049 // Copy in multi-bytes chunks
2050 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2051
2052 __ BIND(L_exit);
2053 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2054 restore_arg_regs_using_thread();
2055 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2056 __ xorptr(rax, rax); // return 0
2057 __ vzeroupper();
2058 __ leave(); // required for proper stackwalking of RuntimeStub frame
2059 __ ret(0);
2060
2061 return start;
2062 }
2063
2064 // Arguments:
2065 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2066 // ignored
2067 // is_oop - true => oop array, so generate store check code
2068 // name - stub name string
2069 //
2070 // Inputs:
2071 // c_rarg0 - source array address
2072 // c_rarg1 - destination array address
2073 // c_rarg2 - element count, treated as ssize_t, can be zero
2074 //
2075 // Side Effects:
2076 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2077 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2078 //
2079 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2080 const char *name, bool dest_uninitialized = false) {
2081 __ align(CodeEntryAlignment);
2082 StubCodeMark mark(this, "StubRoutines", name);
2083 address start = __ pc();
2084
2085 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2086 const Register from = rdi; // source array address
2087 const Register to = rsi; // destination array address
2088 const Register qword_count = rdx; // elements count
2089 const Register end_from = from; // source array end address
2090 const Register end_to = rcx; // destination array end address
2091 const Register saved_count = r11;
2092 // End pointers are inclusive, and if count is not zero they point
2093 // to the last unit copied: end_to[0] := end_from[0]
2094
2095 __ enter(); // required for proper stackwalking of RuntimeStub frame
2096 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2097 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2098
2099 if (entry != NULL) {
2100 *entry = __ pc();
2101 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2102 BLOCK_COMMENT("Entry:");
2103 }
2104
2105 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2106 // r9 is used to save r15_thread
2107 // 'from', 'to' and 'qword_count' are now valid
2108
2109 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2110 if (dest_uninitialized) {
2111 decorators |= IS_DEST_UNINITIALIZED;
2112 }
2113 if (aligned) {
2114 decorators |= ARRAYCOPY_ALIGNED;
2115 }
2116
2117 BasicType type = is_oop ? T_OBJECT : T_LONG;
2118 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2119 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2120
2121 // Copy from low to high addresses. Use 'to' as scratch.
2122 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2123 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2124 __ negptr(qword_count);
2125 __ jmp(L_copy_bytes);
2126
2127 // Copy trailing qwords
2128 __ BIND(L_copy_8_bytes);
2129 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2130 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2131 __ increment(qword_count);
2132 __ jcc(Assembler::notZero, L_copy_8_bytes);
2133
2134 if (is_oop) {
2135 __ jmp(L_exit);
2136 } else {
2137 restore_arg_regs_using_thread();
2138 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2139 __ xorptr(rax, rax); // return 0
2140 __ vzeroupper();
2141 __ leave(); // required for proper stackwalking of RuntimeStub frame
2142 __ ret(0);
2143 }
2144
2145 // Copy in multi-bytes chunks
2146 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2147
2148 __ BIND(L_exit);
2149 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2150 restore_arg_regs_using_thread();
2151 if (is_oop) {
2152 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2153 } else {
2154 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2155 }
2156 __ vzeroupper();
2157 __ xorptr(rax, rax); // return 0
2158 __ leave(); // required for proper stackwalking of RuntimeStub frame
2159 __ ret(0);
2160
2161 return start;
2162 }
2163
2164 // Arguments:
2165 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2166 // ignored
2167 // is_oop - true => oop array, so generate store check code
2168 // name - stub name string
2169 //
2170 // Inputs:
2171 // c_rarg0 - source array address
2172 // c_rarg1 - destination array address
2173 // c_rarg2 - element count, treated as ssize_t, can be zero
2174 //
2175 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2176 address nooverlap_target, address *entry,
2177 const char *name, bool dest_uninitialized = false) {
2178 __ align(CodeEntryAlignment);
2179 StubCodeMark mark(this, "StubRoutines", name);
2180 address start = __ pc();
2181
2182 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2183 const Register from = rdi; // source array address
2184 const Register to = rsi; // destination array address
2185 const Register qword_count = rdx; // elements count
2186 const Register saved_count = rcx;
2187
2188 __ enter(); // required for proper stackwalking of RuntimeStub frame
2189 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2190
2191 if (entry != NULL) {
2192 *entry = __ pc();
2193 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2194 BLOCK_COMMENT("Entry:");
2195 }
2196
2197 array_overlap_test(nooverlap_target, Address::times_8);
2198 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2199 // r9 is used to save r15_thread
2200 // 'from', 'to' and 'qword_count' are now valid
2201
2202 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2203 if (dest_uninitialized) {
2204 decorators |= IS_DEST_UNINITIALIZED;
2205 }
2206 if (aligned) {
2207 decorators |= ARRAYCOPY_ALIGNED;
2208 }
2209
2210 BasicType type = is_oop ? T_OBJECT : T_LONG;
2211 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2212 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2213
2214 __ jmp(L_copy_bytes);
2215
2216 // Copy trailing qwords
2217 __ BIND(L_copy_8_bytes);
2218 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2219 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2220 __ decrement(qword_count);
2221 __ jcc(Assembler::notZero, L_copy_8_bytes);
2222
2223 if (is_oop) {
2224 __ jmp(L_exit);
2225 } else {
2226 restore_arg_regs_using_thread();
2227 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2228 __ xorptr(rax, rax); // return 0
2229 __ vzeroupper();
2230 __ leave(); // required for proper stackwalking of RuntimeStub frame
2231 __ ret(0);
2232 }
2233
2234 // Copy in multi-bytes chunks
2235 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2236
2237 __ BIND(L_exit);
2238 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2239 restore_arg_regs_using_thread();
2240 if (is_oop) {
2241 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2242 } else {
2243 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2244 }
2245 __ vzeroupper();
2246 __ xorptr(rax, rax); // return 0
2247 __ leave(); // required for proper stackwalking of RuntimeStub frame
2248 __ ret(0);
2249
2250 return start;
2251 }
2252
2253
2254 // Helper for generating a dynamic type check.
2255 // Smashes no registers.
2256 void generate_type_check(Register sub_klass,
2257 Register super_check_offset,
2258 Register super_klass,
2259 Label& L_success) {
2260 assert_different_registers(sub_klass, super_check_offset, super_klass);
2261
2262 BLOCK_COMMENT("type_check:");
2263
2264 Label L_miss;
2265
2266 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2267 super_check_offset);
2268 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2269
2270 // Fall through on failure!
2271 __ BIND(L_miss);
2272 }
2273
2274 //
2275 // Generate checkcasting array copy stub
2276 //
2277 // Input:
2278 // c_rarg0 - source array address
2279 // c_rarg1 - destination array address
2280 // c_rarg2 - element count, treated as ssize_t, can be zero
2281 // c_rarg3 - size_t ckoff (super_check_offset)
2282 // not Win64
2283 // c_rarg4 - oop ckval (super_klass)
2284 // Win64
2285 // rsp+40 - oop ckval (super_klass)
2286 //
2287 // Output:
2288 // rax == 0 - success
2289 // rax == -1^K - failure, where K is partial transfer count
2290 //
2291 address generate_checkcast_copy(const char *name, address *entry,
2292 bool dest_uninitialized = false) {
2293
2294 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2295
2296 // Input registers (after setup_arg_regs)
2297 const Register from = rdi; // source array address
2298 const Register to = rsi; // destination array address
2299 const Register length = rdx; // elements count
2300 const Register ckoff = rcx; // super_check_offset
2301 const Register ckval = r8; // super_klass
2302
2303 // Registers used as temps (r13, r14 are save-on-entry)
2304 const Register end_from = from; // source array end address
2305 const Register end_to = r13; // destination array end address
2306 const Register count = rdx; // -(count_remaining)
2307 const Register r14_length = r14; // saved copy of length
2308 // End pointers are inclusive, and if length is not zero they point
2309 // to the last unit copied: end_to[0] := end_from[0]
2310
2311 const Register rax_oop = rax; // actual oop copied
2312 const Register r11_klass = r11; // oop._klass
2313
2314 //---------------------------------------------------------------
2315 // Assembler stub will be used for this call to arraycopy
2316 // if the two arrays are subtypes of Object[] but the
2317 // destination array type is not equal to or a supertype
2318 // of the source type. Each element must be separately
2319 // checked.
2320
2321 __ align(CodeEntryAlignment);
2322 StubCodeMark mark(this, "StubRoutines", name);
2323 address start = __ pc();
2324
2325 __ enter(); // required for proper stackwalking of RuntimeStub frame
2326
2327 #ifdef ASSERT
2328 // caller guarantees that the arrays really are different
2329 // otherwise, we would have to make conjoint checks
2330 { Label L;
2331 array_overlap_test(L, TIMES_OOP);
2332 __ stop("checkcast_copy within a single array");
2333 __ bind(L);
2334 }
2335 #endif //ASSERT
2336
2337 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2338 // ckoff => rcx, ckval => r8
2339 // r9 and r10 may be used to save non-volatile registers
2340 #ifdef _WIN64
2341 // last argument (#4) is on stack on Win64
2342 __ movptr(ckval, Address(rsp, 6 * wordSize));
2343 #endif
2344
2345 // Caller of this entry point must set up the argument registers.
2346 if (entry != NULL) {
2347 *entry = __ pc();
2348 BLOCK_COMMENT("Entry:");
2349 }
2350
2351 // allocate spill slots for r13, r14
2352 enum {
2353 saved_r13_offset,
2354 saved_r14_offset,
2355 saved_r10_offset,
2356 saved_rbp_offset
2357 };
2358 __ subptr(rsp, saved_rbp_offset * wordSize);
2359 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2360 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2361 __ movptr(Address(rsp, saved_r10_offset * wordSize), r10);
2362
2363 #ifdef ASSERT
2364 Label L2;
2365 __ get_thread(r14);
2366 __ cmpptr(r15_thread, r14);
2367 __ jcc(Assembler::equal, L2);
2368 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
2369 __ bind(L2);
2370 #endif // ASSERT
2371
2372 // check that int operands are properly extended to size_t
2373 assert_clean_int(length, rax);
2374 assert_clean_int(ckoff, rax);
2375
2376 #ifdef ASSERT
2377 BLOCK_COMMENT("assert consistent ckoff/ckval");
2378 // The ckoff and ckval must be mutually consistent,
2379 // even though caller generates both.
2380 { Label L;
2381 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2382 __ cmpl(ckoff, Address(ckval, sco_offset));
2383 __ jcc(Assembler::equal, L);
2384 __ stop("super_check_offset inconsistent");
2385 __ bind(L);
2386 }
2387 #endif //ASSERT
2388
2389 // Loop-invariant addresses. They are exclusive end pointers.
2390 Address end_from_addr(from, length, TIMES_OOP, 0);
2391 Address end_to_addr(to, length, TIMES_OOP, 0);
2392 // Loop-variant addresses. They assume post-incremented count < 0.
2393 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2394 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2395
2396 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST | ARRAYCOPY_DISJOINT;
2397 if (dest_uninitialized) {
2398 decorators |= IS_DEST_UNINITIALIZED;
2399 }
2400
2401 BasicType type = T_OBJECT;
2402 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2403 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2404
2405 // Copy from low to high addresses, indexed from the end of each array.
2406 __ lea(end_from, end_from_addr);
2407 __ lea(end_to, end_to_addr);
2408 __ movptr(r14_length, length); // save a copy of the length
2409 assert(length == count, ""); // else fix next line:
2410 __ negptr(count); // negate and test the length
2411 __ jcc(Assembler::notZero, L_load_element);
2412
2413 // Empty array: Nothing to do.
2414 __ xorptr(rax, rax); // return 0 on (trivial) success
2415 __ jmp(L_done);
2416
2417 // ======== begin loop ========
2418 // (Loop is rotated; its entry is L_load_element.)
2419 // Loop control:
2420 // for (count = -count; count != 0; count++)
2421 // Base pointers src, dst are biased by 8*(count-1),to last element.
2422 __ align(OptoLoopAlignment);
2423
2424 __ BIND(L_store_element);
2425 __ store_heap_oop(to_element_addr, rax_oop, noreg, noreg, noreg, AS_RAW); // store the oop
2426 __ increment(count); // increment the count toward zero
2427 __ jcc(Assembler::zero, L_do_card_marks);
2428
2429 // ======== loop entry is here ========
2430 __ BIND(L_load_element);
2431 __ load_heap_oop(rax_oop, from_element_addr, noreg, noreg, AS_RAW); // load the oop
2432 __ testptr(rax_oop, rax_oop);
2433 __ jcc(Assembler::zero, L_store_element);
2434
2435 __ load_klass(r11_klass, rax_oop);// query the object klass
2436 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2437 // ======== end loop ========
2438
2439 // It was a real error; we must depend on the caller to finish the job.
2440 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2441 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2442 // and report their number to the caller.
2443 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2444 Label L_post_barrier;
2445 __ addptr(r14_length, count); // K = (original - remaining) oops
2446 __ movptr(rax, r14_length); // save the value
2447 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2448 __ jccb(Assembler::notZero, L_post_barrier);
2449 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2450
2451 // Come here on success only.
2452 __ BIND(L_do_card_marks);
2453 __ xorptr(rax, rax); // return 0 on success
2454
2455 __ BIND(L_post_barrier);
2456 bs->arraycopy_epilogue(_masm, decorators, type, from, to, r14_length);
2457
2458 // Common exit point (success or failure).
2459 __ BIND(L_done);
2460 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2461 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2462 __ movptr(r10, Address(rsp, saved_r10_offset * wordSize));
2463 restore_arg_regs();
2464 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2465 __ leave(); // required for proper stackwalking of RuntimeStub frame
2466 __ ret(0);
2467
2468 return start;
2469 }
2470
2471 //
2472 // Generate 'unsafe' array copy stub
2473 // Though just as safe as the other stubs, it takes an unscaled
2474 // size_t argument instead of an element count.
2475 //
2476 // Input:
2477 // c_rarg0 - source array address
2478 // c_rarg1 - destination array address
2479 // c_rarg2 - byte count, treated as ssize_t, can be zero
2480 //
2481 // Examines the alignment of the operands and dispatches
2482 // to a long, int, short, or byte copy loop.
2483 //
2484 address generate_unsafe_copy(const char *name,
2485 address byte_copy_entry, address short_copy_entry,
2486 address int_copy_entry, address long_copy_entry) {
2487
2488 Label L_long_aligned, L_int_aligned, L_short_aligned;
2489
2490 // Input registers (before setup_arg_regs)
2491 const Register from = c_rarg0; // source array address
2492 const Register to = c_rarg1; // destination array address
2493 const Register size = c_rarg2; // byte count (size_t)
2494
2495 // Register used as a temp
2496 const Register bits = rax; // test copy of low bits
2497
2498 __ align(CodeEntryAlignment);
2499 StubCodeMark mark(this, "StubRoutines", name);
2500 address start = __ pc();
2501
2502 __ enter(); // required for proper stackwalking of RuntimeStub frame
2503
2504 // bump this on entry, not on exit:
2505 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2506
2507 __ mov(bits, from);
2508 __ orptr(bits, to);
2509 __ orptr(bits, size);
2510
2511 __ testb(bits, BytesPerLong-1);
2512 __ jccb(Assembler::zero, L_long_aligned);
2513
2514 __ testb(bits, BytesPerInt-1);
2515 __ jccb(Assembler::zero, L_int_aligned);
2516
2517 __ testb(bits, BytesPerShort-1);
2518 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2519
2520 __ BIND(L_short_aligned);
2521 __ shrptr(size, LogBytesPerShort); // size => short_count
2522 __ jump(RuntimeAddress(short_copy_entry));
2523
2524 __ BIND(L_int_aligned);
2525 __ shrptr(size, LogBytesPerInt); // size => int_count
2526 __ jump(RuntimeAddress(int_copy_entry));
2527
2528 __ BIND(L_long_aligned);
2529 __ shrptr(size, LogBytesPerLong); // size => qword_count
2530 __ jump(RuntimeAddress(long_copy_entry));
2531
2532 return start;
2533 }
2534
2535 // Perform range checks on the proposed arraycopy.
2536 // Kills temp, but nothing else.
2537 // Also, clean the sign bits of src_pos and dst_pos.
2538 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2539 Register src_pos, // source position (c_rarg1)
2540 Register dst, // destination array oo (c_rarg2)
2541 Register dst_pos, // destination position (c_rarg3)
2542 Register length,
2543 Register temp,
2544 Label& L_failed) {
2545 BLOCK_COMMENT("arraycopy_range_checks:");
2546
2547 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2548 __ movl(temp, length);
2549 __ addl(temp, src_pos); // src_pos + length
2550 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2551 __ jcc(Assembler::above, L_failed);
2552
2553 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2554 __ movl(temp, length);
2555 __ addl(temp, dst_pos); // dst_pos + length
2556 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2557 __ jcc(Assembler::above, L_failed);
2558
2559 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2560 // Move with sign extension can be used since they are positive.
2561 __ movslq(src_pos, src_pos);
2562 __ movslq(dst_pos, dst_pos);
2563
2564 BLOCK_COMMENT("arraycopy_range_checks done");
2565 }
2566
2567 //
2568 // Generate generic array copy stubs
2569 //
2570 // Input:
2571 // c_rarg0 - src oop
2572 // c_rarg1 - src_pos (32-bits)
2573 // c_rarg2 - dst oop
2574 // c_rarg3 - dst_pos (32-bits)
2575 // not Win64
2576 // c_rarg4 - element count (32-bits)
2577 // Win64
2578 // rsp+40 - element count (32-bits)
2579 //
2580 // Output:
2581 // rax == 0 - success
2582 // rax == -1^K - failure, where K is partial transfer count
2583 //
2584 address generate_generic_copy(const char *name,
2585 address byte_copy_entry, address short_copy_entry,
2586 address int_copy_entry, address oop_copy_entry,
2587 address long_copy_entry, address checkcast_copy_entry) {
2588
2589 Label L_failed, L_failed_0, L_objArray;
2590 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2591
2592 // Input registers
2593 const Register src = c_rarg0; // source array oop
2594 const Register src_pos = c_rarg1; // source position
2595 const Register dst = c_rarg2; // destination array oop
2596 const Register dst_pos = c_rarg3; // destination position
2597 #ifndef _WIN64
2598 const Register length = c_rarg4;
2599 #else
2600 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2601 #endif
2602
2603 { int modulus = CodeEntryAlignment;
2604 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2605 int advance = target - (__ offset() % modulus);
2606 if (advance < 0) advance += modulus;
2607 if (advance > 0) __ nop(advance);
2608 }
2609 StubCodeMark mark(this, "StubRoutines", name);
2610
2611 // Short-hop target to L_failed. Makes for denser prologue code.
2612 __ BIND(L_failed_0);
2613 __ jmp(L_failed);
2614 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2615
2616 __ align(CodeEntryAlignment);
2617 address start = __ pc();
2618
2619 __ enter(); // required for proper stackwalking of RuntimeStub frame
2620
2621 // bump this on entry, not on exit:
2622 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2623
2624 //-----------------------------------------------------------------------
2625 // Assembler stub will be used for this call to arraycopy
2626 // if the following conditions are met:
2627 //
2628 // (1) src and dst must not be null.
2629 // (2) src_pos must not be negative.
2630 // (3) dst_pos must not be negative.
2631 // (4) length must not be negative.
2632 // (5) src klass and dst klass should be the same and not NULL.
2633 // (6) src and dst should be arrays.
2634 // (7) src_pos + length must not exceed length of src.
2635 // (8) dst_pos + length must not exceed length of dst.
2636 //
2637
2638 // if (src == NULL) return -1;
2639 __ testptr(src, src); // src oop
2640 size_t j1off = __ offset();
2641 __ jccb(Assembler::zero, L_failed_0);
2642
2643 // if (src_pos < 0) return -1;
2644 __ testl(src_pos, src_pos); // src_pos (32-bits)
2645 __ jccb(Assembler::negative, L_failed_0);
2646
2647 // if (dst == NULL) return -1;
2648 __ testptr(dst, dst); // dst oop
2649 __ jccb(Assembler::zero, L_failed_0);
2650
2651 // if (dst_pos < 0) return -1;
2652 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2653 size_t j4off = __ offset();
2654 __ jccb(Assembler::negative, L_failed_0);
2655
2656 // The first four tests are very dense code,
2657 // but not quite dense enough to put four
2658 // jumps in a 16-byte instruction fetch buffer.
2659 // That's good, because some branch predicters
2660 // do not like jumps so close together.
2661 // Make sure of this.
2662 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2663
2664 // registers used as temp
2665 const Register r11_length = r11; // elements count to copy
2666 const Register r10_src_klass = r10; // array klass
2667
2668 // if (length < 0) return -1;
2669 __ movl(r11_length, length); // length (elements count, 32-bits value)
2670 __ testl(r11_length, r11_length);
2671 __ jccb(Assembler::negative, L_failed_0);
2672
2673 __ load_klass(r10_src_klass, src);
2674 #ifdef ASSERT
2675 // assert(src->klass() != NULL);
2676 {
2677 BLOCK_COMMENT("assert klasses not null {");
2678 Label L1, L2;
2679 __ testptr(r10_src_klass, r10_src_klass);
2680 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2681 __ bind(L1);
2682 __ stop("broken null klass");
2683 __ bind(L2);
2684 __ load_klass(rax, dst);
2685 __ cmpq(rax, 0);
2686 __ jcc(Assembler::equal, L1); // this would be broken also
2687 BLOCK_COMMENT("} assert klasses not null done");
2688 }
2689 #endif
2690
2691 // Load layout helper (32-bits)
2692 //
2693 // |array_tag| | header_size | element_type | |log2_element_size|
2694 // 32 30 24 16 8 2 0
2695 //
2696 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2697 //
2698
2699 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2700
2701 // Handle objArrays completely differently...
2702 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2703 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2704 __ jcc(Assembler::equal, L_objArray);
2705
2706 // if (src->klass() != dst->klass()) return -1;
2707 __ load_klass(rax, dst);
2708 __ cmpq(r10_src_klass, rax);
2709 __ jcc(Assembler::notEqual, L_failed);
2710
2711 const Register rax_lh = rax; // layout helper
2712 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2713
2714 // if (!src->is_Array()) return -1;
2715 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2716 __ jcc(Assembler::greaterEqual, L_failed);
2717
2718 // At this point, it is known to be a typeArray (array_tag 0x3).
2719 #ifdef ASSERT
2720 {
2721 BLOCK_COMMENT("assert primitive array {");
2722 Label L;
2723 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2724 __ jcc(Assembler::greaterEqual, L);
2725 __ stop("must be a primitive array");
2726 __ bind(L);
2727 BLOCK_COMMENT("} assert primitive array done");
2728 }
2729 #endif
2730
2731 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2732 r10, L_failed);
2733
2734 // TypeArrayKlass
2735 //
2736 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2737 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2738 //
2739
2740 const Register r10_offset = r10; // array offset
2741 const Register rax_elsize = rax_lh; // element size
2742
2743 __ movl(r10_offset, rax_lh);
2744 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2745 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2746 __ addptr(src, r10_offset); // src array offset
2747 __ addptr(dst, r10_offset); // dst array offset
2748 BLOCK_COMMENT("choose copy loop based on element size");
2749 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2750
2751 // next registers should be set before the jump to corresponding stub
2752 const Register from = c_rarg0; // source array address
2753 const Register to = c_rarg1; // destination array address
2754 const Register count = c_rarg2; // elements count
2755
2756 // 'from', 'to', 'count' registers should be set in such order
2757 // since they are the same as 'src', 'src_pos', 'dst'.
2758
2759 __ BIND(L_copy_bytes);
2760 __ cmpl(rax_elsize, 0);
2761 __ jccb(Assembler::notEqual, L_copy_shorts);
2762 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2763 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2764 __ movl2ptr(count, r11_length); // length
2765 __ jump(RuntimeAddress(byte_copy_entry));
2766
2767 __ BIND(L_copy_shorts);
2768 __ cmpl(rax_elsize, LogBytesPerShort);
2769 __ jccb(Assembler::notEqual, L_copy_ints);
2770 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2771 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2772 __ movl2ptr(count, r11_length); // length
2773 __ jump(RuntimeAddress(short_copy_entry));
2774
2775 __ BIND(L_copy_ints);
2776 __ cmpl(rax_elsize, LogBytesPerInt);
2777 __ jccb(Assembler::notEqual, L_copy_longs);
2778 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2779 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2780 __ movl2ptr(count, r11_length); // length
2781 __ jump(RuntimeAddress(int_copy_entry));
2782
2783 __ BIND(L_copy_longs);
2784 #ifdef ASSERT
2785 {
2786 BLOCK_COMMENT("assert long copy {");
2787 Label L;
2788 __ cmpl(rax_elsize, LogBytesPerLong);
2789 __ jcc(Assembler::equal, L);
2790 __ stop("must be long copy, but elsize is wrong");
2791 __ bind(L);
2792 BLOCK_COMMENT("} assert long copy done");
2793 }
2794 #endif
2795 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2796 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2797 __ movl2ptr(count, r11_length); // length
2798 __ jump(RuntimeAddress(long_copy_entry));
2799
2800 // ObjArrayKlass
2801 __ BIND(L_objArray);
2802 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2803
2804 Label L_plain_copy, L_checkcast_copy;
2805 // test array classes for subtyping
2806 __ load_klass(rax, dst);
2807 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2808 __ jcc(Assembler::notEqual, L_checkcast_copy);
2809
2810 // Identically typed arrays can be copied without element-wise checks.
2811 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2812 r10, L_failed);
2813
2814 __ lea(from, Address(src, src_pos, TIMES_OOP,
2815 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2816 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2817 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2818 __ movl2ptr(count, r11_length); // length
2819 __ BIND(L_plain_copy);
2820 __ jump(RuntimeAddress(oop_copy_entry));
2821
2822 __ BIND(L_checkcast_copy);
2823 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2824 {
2825 // Before looking at dst.length, make sure dst is also an objArray.
2826 __ cmpl(Address(rax, lh_offset), objArray_lh);
2827 __ jcc(Assembler::notEqual, L_failed);
2828
2829 // It is safe to examine both src.length and dst.length.
2830 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2831 rax, L_failed);
2832
2833 const Register r11_dst_klass = r11;
2834 __ load_klass(r11_dst_klass, dst); // reload
2835
2836 // Marshal the base address arguments now, freeing registers.
2837 __ lea(from, Address(src, src_pos, TIMES_OOP,
2838 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2839 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2840 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2841 __ movl(count, length); // length (reloaded)
2842 Register sco_temp = c_rarg3; // this register is free now
2843 assert_different_registers(from, to, count, sco_temp,
2844 r11_dst_klass, r10_src_klass);
2845 assert_clean_int(count, sco_temp);
2846
2847 // Generate the type check.
2848 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2849 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2850 assert_clean_int(sco_temp, rax);
2851 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2852
2853 // Fetch destination element klass from the ObjArrayKlass header.
2854 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2855 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2856 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2857 assert_clean_int(sco_temp, rax);
2858
2859 // the checkcast_copy loop needs two extra arguments:
2860 assert(c_rarg3 == sco_temp, "#3 already in place");
2861 // Set up arguments for checkcast_copy_entry.
2862 setup_arg_regs(4);
2863 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2864 __ jump(RuntimeAddress(checkcast_copy_entry));
2865 }
2866
2867 __ BIND(L_failed);
2868 __ xorptr(rax, rax);
2869 __ notptr(rax); // return -1
2870 __ leave(); // required for proper stackwalking of RuntimeStub frame
2871 __ ret(0);
2872
2873 return start;
2874 }
2875
2876 void generate_arraycopy_stubs() {
2877 address entry;
2878 address entry_jbyte_arraycopy;
2879 address entry_jshort_arraycopy;
2880 address entry_jint_arraycopy;
2881 address entry_oop_arraycopy;
2882 address entry_jlong_arraycopy;
2883 address entry_checkcast_arraycopy;
2884
2885 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2886 "jbyte_disjoint_arraycopy");
2887 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2888 "jbyte_arraycopy");
2889
2890 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2891 "jshort_disjoint_arraycopy");
2892 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2893 "jshort_arraycopy");
2894
2895 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2896 "jint_disjoint_arraycopy");
2897 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2898 &entry_jint_arraycopy, "jint_arraycopy");
2899
2900 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2901 "jlong_disjoint_arraycopy");
2902 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2903 &entry_jlong_arraycopy, "jlong_arraycopy");
2904
2905
2906 if (UseCompressedOops) {
2907 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2908 "oop_disjoint_arraycopy");
2909 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2910 &entry_oop_arraycopy, "oop_arraycopy");
2911 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2912 "oop_disjoint_arraycopy_uninit",
2913 /*dest_uninitialized*/true);
2914 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2915 NULL, "oop_arraycopy_uninit",
2916 /*dest_uninitialized*/true);
2917 } else {
2918 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2919 "oop_disjoint_arraycopy");
2920 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2921 &entry_oop_arraycopy, "oop_arraycopy");
2922 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2923 "oop_disjoint_arraycopy_uninit",
2924 /*dest_uninitialized*/true);
2925 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2926 NULL, "oop_arraycopy_uninit",
2927 /*dest_uninitialized*/true);
2928 }
2929
2930 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2931 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2932 /*dest_uninitialized*/true);
2933
2934 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2935 entry_jbyte_arraycopy,
2936 entry_jshort_arraycopy,
2937 entry_jint_arraycopy,
2938 entry_jlong_arraycopy);
2939 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2940 entry_jbyte_arraycopy,
2941 entry_jshort_arraycopy,
2942 entry_jint_arraycopy,
2943 entry_oop_arraycopy,
2944 entry_jlong_arraycopy,
2945 entry_checkcast_arraycopy);
2946
2947 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2948 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2949 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2950 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2951 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2952 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2953
2954 // We don't generate specialized code for HeapWord-aligned source
2955 // arrays, so just use the code we've already generated
2956 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2957 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2958
2959 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2960 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2961
2962 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2963 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2964
2965 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2966 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2967
2968 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2969 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2970
2971 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2972 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2973 }
2974
2975 // AES intrinsic stubs
2976 enum {AESBlockSize = 16};
2977
2978 address generate_key_shuffle_mask() {
2979 __ align(16);
2980 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2981 address start = __ pc();
2982 __ emit_data64( 0x0405060700010203, relocInfo::none );
2983 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
2984 return start;
2985 }
2986
2987 address generate_counter_shuffle_mask() {
2988 __ align(16);
2989 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2990 address start = __ pc();
2991 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
2992 __ emit_data64(0x0001020304050607, relocInfo::none);
2993 return start;
2994 }
2995
2996 // Utility routine for loading a 128-bit key word in little endian format
2997 // can optionally specify that the shuffle mask is already in an xmmregister
2998 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2999 __ movdqu(xmmdst, Address(key, offset));
3000 if (xmm_shuf_mask != NULL) {
3001 __ pshufb(xmmdst, xmm_shuf_mask);
3002 } else {
3003 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3004 }
3005 }
3006
3007 // Utility routine for increase 128bit counter (iv in CTR mode)
3008 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
3009 __ pextrq(reg, xmmdst, 0x0);
3010 __ addq(reg, inc_delta);
3011 __ pinsrq(xmmdst, reg, 0x0);
3012 __ jcc(Assembler::carryClear, next_block); // jump if no carry
3013 __ pextrq(reg, xmmdst, 0x01); // Carry
3014 __ addq(reg, 0x01);
3015 __ pinsrq(xmmdst, reg, 0x01); //Carry end
3016 __ BIND(next_block); // next instruction
3017 }
3018
3019 // Arguments:
3020 //
3021 // Inputs:
3022 // c_rarg0 - source byte array address
3023 // c_rarg1 - destination byte array address
3024 // c_rarg2 - K (key) in little endian int array
3025 //
3026 address generate_aescrypt_encryptBlock() {
3027 assert(UseAES, "need AES instructions and misaligned SSE support");
3028 __ align(CodeEntryAlignment);
3029 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
3030 Label L_doLast;
3031 address start = __ pc();
3032
3033 const Register from = c_rarg0; // source array address
3034 const Register to = c_rarg1; // destination array address
3035 const Register key = c_rarg2; // key array address
3036 const Register keylen = rax;
3037
3038 const XMMRegister xmm_result = xmm0;
3039 const XMMRegister xmm_key_shuf_mask = xmm1;
3040 // On win64 xmm6-xmm15 must be preserved so don't use them.
3041 const XMMRegister xmm_temp1 = xmm2;
3042 const XMMRegister xmm_temp2 = xmm3;
3043 const XMMRegister xmm_temp3 = xmm4;
3044 const XMMRegister xmm_temp4 = xmm5;
3045
3046 __ enter(); // required for proper stackwalking of RuntimeStub frame
3047
3048 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3049 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3050
3051 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3052 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
3053
3054 // For encryption, the java expanded key ordering is just what we need
3055 // we don't know if the key is aligned, hence not using load-execute form
3056
3057 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3058 __ pxor(xmm_result, xmm_temp1);
3059
3060 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3061 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3062 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3063 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3064
3065 __ aesenc(xmm_result, xmm_temp1);
3066 __ aesenc(xmm_result, xmm_temp2);
3067 __ aesenc(xmm_result, xmm_temp3);
3068 __ aesenc(xmm_result, xmm_temp4);
3069
3070 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3071 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3072 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3073 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3074
3075 __ aesenc(xmm_result, xmm_temp1);
3076 __ aesenc(xmm_result, xmm_temp2);
3077 __ aesenc(xmm_result, xmm_temp3);
3078 __ aesenc(xmm_result, xmm_temp4);
3079
3080 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3081 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3082
3083 __ cmpl(keylen, 44);
3084 __ jccb(Assembler::equal, L_doLast);
3085
3086 __ aesenc(xmm_result, xmm_temp1);
3087 __ aesenc(xmm_result, xmm_temp2);
3088
3089 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3090 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3091
3092 __ cmpl(keylen, 52);
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, 0xd0, xmm_key_shuf_mask);
3099 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3100
3101 __ BIND(L_doLast);
3102 __ aesenc(xmm_result, xmm_temp1);
3103 __ aesenclast(xmm_result, xmm_temp2);
3104 __ movdqu(Address(to, 0), xmm_result); // store the result
3105 __ xorptr(rax, rax); // return 0
3106 __ leave(); // required for proper stackwalking of RuntimeStub frame
3107 __ ret(0);
3108
3109 return start;
3110 }
3111
3112
3113 // Arguments:
3114 //
3115 // Inputs:
3116 // c_rarg0 - source byte array address
3117 // c_rarg1 - destination byte array address
3118 // c_rarg2 - K (key) in little endian int array
3119 //
3120 address generate_aescrypt_decryptBlock() {
3121 assert(UseAES, "need AES instructions and misaligned SSE support");
3122 __ align(CodeEntryAlignment);
3123 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3124 Label L_doLast;
3125 address start = __ pc();
3126
3127 const Register from = c_rarg0; // source array address
3128 const Register to = c_rarg1; // destination array address
3129 const Register key = c_rarg2; // key array address
3130 const Register keylen = rax;
3131
3132 const XMMRegister xmm_result = xmm0;
3133 const XMMRegister xmm_key_shuf_mask = xmm1;
3134 // On win64 xmm6-xmm15 must be preserved so don't use them.
3135 const XMMRegister xmm_temp1 = xmm2;
3136 const XMMRegister xmm_temp2 = xmm3;
3137 const XMMRegister xmm_temp3 = xmm4;
3138 const XMMRegister xmm_temp4 = xmm5;
3139
3140 __ enter(); // required for proper stackwalking of RuntimeStub frame
3141
3142 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3143 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3144
3145 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3146 __ movdqu(xmm_result, Address(from, 0));
3147
3148 // for decryption java expanded key ordering is rotated one position from what we want
3149 // so we start from 0x10 here and hit 0x00 last
3150 // we don't know if the key is aligned, hence not using load-execute form
3151 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3152 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3153 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3154 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3155
3156 __ pxor (xmm_result, xmm_temp1);
3157 __ aesdec(xmm_result, xmm_temp2);
3158 __ aesdec(xmm_result, xmm_temp3);
3159 __ aesdec(xmm_result, xmm_temp4);
3160
3161 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3162 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3163 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3164 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3165
3166 __ aesdec(xmm_result, xmm_temp1);
3167 __ aesdec(xmm_result, xmm_temp2);
3168 __ aesdec(xmm_result, xmm_temp3);
3169 __ aesdec(xmm_result, xmm_temp4);
3170
3171 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3172 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3173 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3174
3175 __ cmpl(keylen, 44);
3176 __ jccb(Assembler::equal, L_doLast);
3177
3178 __ aesdec(xmm_result, xmm_temp1);
3179 __ aesdec(xmm_result, xmm_temp2);
3180
3181 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3182 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3183
3184 __ cmpl(keylen, 52);
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, 0xd0, xmm_key_shuf_mask);
3191 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3192
3193 __ BIND(L_doLast);
3194 __ aesdec(xmm_result, xmm_temp1);
3195 __ aesdec(xmm_result, xmm_temp2);
3196
3197 // for decryption the aesdeclast operation is always on key+0x00
3198 __ aesdeclast(xmm_result, xmm_temp3);
3199 __ movdqu(Address(to, 0), xmm_result); // store the result
3200 __ xorptr(rax, rax); // return 0
3201 __ leave(); // required for proper stackwalking of RuntimeStub frame
3202 __ ret(0);
3203
3204 return start;
3205 }
3206
3207
3208 // Arguments:
3209 //
3210 // Inputs:
3211 // c_rarg0 - source byte array address
3212 // c_rarg1 - destination byte array address
3213 // c_rarg2 - K (key) in little endian int array
3214 // c_rarg3 - r vector byte array address
3215 // c_rarg4 - input length
3216 //
3217 // Output:
3218 // rax - input length
3219 //
3220 address generate_cipherBlockChaining_encryptAESCrypt() {
3221 assert(UseAES, "need AES instructions and misaligned SSE support");
3222 __ align(CodeEntryAlignment);
3223 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3224 address start = __ pc();
3225
3226 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3227 const Register from = c_rarg0; // source array address
3228 const Register to = c_rarg1; // destination array address
3229 const Register key = c_rarg2; // key array address
3230 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3231 // and left with the results of the last encryption block
3232 #ifndef _WIN64
3233 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3234 #else
3235 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3236 const Register len_reg = r11; // pick the volatile windows register
3237 #endif
3238 const Register pos = rax;
3239
3240 // xmm register assignments for the loops below
3241 const XMMRegister xmm_result = xmm0;
3242 const XMMRegister xmm_temp = xmm1;
3243 // keys 0-10 preloaded into xmm2-xmm12
3244 const int XMM_REG_NUM_KEY_FIRST = 2;
3245 const int XMM_REG_NUM_KEY_LAST = 15;
3246 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3247 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3248 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3249 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3250 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3251
3252 __ enter(); // required for proper stackwalking of RuntimeStub frame
3253
3254 #ifdef _WIN64
3255 // on win64, fill len_reg from stack position
3256 __ movl(len_reg, len_mem);
3257 #else
3258 __ push(len_reg); // Save
3259 #endif
3260
3261 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3262 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3263 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3264 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3265 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3266 offset += 0x10;
3267 }
3268 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3269
3270 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3271 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3272 __ cmpl(rax, 44);
3273 __ jcc(Assembler::notEqual, L_key_192_256);
3274
3275 // 128 bit code follows here
3276 __ movptr(pos, 0);
3277 __ align(OptoLoopAlignment);
3278
3279 __ BIND(L_loopTop_128);
3280 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3281 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3282 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3283 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3284 __ aesenc(xmm_result, as_XMMRegister(rnum));
3285 }
3286 __ aesenclast(xmm_result, xmm_key10);
3287 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3288 // no need to store r to memory until we exit
3289 __ addptr(pos, AESBlockSize);
3290 __ subptr(len_reg, AESBlockSize);
3291 __ jcc(Assembler::notEqual, L_loopTop_128);
3292
3293 __ BIND(L_exit);
3294 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3295
3296 #ifdef _WIN64
3297 __ movl(rax, len_mem);
3298 #else
3299 __ pop(rax); // return length
3300 #endif
3301 __ leave(); // required for proper stackwalking of RuntimeStub frame
3302 __ ret(0);
3303
3304 __ BIND(L_key_192_256);
3305 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3306 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3307 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3308 __ cmpl(rax, 52);
3309 __ jcc(Assembler::notEqual, L_key_256);
3310
3311 // 192-bit code follows here (could be changed to use more xmm registers)
3312 __ movptr(pos, 0);
3313 __ align(OptoLoopAlignment);
3314
3315 __ BIND(L_loopTop_192);
3316 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3317 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3318 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3319 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3320 __ aesenc(xmm_result, as_XMMRegister(rnum));
3321 }
3322 __ aesenclast(xmm_result, xmm_key12);
3323 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3324 // no need to store r to memory until we exit
3325 __ addptr(pos, AESBlockSize);
3326 __ subptr(len_reg, AESBlockSize);
3327 __ jcc(Assembler::notEqual, L_loopTop_192);
3328 __ jmp(L_exit);
3329
3330 __ BIND(L_key_256);
3331 // 256-bit code follows here (could be changed to use more xmm registers)
3332 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3333 __ movptr(pos, 0);
3334 __ align(OptoLoopAlignment);
3335
3336 __ BIND(L_loopTop_256);
3337 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3338 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3339 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3340 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3341 __ aesenc(xmm_result, as_XMMRegister(rnum));
3342 }
3343 load_key(xmm_temp, key, 0xe0);
3344 __ aesenclast(xmm_result, xmm_temp);
3345 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3346 // no need to store r to memory until we exit
3347 __ addptr(pos, AESBlockSize);
3348 __ subptr(len_reg, AESBlockSize);
3349 __ jcc(Assembler::notEqual, L_loopTop_256);
3350 __ jmp(L_exit);
3351
3352 return start;
3353 }
3354
3355 // Safefetch stubs.
3356 void generate_safefetch(const char* name, int size, address* entry,
3357 address* fault_pc, address* continuation_pc) {
3358 // safefetch signatures:
3359 // int SafeFetch32(int* adr, int errValue);
3360 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3361 //
3362 // arguments:
3363 // c_rarg0 = adr
3364 // c_rarg1 = errValue
3365 //
3366 // result:
3367 // PPC_RET = *adr or errValue
3368
3369 StubCodeMark mark(this, "StubRoutines", name);
3370
3371 // Entry point, pc or function descriptor.
3372 *entry = __ pc();
3373
3374 // Load *adr into c_rarg1, may fault.
3375 *fault_pc = __ pc();
3376 switch (size) {
3377 case 4:
3378 // int32_t
3379 __ movl(c_rarg1, Address(c_rarg0, 0));
3380 break;
3381 case 8:
3382 // int64_t
3383 __ movq(c_rarg1, Address(c_rarg0, 0));
3384 break;
3385 default:
3386 ShouldNotReachHere();
3387 }
3388
3389 // return errValue or *adr
3390 *continuation_pc = __ pc();
3391 __ movq(rax, c_rarg1);
3392 __ ret(0);
3393 }
3394
3395 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3396 // to hide instruction latency
3397 //
3398 // Arguments:
3399 //
3400 // Inputs:
3401 // c_rarg0 - source byte array address
3402 // c_rarg1 - destination byte array address
3403 // c_rarg2 - K (key) in little endian int array
3404 // c_rarg3 - r vector byte array address
3405 // c_rarg4 - input length
3406 //
3407 // Output:
3408 // rax - input length
3409 //
3410 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3411 assert(UseAES, "need AES instructions and misaligned SSE support");
3412 __ align(CodeEntryAlignment);
3413 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3414 address start = __ pc();
3415
3416 const Register from = c_rarg0; // source array address
3417 const Register to = c_rarg1; // destination array address
3418 const Register key = c_rarg2; // key array address
3419 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3420 // and left with the results of the last encryption block
3421 #ifndef _WIN64
3422 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3423 #else
3424 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3425 const Register len_reg = r11; // pick the volatile windows register
3426 #endif
3427 const Register pos = rax;
3428
3429 const int PARALLEL_FACTOR = 4;
3430 const int ROUNDS[3] = { 10, 12, 14 }; // aes rounds for key128, key192, key256
3431
3432 Label L_exit;
3433 Label L_singleBlock_loopTopHead[3]; // 128, 192, 256
3434 Label L_singleBlock_loopTopHead2[3]; // 128, 192, 256
3435 Label L_singleBlock_loopTop[3]; // 128, 192, 256
3436 Label L_multiBlock_loopTopHead[3]; // 128, 192, 256
3437 Label L_multiBlock_loopTop[3]; // 128, 192, 256
3438
3439 // keys 0-10 preloaded into xmm5-xmm15
3440 const int XMM_REG_NUM_KEY_FIRST = 5;
3441 const int XMM_REG_NUM_KEY_LAST = 15;
3442 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3443 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3444
3445 __ enter(); // required for proper stackwalking of RuntimeStub frame
3446
3447 #ifdef _WIN64
3448 // on win64, fill len_reg from stack position
3449 __ movl(len_reg, len_mem);
3450 #else
3451 __ push(len_reg); // Save
3452 #endif
3453 __ push(rbx);
3454 // the java expanded key ordering is rotated one position from what we want
3455 // so we start from 0x10 here and hit 0x00 last
3456 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3457 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3458 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3459 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3460 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3461 offset += 0x10;
3462 }
3463 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3464
3465 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3466
3467 // registers holding the four results in the parallelized loop
3468 const XMMRegister xmm_result0 = xmm0;
3469 const XMMRegister xmm_result1 = xmm2;
3470 const XMMRegister xmm_result2 = xmm3;
3471 const XMMRegister xmm_result3 = xmm4;
3472
3473 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3474
3475 __ xorptr(pos, pos);
3476
3477 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3478 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3479 __ cmpl(rbx, 52);
3480 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[1]);
3481 __ cmpl(rbx, 60);
3482 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[2]);
3483
3484 #define DoFour(opc, src_reg) \
3485 __ opc(xmm_result0, src_reg); \
3486 __ opc(xmm_result1, src_reg); \
3487 __ opc(xmm_result2, src_reg); \
3488 __ opc(xmm_result3, src_reg); \
3489
3490 for (int k = 0; k < 3; ++k) {
3491 __ BIND(L_multiBlock_loopTopHead[k]);
3492 if (k != 0) {
3493 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3494 __ jcc(Assembler::less, L_singleBlock_loopTopHead2[k]);
3495 }
3496 if (k == 1) {
3497 __ subptr(rsp, 6 * wordSize);
3498 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3499 load_key(xmm15, key, 0xb0); // 0xb0; 192-bit key goes up to 0xc0
3500 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3501 load_key(xmm1, key, 0xc0); // 0xc0;
3502 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3503 } else if (k == 2) {
3504 __ subptr(rsp, 10 * wordSize);
3505 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3506 load_key(xmm15, key, 0xd0); // 0xd0; 256-bit key goes upto 0xe0
3507 __ movdqu(Address(rsp, 6 * wordSize), xmm15);
3508 load_key(xmm1, key, 0xe0); // 0xe0;
3509 __ movdqu(Address(rsp, 8 * wordSize), xmm1);
3510 load_key(xmm15, key, 0xb0); // 0xb0;
3511 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3512 load_key(xmm1, key, 0xc0); // 0xc0;
3513 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3514 }
3515 __ align(OptoLoopAlignment);
3516 __ BIND(L_multiBlock_loopTop[k]);
3517 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3518 __ jcc(Assembler::less, L_singleBlock_loopTopHead[k]);
3519
3520 if (k != 0) {
3521 __ movdqu(xmm15, Address(rsp, 2 * wordSize));
3522 __ movdqu(xmm1, Address(rsp, 4 * wordSize));
3523 }
3524
3525 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
3526 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3527 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3528 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3529
3530 DoFour(pxor, xmm_key_first);
3531 if (k == 0) {
3532 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
3533 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3534 }
3535 DoFour(aesdeclast, xmm_key_last);
3536 } else if (k == 1) {
3537 for (int rnum = 1; rnum <= ROUNDS[k]-2; rnum++) {
3538 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3539 }
3540 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3541 DoFour(aesdec, xmm1); // key : 0xc0
3542 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3543 DoFour(aesdeclast, xmm_key_last);
3544 } else if (k == 2) {
3545 for (int rnum = 1; rnum <= ROUNDS[k] - 4; rnum++) {
3546 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3547 }
3548 DoFour(aesdec, xmm1); // key : 0xc0
3549 __ movdqu(xmm15, Address(rsp, 6 * wordSize));
3550 __ movdqu(xmm1, Address(rsp, 8 * wordSize));
3551 DoFour(aesdec, xmm15); // key : 0xd0
3552 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3553 DoFour(aesdec, xmm1); // key : 0xe0
3554 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3555 DoFour(aesdeclast, xmm_key_last);
3556 }
3557
3558 // for each result, xor with the r vector of previous cipher block
3559 __ pxor(xmm_result0, xmm_prev_block_cipher);
3560 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3561 __ pxor(xmm_result1, xmm_prev_block_cipher);
3562 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3563 __ pxor(xmm_result2, xmm_prev_block_cipher);
3564 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3565 __ pxor(xmm_result3, xmm_prev_block_cipher);
3566 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
3567 if (k != 0) {
3568 __ movdqu(Address(rvec, 0x00), xmm_prev_block_cipher);
3569 }
3570
3571 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3572 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3573 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3574 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3575
3576 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize);
3577 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize);
3578 __ jmp(L_multiBlock_loopTop[k]);
3579
3580 // registers used in the non-parallelized loops
3581 // xmm register assignments for the loops below
3582 const XMMRegister xmm_result = xmm0;
3583 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3584 const XMMRegister xmm_key11 = xmm3;
3585 const XMMRegister xmm_key12 = xmm4;
3586 const XMMRegister key_tmp = xmm4;
3587
3588 __ BIND(L_singleBlock_loopTopHead[k]);
3589 if (k == 1) {
3590 __ addptr(rsp, 6 * wordSize);
3591 } else if (k == 2) {
3592 __ addptr(rsp, 10 * wordSize);
3593 }
3594 __ cmpptr(len_reg, 0); // any blocks left??
3595 __ jcc(Assembler::equal, L_exit);
3596 __ BIND(L_singleBlock_loopTopHead2[k]);
3597 if (k == 1) {
3598 load_key(xmm_key11, key, 0xb0); // 0xb0; 192-bit key goes upto 0xc0
3599 load_key(xmm_key12, key, 0xc0); // 0xc0; 192-bit key goes upto 0xc0
3600 }
3601 if (k == 2) {
3602 load_key(xmm_key11, key, 0xb0); // 0xb0; 256-bit key goes upto 0xe0
3603 }
3604 __ align(OptoLoopAlignment);
3605 __ BIND(L_singleBlock_loopTop[k]);
3606 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3607 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3608 __ pxor(xmm_result, xmm_key_first); // do the aes dec rounds
3609 for (int rnum = 1; rnum <= 9 ; rnum++) {
3610 __ aesdec(xmm_result, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3611 }
3612 if (k == 1) {
3613 __ aesdec(xmm_result, xmm_key11);
3614 __ aesdec(xmm_result, xmm_key12);
3615 }
3616 if (k == 2) {
3617 __ aesdec(xmm_result, xmm_key11);
3618 load_key(key_tmp, key, 0xc0);
3619 __ aesdec(xmm_result, key_tmp);
3620 load_key(key_tmp, key, 0xd0);
3621 __ aesdec(xmm_result, key_tmp);
3622 load_key(key_tmp, key, 0xe0);
3623 __ aesdec(xmm_result, key_tmp);
3624 }
3625
3626 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3627 __ pxor(xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3628 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3629 // no need to store r to memory until we exit
3630 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3631 __ addptr(pos, AESBlockSize);
3632 __ subptr(len_reg, AESBlockSize);
3633 __ jcc(Assembler::notEqual, L_singleBlock_loopTop[k]);
3634 if (k != 2) {
3635 __ jmp(L_exit);
3636 }
3637 } //for 128/192/256
3638
3639 __ BIND(L_exit);
3640 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3641 __ pop(rbx);
3642 #ifdef _WIN64
3643 __ movl(rax, len_mem);
3644 #else
3645 __ pop(rax); // return length
3646 #endif
3647 __ leave(); // required for proper stackwalking of RuntimeStub frame
3648 __ ret(0);
3649 return start;
3650 }
3651
3652 address generate_upper_word_mask() {
3653 __ align(64);
3654 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3655 address start = __ pc();
3656 __ emit_data64(0x0000000000000000, relocInfo::none);
3657 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
3658 return start;
3659 }
3660
3661 address generate_shuffle_byte_flip_mask() {
3662 __ align(64);
3663 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3664 address start = __ pc();
3665 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3666 __ emit_data64(0x0001020304050607, relocInfo::none);
3667 return start;
3668 }
3669
3670 // ofs and limit are use for multi-block byte array.
3671 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3672 address generate_sha1_implCompress(bool multi_block, const char *name) {
3673 __ align(CodeEntryAlignment);
3674 StubCodeMark mark(this, "StubRoutines", name);
3675 address start = __ pc();
3676
3677 Register buf = c_rarg0;
3678 Register state = c_rarg1;
3679 Register ofs = c_rarg2;
3680 Register limit = c_rarg3;
3681
3682 const XMMRegister abcd = xmm0;
3683 const XMMRegister e0 = xmm1;
3684 const XMMRegister e1 = xmm2;
3685 const XMMRegister msg0 = xmm3;
3686
3687 const XMMRegister msg1 = xmm4;
3688 const XMMRegister msg2 = xmm5;
3689 const XMMRegister msg3 = xmm6;
3690 const XMMRegister shuf_mask = xmm7;
3691
3692 __ enter();
3693
3694 __ subptr(rsp, 4 * wordSize);
3695
3696 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3697 buf, state, ofs, limit, rsp, multi_block);
3698
3699 __ addptr(rsp, 4 * wordSize);
3700
3701 __ leave();
3702 __ ret(0);
3703 return start;
3704 }
3705
3706 address generate_pshuffle_byte_flip_mask() {
3707 __ align(64);
3708 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3709 address start = __ pc();
3710 __ emit_data64(0x0405060700010203, relocInfo::none);
3711 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3712
3713 if (VM_Version::supports_avx2()) {
3714 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
3715 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3716 // _SHUF_00BA
3717 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3718 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3719 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3720 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3721 // _SHUF_DC00
3722 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3723 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3724 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3725 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3726 }
3727
3728 return start;
3729 }
3730
3731 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
3732 address generate_pshuffle_byte_flip_mask_sha512() {
3733 __ align(32);
3734 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
3735 address start = __ pc();
3736 if (VM_Version::supports_avx2()) {
3737 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
3738 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3739 __ emit_data64(0x1011121314151617, relocInfo::none);
3740 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
3741 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
3742 __ emit_data64(0x0000000000000000, relocInfo::none);
3743 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3744 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3745 }
3746
3747 return start;
3748 }
3749
3750 // ofs and limit are use for multi-block byte array.
3751 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3752 address generate_sha256_implCompress(bool multi_block, const char *name) {
3753 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
3754 __ align(CodeEntryAlignment);
3755 StubCodeMark mark(this, "StubRoutines", name);
3756 address start = __ pc();
3757
3758 Register buf = c_rarg0;
3759 Register state = c_rarg1;
3760 Register ofs = c_rarg2;
3761 Register limit = c_rarg3;
3762
3763 const XMMRegister msg = xmm0;
3764 const XMMRegister state0 = xmm1;
3765 const XMMRegister state1 = xmm2;
3766 const XMMRegister msgtmp0 = xmm3;
3767
3768 const XMMRegister msgtmp1 = xmm4;
3769 const XMMRegister msgtmp2 = xmm5;
3770 const XMMRegister msgtmp3 = xmm6;
3771 const XMMRegister msgtmp4 = xmm7;
3772
3773 const XMMRegister shuf_mask = xmm8;
3774
3775 __ enter();
3776
3777 __ subptr(rsp, 4 * wordSize);
3778
3779 if (VM_Version::supports_sha()) {
3780 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3781 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3782 } else if (VM_Version::supports_avx2()) {
3783 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3784 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3785 }
3786 __ addptr(rsp, 4 * wordSize);
3787 __ vzeroupper();
3788 __ leave();
3789 __ ret(0);
3790 return start;
3791 }
3792
3793 address generate_sha512_implCompress(bool multi_block, const char *name) {
3794 assert(VM_Version::supports_avx2(), "");
3795 assert(VM_Version::supports_bmi2(), "");
3796 __ align(CodeEntryAlignment);
3797 StubCodeMark mark(this, "StubRoutines", name);
3798 address start = __ pc();
3799
3800 Register buf = c_rarg0;
3801 Register state = c_rarg1;
3802 Register ofs = c_rarg2;
3803 Register limit = c_rarg3;
3804
3805 const XMMRegister msg = xmm0;
3806 const XMMRegister state0 = xmm1;
3807 const XMMRegister state1 = xmm2;
3808 const XMMRegister msgtmp0 = xmm3;
3809 const XMMRegister msgtmp1 = xmm4;
3810 const XMMRegister msgtmp2 = xmm5;
3811 const XMMRegister msgtmp3 = xmm6;
3812 const XMMRegister msgtmp4 = xmm7;
3813
3814 const XMMRegister shuf_mask = xmm8;
3815
3816 __ enter();
3817
3818 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3819 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3820
3821 __ vzeroupper();
3822 __ leave();
3823 __ ret(0);
3824 return start;
3825 }
3826
3827 // This is a version of CTR/AES crypt which does 6 blocks in a loop at a time
3828 // to hide instruction latency
3829 //
3830 // Arguments:
3831 //
3832 // Inputs:
3833 // c_rarg0 - source byte array address
3834 // c_rarg1 - destination byte array address
3835 // c_rarg2 - K (key) in little endian int array
3836 // c_rarg3 - counter vector byte array address
3837 // Linux
3838 // c_rarg4 - input length
3839 // c_rarg5 - saved encryptedCounter start
3840 // rbp + 6 * wordSize - saved used length
3841 // Windows
3842 // rbp + 6 * wordSize - input length
3843 // rbp + 7 * wordSize - saved encryptedCounter start
3844 // rbp + 8 * wordSize - saved used length
3845 //
3846 // Output:
3847 // rax - input length
3848 //
3849 address generate_counterMode_AESCrypt_Parallel() {
3850 assert(UseAES, "need AES instructions and misaligned SSE support");
3851 __ align(CodeEntryAlignment);
3852 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
3853 address start = __ pc();
3854 const Register from = c_rarg0; // source array address
3855 const Register to = c_rarg1; // destination array address
3856 const Register key = c_rarg2; // key array address
3857 const Register counter = c_rarg3; // counter byte array initialized from counter array address
3858 // and updated with the incremented counter in the end
3859 #ifndef _WIN64
3860 const Register len_reg = c_rarg4;
3861 const Register saved_encCounter_start = c_rarg5;
3862 const Register used_addr = r10;
3863 const Address used_mem(rbp, 2 * wordSize);
3864 const Register used = r11;
3865 #else
3866 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3867 const Address saved_encCounter_mem(rbp, 7 * wordSize); // length is on stack on Win64
3868 const Address used_mem(rbp, 8 * wordSize); // length is on stack on Win64
3869 const Register len_reg = r10; // pick the first volatile windows register
3870 const Register saved_encCounter_start = r11;
3871 const Register used_addr = r13;
3872 const Register used = r14;
3873 #endif
3874 const Register pos = rax;
3875
3876 const int PARALLEL_FACTOR = 6;
3877 const XMMRegister xmm_counter_shuf_mask = xmm0;
3878 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3879 const XMMRegister xmm_curr_counter = xmm2;
3880
3881 const XMMRegister xmm_key_tmp0 = xmm3;
3882 const XMMRegister xmm_key_tmp1 = xmm4;
3883
3884 // registers holding the four results in the parallelized loop
3885 const XMMRegister xmm_result0 = xmm5;
3886 const XMMRegister xmm_result1 = xmm6;
3887 const XMMRegister xmm_result2 = xmm7;
3888 const XMMRegister xmm_result3 = xmm8;
3889 const XMMRegister xmm_result4 = xmm9;
3890 const XMMRegister xmm_result5 = xmm10;
3891
3892 const XMMRegister xmm_from0 = xmm11;
3893 const XMMRegister xmm_from1 = xmm12;
3894 const XMMRegister xmm_from2 = xmm13;
3895 const XMMRegister xmm_from3 = xmm14; //the last one is xmm14. we have to preserve it on WIN64.
3896 const XMMRegister xmm_from4 = xmm3; //reuse xmm3~4. Because xmm_key_tmp0~1 are useless when loading input text
3897 const XMMRegister xmm_from5 = xmm4;
3898
3899 //for key_128, key_192, key_256
3900 const int rounds[3] = {10, 12, 14};
3901 Label L_exit_preLoop, L_preLoop_start;
3902 Label L_multiBlock_loopTop[3];
3903 Label L_singleBlockLoopTop[3];
3904 Label L__incCounter[3][6]; //for 6 blocks
3905 Label L__incCounter_single[3]; //for single block, key128, key192, key256
3906 Label L_processTail_insr[3], L_processTail_4_insr[3], L_processTail_2_insr[3], L_processTail_1_insr[3], L_processTail_exit_insr[3];
3907 Label L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
3908
3909 Label L_exit;
3910
3911 __ enter(); // required for proper stackwalking of RuntimeStub frame
3912
3913 #ifdef _WIN64
3914 // allocate spill slots for r13, r14
3915 enum {
3916 saved_r13_offset,
3917 saved_r14_offset
3918 };
3919 __ subptr(rsp, 2 * wordSize);
3920 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
3921 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
3922
3923 // on win64, fill len_reg from stack position
3924 __ movl(len_reg, len_mem);
3925 __ movptr(saved_encCounter_start, saved_encCounter_mem);
3926 __ movptr(used_addr, used_mem);
3927 __ movl(used, Address(used_addr, 0));
3928 #else
3929 __ push(len_reg); // Save
3930 __ movptr(used_addr, used_mem);
3931 __ movl(used, Address(used_addr, 0));
3932 #endif
3933
3934 __ push(rbx); // Save RBX
3935 __ movdqu(xmm_curr_counter, Address(counter, 0x00)); // initialize counter with initial counter
3936 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()), pos); // pos as scratch
3937 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled
3938 __ movptr(pos, 0);
3939
3940 // Use the partially used encrpyted counter from last invocation
3941 __ BIND(L_preLoop_start);
3942 __ cmpptr(used, 16);
3943 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
3944 __ cmpptr(len_reg, 0);
3945 __ jcc(Assembler::lessEqual, L_exit_preLoop);
3946 __ movb(rbx, Address(saved_encCounter_start, used));
3947 __ xorb(rbx, Address(from, pos));
3948 __ movb(Address(to, pos), rbx);
3949 __ addptr(pos, 1);
3950 __ addptr(used, 1);
3951 __ subptr(len_reg, 1);
3952
3953 __ jmp(L_preLoop_start);
3954
3955 __ BIND(L_exit_preLoop);
3956 __ movl(Address(used_addr, 0), used);
3957
3958 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
3959 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()), rbx); // rbx as scratch
3960 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3961 __ cmpl(rbx, 52);
3962 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
3963 __ cmpl(rbx, 60);
3964 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
3965
3966 #define CTR_DoSix(opc, src_reg) \
3967 __ opc(xmm_result0, src_reg); \
3968 __ opc(xmm_result1, src_reg); \
3969 __ opc(xmm_result2, src_reg); \
3970 __ opc(xmm_result3, src_reg); \
3971 __ opc(xmm_result4, src_reg); \
3972 __ opc(xmm_result5, src_reg);
3973
3974 // k == 0 : generate code for key_128
3975 // k == 1 : generate code for key_192
3976 // k == 2 : generate code for key_256
3977 for (int k = 0; k < 3; ++k) {
3978 //multi blocks starts here
3979 __ align(OptoLoopAlignment);
3980 __ BIND(L_multiBlock_loopTop[k]);
3981 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
3982 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
3983 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
3984
3985 //load, then increase counters
3986 CTR_DoSix(movdqa, xmm_curr_counter);
3987 inc_counter(rbx, xmm_result1, 0x01, L__incCounter[k][0]);
3988 inc_counter(rbx, xmm_result2, 0x02, L__incCounter[k][1]);
3989 inc_counter(rbx, xmm_result3, 0x03, L__incCounter[k][2]);
3990 inc_counter(rbx, xmm_result4, 0x04, L__incCounter[k][3]);
3991 inc_counter(rbx, xmm_result5, 0x05, L__incCounter[k][4]);
3992 inc_counter(rbx, xmm_curr_counter, 0x06, L__incCounter[k][5]);
3993 CTR_DoSix(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
3994 CTR_DoSix(pxor, xmm_key_tmp0); //PXOR with Round 0 key
3995
3996 //load two ROUND_KEYs at a time
3997 for (int i = 1; i < rounds[k]; ) {
3998 load_key(xmm_key_tmp1, key, (0x10 * i), xmm_key_shuf_mask);
3999 load_key(xmm_key_tmp0, key, (0x10 * (i+1)), xmm_key_shuf_mask);
4000 CTR_DoSix(aesenc, xmm_key_tmp1);
4001 i++;
4002 if (i != rounds[k]) {
4003 CTR_DoSix(aesenc, xmm_key_tmp0);
4004 } else {
4005 CTR_DoSix(aesenclast, xmm_key_tmp0);
4006 }
4007 i++;
4008 }
4009
4010 // get next PARALLEL_FACTOR blocks into xmm_result registers
4011 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4012 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
4013 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
4014 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
4015 __ movdqu(xmm_from4, Address(from, pos, Address::times_1, 4 * AESBlockSize));
4016 __ movdqu(xmm_from5, Address(from, pos, Address::times_1, 5 * AESBlockSize));
4017
4018 __ pxor(xmm_result0, xmm_from0);
4019 __ pxor(xmm_result1, xmm_from1);
4020 __ pxor(xmm_result2, xmm_from2);
4021 __ pxor(xmm_result3, xmm_from3);
4022 __ pxor(xmm_result4, xmm_from4);
4023 __ pxor(xmm_result5, xmm_from5);
4024
4025 // store 6 results into the next 64 bytes of output
4026 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4027 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
4028 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
4029 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
4030 __ movdqu(Address(to, pos, Address::times_1, 4 * AESBlockSize), xmm_result4);
4031 __ movdqu(Address(to, pos, Address::times_1, 5 * AESBlockSize), xmm_result5);
4032
4033 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
4034 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
4035 __ jmp(L_multiBlock_loopTop[k]);
4036
4037 // singleBlock starts here
4038 __ align(OptoLoopAlignment);
4039 __ BIND(L_singleBlockLoopTop[k]);
4040 __ cmpptr(len_reg, 0);
4041 __ jcc(Assembler::lessEqual, L_exit);
4042 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4043 __ movdqa(xmm_result0, xmm_curr_counter);
4044 inc_counter(rbx, xmm_curr_counter, 0x01, L__incCounter_single[k]);
4045 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
4046 __ pxor(xmm_result0, xmm_key_tmp0);
4047 for (int i = 1; i < rounds[k]; i++) {
4048 load_key(xmm_key_tmp0, key, (0x10 * i), xmm_key_shuf_mask);
4049 __ aesenc(xmm_result0, xmm_key_tmp0);
4050 }
4051 load_key(xmm_key_tmp0, key, (rounds[k] * 0x10), xmm_key_shuf_mask);
4052 __ aesenclast(xmm_result0, xmm_key_tmp0);
4053 __ cmpptr(len_reg, AESBlockSize);
4054 __ jcc(Assembler::less, L_processTail_insr[k]);
4055 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4056 __ pxor(xmm_result0, xmm_from0);
4057 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4058 __ addptr(pos, AESBlockSize);
4059 __ subptr(len_reg, AESBlockSize);
4060 __ jmp(L_singleBlockLoopTop[k]);
4061 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
4062 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
4063 __ testptr(len_reg, 8);
4064 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
4065 __ subptr(pos,8);
4066 __ pinsrq(xmm_from0, Address(from, pos), 0);
4067 __ BIND(L_processTail_4_insr[k]);
4068 __ testptr(len_reg, 4);
4069 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
4070 __ subptr(pos,4);
4071 __ pslldq(xmm_from0, 4);
4072 __ pinsrd(xmm_from0, Address(from, pos), 0);
4073 __ BIND(L_processTail_2_insr[k]);
4074 __ testptr(len_reg, 2);
4075 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
4076 __ subptr(pos, 2);
4077 __ pslldq(xmm_from0, 2);
4078 __ pinsrw(xmm_from0, Address(from, pos), 0);
4079 __ BIND(L_processTail_1_insr[k]);
4080 __ testptr(len_reg, 1);
4081 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
4082 __ subptr(pos, 1);
4083 __ pslldq(xmm_from0, 1);
4084 __ pinsrb(xmm_from0, Address(from, pos), 0);
4085 __ BIND(L_processTail_exit_insr[k]);
4086
4087 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
4088 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
4089
4090 __ testptr(len_reg, 8);
4091 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
4092 __ pextrq(Address(to, pos), xmm_result0, 0);
4093 __ psrldq(xmm_result0, 8);
4094 __ addptr(pos, 8);
4095 __ BIND(L_processTail_4_extr[k]);
4096 __ testptr(len_reg, 4);
4097 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
4098 __ pextrd(Address(to, pos), xmm_result0, 0);
4099 __ psrldq(xmm_result0, 4);
4100 __ addptr(pos, 4);
4101 __ BIND(L_processTail_2_extr[k]);
4102 __ testptr(len_reg, 2);
4103 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
4104 __ pextrw(Address(to, pos), xmm_result0, 0);
4105 __ psrldq(xmm_result0, 2);
4106 __ addptr(pos, 2);
4107 __ BIND(L_processTail_1_extr[k]);
4108 __ testptr(len_reg, 1);
4109 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
4110 __ pextrb(Address(to, pos), xmm_result0, 0);
4111
4112 __ BIND(L_processTail_exit_extr[k]);
4113 __ movl(Address(used_addr, 0), len_reg);
4114 __ jmp(L_exit);
4115
4116 }
4117
4118 __ BIND(L_exit);
4119 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
4120 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
4121 __ pop(rbx); // pop the saved RBX.
4122 #ifdef _WIN64
4123 __ movl(rax, len_mem);
4124 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
4125 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
4126 __ addptr(rsp, 2 * wordSize);
4127 #else
4128 __ pop(rax); // return 'len'
4129 #endif
4130 __ leave(); // required for proper stackwalking of RuntimeStub frame
4131 __ ret(0);
4132 return start;
4133 }
4134
4135 void roundDec(XMMRegister xmm_reg) {
4136 __ vaesdec(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4137 __ vaesdec(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4138 __ vaesdec(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4139 __ vaesdec(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4140 __ vaesdec(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4141 __ vaesdec(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4142 __ vaesdec(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4143 __ vaesdec(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4144 }
4145
4146 void roundDeclast(XMMRegister xmm_reg) {
4147 __ vaesdeclast(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4148 __ vaesdeclast(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4149 __ vaesdeclast(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4150 __ vaesdeclast(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4151 __ vaesdeclast(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4152 __ vaesdeclast(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4153 __ vaesdeclast(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4154 __ vaesdeclast(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4155 }
4156
4157 void ev_load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask = NULL) {
4158 __ movdqu(xmmdst, Address(key, offset));
4159 if (xmm_shuf_mask != NULL) {
4160 __ pshufb(xmmdst, xmm_shuf_mask);
4161 } else {
4162 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4163 }
4164 __ evshufi64x2(xmmdst, xmmdst, xmmdst, 0x0, Assembler::AVX_512bit);
4165
4166 }
4167
4168 address generate_cipherBlockChaining_decryptVectorAESCrypt() {
4169 assert(VM_Version::supports_vaes(), "need AES instructions and misaligned SSE support");
4170 __ align(CodeEntryAlignment);
4171 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
4172 address start = __ pc();
4173
4174 const Register from = c_rarg0; // source array address
4175 const Register to = c_rarg1; // destination array address
4176 const Register key = c_rarg2; // key array address
4177 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
4178 // and left with the results of the last encryption block
4179 #ifndef _WIN64
4180 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
4181 #else
4182 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4183 const Register len_reg = r11; // pick the volatile windows register
4184 #endif
4185
4186 Label Loop, Loop1, L_128, L_256, L_192, KEY_192, KEY_256, Loop2, Lcbc_dec_rem_loop,
4187 Lcbc_dec_rem_last, Lcbc_dec_ret, Lcbc_dec_rem, Lcbc_exit;
4188
4189 __ enter();
4190
4191 #ifdef _WIN64
4192 // on win64, fill len_reg from stack position
4193 __ movl(len_reg, len_mem);
4194 #else
4195 __ push(len_reg); // Save
4196 #endif
4197 __ push(rbx);
4198 __ vzeroupper();
4199
4200 // Temporary variable declaration for swapping key bytes
4201 const XMMRegister xmm_key_shuf_mask = xmm1;
4202 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4203
4204 // Calculate number of rounds from key size: 44 for 10-rounds, 52 for 12-rounds, 60 for 14-rounds
4205 const Register rounds = rbx;
4206 __ movl(rounds, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4207
4208 const XMMRegister IV = xmm0;
4209 // Load IV and broadcast value to 512-bits
4210 __ evbroadcasti64x2(IV, Address(rvec, 0), Assembler::AVX_512bit);
4211
4212 // Temporary variables for storing round keys
4213 const XMMRegister RK0 = xmm30;
4214 const XMMRegister RK1 = xmm9;
4215 const XMMRegister RK2 = xmm18;
4216 const XMMRegister RK3 = xmm19;
4217 const XMMRegister RK4 = xmm20;
4218 const XMMRegister RK5 = xmm21;
4219 const XMMRegister RK6 = xmm22;
4220 const XMMRegister RK7 = xmm23;
4221 const XMMRegister RK8 = xmm24;
4222 const XMMRegister RK9 = xmm25;
4223 const XMMRegister RK10 = xmm26;
4224
4225 // Load and shuffle key
4226 // the java expanded key ordering is rotated one position from what we want
4227 // so we start from 1*16 here and hit 0*16 last
4228 ev_load_key(RK1, key, 1 * 16, xmm_key_shuf_mask);
4229 ev_load_key(RK2, key, 2 * 16, xmm_key_shuf_mask);
4230 ev_load_key(RK3, key, 3 * 16, xmm_key_shuf_mask);
4231 ev_load_key(RK4, key, 4 * 16, xmm_key_shuf_mask);
4232 ev_load_key(RK5, key, 5 * 16, xmm_key_shuf_mask);
4233 ev_load_key(RK6, key, 6 * 16, xmm_key_shuf_mask);
4234 ev_load_key(RK7, key, 7 * 16, xmm_key_shuf_mask);
4235 ev_load_key(RK8, key, 8 * 16, xmm_key_shuf_mask);
4236 ev_load_key(RK9, key, 9 * 16, xmm_key_shuf_mask);
4237 ev_load_key(RK10, key, 10 * 16, xmm_key_shuf_mask);
4238 ev_load_key(RK0, key, 0*16, xmm_key_shuf_mask);
4239
4240 // Variables for storing source cipher text
4241 const XMMRegister S0 = xmm10;
4242 const XMMRegister S1 = xmm11;
4243 const XMMRegister S2 = xmm12;
4244 const XMMRegister S3 = xmm13;
4245 const XMMRegister S4 = xmm14;
4246 const XMMRegister S5 = xmm15;
4247 const XMMRegister S6 = xmm16;
4248 const XMMRegister S7 = xmm17;
4249
4250 // Variables for storing decrypted text
4251 const XMMRegister B0 = xmm1;
4252 const XMMRegister B1 = xmm2;
4253 const XMMRegister B2 = xmm3;
4254 const XMMRegister B3 = xmm4;
4255 const XMMRegister B4 = xmm5;
4256 const XMMRegister B5 = xmm6;
4257 const XMMRegister B6 = xmm7;
4258 const XMMRegister B7 = xmm8;
4259
4260 __ cmpl(rounds, 44);
4261 __ jcc(Assembler::greater, KEY_192);
4262 __ jmp(Loop);
4263
4264 __ BIND(KEY_192);
4265 const XMMRegister RK11 = xmm27;
4266 const XMMRegister RK12 = xmm28;
4267 ev_load_key(RK11, key, 11*16, xmm_key_shuf_mask);
4268 ev_load_key(RK12, key, 12*16, xmm_key_shuf_mask);
4269
4270 __ cmpl(rounds, 52);
4271 __ jcc(Assembler::greater, KEY_256);
4272 __ jmp(Loop);
4273
4274 __ BIND(KEY_256);
4275 const XMMRegister RK13 = xmm29;
4276 const XMMRegister RK14 = xmm31;
4277 ev_load_key(RK13, key, 13*16, xmm_key_shuf_mask);
4278 ev_load_key(RK14, key, 14*16, xmm_key_shuf_mask);
4279
4280 __ BIND(Loop);
4281 __ cmpl(len_reg, 512);
4282 __ jcc(Assembler::below, Lcbc_dec_rem);
4283 __ BIND(Loop1);
4284 __ subl(len_reg, 512);
4285 __ evmovdquq(S0, Address(from, 0 * 64), Assembler::AVX_512bit);
4286 __ evmovdquq(S1, Address(from, 1 * 64), Assembler::AVX_512bit);
4287 __ evmovdquq(S2, Address(from, 2 * 64), Assembler::AVX_512bit);
4288 __ evmovdquq(S3, Address(from, 3 * 64), Assembler::AVX_512bit);
4289 __ evmovdquq(S4, Address(from, 4 * 64), Assembler::AVX_512bit);
4290 __ evmovdquq(S5, Address(from, 5 * 64), Assembler::AVX_512bit);
4291 __ evmovdquq(S6, Address(from, 6 * 64), Assembler::AVX_512bit);
4292 __ evmovdquq(S7, Address(from, 7 * 64), Assembler::AVX_512bit);
4293 __ leaq(from, Address(from, 8 * 64));
4294
4295 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4296 __ evpxorq(B1, S1, RK1, Assembler::AVX_512bit);
4297 __ evpxorq(B2, S2, RK1, Assembler::AVX_512bit);
4298 __ evpxorq(B3, S3, RK1, Assembler::AVX_512bit);
4299 __ evpxorq(B4, S4, RK1, Assembler::AVX_512bit);
4300 __ evpxorq(B5, S5, RK1, Assembler::AVX_512bit);
4301 __ evpxorq(B6, S6, RK1, Assembler::AVX_512bit);
4302 __ evpxorq(B7, S7, RK1, Assembler::AVX_512bit);
4303
4304 __ evalignq(IV, S0, IV, 0x06);
4305 __ evalignq(S0, S1, S0, 0x06);
4306 __ evalignq(S1, S2, S1, 0x06);
4307 __ evalignq(S2, S3, S2, 0x06);
4308 __ evalignq(S3, S4, S3, 0x06);
4309 __ evalignq(S4, S5, S4, 0x06);
4310 __ evalignq(S5, S6, S5, 0x06);
4311 __ evalignq(S6, S7, S6, 0x06);
4312
4313 roundDec(RK2);
4314 roundDec(RK3);
4315 roundDec(RK4);
4316 roundDec(RK5);
4317 roundDec(RK6);
4318 roundDec(RK7);
4319 roundDec(RK8);
4320 roundDec(RK9);
4321 roundDec(RK10);
4322
4323 __ cmpl(rounds, 44);
4324 __ jcc(Assembler::belowEqual, L_128);
4325 roundDec(RK11);
4326 roundDec(RK12);
4327
4328 __ cmpl(rounds, 52);
4329 __ jcc(Assembler::belowEqual, L_192);
4330 roundDec(RK13);
4331 roundDec(RK14);
4332
4333 __ BIND(L_256);
4334 roundDeclast(RK0);
4335 __ jmp(Loop2);
4336
4337 __ BIND(L_128);
4338 roundDeclast(RK0);
4339 __ jmp(Loop2);
4340
4341 __ BIND(L_192);
4342 roundDeclast(RK0);
4343
4344 __ BIND(Loop2);
4345 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4346 __ evpxorq(B1, B1, S0, Assembler::AVX_512bit);
4347 __ evpxorq(B2, B2, S1, Assembler::AVX_512bit);
4348 __ evpxorq(B3, B3, S2, Assembler::AVX_512bit);
4349 __ evpxorq(B4, B4, S3, Assembler::AVX_512bit);
4350 __ evpxorq(B5, B5, S4, Assembler::AVX_512bit);
4351 __ evpxorq(B6, B6, S5, Assembler::AVX_512bit);
4352 __ evpxorq(B7, B7, S6, Assembler::AVX_512bit);
4353 __ evmovdquq(IV, S7, Assembler::AVX_512bit);
4354
4355 __ evmovdquq(Address(to, 0 * 64), B0, Assembler::AVX_512bit);
4356 __ evmovdquq(Address(to, 1 * 64), B1, Assembler::AVX_512bit);
4357 __ evmovdquq(Address(to, 2 * 64), B2, Assembler::AVX_512bit);
4358 __ evmovdquq(Address(to, 3 * 64), B3, Assembler::AVX_512bit);
4359 __ evmovdquq(Address(to, 4 * 64), B4, Assembler::AVX_512bit);
4360 __ evmovdquq(Address(to, 5 * 64), B5, Assembler::AVX_512bit);
4361 __ evmovdquq(Address(to, 6 * 64), B6, Assembler::AVX_512bit);
4362 __ evmovdquq(Address(to, 7 * 64), B7, Assembler::AVX_512bit);
4363 __ leaq(to, Address(to, 8 * 64));
4364 __ jmp(Loop);
4365
4366 __ BIND(Lcbc_dec_rem);
4367 __ evshufi64x2(IV, IV, IV, 0x03, Assembler::AVX_512bit);
4368
4369 __ BIND(Lcbc_dec_rem_loop);
4370 __ subl(len_reg, 16);
4371 __ jcc(Assembler::carrySet, Lcbc_dec_ret);
4372
4373 __ movdqu(S0, Address(from, 0));
4374 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4375 __ vaesdec(B0, B0, RK2, Assembler::AVX_512bit);
4376 __ vaesdec(B0, B0, RK3, Assembler::AVX_512bit);
4377 __ vaesdec(B0, B0, RK4, Assembler::AVX_512bit);
4378 __ vaesdec(B0, B0, RK5, Assembler::AVX_512bit);
4379 __ vaesdec(B0, B0, RK6, Assembler::AVX_512bit);
4380 __ vaesdec(B0, B0, RK7, Assembler::AVX_512bit);
4381 __ vaesdec(B0, B0, RK8, Assembler::AVX_512bit);
4382 __ vaesdec(B0, B0, RK9, Assembler::AVX_512bit);
4383 __ vaesdec(B0, B0, RK10, Assembler::AVX_512bit);
4384 __ cmpl(rounds, 44);
4385 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4386
4387 __ vaesdec(B0, B0, RK11, Assembler::AVX_512bit);
4388 __ vaesdec(B0, B0, RK12, Assembler::AVX_512bit);
4389 __ cmpl(rounds, 52);
4390 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4391
4392 __ vaesdec(B0, B0, RK13, Assembler::AVX_512bit);
4393 __ vaesdec(B0, B0, RK14, Assembler::AVX_512bit);
4394
4395 __ BIND(Lcbc_dec_rem_last);
4396 __ vaesdeclast(B0, B0, RK0, Assembler::AVX_512bit);
4397
4398 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4399 __ evmovdquq(IV, S0, Assembler::AVX_512bit);
4400 __ movdqu(Address(to, 0), B0);
4401 __ leaq(from, Address(from, 16));
4402 __ leaq(to, Address(to, 16));
4403 __ jmp(Lcbc_dec_rem_loop);
4404
4405 __ BIND(Lcbc_dec_ret);
4406 __ movdqu(Address(rvec, 0), IV);
4407
4408 // Zero out the round keys
4409 __ evpxorq(RK0, RK0, RK0, Assembler::AVX_512bit);
4410 __ evpxorq(RK1, RK1, RK1, Assembler::AVX_512bit);
4411 __ evpxorq(RK2, RK2, RK2, Assembler::AVX_512bit);
4412 __ evpxorq(RK3, RK3, RK3, Assembler::AVX_512bit);
4413 __ evpxorq(RK4, RK4, RK4, Assembler::AVX_512bit);
4414 __ evpxorq(RK5, RK5, RK5, Assembler::AVX_512bit);
4415 __ evpxorq(RK6, RK6, RK6, Assembler::AVX_512bit);
4416 __ evpxorq(RK7, RK7, RK7, Assembler::AVX_512bit);
4417 __ evpxorq(RK8, RK8, RK8, Assembler::AVX_512bit);
4418 __ evpxorq(RK9, RK9, RK9, Assembler::AVX_512bit);
4419 __ evpxorq(RK10, RK10, RK10, Assembler::AVX_512bit);
4420 __ cmpl(rounds, 44);
4421 __ jcc(Assembler::belowEqual, Lcbc_exit);
4422 __ evpxorq(RK11, RK11, RK11, Assembler::AVX_512bit);
4423 __ evpxorq(RK12, RK12, RK12, Assembler::AVX_512bit);
4424 __ cmpl(rounds, 52);
4425 __ jcc(Assembler::belowEqual, Lcbc_exit);
4426 __ evpxorq(RK13, RK13, RK13, Assembler::AVX_512bit);
4427 __ evpxorq(RK14, RK14, RK14, Assembler::AVX_512bit);
4428
4429 __ BIND(Lcbc_exit);
4430 __ pop(rbx);
4431 #ifdef _WIN64
4432 __ movl(rax, len_mem);
4433 #else
4434 __ pop(rax); // return length
4435 #endif
4436 __ leave(); // required for proper stackwalking of RuntimeStub frame
4437 __ ret(0);
4438 return start;
4439 }
4440
4441 // Polynomial x^128+x^127+x^126+x^121+1
4442 address ghash_polynomial_addr() {
4443 __ align(CodeEntryAlignment);
4444 StubCodeMark mark(this, "StubRoutines", "_ghash_poly_addr");
4445 address start = __ pc();
4446 __ emit_data64(0x0000000000000001, relocInfo::none);
4447 __ emit_data64(0xc200000000000000, relocInfo::none);
4448 return start;
4449 }
4450
4451 address ghash_shufflemask_addr() {
4452 __ align(CodeEntryAlignment);
4453 StubCodeMark mark(this, "StubRoutines", "_ghash_shuffmask_addr");
4454 address start = __ pc();
4455 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4456 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4457 return start;
4458 }
4459
4460 // Ghash single and multi block operations using AVX instructions
4461 address generate_avx_ghash_processBlocks() {
4462 __ align(CodeEntryAlignment);
4463
4464 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4465 address start = __ pc();
4466
4467 // arguments
4468 const Register state = c_rarg0;
4469 const Register htbl = c_rarg1;
4470 const Register data = c_rarg2;
4471 const Register blocks = c_rarg3;
4472 __ enter();
4473 // Save state before entering routine
4474 __ avx_ghash(state, htbl, data, blocks);
4475 __ leave(); // required for proper stackwalking of RuntimeStub frame
4476 __ ret(0);
4477 return start;
4478 }
4479
4480 // byte swap x86 long
4481 address generate_ghash_long_swap_mask() {
4482 __ align(CodeEntryAlignment);
4483 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
4484 address start = __ pc();
4485 __ emit_data64(0x0f0e0d0c0b0a0908, relocInfo::none );
4486 __ emit_data64(0x0706050403020100, relocInfo::none );
4487 return start;
4488 }
4489
4490 // byte swap x86 byte array
4491 address generate_ghash_byte_swap_mask() {
4492 __ align(CodeEntryAlignment);
4493 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
4494 address start = __ pc();
4495 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none );
4496 __ emit_data64(0x0001020304050607, relocInfo::none );
4497 return start;
4498 }
4499
4500 /* Single and multi-block ghash operations */
4501 address generate_ghash_processBlocks() {
4502 __ align(CodeEntryAlignment);
4503 Label L_ghash_loop, L_exit;
4504 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4505 address start = __ pc();
4506
4507 const Register state = c_rarg0;
4508 const Register subkeyH = c_rarg1;
4509 const Register data = c_rarg2;
4510 const Register blocks = c_rarg3;
4511
4512 const XMMRegister xmm_temp0 = xmm0;
4513 const XMMRegister xmm_temp1 = xmm1;
4514 const XMMRegister xmm_temp2 = xmm2;
4515 const XMMRegister xmm_temp3 = xmm3;
4516 const XMMRegister xmm_temp4 = xmm4;
4517 const XMMRegister xmm_temp5 = xmm5;
4518 const XMMRegister xmm_temp6 = xmm6;
4519 const XMMRegister xmm_temp7 = xmm7;
4520 const XMMRegister xmm_temp8 = xmm8;
4521 const XMMRegister xmm_temp9 = xmm9;
4522 const XMMRegister xmm_temp10 = xmm10;
4523
4524 __ enter();
4525
4526 __ movdqu(xmm_temp10, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
4527
4528 __ movdqu(xmm_temp0, Address(state, 0));
4529 __ pshufb(xmm_temp0, xmm_temp10);
4530
4531
4532 __ BIND(L_ghash_loop);
4533 __ movdqu(xmm_temp2, Address(data, 0));
4534 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
4535
4536 __ movdqu(xmm_temp1, Address(subkeyH, 0));
4537 __ pshufb(xmm_temp1, xmm_temp10);
4538
4539 __ pxor(xmm_temp0, xmm_temp2);
4540
4541 //
4542 // Multiply with the hash key
4543 //
4544 __ movdqu(xmm_temp3, xmm_temp0);
4545 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
4546 __ movdqu(xmm_temp4, xmm_temp0);
4547 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
4548
4549 __ movdqu(xmm_temp5, xmm_temp0);
4550 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
4551 __ movdqu(xmm_temp6, xmm_temp0);
4552 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
4553
4554 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
4555
4556 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
4557 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
4558 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
4559 __ pxor(xmm_temp3, xmm_temp5);
4560 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
4561 // of the carry-less multiplication of
4562 // xmm0 by xmm1.
4563
4564 // We shift the result of the multiplication by one bit position
4565 // to the left to cope for the fact that the bits are reversed.
4566 __ movdqu(xmm_temp7, xmm_temp3);
4567 __ movdqu(xmm_temp8, xmm_temp6);
4568 __ pslld(xmm_temp3, 1);
4569 __ pslld(xmm_temp6, 1);
4570 __ psrld(xmm_temp7, 31);
4571 __ psrld(xmm_temp8, 31);
4572 __ movdqu(xmm_temp9, xmm_temp7);
4573 __ pslldq(xmm_temp8, 4);
4574 __ pslldq(xmm_temp7, 4);
4575 __ psrldq(xmm_temp9, 12);
4576 __ por(xmm_temp3, xmm_temp7);
4577 __ por(xmm_temp6, xmm_temp8);
4578 __ por(xmm_temp6, xmm_temp9);
4579
4580 //
4581 // First phase of the reduction
4582 //
4583 // Move xmm3 into xmm7, xmm8, xmm9 in order to perform the shifts
4584 // independently.
4585 __ movdqu(xmm_temp7, xmm_temp3);
4586 __ movdqu(xmm_temp8, xmm_temp3);
4587 __ movdqu(xmm_temp9, xmm_temp3);
4588 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
4589 __ pslld(xmm_temp8, 30); // packed right shift shifting << 30
4590 __ pslld(xmm_temp9, 25); // packed right shift shifting << 25
4591 __ pxor(xmm_temp7, xmm_temp8); // xor the shifted versions
4592 __ pxor(xmm_temp7, xmm_temp9);
4593 __ movdqu(xmm_temp8, xmm_temp7);
4594 __ pslldq(xmm_temp7, 12);
4595 __ psrldq(xmm_temp8, 4);
4596 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
4597
4598 //
4599 // Second phase of the reduction
4600 //
4601 // Make 3 copies of xmm3 in xmm2, xmm4, xmm5 for doing these
4602 // shift operations.
4603 __ movdqu(xmm_temp2, xmm_temp3);
4604 __ movdqu(xmm_temp4, xmm_temp3);
4605 __ movdqu(xmm_temp5, xmm_temp3);
4606 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
4607 __ psrld(xmm_temp4, 2); // packed left shifting >> 2
4608 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
4609 __ pxor(xmm_temp2, xmm_temp4); // xor the shifted versions
4610 __ pxor(xmm_temp2, xmm_temp5);
4611 __ pxor(xmm_temp2, xmm_temp8);
4612 __ pxor(xmm_temp3, xmm_temp2);
4613 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
4614
4615 __ decrement(blocks);
4616 __ jcc(Assembler::zero, L_exit);
4617 __ movdqu(xmm_temp0, xmm_temp6);
4618 __ addptr(data, 16);
4619 __ jmp(L_ghash_loop);
4620
4621 __ BIND(L_exit);
4622 __ pshufb(xmm_temp6, xmm_temp10); // Byte swap 16-byte result
4623 __ movdqu(Address(state, 0), xmm_temp6); // store the result
4624 __ leave();
4625 __ ret(0);
4626 return start;
4627 }
4628
4629 //base64 character set
4630 address base64_charset_addr() {
4631 __ align(CodeEntryAlignment);
4632 StubCodeMark mark(this, "StubRoutines", "base64_charset");
4633 address start = __ pc();
4634 __ emit_data64(0x0000004200000041, relocInfo::none);
4635 __ emit_data64(0x0000004400000043, relocInfo::none);
4636 __ emit_data64(0x0000004600000045, relocInfo::none);
4637 __ emit_data64(0x0000004800000047, relocInfo::none);
4638 __ emit_data64(0x0000004a00000049, relocInfo::none);
4639 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4640 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4641 __ emit_data64(0x000000500000004f, relocInfo::none);
4642 __ emit_data64(0x0000005200000051, relocInfo::none);
4643 __ emit_data64(0x0000005400000053, relocInfo::none);
4644 __ emit_data64(0x0000005600000055, relocInfo::none);
4645 __ emit_data64(0x0000005800000057, relocInfo::none);
4646 __ emit_data64(0x0000005a00000059, relocInfo::none);
4647 __ emit_data64(0x0000006200000061, relocInfo::none);
4648 __ emit_data64(0x0000006400000063, relocInfo::none);
4649 __ emit_data64(0x0000006600000065, relocInfo::none);
4650 __ emit_data64(0x0000006800000067, relocInfo::none);
4651 __ emit_data64(0x0000006a00000069, relocInfo::none);
4652 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4653 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4654 __ emit_data64(0x000000700000006f, relocInfo::none);
4655 __ emit_data64(0x0000007200000071, relocInfo::none);
4656 __ emit_data64(0x0000007400000073, relocInfo::none);
4657 __ emit_data64(0x0000007600000075, relocInfo::none);
4658 __ emit_data64(0x0000007800000077, relocInfo::none);
4659 __ emit_data64(0x0000007a00000079, relocInfo::none);
4660 __ emit_data64(0x0000003100000030, relocInfo::none);
4661 __ emit_data64(0x0000003300000032, relocInfo::none);
4662 __ emit_data64(0x0000003500000034, relocInfo::none);
4663 __ emit_data64(0x0000003700000036, relocInfo::none);
4664 __ emit_data64(0x0000003900000038, relocInfo::none);
4665 __ emit_data64(0x0000002f0000002b, relocInfo::none);
4666 return start;
4667 }
4668
4669 //base64 url character set
4670 address base64url_charset_addr() {
4671 __ align(CodeEntryAlignment);
4672 StubCodeMark mark(this, "StubRoutines", "base64url_charset");
4673 address start = __ pc();
4674 __ emit_data64(0x0000004200000041, relocInfo::none);
4675 __ emit_data64(0x0000004400000043, relocInfo::none);
4676 __ emit_data64(0x0000004600000045, relocInfo::none);
4677 __ emit_data64(0x0000004800000047, relocInfo::none);
4678 __ emit_data64(0x0000004a00000049, relocInfo::none);
4679 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4680 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4681 __ emit_data64(0x000000500000004f, relocInfo::none);
4682 __ emit_data64(0x0000005200000051, relocInfo::none);
4683 __ emit_data64(0x0000005400000053, relocInfo::none);
4684 __ emit_data64(0x0000005600000055, relocInfo::none);
4685 __ emit_data64(0x0000005800000057, relocInfo::none);
4686 __ emit_data64(0x0000005a00000059, relocInfo::none);
4687 __ emit_data64(0x0000006200000061, relocInfo::none);
4688 __ emit_data64(0x0000006400000063, relocInfo::none);
4689 __ emit_data64(0x0000006600000065, relocInfo::none);
4690 __ emit_data64(0x0000006800000067, relocInfo::none);
4691 __ emit_data64(0x0000006a00000069, relocInfo::none);
4692 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4693 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4694 __ emit_data64(0x000000700000006f, relocInfo::none);
4695 __ emit_data64(0x0000007200000071, relocInfo::none);
4696 __ emit_data64(0x0000007400000073, relocInfo::none);
4697 __ emit_data64(0x0000007600000075, relocInfo::none);
4698 __ emit_data64(0x0000007800000077, relocInfo::none);
4699 __ emit_data64(0x0000007a00000079, relocInfo::none);
4700 __ emit_data64(0x0000003100000030, relocInfo::none);
4701 __ emit_data64(0x0000003300000032, relocInfo::none);
4702 __ emit_data64(0x0000003500000034, relocInfo::none);
4703 __ emit_data64(0x0000003700000036, relocInfo::none);
4704 __ emit_data64(0x0000003900000038, relocInfo::none);
4705 __ emit_data64(0x0000005f0000002d, relocInfo::none);
4706
4707 return start;
4708 }
4709
4710 address base64_bswap_mask_addr() {
4711 __ align(CodeEntryAlignment);
4712 StubCodeMark mark(this, "StubRoutines", "bswap_mask_base64");
4713 address start = __ pc();
4714 __ emit_data64(0x0504038002010080, relocInfo::none);
4715 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4716 __ emit_data64(0x0908078006050480, relocInfo::none);
4717 __ emit_data64(0x0f0e0d800c0b0a80, relocInfo::none);
4718 __ emit_data64(0x0605048003020180, relocInfo::none);
4719 __ emit_data64(0x0c0b0a8009080780, relocInfo::none);
4720 __ emit_data64(0x0504038002010080, relocInfo::none);
4721 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4722
4723 return start;
4724 }
4725
4726 address base64_right_shift_mask_addr() {
4727 __ align(CodeEntryAlignment);
4728 StubCodeMark mark(this, "StubRoutines", "right_shift_mask");
4729 address start = __ pc();
4730 __ emit_data64(0x0006000400020000, relocInfo::none);
4731 __ emit_data64(0x0006000400020000, relocInfo::none);
4732 __ emit_data64(0x0006000400020000, relocInfo::none);
4733 __ emit_data64(0x0006000400020000, relocInfo::none);
4734 __ emit_data64(0x0006000400020000, relocInfo::none);
4735 __ emit_data64(0x0006000400020000, relocInfo::none);
4736 __ emit_data64(0x0006000400020000, relocInfo::none);
4737 __ emit_data64(0x0006000400020000, relocInfo::none);
4738
4739 return start;
4740 }
4741
4742 address base64_left_shift_mask_addr() {
4743 __ align(CodeEntryAlignment);
4744 StubCodeMark mark(this, "StubRoutines", "left_shift_mask");
4745 address start = __ pc();
4746 __ emit_data64(0x0000000200040000, relocInfo::none);
4747 __ emit_data64(0x0000000200040000, relocInfo::none);
4748 __ emit_data64(0x0000000200040000, relocInfo::none);
4749 __ emit_data64(0x0000000200040000, relocInfo::none);
4750 __ emit_data64(0x0000000200040000, relocInfo::none);
4751 __ emit_data64(0x0000000200040000, relocInfo::none);
4752 __ emit_data64(0x0000000200040000, relocInfo::none);
4753 __ emit_data64(0x0000000200040000, relocInfo::none);
4754
4755 return start;
4756 }
4757
4758 address base64_and_mask_addr() {
4759 __ align(CodeEntryAlignment);
4760 StubCodeMark mark(this, "StubRoutines", "and_mask");
4761 address start = __ pc();
4762 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4763 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4764 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4765 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4766 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4767 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4768 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4769 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4770 return start;
4771 }
4772
4773 address base64_gather_mask_addr() {
4774 __ align(CodeEntryAlignment);
4775 StubCodeMark mark(this, "StubRoutines", "gather_mask");
4776 address start = __ pc();
4777 __ emit_data64(0xffffffffffffffff, relocInfo::none);
4778 return start;
4779 }
4780
4781 // Code for generating Base64 encoding.
4782 // Intrinsic function prototype in Base64.java:
4783 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL) {
4784 address generate_base64_encodeBlock() {
4785 __ align(CodeEntryAlignment);
4786 StubCodeMark mark(this, "StubRoutines", "implEncode");
4787 address start = __ pc();
4788 __ enter();
4789
4790 // Save callee-saved registers before using them
4791 __ push(r12);
4792 __ push(r13);
4793 __ push(r14);
4794 __ push(r15);
4795
4796 // arguments
4797 const Register source = c_rarg0; // Source Array
4798 const Register start_offset = c_rarg1; // start offset
4799 const Register end_offset = c_rarg2; // end offset
4800 const Register dest = c_rarg3; // destination array
4801
4802 #ifndef _WIN64
4803 const Register dp = c_rarg4; // Position for writing to dest array
4804 const Register isURL = c_rarg5;// Base64 or URL character set
4805 #else
4806 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
4807 const Address isURL_mem(rbp, 7 * wordSize);
4808 const Register isURL = r10; // pick the volatile windows register
4809 const Register dp = r12;
4810 __ movl(dp, dp_mem);
4811 __ movl(isURL, isURL_mem);
4812 #endif
4813
4814 const Register length = r14;
4815 Label L_process80, L_process32, L_process3, L_exit, L_processdata;
4816
4817 // calculate length from offsets
4818 __ movl(length, end_offset);
4819 __ subl(length, start_offset);
4820 __ cmpl(length, 0);
4821 __ jcc(Assembler::lessEqual, L_exit);
4822
4823 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_charset_addr()));
4824 // check if base64 charset(isURL=0) or base64 url charset(isURL=1) needs to be loaded
4825 __ cmpl(isURL, 0);
4826 __ jcc(Assembler::equal, L_processdata);
4827 __ lea(r11, ExternalAddress(StubRoutines::x86::base64url_charset_addr()));
4828
4829 // load masks required for encoding data
4830 __ BIND(L_processdata);
4831 __ movdqu(xmm16, ExternalAddress(StubRoutines::x86::base64_gather_mask_addr()));
4832 // Set 64 bits of K register.
4833 __ evpcmpeqb(k3, xmm16, xmm16, Assembler::AVX_512bit);
4834 __ evmovdquq(xmm12, ExternalAddress(StubRoutines::x86::base64_bswap_mask_addr()), Assembler::AVX_256bit, r13);
4835 __ evmovdquq(xmm13, ExternalAddress(StubRoutines::x86::base64_right_shift_mask_addr()), Assembler::AVX_512bit, r13);
4836 __ evmovdquq(xmm14, ExternalAddress(StubRoutines::x86::base64_left_shift_mask_addr()), Assembler::AVX_512bit, r13);
4837 __ evmovdquq(xmm15, ExternalAddress(StubRoutines::x86::base64_and_mask_addr()), Assembler::AVX_512bit, r13);
4838
4839 // Vector Base64 implementation, producing 96 bytes of encoded data
4840 __ BIND(L_process80);
4841 __ cmpl(length, 80);
4842 __ jcc(Assembler::below, L_process32);
4843 __ evmovdquq(xmm0, Address(source, start_offset, Address::times_1, 0), Assembler::AVX_256bit);
4844 __ evmovdquq(xmm1, Address(source, start_offset, Address::times_1, 24), Assembler::AVX_256bit);
4845 __ evmovdquq(xmm2, Address(source, start_offset, Address::times_1, 48), Assembler::AVX_256bit);
4846
4847 //permute the input data in such a manner that we have continuity of the source
4848 __ vpermq(xmm3, xmm0, 148, Assembler::AVX_256bit);
4849 __ vpermq(xmm4, xmm1, 148, Assembler::AVX_256bit);
4850 __ vpermq(xmm5, xmm2, 148, Assembler::AVX_256bit);
4851
4852 //shuffle input and group 3 bytes of data and to it add 0 as the 4th byte.
4853 //we can deal with 12 bytes at a time in a 128 bit register
4854 __ vpshufb(xmm3, xmm3, xmm12, Assembler::AVX_256bit);
4855 __ vpshufb(xmm4, xmm4, xmm12, Assembler::AVX_256bit);
4856 __ vpshufb(xmm5, xmm5, xmm12, Assembler::AVX_256bit);
4857
4858 //convert byte to word. Each 128 bit register will have 6 bytes for processing
4859 __ vpmovzxbw(xmm3, xmm3, Assembler::AVX_512bit);
4860 __ vpmovzxbw(xmm4, xmm4, Assembler::AVX_512bit);
4861 __ vpmovzxbw(xmm5, xmm5, Assembler::AVX_512bit);
4862
4863 // Extract bits in the following pattern 6, 4+2, 2+4, 6 to convert 3, 8 bit numbers to 4, 6 bit numbers
4864 __ evpsrlvw(xmm0, xmm3, xmm13, Assembler::AVX_512bit);
4865 __ evpsrlvw(xmm1, xmm4, xmm13, Assembler::AVX_512bit);
4866 __ evpsrlvw(xmm2, xmm5, xmm13, Assembler::AVX_512bit);
4867
4868 __ evpsllvw(xmm3, xmm3, xmm14, Assembler::AVX_512bit);
4869 __ evpsllvw(xmm4, xmm4, xmm14, Assembler::AVX_512bit);
4870 __ evpsllvw(xmm5, xmm5, xmm14, Assembler::AVX_512bit);
4871
4872 __ vpsrlq(xmm0, xmm0, 8, Assembler::AVX_512bit);
4873 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
4874 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
4875
4876 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4877 __ vpsllq(xmm4, xmm4, 8, Assembler::AVX_512bit);
4878 __ vpsllq(xmm5, xmm5, 8, Assembler::AVX_512bit);
4879
4880 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
4881 __ vpandq(xmm4, xmm4, xmm15, Assembler::AVX_512bit);
4882 __ vpandq(xmm5, xmm5, xmm15, Assembler::AVX_512bit);
4883
4884 // Get the final 4*6 bits base64 encoding
4885 __ vporq(xmm3, xmm3, xmm0, Assembler::AVX_512bit);
4886 __ vporq(xmm4, xmm4, xmm1, Assembler::AVX_512bit);
4887 __ vporq(xmm5, xmm5, xmm2, Assembler::AVX_512bit);
4888
4889 // Shift
4890 __ vpsrlq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4891 __ vpsrlq(xmm4, xmm4, 8, Assembler::AVX_512bit);
4892 __ vpsrlq(xmm5, xmm5, 8, Assembler::AVX_512bit);
4893
4894 // look up 6 bits in the base64 character set to fetch the encoding
4895 // we are converting word to dword as gather instructions need dword indices for looking up encoding
4896 __ vextracti64x4(xmm6, xmm3, 0);
4897 __ vpmovzxwd(xmm0, xmm6, Assembler::AVX_512bit);
4898 __ vextracti64x4(xmm6, xmm3, 1);
4899 __ vpmovzxwd(xmm1, xmm6, Assembler::AVX_512bit);
4900
4901 __ vextracti64x4(xmm6, xmm4, 0);
4902 __ vpmovzxwd(xmm2, xmm6, Assembler::AVX_512bit);
4903 __ vextracti64x4(xmm6, xmm4, 1);
4904 __ vpmovzxwd(xmm3, xmm6, Assembler::AVX_512bit);
4905
4906 __ vextracti64x4(xmm4, xmm5, 0);
4907 __ vpmovzxwd(xmm6, xmm4, Assembler::AVX_512bit);
4908
4909 __ vextracti64x4(xmm4, xmm5, 1);
4910 __ vpmovzxwd(xmm7, xmm4, Assembler::AVX_512bit);
4911
4912 __ kmovql(k2, k3);
4913 __ evpgatherdd(xmm4, k2, Address(r11, xmm0, Address::times_4, 0), Assembler::AVX_512bit);
4914 __ kmovql(k2, k3);
4915 __ evpgatherdd(xmm5, k2, Address(r11, xmm1, Address::times_4, 0), Assembler::AVX_512bit);
4916 __ kmovql(k2, k3);
4917 __ evpgatherdd(xmm8, k2, Address(r11, xmm2, Address::times_4, 0), Assembler::AVX_512bit);
4918 __ kmovql(k2, k3);
4919 __ evpgatherdd(xmm9, k2, Address(r11, xmm3, Address::times_4, 0), Assembler::AVX_512bit);
4920 __ kmovql(k2, k3);
4921 __ evpgatherdd(xmm10, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
4922 __ kmovql(k2, k3);
4923 __ evpgatherdd(xmm11, k2, Address(r11, xmm7, Address::times_4, 0), Assembler::AVX_512bit);
4924
4925 //Down convert dword to byte. Final output is 16*6 = 96 bytes long
4926 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm4, Assembler::AVX_512bit);
4927 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm5, Assembler::AVX_512bit);
4928 __ evpmovdb(Address(dest, dp, Address::times_1, 32), xmm8, Assembler::AVX_512bit);
4929 __ evpmovdb(Address(dest, dp, Address::times_1, 48), xmm9, Assembler::AVX_512bit);
4930 __ evpmovdb(Address(dest, dp, Address::times_1, 64), xmm10, Assembler::AVX_512bit);
4931 __ evpmovdb(Address(dest, dp, Address::times_1, 80), xmm11, Assembler::AVX_512bit);
4932
4933 __ addq(dest, 96);
4934 __ addq(source, 72);
4935 __ subq(length, 72);
4936 __ jmp(L_process80);
4937
4938 // Vector Base64 implementation generating 32 bytes of encoded data
4939 __ BIND(L_process32);
4940 __ cmpl(length, 32);
4941 __ jcc(Assembler::below, L_process3);
4942 __ evmovdquq(xmm0, Address(source, start_offset), Assembler::AVX_256bit);
4943 __ vpermq(xmm0, xmm0, 148, Assembler::AVX_256bit);
4944 __ vpshufb(xmm6, xmm0, xmm12, Assembler::AVX_256bit);
4945 __ vpmovzxbw(xmm6, xmm6, Assembler::AVX_512bit);
4946 __ evpsrlvw(xmm2, xmm6, xmm13, Assembler::AVX_512bit);
4947 __ evpsllvw(xmm3, xmm6, xmm14, Assembler::AVX_512bit);
4948
4949 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
4950 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4951 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
4952 __ vporq(xmm1, xmm2, xmm3, Assembler::AVX_512bit);
4953 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
4954 __ vextracti64x4(xmm9, xmm1, 0);
4955 __ vpmovzxwd(xmm6, xmm9, Assembler::AVX_512bit);
4956 __ vextracti64x4(xmm9, xmm1, 1);
4957 __ vpmovzxwd(xmm5, xmm9, Assembler::AVX_512bit);
4958 __ kmovql(k2, k3);
4959 __ evpgatherdd(xmm8, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
4960 __ kmovql(k2, k3);
4961 __ evpgatherdd(xmm10, k2, Address(r11, xmm5, Address::times_4, 0), Assembler::AVX_512bit);
4962 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm8, Assembler::AVX_512bit);
4963 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm10, Assembler::AVX_512bit);
4964 __ subq(length, 24);
4965 __ addq(dest, 32);
4966 __ addq(source, 24);
4967 __ jmp(L_process32);
4968
4969 // Scalar data processing takes 3 bytes at a time and produces 4 bytes of encoded data
4970 /* This code corresponds to the scalar version of the following snippet in Base64.java
4971 ** int bits = (src[sp0++] & 0xff) << 16 |(src[sp0++] & 0xff) << 8 |(src[sp0++] & 0xff);
4972 ** dst[dp0++] = (byte)base64[(bits >> > 18) & 0x3f];
4973 ** dst[dp0++] = (byte)base64[(bits >> > 12) & 0x3f];
4974 ** dst[dp0++] = (byte)base64[(bits >> > 6) & 0x3f];
4975 ** dst[dp0++] = (byte)base64[bits & 0x3f];*/
4976 __ BIND(L_process3);
4977 __ cmpl(length, 3);
4978 __ jcc(Assembler::below, L_exit);
4979 // Read 1 byte at a time
4980 __ movzbl(rax, Address(source, start_offset));
4981 __ shll(rax, 0x10);
4982 __ movl(r15, rax);
4983 __ movzbl(rax, Address(source, start_offset, Address::times_1, 1));
4984 __ shll(rax, 0x8);
4985 __ movzwl(rax, rax);
4986 __ orl(r15, rax);
4987 __ movzbl(rax, Address(source, start_offset, Address::times_1, 2));
4988 __ orl(rax, r15);
4989 // Save 3 bytes read in r15
4990 __ movl(r15, rax);
4991 __ shrl(rax, 0x12);
4992 __ andl(rax, 0x3f);
4993 // rax contains the index, r11 contains base64 lookup table
4994 __ movb(rax, Address(r11, rax, Address::times_4));
4995 // Write the encoded byte to destination
4996 __ movb(Address(dest, dp, Address::times_1, 0), rax);
4997 __ movl(rax, r15);
4998 __ shrl(rax, 0xc);
4999 __ andl(rax, 0x3f);
5000 __ movb(rax, Address(r11, rax, Address::times_4));
5001 __ movb(Address(dest, dp, Address::times_1, 1), rax);
5002 __ movl(rax, r15);
5003 __ shrl(rax, 0x6);
5004 __ andl(rax, 0x3f);
5005 __ movb(rax, Address(r11, rax, Address::times_4));
5006 __ movb(Address(dest, dp, Address::times_1, 2), rax);
5007 __ movl(rax, r15);
5008 __ andl(rax, 0x3f);
5009 __ movb(rax, Address(r11, rax, Address::times_4));
5010 __ movb(Address(dest, dp, Address::times_1, 3), rax);
5011 __ subl(length, 3);
5012 __ addq(dest, 4);
5013 __ addq(source, 3);
5014 __ jmp(L_process3);
5015 __ BIND(L_exit);
5016 __ pop(r15);
5017 __ pop(r14);
5018 __ pop(r13);
5019 __ pop(r12);
5020 __ leave();
5021 __ ret(0);
5022 return start;
5023 }
5024
5025 /**
5026 * Arguments:
5027 *
5028 * Inputs:
5029 * c_rarg0 - int crc
5030 * c_rarg1 - byte* buf
5031 * c_rarg2 - int length
5032 *
5033 * Ouput:
5034 * rax - int crc result
5035 */
5036 address generate_updateBytesCRC32() {
5037 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
5038
5039 __ align(CodeEntryAlignment);
5040 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
5041
5042 address start = __ pc();
5043 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5044 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5045 // rscratch1: r10
5046 const Register crc = c_rarg0; // crc
5047 const Register buf = c_rarg1; // source java byte array address
5048 const Register len = c_rarg2; // length
5049 const Register table = c_rarg3; // crc_table address (reuse register)
5050 const Register tmp = r11;
5051 assert_different_registers(crc, buf, len, table, tmp, rax);
5052
5053 BLOCK_COMMENT("Entry:");
5054 __ enter(); // required for proper stackwalking of RuntimeStub frame
5055
5056 __ kernel_crc32(crc, buf, len, table, tmp);
5057
5058 __ movl(rax, crc);
5059 __ vzeroupper();
5060 __ leave(); // required for proper stackwalking of RuntimeStub frame
5061 __ ret(0);
5062
5063 return start;
5064 }
5065
5066 /**
5067 * Arguments:
5068 *
5069 * Inputs:
5070 * c_rarg0 - int crc
5071 * c_rarg1 - byte* buf
5072 * c_rarg2 - long length
5073 * c_rarg3 - table_start - optional (present only when doing a library_call,
5074 * not used by x86 algorithm)
5075 *
5076 * Ouput:
5077 * rax - int crc result
5078 */
5079 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
5080 assert(UseCRC32CIntrinsics, "need SSE4_2");
5081 __ align(CodeEntryAlignment);
5082 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
5083 address start = __ pc();
5084 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
5085 //Windows RCX RDX R8 R9 none none XMM0..XMM3
5086 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
5087 const Register crc = c_rarg0; // crc
5088 const Register buf = c_rarg1; // source java byte array address
5089 const Register len = c_rarg2; // length
5090 const Register a = rax;
5091 const Register j = r9;
5092 const Register k = r10;
5093 const Register l = r11;
5094 #ifdef _WIN64
5095 const Register y = rdi;
5096 const Register z = rsi;
5097 #else
5098 const Register y = rcx;
5099 const Register z = r8;
5100 #endif
5101 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
5102
5103 BLOCK_COMMENT("Entry:");
5104 __ enter(); // required for proper stackwalking of RuntimeStub frame
5105 #ifdef _WIN64
5106 __ push(y);
5107 __ push(z);
5108 #endif
5109 __ crc32c_ipl_alg2_alt2(crc, buf, len,
5110 a, j, k,
5111 l, y, z,
5112 c_farg0, c_farg1, c_farg2,
5113 is_pclmulqdq_supported);
5114 __ movl(rax, crc);
5115 #ifdef _WIN64
5116 __ pop(z);
5117 __ pop(y);
5118 #endif
5119 __ vzeroupper();
5120 __ leave(); // required for proper stackwalking of RuntimeStub frame
5121 __ ret(0);
5122
5123 return start;
5124 }
5125
5126 /**
5127 * Arguments:
5128 *
5129 * Input:
5130 * c_rarg0 - x address
5131 * c_rarg1 - x length
5132 * c_rarg2 - y address
5133 * c_rarg3 - y length
5134 * not Win64
5135 * c_rarg4 - z address
5136 * c_rarg5 - z length
5137 * Win64
5138 * rsp+40 - z address
5139 * rsp+48 - z length
5140 */
5141 address generate_multiplyToLen() {
5142 __ align(CodeEntryAlignment);
5143 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
5144
5145 address start = __ pc();
5146 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5147 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5148 const Register x = rdi;
5149 const Register xlen = rax;
5150 const Register y = rsi;
5151 const Register ylen = rcx;
5152 const Register z = r8;
5153 const Register zlen = r11;
5154
5155 // Next registers will be saved on stack in multiply_to_len().
5156 const Register tmp1 = r12;
5157 const Register tmp2 = r13;
5158 const Register tmp3 = r14;
5159 const Register tmp4 = r15;
5160 const Register tmp5 = rbx;
5161
5162 BLOCK_COMMENT("Entry:");
5163 __ enter(); // required for proper stackwalking of RuntimeStub frame
5164
5165 #ifndef _WIN64
5166 __ movptr(zlen, r9); // Save r9 in r11 - zlen
5167 #endif
5168 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
5169 // ylen => rcx, z => r8, zlen => r11
5170 // r9 and r10 may be used to save non-volatile registers
5171 #ifdef _WIN64
5172 // last 2 arguments (#4, #5) are on stack on Win64
5173 __ movptr(z, Address(rsp, 6 * wordSize));
5174 __ movptr(zlen, Address(rsp, 7 * wordSize));
5175 #endif
5176
5177 __ movptr(xlen, rsi);
5178 __ movptr(y, rdx);
5179 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
5180
5181 restore_arg_regs();
5182
5183 __ leave(); // required for proper stackwalking of RuntimeStub frame
5184 __ ret(0);
5185
5186 return start;
5187 }
5188
5189 /**
5190 * Arguments:
5191 *
5192 * Input:
5193 * c_rarg0 - obja address
5194 * c_rarg1 - objb address
5195 * c_rarg3 - length length
5196 * c_rarg4 - scale log2_array_indxscale
5197 *
5198 * Output:
5199 * rax - int >= mismatched index, < 0 bitwise complement of tail
5200 */
5201 address generate_vectorizedMismatch() {
5202 __ align(CodeEntryAlignment);
5203 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
5204 address start = __ pc();
5205
5206 BLOCK_COMMENT("Entry:");
5207 __ enter();
5208
5209 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5210 const Register scale = c_rarg0; //rcx, will exchange with r9
5211 const Register objb = c_rarg1; //rdx
5212 const Register length = c_rarg2; //r8
5213 const Register obja = c_rarg3; //r9
5214 __ xchgq(obja, scale); //now obja and scale contains the correct contents
5215
5216 const Register tmp1 = r10;
5217 const Register tmp2 = r11;
5218 #endif
5219 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5220 const Register obja = c_rarg0; //U:rdi
5221 const Register objb = c_rarg1; //U:rsi
5222 const Register length = c_rarg2; //U:rdx
5223 const Register scale = c_rarg3; //U:rcx
5224 const Register tmp1 = r8;
5225 const Register tmp2 = r9;
5226 #endif
5227 const Register result = rax; //return value
5228 const XMMRegister vec0 = xmm0;
5229 const XMMRegister vec1 = xmm1;
5230 const XMMRegister vec2 = xmm2;
5231
5232 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
5233
5234 __ vzeroupper();
5235 __ leave();
5236 __ ret(0);
5237
5238 return start;
5239 }
5240
5241 /**
5242 * Arguments:
5243 *
5244 // Input:
5245 // c_rarg0 - x address
5246 // c_rarg1 - x length
5247 // c_rarg2 - z address
5248 // c_rarg3 - z lenth
5249 *
5250 */
5251 address generate_squareToLen() {
5252
5253 __ align(CodeEntryAlignment);
5254 StubCodeMark mark(this, "StubRoutines", "squareToLen");
5255
5256 address start = __ pc();
5257 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5258 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
5259 const Register x = rdi;
5260 const Register len = rsi;
5261 const Register z = r8;
5262 const Register zlen = rcx;
5263
5264 const Register tmp1 = r12;
5265 const Register tmp2 = r13;
5266 const Register tmp3 = r14;
5267 const Register tmp4 = r15;
5268 const Register tmp5 = rbx;
5269
5270 BLOCK_COMMENT("Entry:");
5271 __ enter(); // required for proper stackwalking of RuntimeStub frame
5272
5273 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
5274 // zlen => rcx
5275 // r9 and r10 may be used to save non-volatile registers
5276 __ movptr(r8, rdx);
5277 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5278
5279 restore_arg_regs();
5280
5281 __ leave(); // required for proper stackwalking of RuntimeStub frame
5282 __ ret(0);
5283
5284 return start;
5285 }
5286
5287 address generate_method_entry_barrier() {
5288 __ align(CodeEntryAlignment);
5289 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
5290
5291 Label deoptimize_label;
5292
5293 address start = __ pc();
5294
5295 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
5296
5297 BLOCK_COMMENT("Entry:");
5298 __ enter(); // save rbp
5299
5300 // save c_rarg0, because we want to use that value.
5301 // We could do without it but then we depend on the number of slots used by pusha
5302 __ push(c_rarg0);
5303
5304 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
5305
5306 __ pusha();
5307
5308 // The method may have floats as arguments, and we must spill them before calling
5309 // the VM runtime.
5310 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
5311 const int xmm_size = wordSize * 2;
5312 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
5313 __ subptr(rsp, xmm_spill_size);
5314 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
5315 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
5316 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
5317 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
5318 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
5319 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
5320 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
5321 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
5322
5323 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
5324
5325 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
5326 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
5327 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
5328 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
5329 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
5330 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
5331 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
5332 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
5333 __ addptr(rsp, xmm_spill_size);
5334
5335 __ cmpl(rax, 1); // 1 means deoptimize
5336 __ jcc(Assembler::equal, deoptimize_label);
5337
5338 __ popa();
5339 __ pop(c_rarg0);
5340
5341 __ leave();
5342
5343 __ addptr(rsp, 1 * wordSize); // cookie
5344 __ ret(0);
5345
5346
5347 __ BIND(deoptimize_label);
5348
5349 __ popa();
5350 __ pop(c_rarg0);
5351
5352 __ leave();
5353
5354 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
5355 // here while still having a correct stack is valuable
5356 __ testptr(rsp, Address(rsp, 0));
5357
5358 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
5359 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
5360
5361 return start;
5362 }
5363
5364 /**
5365 * Arguments:
5366 *
5367 * Input:
5368 * c_rarg0 - out address
5369 * c_rarg1 - in address
5370 * c_rarg2 - offset
5371 * c_rarg3 - len
5372 * not Win64
5373 * c_rarg4 - k
5374 * Win64
5375 * rsp+40 - k
5376 */
5377 address generate_mulAdd() {
5378 __ align(CodeEntryAlignment);
5379 StubCodeMark mark(this, "StubRoutines", "mulAdd");
5380
5381 address start = __ pc();
5382 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5383 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5384 const Register out = rdi;
5385 const Register in = rsi;
5386 const Register offset = r11;
5387 const Register len = rcx;
5388 const Register k = r8;
5389
5390 // Next registers will be saved on stack in mul_add().
5391 const Register tmp1 = r12;
5392 const Register tmp2 = r13;
5393 const Register tmp3 = r14;
5394 const Register tmp4 = r15;
5395 const Register tmp5 = rbx;
5396
5397 BLOCK_COMMENT("Entry:");
5398 __ enter(); // required for proper stackwalking of RuntimeStub frame
5399
5400 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
5401 // len => rcx, k => r8
5402 // r9 and r10 may be used to save non-volatile registers
5403 #ifdef _WIN64
5404 // last argument is on stack on Win64
5405 __ movl(k, Address(rsp, 6 * wordSize));
5406 #endif
5407 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
5408 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5409
5410 restore_arg_regs();
5411
5412 __ leave(); // required for proper stackwalking of RuntimeStub frame
5413 __ ret(0);
5414
5415 return start;
5416 }
5417
5418 address generate_libmExp() {
5419 StubCodeMark mark(this, "StubRoutines", "libmExp");
5420
5421 address start = __ pc();
5422
5423 const XMMRegister x0 = xmm0;
5424 const XMMRegister x1 = xmm1;
5425 const XMMRegister x2 = xmm2;
5426 const XMMRegister x3 = xmm3;
5427
5428 const XMMRegister x4 = xmm4;
5429 const XMMRegister x5 = xmm5;
5430 const XMMRegister x6 = xmm6;
5431 const XMMRegister x7 = xmm7;
5432
5433 const Register tmp = r11;
5434
5435 BLOCK_COMMENT("Entry:");
5436 __ enter(); // required for proper stackwalking of RuntimeStub frame
5437
5438 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5439
5440 __ leave(); // required for proper stackwalking of RuntimeStub frame
5441 __ ret(0);
5442
5443 return start;
5444
5445 }
5446
5447 address generate_libmLog() {
5448 StubCodeMark mark(this, "StubRoutines", "libmLog");
5449
5450 address start = __ pc();
5451
5452 const XMMRegister x0 = xmm0;
5453 const XMMRegister x1 = xmm1;
5454 const XMMRegister x2 = xmm2;
5455 const XMMRegister x3 = xmm3;
5456
5457 const XMMRegister x4 = xmm4;
5458 const XMMRegister x5 = xmm5;
5459 const XMMRegister x6 = xmm6;
5460 const XMMRegister x7 = xmm7;
5461
5462 const Register tmp1 = r11;
5463 const Register tmp2 = r8;
5464
5465 BLOCK_COMMENT("Entry:");
5466 __ enter(); // required for proper stackwalking of RuntimeStub frame
5467
5468 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2);
5469
5470 __ leave(); // required for proper stackwalking of RuntimeStub frame
5471 __ ret(0);
5472
5473 return start;
5474
5475 }
5476
5477 address generate_libmLog10() {
5478 StubCodeMark mark(this, "StubRoutines", "libmLog10");
5479
5480 address start = __ pc();
5481
5482 const XMMRegister x0 = xmm0;
5483 const XMMRegister x1 = xmm1;
5484 const XMMRegister x2 = xmm2;
5485 const XMMRegister x3 = xmm3;
5486
5487 const XMMRegister x4 = xmm4;
5488 const XMMRegister x5 = xmm5;
5489 const XMMRegister x6 = xmm6;
5490 const XMMRegister x7 = xmm7;
5491
5492 const Register tmp = r11;
5493
5494 BLOCK_COMMENT("Entry:");
5495 __ enter(); // required for proper stackwalking of RuntimeStub frame
5496
5497 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5498
5499 __ leave(); // required for proper stackwalking of RuntimeStub frame
5500 __ ret(0);
5501
5502 return start;
5503
5504 }
5505
5506 address generate_libmPow() {
5507 StubCodeMark mark(this, "StubRoutines", "libmPow");
5508
5509 address start = __ pc();
5510
5511 const XMMRegister x0 = xmm0;
5512 const XMMRegister x1 = xmm1;
5513 const XMMRegister x2 = xmm2;
5514 const XMMRegister x3 = xmm3;
5515
5516 const XMMRegister x4 = xmm4;
5517 const XMMRegister x5 = xmm5;
5518 const XMMRegister x6 = xmm6;
5519 const XMMRegister x7 = xmm7;
5520
5521 const Register tmp1 = r8;
5522 const Register tmp2 = r9;
5523 const Register tmp3 = r10;
5524 const Register tmp4 = r11;
5525
5526 BLOCK_COMMENT("Entry:");
5527 __ enter(); // required for proper stackwalking of RuntimeStub frame
5528
5529 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5530
5531 __ leave(); // required for proper stackwalking of RuntimeStub frame
5532 __ ret(0);
5533
5534 return start;
5535
5536 }
5537
5538 address generate_libmSin() {
5539 StubCodeMark mark(this, "StubRoutines", "libmSin");
5540
5541 address start = __ pc();
5542
5543 const XMMRegister x0 = xmm0;
5544 const XMMRegister x1 = xmm1;
5545 const XMMRegister x2 = xmm2;
5546 const XMMRegister x3 = xmm3;
5547
5548 const XMMRegister x4 = xmm4;
5549 const XMMRegister x5 = xmm5;
5550 const XMMRegister x6 = xmm6;
5551 const XMMRegister x7 = xmm7;
5552
5553 const Register tmp1 = r8;
5554 const Register tmp2 = r9;
5555 const Register tmp3 = r10;
5556 const Register tmp4 = r11;
5557
5558 BLOCK_COMMENT("Entry:");
5559 __ enter(); // required for proper stackwalking of RuntimeStub frame
5560
5561 #ifdef _WIN64
5562 __ push(rsi);
5563 __ push(rdi);
5564 #endif
5565 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5566
5567 #ifdef _WIN64
5568 __ pop(rdi);
5569 __ pop(rsi);
5570 #endif
5571
5572 __ leave(); // required for proper stackwalking of RuntimeStub frame
5573 __ ret(0);
5574
5575 return start;
5576
5577 }
5578
5579 address generate_libmCos() {
5580 StubCodeMark mark(this, "StubRoutines", "libmCos");
5581
5582 address start = __ pc();
5583
5584 const XMMRegister x0 = xmm0;
5585 const XMMRegister x1 = xmm1;
5586 const XMMRegister x2 = xmm2;
5587 const XMMRegister x3 = xmm3;
5588
5589 const XMMRegister x4 = xmm4;
5590 const XMMRegister x5 = xmm5;
5591 const XMMRegister x6 = xmm6;
5592 const XMMRegister x7 = xmm7;
5593
5594 const Register tmp1 = r8;
5595 const Register tmp2 = r9;
5596 const Register tmp3 = r10;
5597 const Register tmp4 = r11;
5598
5599 BLOCK_COMMENT("Entry:");
5600 __ enter(); // required for proper stackwalking of RuntimeStub frame
5601
5602 #ifdef _WIN64
5603 __ push(rsi);
5604 __ push(rdi);
5605 #endif
5606 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5607
5608 #ifdef _WIN64
5609 __ pop(rdi);
5610 __ pop(rsi);
5611 #endif
5612
5613 __ leave(); // required for proper stackwalking of RuntimeStub frame
5614 __ ret(0);
5615
5616 return start;
5617
5618 }
5619
5620 address generate_libmTan() {
5621 StubCodeMark mark(this, "StubRoutines", "libmTan");
5622
5623 address start = __ pc();
5624
5625 const XMMRegister x0 = xmm0;
5626 const XMMRegister x1 = xmm1;
5627 const XMMRegister x2 = xmm2;
5628 const XMMRegister x3 = xmm3;
5629
5630 const XMMRegister x4 = xmm4;
5631 const XMMRegister x5 = xmm5;
5632 const XMMRegister x6 = xmm6;
5633 const XMMRegister x7 = xmm7;
5634
5635 const Register tmp1 = r8;
5636 const Register tmp2 = r9;
5637 const Register tmp3 = r10;
5638 const Register tmp4 = r11;
5639
5640 BLOCK_COMMENT("Entry:");
5641 __ enter(); // required for proper stackwalking of RuntimeStub frame
5642
5643 #ifdef _WIN64
5644 __ push(rsi);
5645 __ push(rdi);
5646 #endif
5647 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5648
5649 #ifdef _WIN64
5650 __ pop(rdi);
5651 __ pop(rsi);
5652 #endif
5653
5654 __ leave(); // required for proper stackwalking of RuntimeStub frame
5655 __ ret(0);
5656
5657 return start;
5658
5659 }
5660
5661 #undef __
5662 #define __ masm->
5663
5664 // Continuation point for throwing of implicit exceptions that are
5665 // not handled in the current activation. Fabricates an exception
5666 // oop and initiates normal exception dispatching in this
5667 // frame. Since we need to preserve callee-saved values (currently
5668 // only for C2, but done for C1 as well) we need a callee-saved oop
5669 // map and therefore have to make these stubs into RuntimeStubs
5670 // rather than BufferBlobs. If the compiler needs all registers to
5671 // be preserved between the fault point and the exception handler
5672 // then it must assume responsibility for that in
5673 // AbstractCompiler::continuation_for_implicit_null_exception or
5674 // continuation_for_implicit_division_by_zero_exception. All other
5675 // implicit exceptions (e.g., NullPointerException or
5676 // AbstractMethodError on entry) are either at call sites or
5677 // otherwise assume that stack unwinding will be initiated, so
5678 // caller saved registers were assumed volatile in the compiler.
5679 address generate_throw_exception(const char* name,
5680 address runtime_entry,
5681 Register arg1 = noreg,
5682 Register arg2 = noreg) {
5683 // Information about frame layout at time of blocking runtime call.
5684 // Note that we only have to preserve callee-saved registers since
5685 // the compilers are responsible for supplying a continuation point
5686 // if they expect all registers to be preserved.
5687 enum layout {
5688 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
5689 rbp_off2,
5690 return_off,
5691 return_off2,
5692 framesize // inclusive of return address
5693 };
5694
5695 int insts_size = 512;
5696 int locs_size = 64;
5697
5698 CodeBuffer code(name, insts_size, locs_size);
5699 OopMapSet* oop_maps = new OopMapSet();
5700 MacroAssembler* masm = new MacroAssembler(&code);
5701
5702 address start = __ pc();
5703
5704 // This is an inlined and slightly modified version of call_VM
5705 // which has the ability to fetch the return PC out of
5706 // thread-local storage and also sets up last_Java_sp slightly
5707 // differently than the real call_VM
5708
5709 __ enter(); // required for proper stackwalking of RuntimeStub frame
5710
5711 assert(is_even(framesize/2), "sp not 16-byte aligned");
5712
5713 // return address and rbp are already in place
5714 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
5715
5716 int frame_complete = __ pc() - start;
5717
5718 // Set up last_Java_sp and last_Java_fp
5719 address the_pc = __ pc();
5720 __ set_last_Java_frame(rsp, rbp, the_pc);
5721 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
5722
5723 // Call runtime
5724 if (arg1 != noreg) {
5725 assert(arg2 != c_rarg1, "clobbered");
5726 __ movptr(c_rarg1, arg1);
5727 }
5728 if (arg2 != noreg) {
5729 __ movptr(c_rarg2, arg2);
5730 }
5731 __ movptr(c_rarg0, r15_thread);
5732 BLOCK_COMMENT("call runtime_entry");
5733 __ call(RuntimeAddress(runtime_entry));
5734
5735 // Generate oop map
5736 OopMap* map = new OopMap(framesize, 0);
5737
5738 oop_maps->add_gc_map(the_pc - start, map);
5739
5740 __ reset_last_Java_frame(true);
5741
5742 __ leave(); // required for proper stackwalking of RuntimeStub frame
5743
5744 // check for pending exceptions
5745 #ifdef ASSERT
5746 Label L;
5747 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
5748 (int32_t) NULL_WORD);
5749 __ jcc(Assembler::notEqual, L);
5750 __ should_not_reach_here();
5751 __ bind(L);
5752 #endif // ASSERT
5753 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5754
5755
5756 // codeBlob framesize is in words (not VMRegImpl::slot_size)
5757 RuntimeStub* stub =
5758 RuntimeStub::new_runtime_stub(name,
5759 &code,
5760 frame_complete,
5761 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
5762 oop_maps, false);
5763 return stub->entry_point();
5764 }
5765
5766 void create_control_words() {
5767 // Round to nearest, 53-bit mode, exceptions masked
5768 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
5769 // Round to zero, 53-bit mode, exception mased
5770 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
5771 // Round to nearest, 24-bit mode, exceptions masked
5772 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
5773 // Round to nearest, 64-bit mode, exceptions masked
5774 StubRoutines::_mxcsr_std = 0x1F80;
5775 // Note: the following two constants are 80-bit values
5776 // layout is critical for correct loading by FPU.
5777 // Bias for strict fp multiply/divide
5778 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
5779 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
5780 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
5781 // Un-Bias for strict fp multiply/divide
5782 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
5783 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
5784 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
5785 }
5786
5787 // Call here from the interpreter or compiled code to either load
5788 // multiple returned values from the value type instance being
5789 // returned to registers or to store returned values to a newly
5790 // allocated value type instance.
5791 address generate_return_value_stub(address destination, const char* name, bool has_res) {
5792 // We need to save all registers the calling convention may use so
5793 // the runtime calls read or update those registers. This needs to
5794 // be in sync with SharedRuntime::java_return_convention().
5795 enum layout {
5796 pad_off = frame::arg_reg_save_area_bytes/BytesPerInt, pad_off_2,
5797 rax_off, rax_off_2,
5798 j_rarg5_off, j_rarg5_2,
5799 j_rarg4_off, j_rarg4_2,
5800 j_rarg3_off, j_rarg3_2,
5801 j_rarg2_off, j_rarg2_2,
5802 j_rarg1_off, j_rarg1_2,
5803 j_rarg0_off, j_rarg0_2,
5804 j_farg0_off, j_farg0_2,
5805 j_farg1_off, j_farg1_2,
5806 j_farg2_off, j_farg2_2,
5807 j_farg3_off, j_farg3_2,
5808 j_farg4_off, j_farg4_2,
5809 j_farg5_off, j_farg5_2,
5810 j_farg6_off, j_farg6_2,
5811 j_farg7_off, j_farg7_2,
5812 rbp_off, rbp_off_2,
5813 return_off, return_off_2,
5814
5815 framesize
5816 };
5817
5818 CodeBuffer buffer(name, 1000, 512);
5819 MacroAssembler* masm = new MacroAssembler(&buffer);
5820
5821 int frame_size_in_bytes = align_up(framesize*BytesPerInt, 16);
5822 assert(frame_size_in_bytes == framesize*BytesPerInt, "misaligned");
5823 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
5824 int frame_size_in_words = frame_size_in_bytes / wordSize;
5825
5826 OopMapSet *oop_maps = new OopMapSet();
5827 OopMap* map = new OopMap(frame_size_in_slots, 0);
5828
5829 map->set_callee_saved(VMRegImpl::stack2reg(rax_off), rax->as_VMReg());
5830 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg5_off), j_rarg5->as_VMReg());
5831 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg4_off), j_rarg4->as_VMReg());
5832 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg3_off), j_rarg3->as_VMReg());
5833 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg2_off), j_rarg2->as_VMReg());
5834 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg1_off), j_rarg1->as_VMReg());
5835 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg0_off), j_rarg0->as_VMReg());
5836 map->set_callee_saved(VMRegImpl::stack2reg(j_farg0_off), j_farg0->as_VMReg());
5837 map->set_callee_saved(VMRegImpl::stack2reg(j_farg1_off), j_farg1->as_VMReg());
5838 map->set_callee_saved(VMRegImpl::stack2reg(j_farg2_off), j_farg2->as_VMReg());
5839 map->set_callee_saved(VMRegImpl::stack2reg(j_farg3_off), j_farg3->as_VMReg());
5840 map->set_callee_saved(VMRegImpl::stack2reg(j_farg4_off), j_farg4->as_VMReg());
5841 map->set_callee_saved(VMRegImpl::stack2reg(j_farg5_off), j_farg5->as_VMReg());
5842 map->set_callee_saved(VMRegImpl::stack2reg(j_farg6_off), j_farg6->as_VMReg());
5843 map->set_callee_saved(VMRegImpl::stack2reg(j_farg7_off), j_farg7->as_VMReg());
5844
5845 int start = __ offset();
5846
5847 __ subptr(rsp, frame_size_in_bytes - 8 /* return address*/);
5848
5849 __ movptr(Address(rsp, rbp_off * BytesPerInt), rbp);
5850 __ movdbl(Address(rsp, j_farg7_off * BytesPerInt), j_farg7);
5851 __ movdbl(Address(rsp, j_farg6_off * BytesPerInt), j_farg6);
5852 __ movdbl(Address(rsp, j_farg5_off * BytesPerInt), j_farg5);
5853 __ movdbl(Address(rsp, j_farg4_off * BytesPerInt), j_farg4);
5854 __ movdbl(Address(rsp, j_farg3_off * BytesPerInt), j_farg3);
5855 __ movdbl(Address(rsp, j_farg2_off * BytesPerInt), j_farg2);
5856 __ movdbl(Address(rsp, j_farg1_off * BytesPerInt), j_farg1);
5857 __ movdbl(Address(rsp, j_farg0_off * BytesPerInt), j_farg0);
5858
5859 __ movptr(Address(rsp, j_rarg0_off * BytesPerInt), j_rarg0);
5860 __ movptr(Address(rsp, j_rarg1_off * BytesPerInt), j_rarg1);
5861 __ movptr(Address(rsp, j_rarg2_off * BytesPerInt), j_rarg2);
5862 __ movptr(Address(rsp, j_rarg3_off * BytesPerInt), j_rarg3);
5863 __ movptr(Address(rsp, j_rarg4_off * BytesPerInt), j_rarg4);
5864 __ movptr(Address(rsp, j_rarg5_off * BytesPerInt), j_rarg5);
5865 __ movptr(Address(rsp, rax_off * BytesPerInt), rax);
5866
5867 int frame_complete = __ offset();
5868
5869 __ set_last_Java_frame(noreg, noreg, NULL);
5870
5871 __ mov(c_rarg0, r15_thread);
5872 __ mov(c_rarg1, rax);
5873
5874 __ call(RuntimeAddress(destination));
5875
5876 // Set an oopmap for the call site.
5877
5878 oop_maps->add_gc_map( __ offset() - start, map);
5879
5880 // clear last_Java_sp
5881 __ reset_last_Java_frame(false);
5882
5883 __ movptr(rbp, Address(rsp, rbp_off * BytesPerInt));
5884 __ movdbl(j_farg7, Address(rsp, j_farg7_off * BytesPerInt));
5885 __ movdbl(j_farg6, Address(rsp, j_farg6_off * BytesPerInt));
5886 __ movdbl(j_farg5, Address(rsp, j_farg5_off * BytesPerInt));
5887 __ movdbl(j_farg4, Address(rsp, j_farg4_off * BytesPerInt));
5888 __ movdbl(j_farg3, Address(rsp, j_farg3_off * BytesPerInt));
5889 __ movdbl(j_farg2, Address(rsp, j_farg2_off * BytesPerInt));
5890 __ movdbl(j_farg1, Address(rsp, j_farg1_off * BytesPerInt));
5891 __ movdbl(j_farg0, Address(rsp, j_farg0_off * BytesPerInt));
5892
5893 __ movptr(j_rarg0, Address(rsp, j_rarg0_off * BytesPerInt));
5894 __ movptr(j_rarg1, Address(rsp, j_rarg1_off * BytesPerInt));
5895 __ movptr(j_rarg2, Address(rsp, j_rarg2_off * BytesPerInt));
5896 __ movptr(j_rarg3, Address(rsp, j_rarg3_off * BytesPerInt));
5897 __ movptr(j_rarg4, Address(rsp, j_rarg4_off * BytesPerInt));
5898 __ movptr(j_rarg5, Address(rsp, j_rarg5_off * BytesPerInt));
5899 __ movptr(rax, Address(rsp, rax_off * BytesPerInt));
5900
5901 __ addptr(rsp, frame_size_in_bytes-8);
5902
5903 // check for pending exceptions
5904 Label pending;
5905 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
5906 __ jcc(Assembler::notEqual, pending);
5907
5908 if (has_res) {
5909 __ get_vm_result(rax, r15_thread);
5910 }
5911
5912 __ ret(0);
5913
5914 __ bind(pending);
5915
5916 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
5917 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5918
5919 // -------------
5920 // make sure all code is generated
5921 masm->flush();
5922
5923 // The caller may not know the register mapping of the fields of the returned value
5924 // object, so it won't generate a valid oopmap for the call site. Hence, we can't
5925 // do InterfaceSupport::verify_stack().
5926 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, false,
5927 /*can_verify_stack =*/false);
5928 return stub->entry_point();
5929 }
5930
5931 // Initialization
5932 void generate_initial() {
5933 // Generates all stubs and initializes the entry points
5934
5935 // This platform-specific settings are needed by generate_call_stub()
5936 create_control_words();
5937
5938 // entry points that exist in all platforms Note: This is code
5939 // that could be shared among different platforms - however the
5940 // benefit seems to be smaller than the disadvantage of having a
5941 // much more complicated generator structure. See also comment in
5942 // stubRoutines.hpp.
5943
5944 StubRoutines::_forward_exception_entry = generate_forward_exception();
5945
5946 // Generate these first because they are called from other stubs
5947 StubRoutines::_load_value_type_fields_in_regs = generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::load_value_type_fields_in_regs), "load_value_type_fields_in_regs", false);
5948 StubRoutines::_store_value_type_fields_to_buf = generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::store_value_type_fields_to_buf), "store_value_type_fields_to_buf", true);
5949
5950 StubRoutines::_call_stub_entry = generate_call_stub(StubRoutines::_call_stub_return_address);
5951
5952 // is referenced by megamorphic call
5953 StubRoutines::_catch_exception_entry = generate_catch_exception();
5954
5955 // atomic calls
5956 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
5957 StubRoutines::_atomic_xchg_long_entry = generate_atomic_xchg_long();
5958 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
5959 StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
5960 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
5961 StubRoutines::_atomic_add_entry = generate_atomic_add();
5962 StubRoutines::_atomic_add_long_entry = generate_atomic_add_long();
5963 StubRoutines::_fence_entry = generate_orderaccess_fence();
5964
5965 // platform dependent
5966 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
5967 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
5968
5969 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
5970
5971 // Build this early so it's available for the interpreter.
5972 StubRoutines::_throw_StackOverflowError_entry =
5973 generate_throw_exception("StackOverflowError throw_exception",
5974 CAST_FROM_FN_PTR(address,
5975 SharedRuntime::
5976 throw_StackOverflowError));
5977 StubRoutines::_throw_delayed_StackOverflowError_entry =
5978 generate_throw_exception("delayed StackOverflowError throw_exception",
5979 CAST_FROM_FN_PTR(address,
5980 SharedRuntime::
5981 throw_delayed_StackOverflowError));
5982 if (UseCRC32Intrinsics) {
5983 // set table address before stub generation which use it
5984 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
5985 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
5986 }
5987
5988 if (UseCRC32CIntrinsics) {
5989 bool supports_clmul = VM_Version::supports_clmul();
5990 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
5991 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
5992 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
5993 }
5994 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
5995 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
5996 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
5997 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
5998 StubRoutines::x86::_ONEHALF_adr = (address)StubRoutines::x86::_ONEHALF;
5999 StubRoutines::x86::_P_2_adr = (address)StubRoutines::x86::_P_2;
6000 StubRoutines::x86::_SC_4_adr = (address)StubRoutines::x86::_SC_4;
6001 StubRoutines::x86::_Ctable_adr = (address)StubRoutines::x86::_Ctable;
6002 StubRoutines::x86::_SC_2_adr = (address)StubRoutines::x86::_SC_2;
6003 StubRoutines::x86::_SC_3_adr = (address)StubRoutines::x86::_SC_3;
6004 StubRoutines::x86::_SC_1_adr = (address)StubRoutines::x86::_SC_1;
6005 StubRoutines::x86::_PI_INV_TABLE_adr = (address)StubRoutines::x86::_PI_INV_TABLE;
6006 StubRoutines::x86::_PI_4_adr = (address)StubRoutines::x86::_PI_4;
6007 StubRoutines::x86::_PI32INV_adr = (address)StubRoutines::x86::_PI32INV;
6008 StubRoutines::x86::_SIGN_MASK_adr = (address)StubRoutines::x86::_SIGN_MASK;
6009 StubRoutines::x86::_P_1_adr = (address)StubRoutines::x86::_P_1;
6010 StubRoutines::x86::_P_3_adr = (address)StubRoutines::x86::_P_3;
6011 StubRoutines::x86::_NEG_ZERO_adr = (address)StubRoutines::x86::_NEG_ZERO;
6012 }
6013 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
6014 StubRoutines::_dexp = generate_libmExp();
6015 }
6016 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
6017 StubRoutines::_dlog = generate_libmLog();
6018 }
6019 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
6020 StubRoutines::_dlog10 = generate_libmLog10();
6021 }
6022 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
6023 StubRoutines::_dpow = generate_libmPow();
6024 }
6025 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
6026 StubRoutines::_dsin = generate_libmSin();
6027 }
6028 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
6029 StubRoutines::_dcos = generate_libmCos();
6030 }
6031 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6032 StubRoutines::_dtan = generate_libmTan();
6033 }
6034 }
6035 }
6036
6037 void generate_all() {
6038 // Generates all stubs and initializes the entry points
6039
6040 // These entry points require SharedInfo::stack0 to be set up in
6041 // non-core builds and need to be relocatable, so they each
6042 // fabricate a RuntimeStub internally.
6043 StubRoutines::_throw_AbstractMethodError_entry =
6044 generate_throw_exception("AbstractMethodError throw_exception",
6045 CAST_FROM_FN_PTR(address,
6046 SharedRuntime::
6047 throw_AbstractMethodError));
6048
6049 StubRoutines::_throw_IncompatibleClassChangeError_entry =
6050 generate_throw_exception("IncompatibleClassChangeError throw_exception",
6051 CAST_FROM_FN_PTR(address,
6052 SharedRuntime::
6053 throw_IncompatibleClassChangeError));
6054
6055 StubRoutines::_throw_NullPointerException_at_call_entry =
6056 generate_throw_exception("NullPointerException at call throw_exception",
6057 CAST_FROM_FN_PTR(address,
6058 SharedRuntime::
6059 throw_NullPointerException_at_call));
6060
6061 // entry points that are platform specific
6062 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
6063 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
6064 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
6065 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
6066
6067 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6068 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
6069 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6070 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
6071 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6072 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
6073 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6074 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
6075 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
6076 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
6077 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
6078
6079 // support for verify_oop (must happen after universe_init)
6080 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
6081
6082 // arraycopy stubs used by compilers
6083 generate_arraycopy_stubs();
6084
6085 // don't bother generating these AES intrinsic stubs unless global flag is set
6086 if (UseAESIntrinsics) {
6087 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
6088 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
6089 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
6090 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
6091 if (VM_Version::supports_vaes() && VM_Version::supports_avx512vl() && VM_Version::supports_avx512dq() ) {
6092 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptVectorAESCrypt();
6093 } else {
6094 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
6095 }
6096 }
6097 if (UseAESCTRIntrinsics){
6098 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
6099 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
6100 }
6101
6102 if (UseSHA1Intrinsics) {
6103 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
6104 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
6105 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
6106 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
6107 }
6108 if (UseSHA256Intrinsics) {
6109 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
6110 char* dst = (char*)StubRoutines::x86::_k256_W;
6111 char* src = (char*)StubRoutines::x86::_k256;
6112 for (int ii = 0; ii < 16; ++ii) {
6113 memcpy(dst + 32 * ii, src + 16 * ii, 16);
6114 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
6115 }
6116 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
6117 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
6118 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
6119 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
6120 }
6121 if (UseSHA512Intrinsics) {
6122 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
6123 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
6124 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
6125 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
6126 }
6127
6128 // Generate GHASH intrinsics code
6129 if (UseGHASHIntrinsics) {
6130 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
6131 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
6132 if (VM_Version::supports_avx()) {
6133 StubRoutines::x86::_ghash_shuffmask_addr = ghash_shufflemask_addr();
6134 StubRoutines::x86::_ghash_poly_addr = ghash_polynomial_addr();
6135 StubRoutines::_ghash_processBlocks = generate_avx_ghash_processBlocks();
6136 } else {
6137 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
6138 }
6139 }
6140
6141 if (UseBASE64Intrinsics) {
6142 StubRoutines::x86::_and_mask = base64_and_mask_addr();
6143 StubRoutines::x86::_bswap_mask = base64_bswap_mask_addr();
6144 StubRoutines::x86::_base64_charset = base64_charset_addr();
6145 StubRoutines::x86::_url_charset = base64url_charset_addr();
6146 StubRoutines::x86::_gather_mask = base64_gather_mask_addr();
6147 StubRoutines::x86::_left_shift_mask = base64_left_shift_mask_addr();
6148 StubRoutines::x86::_right_shift_mask = base64_right_shift_mask_addr();
6149 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
6150 }
6151
6152 // Safefetch stubs.
6153 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
6154 &StubRoutines::_safefetch32_fault_pc,
6155 &StubRoutines::_safefetch32_continuation_pc);
6156 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
6157 &StubRoutines::_safefetchN_fault_pc,
6158 &StubRoutines::_safefetchN_continuation_pc);
6159
6160 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
6161 if (bs_nm != NULL) {
6162 StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
6163 }
6164 #ifdef COMPILER2
6165 if (UseMultiplyToLenIntrinsic) {
6166 StubRoutines::_multiplyToLen = generate_multiplyToLen();
6167 }
6168 if (UseSquareToLenIntrinsic) {
6169 StubRoutines::_squareToLen = generate_squareToLen();
6170 }
6171 if (UseMulAddIntrinsic) {
6172 StubRoutines::_mulAdd = generate_mulAdd();
6173 }
6174 #ifndef _WINDOWS
6175 if (UseMontgomeryMultiplyIntrinsic) {
6176 StubRoutines::_montgomeryMultiply
6177 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
6178 }
6179 if (UseMontgomerySquareIntrinsic) {
6180 StubRoutines::_montgomerySquare
6181 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
6182 }
6183 #endif // WINDOWS
6184 #endif // COMPILER2
6185
6186 if (UseVectorizedMismatchIntrinsic) {
6187 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
6188 }
6189 }
6190
6191 public:
6192 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
6193 if (all) {
6194 generate_all();
6195 } else {
6196 generate_initial();
6197 }
6198 }
6199 }; // end class declaration
6200
6201 void StubGenerator_generate(CodeBuffer* code, bool all) {
6202 StubGenerator g(code, all);
6203 }