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_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
340 __ movptr(c_rarg0, result);
341 Label is_long, is_float, is_double, exit;
342 __ movl(c_rarg1, result_type);
343 __ cmpl(c_rarg1, T_OBJECT);
344 __ jcc(Assembler::equal, is_long);
345 __ cmpl(c_rarg1, T_LONG);
346 __ jcc(Assembler::equal, is_long);
347 __ cmpl(c_rarg1, T_FLOAT);
348 __ jcc(Assembler::equal, is_float);
349 __ cmpl(c_rarg1, T_DOUBLE);
350 __ jcc(Assembler::equal, is_double);
351
352 // handle T_INT case
353 __ movl(Address(c_rarg0, 0), rax);
354
355 __ BIND(exit);
356
357 // pop parameters
358 __ lea(rsp, rsp_after_call);
359
360 #ifdef ASSERT
361 // verify that threads correspond
362 {
363 Label L1, L2, L3;
364 __ cmpptr(r15_thread, thread);
365 __ jcc(Assembler::equal, L1);
366 __ stop("StubRoutines::call_stub: r15_thread is corrupted");
367 __ bind(L1);
368 __ get_thread(rbx);
369 __ cmpptr(r15_thread, thread);
370 __ jcc(Assembler::equal, L2);
371 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
372 __ bind(L2);
373 __ cmpptr(r15_thread, rbx);
374 __ jcc(Assembler::equal, L3);
375 __ stop("StubRoutines::call_stub: threads must correspond");
376 __ bind(L3);
377 }
378 #endif
379
380 // restore regs belonging to calling function
381 #ifdef _WIN64
382 // emit the restores for xmm regs
383 if (VM_Version::supports_evex()) {
384 for (int i = xmm_save_first; i <= last_reg; i++) {
385 __ vinsertf32x4(as_XMMRegister(i), as_XMMRegister(i), xmm_save(i), 0);
386 }
387 } else {
388 for (int i = xmm_save_first; i <= last_reg; i++) {
389 __ movdqu(as_XMMRegister(i), xmm_save(i));
390 }
391 }
392 #endif
393 __ movptr(r15, r15_save);
394 __ movptr(r14, r14_save);
395 __ movptr(r13, r13_save);
396 __ movptr(r12, r12_save);
397 __ movptr(rbx, rbx_save);
398
399 #ifdef _WIN64
400 __ movptr(rdi, rdi_save);
401 __ movptr(rsi, rsi_save);
402 #else
403 __ ldmxcsr(mxcsr_save);
404 #endif
405
406 // restore rsp
407 __ addptr(rsp, -rsp_after_call_off * wordSize);
408
409 // return
410 __ vzeroupper();
411 __ pop(rbp);
412 __ ret(0);
413
414 // handle return types different from T_INT
415 __ BIND(is_long);
416 __ movq(Address(c_rarg0, 0), rax);
417 __ jmp(exit);
418
419 __ BIND(is_float);
420 __ movflt(Address(c_rarg0, 0), xmm0);
421 __ jmp(exit);
422
423 __ BIND(is_double);
424 __ movdbl(Address(c_rarg0, 0), xmm0);
425 __ jmp(exit);
426
427 return start;
428 }
429
430 // Return point for a Java call if there's an exception thrown in
431 // Java code. The exception is caught and transformed into a
432 // pending exception stored in JavaThread that can be tested from
433 // within the VM.
434 //
435 // Note: Usually the parameters are removed by the callee. In case
436 // of an exception crossing an activation frame boundary, that is
437 // not the case if the callee is compiled code => need to setup the
438 // rsp.
439 //
440 // rax: exception oop
441
442 address generate_catch_exception() {
443 StubCodeMark mark(this, "StubRoutines", "catch_exception");
444 address start = __ pc();
445
446 // same as in generate_call_stub():
447 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
448 const Address thread (rbp, thread_off * wordSize);
449
450 #ifdef ASSERT
451 // verify that threads correspond
452 {
453 Label L1, L2, L3;
454 __ cmpptr(r15_thread, thread);
455 __ jcc(Assembler::equal, L1);
456 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
457 __ bind(L1);
458 __ get_thread(rbx);
459 __ cmpptr(r15_thread, thread);
460 __ jcc(Assembler::equal, L2);
461 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
462 __ bind(L2);
463 __ cmpptr(r15_thread, rbx);
464 __ jcc(Assembler::equal, L3);
465 __ stop("StubRoutines::catch_exception: threads must correspond");
466 __ bind(L3);
467 }
468 #endif
469
470 // set pending exception
471 __ verify_oop(rax);
472
473 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
474 __ lea(rscratch1, ExternalAddress((address)__FILE__));
475 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
476 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
477
478 // complete return to VM
479 assert(StubRoutines::_call_stub_return_address != NULL,
480 "_call_stub_return_address must have been generated before");
481 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
482
483 return start;
484 }
485
486 // Continuation point for runtime calls returning with a pending
487 // exception. The pending exception check happened in the runtime
488 // or native call stub. The pending exception in Thread is
489 // converted into a Java-level exception.
490 //
491 // Contract with Java-level exception handlers:
492 // rax: exception
493 // rdx: throwing pc
494 //
495 // NOTE: At entry of this stub, exception-pc must be on stack !!
496
497 address generate_forward_exception() {
498 StubCodeMark mark(this, "StubRoutines", "forward exception");
499 address start = __ pc();
500
501 // Upon entry, the sp points to the return address returning into
502 // Java (interpreted or compiled) code; i.e., the return address
503 // becomes the throwing pc.
504 //
505 // Arguments pushed before the runtime call are still on the stack
506 // but the exception handler will reset the stack pointer ->
507 // ignore them. A potential result in registers can be ignored as
508 // well.
509
510 #ifdef ASSERT
511 // make sure this code is only executed if there is a pending exception
512 {
513 Label L;
514 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
515 __ jcc(Assembler::notEqual, L);
516 __ stop("StubRoutines::forward exception: no pending exception (1)");
517 __ bind(L);
518 }
519 #endif
520
521 // compute exception handler into rbx
522 __ movptr(c_rarg0, Address(rsp, 0));
523 BLOCK_COMMENT("call exception_handler_for_return_address");
524 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
525 SharedRuntime::exception_handler_for_return_address),
526 r15_thread, c_rarg0);
527 __ mov(rbx, rax);
528
529 // setup rax & rdx, remove return address & clear pending exception
530 __ pop(rdx);
531 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
532 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
533
534 #ifdef ASSERT
535 // make sure exception is set
536 {
537 Label L;
538 __ testptr(rax, rax);
539 __ jcc(Assembler::notEqual, L);
540 __ stop("StubRoutines::forward exception: no pending exception (2)");
541 __ bind(L);
542 }
543 #endif
544
545 // continue at exception handler (return address removed)
546 // rax: exception
547 // rbx: exception handler
548 // rdx: throwing pc
549 __ verify_oop(rax);
550 __ jmp(rbx);
551
552 return start;
553 }
554
555 // Implementation of jint atomic_xchg(jint add_value, volatile jint* dest)
556 // used by Atomic::xchg(volatile jint* dest, jint exchange_value)
557 //
558 // Arguments :
559 // c_rarg0: exchange_value
560 // c_rarg0: dest
561 //
562 // Result:
563 // *dest <- ex, return (orig *dest)
564 address generate_atomic_xchg() {
565 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
566 address start = __ pc();
567
568 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
569 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
570 __ ret(0);
571
572 return start;
573 }
574
575 // Implementation of intptr_t atomic_xchg(jlong add_value, volatile jlong* dest)
576 // used by Atomic::xchg(volatile jlong* dest, jlong exchange_value)
577 //
578 // Arguments :
579 // c_rarg0: exchange_value
580 // c_rarg1: dest
581 //
582 // Result:
583 // *dest <- ex, return (orig *dest)
584 address generate_atomic_xchg_long() {
585 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_long");
586 address start = __ pc();
587
588 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
589 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
590 __ ret(0);
591
592 return start;
593 }
594
595 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
596 // jint compare_value)
597 //
598 // Arguments :
599 // c_rarg0: exchange_value
600 // c_rarg1: dest
601 // c_rarg2: compare_value
602 //
603 // Result:
604 // if ( compare_value == *dest ) {
605 // *dest = exchange_value
606 // return compare_value;
607 // else
608 // return *dest;
609 address generate_atomic_cmpxchg() {
610 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
611 address start = __ pc();
612
613 __ movl(rax, c_rarg2);
614 __ lock();
615 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
616 __ ret(0);
617
618 return start;
619 }
620
621 // Support for int8_t atomic::atomic_cmpxchg(int8_t exchange_value, volatile int8_t* dest,
622 // int8_t compare_value)
623 //
624 // Arguments :
625 // c_rarg0: exchange_value
626 // c_rarg1: dest
627 // c_rarg2: compare_value
628 //
629 // Result:
630 // if ( compare_value == *dest ) {
631 // *dest = exchange_value
632 // return compare_value;
633 // else
634 // return *dest;
635 address generate_atomic_cmpxchg_byte() {
636 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_byte");
637 address start = __ pc();
638
639 __ movsbq(rax, c_rarg2);
640 __ lock();
641 __ cmpxchgb(c_rarg0, Address(c_rarg1, 0));
642 __ ret(0);
643
644 return start;
645 }
646
647 // Support for int64_t atomic::atomic_cmpxchg(int64_t exchange_value,
648 // volatile int64_t* dest,
649 // int64_t compare_value)
650 // Arguments :
651 // c_rarg0: exchange_value
652 // c_rarg1: dest
653 // c_rarg2: compare_value
654 //
655 // Result:
656 // if ( compare_value == *dest ) {
657 // *dest = exchange_value
658 // return compare_value;
659 // else
660 // return *dest;
661 address generate_atomic_cmpxchg_long() {
662 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
663 address start = __ pc();
664
665 __ movq(rax, c_rarg2);
666 __ lock();
667 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
668 __ ret(0);
669
670 return start;
671 }
672
673 // Implementation of jint atomic_add(jint add_value, volatile jint* dest)
674 // used by Atomic::add(volatile jint* dest, jint add_value)
675 //
676 // Arguments :
677 // c_rarg0: add_value
678 // c_rarg1: dest
679 //
680 // Result:
681 // *dest += add_value
682 // return *dest;
683 address generate_atomic_add() {
684 StubCodeMark mark(this, "StubRoutines", "atomic_add");
685 address start = __ pc();
686
687 __ movl(rax, c_rarg0);
688 __ lock();
689 __ xaddl(Address(c_rarg1, 0), c_rarg0);
690 __ addl(rax, c_rarg0);
691 __ ret(0);
692
693 return start;
694 }
695
696 // Implementation of intptr_t atomic_add(intptr_t add_value, volatile intptr_t* dest)
697 // used by Atomic::add(volatile intptr_t* dest, intptr_t add_value)
698 //
699 // Arguments :
700 // c_rarg0: add_value
701 // c_rarg1: dest
702 //
703 // Result:
704 // *dest += add_value
705 // return *dest;
706 address generate_atomic_add_long() {
707 StubCodeMark mark(this, "StubRoutines", "atomic_add_long");
708 address start = __ pc();
709
710 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
711 __ lock();
712 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
713 __ addptr(rax, c_rarg0);
714 __ ret(0);
715
716 return start;
717 }
718
719 // Support for intptr_t OrderAccess::fence()
720 //
721 // Arguments :
722 //
723 // Result:
724 address generate_orderaccess_fence() {
725 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
726 address start = __ pc();
727 __ membar(Assembler::StoreLoad);
728 __ ret(0);
729
730 return start;
731 }
732
733 // Support for intptr_t get_previous_fp()
734 //
735 // This routine is used to find the previous frame pointer for the
736 // caller (current_frame_guess). This is used as part of debugging
737 // ps() is seemingly lost trying to find frames.
738 // This code assumes that caller current_frame_guess) has a frame.
739 address generate_get_previous_fp() {
740 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
741 const Address old_fp(rbp, 0);
742 const Address older_fp(rax, 0);
743 address start = __ pc();
744
745 __ enter();
746 __ movptr(rax, old_fp); // callers fp
747 __ movptr(rax, older_fp); // the frame for ps()
748 __ pop(rbp);
749 __ ret(0);
750
751 return start;
752 }
753
754 // Support for intptr_t get_previous_sp()
755 //
756 // This routine is used to find the previous stack pointer for the
757 // caller.
758 address generate_get_previous_sp() {
759 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
760 address start = __ pc();
761
762 __ movptr(rax, rsp);
763 __ addptr(rax, 8); // return address is at the top of the stack.
764 __ ret(0);
765
766 return start;
767 }
768
769 //----------------------------------------------------------------------------------------------------
770 // Support for void verify_mxcsr()
771 //
772 // This routine is used with -Xcheck:jni to verify that native
773 // JNI code does not return to Java code without restoring the
774 // MXCSR register to our expected state.
775
776 address generate_verify_mxcsr() {
777 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
778 address start = __ pc();
779
780 const Address mxcsr_save(rsp, 0);
781
782 if (CheckJNICalls) {
783 Label ok_ret;
784 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
785 __ push(rax);
786 __ subptr(rsp, wordSize); // allocate a temp location
787 __ stmxcsr(mxcsr_save);
788 __ movl(rax, mxcsr_save);
789 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
790 __ cmp32(rax, mxcsr_std);
791 __ jcc(Assembler::equal, ok_ret);
792
793 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
794
795 __ ldmxcsr(mxcsr_std);
796
797 __ bind(ok_ret);
798 __ addptr(rsp, wordSize);
799 __ pop(rax);
800 }
801
802 __ ret(0);
803
804 return start;
805 }
806
807 address generate_f2i_fixup() {
808 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
809 Address inout(rsp, 5 * wordSize); // return address + 4 saves
810
811 address start = __ pc();
812
813 Label L;
814
815 __ push(rax);
816 __ push(c_rarg3);
817 __ push(c_rarg2);
818 __ push(c_rarg1);
819
820 __ movl(rax, 0x7f800000);
821 __ xorl(c_rarg3, c_rarg3);
822 __ movl(c_rarg2, inout);
823 __ movl(c_rarg1, c_rarg2);
824 __ andl(c_rarg1, 0x7fffffff);
825 __ cmpl(rax, c_rarg1); // NaN? -> 0
826 __ jcc(Assembler::negative, L);
827 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
828 __ movl(c_rarg3, 0x80000000);
829 __ movl(rax, 0x7fffffff);
830 __ cmovl(Assembler::positive, c_rarg3, rax);
831
832 __ bind(L);
833 __ movptr(inout, c_rarg3);
834
835 __ pop(c_rarg1);
836 __ pop(c_rarg2);
837 __ pop(c_rarg3);
838 __ pop(rax);
839
840 __ ret(0);
841
842 return start;
843 }
844
845 address generate_f2l_fixup() {
846 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
847 Address inout(rsp, 5 * wordSize); // return address + 4 saves
848 address start = __ pc();
849
850 Label L;
851
852 __ push(rax);
853 __ push(c_rarg3);
854 __ push(c_rarg2);
855 __ push(c_rarg1);
856
857 __ movl(rax, 0x7f800000);
858 __ xorl(c_rarg3, c_rarg3);
859 __ movl(c_rarg2, inout);
860 __ movl(c_rarg1, c_rarg2);
861 __ andl(c_rarg1, 0x7fffffff);
862 __ cmpl(rax, c_rarg1); // NaN? -> 0
863 __ jcc(Assembler::negative, L);
864 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
865 __ mov64(c_rarg3, 0x8000000000000000);
866 __ mov64(rax, 0x7fffffffffffffff);
867 __ cmov(Assembler::positive, c_rarg3, rax);
868
869 __ bind(L);
870 __ movptr(inout, c_rarg3);
871
872 __ pop(c_rarg1);
873 __ pop(c_rarg2);
874 __ pop(c_rarg3);
875 __ pop(rax);
876
877 __ ret(0);
878
879 return start;
880 }
881
882 address generate_d2i_fixup() {
883 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
884 Address inout(rsp, 6 * wordSize); // return address + 5 saves
885
886 address start = __ pc();
887
888 Label L;
889
890 __ push(rax);
891 __ push(c_rarg3);
892 __ push(c_rarg2);
893 __ push(c_rarg1);
894 __ push(c_rarg0);
895
896 __ movl(rax, 0x7ff00000);
897 __ movq(c_rarg2, inout);
898 __ movl(c_rarg3, c_rarg2);
899 __ mov(c_rarg1, c_rarg2);
900 __ mov(c_rarg0, c_rarg2);
901 __ negl(c_rarg3);
902 __ shrptr(c_rarg1, 0x20);
903 __ orl(c_rarg3, c_rarg2);
904 __ andl(c_rarg1, 0x7fffffff);
905 __ xorl(c_rarg2, c_rarg2);
906 __ shrl(c_rarg3, 0x1f);
907 __ orl(c_rarg1, c_rarg3);
908 __ cmpl(rax, c_rarg1);
909 __ jcc(Assembler::negative, L); // NaN -> 0
910 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
911 __ movl(c_rarg2, 0x80000000);
912 __ movl(rax, 0x7fffffff);
913 __ cmov(Assembler::positive, c_rarg2, rax);
914
915 __ bind(L);
916 __ movptr(inout, c_rarg2);
917
918 __ pop(c_rarg0);
919 __ pop(c_rarg1);
920 __ pop(c_rarg2);
921 __ pop(c_rarg3);
922 __ pop(rax);
923
924 __ ret(0);
925
926 return start;
927 }
928
929 address generate_d2l_fixup() {
930 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
931 Address inout(rsp, 6 * wordSize); // return address + 5 saves
932
933 address start = __ pc();
934
935 Label L;
936
937 __ push(rax);
938 __ push(c_rarg3);
939 __ push(c_rarg2);
940 __ push(c_rarg1);
941 __ push(c_rarg0);
942
943 __ movl(rax, 0x7ff00000);
944 __ movq(c_rarg2, inout);
945 __ movl(c_rarg3, c_rarg2);
946 __ mov(c_rarg1, c_rarg2);
947 __ mov(c_rarg0, c_rarg2);
948 __ negl(c_rarg3);
949 __ shrptr(c_rarg1, 0x20);
950 __ orl(c_rarg3, c_rarg2);
951 __ andl(c_rarg1, 0x7fffffff);
952 __ xorl(c_rarg2, c_rarg2);
953 __ shrl(c_rarg3, 0x1f);
954 __ orl(c_rarg1, c_rarg3);
955 __ cmpl(rax, c_rarg1);
956 __ jcc(Assembler::negative, L); // NaN -> 0
957 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
958 __ mov64(c_rarg2, 0x8000000000000000);
959 __ mov64(rax, 0x7fffffffffffffff);
960 __ cmovq(Assembler::positive, c_rarg2, rax);
961
962 __ bind(L);
963 __ movq(inout, c_rarg2);
964
965 __ pop(c_rarg0);
966 __ pop(c_rarg1);
967 __ pop(c_rarg2);
968 __ pop(c_rarg3);
969 __ pop(rax);
970
971 __ ret(0);
972
973 return start;
974 }
975
976 address generate_iota_indices(const char *stub_name) {
977 __ align(CodeEntryAlignment);
978 StubCodeMark mark(this, "StubRoutines", stub_name);
979 address start = __ pc();
980 __ emit_data64(0x0706050403020100, relocInfo::none);
981 __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
982 __ emit_data64(0x1716151413121110, relocInfo::none);
983 __ emit_data64(0x1F1E1D1C1B1A1918, relocInfo::none);
984 __ emit_data64(0x2726252423222120, relocInfo::none);
985 __ emit_data64(0x2F2E2D2C2B2A2928, relocInfo::none);
986 __ emit_data64(0x3736353433323130, relocInfo::none);
987 __ emit_data64(0x3F3E3D3C3B3A3938, relocInfo::none);
988 return start;
989 }
990
991 address generate_fp_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
999 return start;
1000 }
1001
1002 address generate_vector_mask(const char *stub_name, int64_t mask) {
1003 __ align(CodeEntryAlignment);
1004 StubCodeMark mark(this, "StubRoutines", stub_name);
1005 address start = __ pc();
1006
1007 __ emit_data64(mask, relocInfo::none);
1008 __ emit_data64(mask, relocInfo::none);
1009 __ emit_data64(mask, relocInfo::none);
1010 __ emit_data64(mask, relocInfo::none);
1011 __ emit_data64(mask, relocInfo::none);
1012 __ emit_data64(mask, relocInfo::none);
1013 __ emit_data64(mask, relocInfo::none);
1014 __ emit_data64(mask, relocInfo::none);
1015
1016 return start;
1017 }
1018
1019 address generate_vector_byte_perm_mask(const char *stub_name) {
1020 __ align(CodeEntryAlignment);
1021 StubCodeMark mark(this, "StubRoutines", stub_name);
1022 address start = __ pc();
1023
1024 __ emit_data64(0x0000000000000001, relocInfo::none);
1025 __ emit_data64(0x0000000000000003, relocInfo::none);
1026 __ emit_data64(0x0000000000000005, relocInfo::none);
1027 __ emit_data64(0x0000000000000007, relocInfo::none);
1028 __ emit_data64(0x0000000000000000, relocInfo::none);
1029 __ emit_data64(0x0000000000000002, relocInfo::none);
1030 __ emit_data64(0x0000000000000004, relocInfo::none);
1031 __ emit_data64(0x0000000000000006, relocInfo::none);
1032
1033 return start;
1034 }
1035
1036 address generate_vector_fp_mask(const char *stub_name, int64_t mask) {
1037 __ align(CodeEntryAlignment);
1038 StubCodeMark mark(this, "StubRoutines", stub_name);
1039 address start = __ pc();
1040
1041 __ emit_data64(mask, relocInfo::none);
1042 __ emit_data64(mask, relocInfo::none);
1043 __ emit_data64(mask, relocInfo::none);
1044 __ emit_data64(mask, relocInfo::none);
1045 __ emit_data64(mask, relocInfo::none);
1046 __ emit_data64(mask, relocInfo::none);
1047 __ emit_data64(mask, relocInfo::none);
1048 __ emit_data64(mask, relocInfo::none);
1049
1050 return start;
1051 }
1052
1053 address generate_vector_custom_i32(const char *stub_name, Assembler::AvxVectorLen len,
1054 int32_t val0, int32_t val1, int32_t val2, int32_t val3,
1055 int32_t val4 = 0, int32_t val5 = 0, int32_t val6 = 0, int32_t val7 = 0,
1056 int32_t val8 = 0, int32_t val9 = 0, int32_t val10 = 0, int32_t val11 = 0,
1057 int32_t val12 = 0, int32_t val13 = 0, int32_t val14 = 0, int32_t val15 = 0) {
1058 __ align(CodeEntryAlignment);
1059 StubCodeMark mark(this, "StubRoutines", stub_name);
1060 address start = __ pc();
1061
1062 assert(len != Assembler::AVX_NoVec, "vector len must be specified");
1063 __ emit_data(val0, relocInfo::none, 0);
1064 __ emit_data(val1, relocInfo::none, 0);
1065 __ emit_data(val2, relocInfo::none, 0);
1066 __ emit_data(val3, relocInfo::none, 0);
1067 if (len >= Assembler::AVX_256bit) {
1068 __ emit_data(val4, relocInfo::none, 0);
1069 __ emit_data(val5, relocInfo::none, 0);
1070 __ emit_data(val6, relocInfo::none, 0);
1071 __ emit_data(val7, relocInfo::none, 0);
1072 if (len >= Assembler::AVX_512bit) {
1073 __ emit_data(val8, relocInfo::none, 0);
1074 __ emit_data(val9, relocInfo::none, 0);
1075 __ emit_data(val10, relocInfo::none, 0);
1076 __ emit_data(val11, relocInfo::none, 0);
1077 __ emit_data(val12, relocInfo::none, 0);
1078 __ emit_data(val13, relocInfo::none, 0);
1079 __ emit_data(val14, relocInfo::none, 0);
1080 __ emit_data(val15, relocInfo::none, 0);
1081 }
1082 }
1083
1084 return start;
1085 }
1086
1087 // Non-destructive plausibility checks for oops
1088 //
1089 // Arguments:
1090 // all args on stack!
1091 //
1092 // Stack after saving c_rarg3:
1093 // [tos + 0]: saved c_rarg3
1094 // [tos + 1]: saved c_rarg2
1095 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1096 // [tos + 3]: saved flags
1097 // [tos + 4]: return address
1098 // * [tos + 5]: error message (char*)
1099 // * [tos + 6]: object to verify (oop)
1100 // * [tos + 7]: saved rax - saved by caller and bashed
1101 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1102 // * = popped on exit
1103 address generate_verify_oop() {
1104 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1105 address start = __ pc();
1106
1107 Label exit, error;
1108
1109 __ pushf();
1110 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1111
1112 __ push(r12);
1113
1114 // save c_rarg2 and c_rarg3
1115 __ push(c_rarg2);
1116 __ push(c_rarg3);
1117
1118 enum {
1119 // After previous pushes.
1120 oop_to_verify = 6 * wordSize,
1121 saved_rax = 7 * wordSize,
1122 saved_r10 = 8 * wordSize,
1123
1124 // Before the call to MacroAssembler::debug(), see below.
1125 return_addr = 16 * wordSize,
1126 error_msg = 17 * wordSize
1127 };
1128
1129 // get object
1130 __ movptr(rax, Address(rsp, oop_to_verify));
1131
1132 // make sure object is 'reasonable'
1133 __ testptr(rax, rax);
1134 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1135
1136 #if INCLUDE_ZGC
1137 if (UseZGC) {
1138 // Check if metadata bits indicate a bad oop
1139 __ testptr(rax, Address(r15_thread, ZThreadLocalData::address_bad_mask_offset()));
1140 __ jcc(Assembler::notZero, error);
1141 }
1142 #endif
1143
1144 // Check if the oop is in the right area of memory
1145 __ movptr(c_rarg2, rax);
1146 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1147 __ andptr(c_rarg2, c_rarg3);
1148 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1149 __ cmpptr(c_rarg2, c_rarg3);
1150 __ jcc(Assembler::notZero, error);
1151
1152 // make sure klass is 'reasonable', which is not zero.
1153 __ load_klass(rax, rax, rscratch1); // get klass
1154 __ testptr(rax, rax);
1155 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1156
1157 // return if everything seems ok
1158 __ bind(exit);
1159 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1160 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1161 __ pop(c_rarg3); // restore c_rarg3
1162 __ pop(c_rarg2); // restore c_rarg2
1163 __ pop(r12); // restore r12
1164 __ popf(); // restore flags
1165 __ ret(4 * wordSize); // pop caller saved stuff
1166
1167 // handle errors
1168 __ bind(error);
1169 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1170 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1171 __ pop(c_rarg3); // get saved c_rarg3 back
1172 __ pop(c_rarg2); // get saved c_rarg2 back
1173 __ pop(r12); // get saved r12 back
1174 __ popf(); // get saved flags off stack --
1175 // will be ignored
1176
1177 __ pusha(); // push registers
1178 // (rip is already
1179 // already pushed)
1180 // debug(char* msg, int64_t pc, int64_t regs[])
1181 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1182 // pushed all the registers, so now the stack looks like:
1183 // [tos + 0] 16 saved registers
1184 // [tos + 16] return address
1185 // * [tos + 17] error message (char*)
1186 // * [tos + 18] object to verify (oop)
1187 // * [tos + 19] saved rax - saved by caller and bashed
1188 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1189 // * = popped on exit
1190
1191 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1192 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1193 __ movq(c_rarg2, rsp); // pass address of regs on stack
1194 __ mov(r12, rsp); // remember rsp
1195 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1196 __ andptr(rsp, -16); // align stack as required by ABI
1197 BLOCK_COMMENT("call MacroAssembler::debug");
1198 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1199 __ hlt();
1200 return start;
1201 }
1202
1203 //
1204 // Verify that a register contains clean 32-bits positive value
1205 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1206 //
1207 // Input:
1208 // Rint - 32-bits value
1209 // Rtmp - scratch
1210 //
1211 void assert_clean_int(Register Rint, Register Rtmp) {
1212 #ifdef ASSERT
1213 Label L;
1214 assert_different_registers(Rtmp, Rint);
1215 __ movslq(Rtmp, Rint);
1216 __ cmpq(Rtmp, Rint);
1217 __ jcc(Assembler::equal, L);
1218 __ stop("high 32-bits of int value are not 0");
1219 __ bind(L);
1220 #endif
1221 }
1222
1223 // Generate overlap test for array copy stubs
1224 //
1225 // Input:
1226 // c_rarg0 - from
1227 // c_rarg1 - to
1228 // c_rarg2 - element count
1229 //
1230 // Output:
1231 // rax - &from[element count - 1]
1232 //
1233 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1234 assert(no_overlap_target != NULL, "must be generated");
1235 array_overlap_test(no_overlap_target, NULL, sf);
1236 }
1237 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1238 array_overlap_test(NULL, &L_no_overlap, sf);
1239 }
1240 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1241 const Register from = c_rarg0;
1242 const Register to = c_rarg1;
1243 const Register count = c_rarg2;
1244 const Register end_from = rax;
1245
1246 __ cmpptr(to, from);
1247 __ lea(end_from, Address(from, count, sf, 0));
1248 if (NOLp == NULL) {
1249 ExternalAddress no_overlap(no_overlap_target);
1250 __ jump_cc(Assembler::belowEqual, no_overlap);
1251 __ cmpptr(to, end_from);
1252 __ jump_cc(Assembler::aboveEqual, no_overlap);
1253 } else {
1254 __ jcc(Assembler::belowEqual, (*NOLp));
1255 __ cmpptr(to, end_from);
1256 __ jcc(Assembler::aboveEqual, (*NOLp));
1257 }
1258 }
1259
1260 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1261 //
1262 // Outputs:
1263 // rdi - rcx
1264 // rsi - rdx
1265 // rdx - r8
1266 // rcx - r9
1267 //
1268 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1269 // are non-volatile. r9 and r10 should not be used by the caller.
1270 //
1271 DEBUG_ONLY(bool regs_in_thread;)
1272
1273 void setup_arg_regs(int nargs = 3) {
1274 const Register saved_rdi = r9;
1275 const Register saved_rsi = r10;
1276 assert(nargs == 3 || nargs == 4, "else fix");
1277 #ifdef _WIN64
1278 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1279 "unexpected argument registers");
1280 if (nargs >= 4)
1281 __ mov(rax, r9); // r9 is also saved_rdi
1282 __ movptr(saved_rdi, rdi);
1283 __ movptr(saved_rsi, rsi);
1284 __ mov(rdi, rcx); // c_rarg0
1285 __ mov(rsi, rdx); // c_rarg1
1286 __ mov(rdx, r8); // c_rarg2
1287 if (nargs >= 4)
1288 __ mov(rcx, rax); // c_rarg3 (via rax)
1289 #else
1290 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1291 "unexpected argument registers");
1292 #endif
1293 DEBUG_ONLY(regs_in_thread = false;)
1294 }
1295
1296 void restore_arg_regs() {
1297 assert(!regs_in_thread, "wrong call to restore_arg_regs");
1298 const Register saved_rdi = r9;
1299 const Register saved_rsi = r10;
1300 #ifdef _WIN64
1301 __ movptr(rdi, saved_rdi);
1302 __ movptr(rsi, saved_rsi);
1303 #endif
1304 }
1305
1306 // This is used in places where r10 is a scratch register, and can
1307 // be adapted if r9 is needed also.
1308 void setup_arg_regs_using_thread() {
1309 const Register saved_r15 = r9;
1310 #ifdef _WIN64
1311 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1312 __ get_thread(r15_thread);
1313 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1314 "unexpected argument registers");
1315 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1316 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1317
1318 __ mov(rdi, rcx); // c_rarg0
1319 __ mov(rsi, rdx); // c_rarg1
1320 __ mov(rdx, r8); // c_rarg2
1321 #else
1322 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1323 "unexpected argument registers");
1324 #endif
1325 DEBUG_ONLY(regs_in_thread = true;)
1326 }
1327
1328 void restore_arg_regs_using_thread() {
1329 assert(regs_in_thread, "wrong call to restore_arg_regs");
1330 const Register saved_r15 = r9;
1331 #ifdef _WIN64
1332 __ get_thread(r15_thread);
1333 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1334 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1335 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1336 #endif
1337 }
1338
1339 // Copy big chunks forward
1340 //
1341 // Inputs:
1342 // end_from - source arrays end address
1343 // end_to - destination array end address
1344 // qword_count - 64-bits element count, negative
1345 // to - scratch
1346 // L_copy_bytes - entry label
1347 // L_copy_8_bytes - exit label
1348 //
1349 void copy_bytes_forward(Register end_from, Register end_to,
1350 Register qword_count, Register to,
1351 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1352 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1353 Label L_loop;
1354 __ align(OptoLoopAlignment);
1355 if (UseUnalignedLoadStores) {
1356 Label L_end;
1357 // Copy 64-bytes per iteration
1358 if (UseAVX > 2) {
1359 Label L_loop_avx512, L_loop_avx2, L_32_byte_head, L_above_threshold, L_below_threshold;
1360
1361 __ BIND(L_copy_bytes);
1362 __ cmpptr(qword_count, (-1 * AVX3Threshold / 8));
1363 __ jccb(Assembler::less, L_above_threshold);
1364 __ jmpb(L_below_threshold);
1365
1366 __ bind(L_loop_avx512);
1367 __ evmovdqul(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1368 __ evmovdqul(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1369 __ bind(L_above_threshold);
1370 __ addptr(qword_count, 8);
1371 __ jcc(Assembler::lessEqual, L_loop_avx512);
1372 __ jmpb(L_32_byte_head);
1373
1374 __ bind(L_loop_avx2);
1375 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1376 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1377 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1378 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1379 __ bind(L_below_threshold);
1380 __ addptr(qword_count, 8);
1381 __ jcc(Assembler::lessEqual, L_loop_avx2);
1382
1383 __ bind(L_32_byte_head);
1384 __ subptr(qword_count, 4); // sub(8) and add(4)
1385 __ jccb(Assembler::greater, L_end);
1386 } else {
1387 __ BIND(L_loop);
1388 if (UseAVX == 2) {
1389 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1390 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1391 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1392 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1393 } else {
1394 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1395 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1396 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1397 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1398 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1399 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1400 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1401 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1402 }
1403
1404 __ BIND(L_copy_bytes);
1405 __ addptr(qword_count, 8);
1406 __ jcc(Assembler::lessEqual, L_loop);
1407 __ subptr(qword_count, 4); // sub(8) and add(4)
1408 __ jccb(Assembler::greater, L_end);
1409 }
1410 // Copy trailing 32 bytes
1411 if (UseAVX >= 2) {
1412 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1413 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1414 } else {
1415 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1416 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1417 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1418 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1419 }
1420 __ addptr(qword_count, 4);
1421 __ BIND(L_end);
1422 if (UseAVX >= 2) {
1423 // clean upper bits of YMM registers
1424 __ vpxor(xmm0, xmm0);
1425 __ vpxor(xmm1, xmm1);
1426 }
1427 } else {
1428 // Copy 32-bytes per iteration
1429 __ BIND(L_loop);
1430 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1431 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1432 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1433 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1434 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1435 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1436 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1437 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1438
1439 __ BIND(L_copy_bytes);
1440 __ addptr(qword_count, 4);
1441 __ jcc(Assembler::lessEqual, L_loop);
1442 }
1443 __ subptr(qword_count, 4);
1444 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1445 }
1446
1447 // Copy big chunks backward
1448 //
1449 // Inputs:
1450 // from - source arrays address
1451 // dest - destination array address
1452 // qword_count - 64-bits element count
1453 // to - scratch
1454 // L_copy_bytes - entry label
1455 // L_copy_8_bytes - exit label
1456 //
1457 void copy_bytes_backward(Register from, Register dest,
1458 Register qword_count, Register to,
1459 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1460 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1461 Label L_loop;
1462 __ align(OptoLoopAlignment);
1463 if (UseUnalignedLoadStores) {
1464 Label L_end;
1465 // Copy 64-bytes per iteration
1466 if (UseAVX > 2) {
1467 Label L_loop_avx512, L_loop_avx2, L_32_byte_head, L_above_threshold, L_below_threshold;
1468
1469 __ BIND(L_copy_bytes);
1470 __ cmpptr(qword_count, (AVX3Threshold / 8));
1471 __ jccb(Assembler::greater, L_above_threshold);
1472 __ jmpb(L_below_threshold);
1473
1474 __ BIND(L_loop_avx512);
1475 __ evmovdqul(xmm0, Address(from, qword_count, Address::times_8, 0), Assembler::AVX_512bit);
1476 __ evmovdqul(Address(dest, qword_count, Address::times_8, 0), xmm0, Assembler::AVX_512bit);
1477 __ bind(L_above_threshold);
1478 __ subptr(qword_count, 8);
1479 __ jcc(Assembler::greaterEqual, L_loop_avx512);
1480 __ jmpb(L_32_byte_head);
1481
1482 __ bind(L_loop_avx2);
1483 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1484 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1485 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1486 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1487 __ bind(L_below_threshold);
1488 __ subptr(qword_count, 8);
1489 __ jcc(Assembler::greaterEqual, L_loop_avx2);
1490
1491 __ bind(L_32_byte_head);
1492 __ addptr(qword_count, 4); // add(8) and sub(4)
1493 __ jccb(Assembler::less, L_end);
1494 } else {
1495 __ BIND(L_loop);
1496 if (UseAVX == 2) {
1497 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1498 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1499 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1500 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1501 } else {
1502 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1503 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1504 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1505 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1506 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1507 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1508 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1509 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1510 }
1511
1512 __ BIND(L_copy_bytes);
1513 __ subptr(qword_count, 8);
1514 __ jcc(Assembler::greaterEqual, L_loop);
1515
1516 __ addptr(qword_count, 4); // add(8) and sub(4)
1517 __ jccb(Assembler::less, L_end);
1518 }
1519 // Copy trailing 32 bytes
1520 if (UseAVX >= 2) {
1521 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1522 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1523 } else {
1524 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1525 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1526 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1527 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1528 }
1529 __ subptr(qword_count, 4);
1530 __ BIND(L_end);
1531 if (UseAVX >= 2) {
1532 // clean upper bits of YMM registers
1533 __ vpxor(xmm0, xmm0);
1534 __ vpxor(xmm1, xmm1);
1535 }
1536 } else {
1537 // Copy 32-bytes per iteration
1538 __ BIND(L_loop);
1539 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1540 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1541 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1542 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1543 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1544 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1545 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1546 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1547
1548 __ BIND(L_copy_bytes);
1549 __ subptr(qword_count, 4);
1550 __ jcc(Assembler::greaterEqual, L_loop);
1551 }
1552 __ addptr(qword_count, 4);
1553 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1554 }
1555
1556 // Arguments:
1557 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1558 // ignored
1559 // name - stub name string
1560 //
1561 // Inputs:
1562 // c_rarg0 - source array address
1563 // c_rarg1 - destination array address
1564 // c_rarg2 - element count, treated as ssize_t, can be zero
1565 //
1566 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1567 // we let the hardware handle it. The one to eight bytes within words,
1568 // dwords or qwords that span cache line boundaries will still be loaded
1569 // and stored atomically.
1570 //
1571 // Side Effects:
1572 // disjoint_byte_copy_entry is set to the no-overlap entry point
1573 // used by generate_conjoint_byte_copy().
1574 //
1575 address generate_disjoint_byte_copy(bool aligned, 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 Label L_copy_byte, L_exit;
1582 const Register from = rdi; // source array address
1583 const Register to = rsi; // destination array address
1584 const Register count = rdx; // elements count
1585 const Register byte_count = rcx;
1586 const Register qword_count = count;
1587 const Register end_from = from; // source array end address
1588 const Register end_to = to; // destination array end address
1589 // End pointers are inclusive, and if count is not zero they point
1590 // to the last unit copied: end_to[0] := end_from[0]
1591
1592 __ enter(); // required for proper stackwalking of RuntimeStub frame
1593 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1594
1595 if (entry != NULL) {
1596 *entry = __ pc();
1597 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1598 BLOCK_COMMENT("Entry:");
1599 }
1600
1601 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1602 // r9 and r10 may be used to save non-volatile registers
1603
1604 {
1605 // UnsafeCopyMemory page error: continue after ucm
1606 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1607 // 'from', 'to' and 'count' are now valid
1608 __ movptr(byte_count, count);
1609 __ shrptr(count, 3); // count => qword_count
1610
1611 // Copy from low to high addresses. Use 'to' as scratch.
1612 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1613 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1614 __ negptr(qword_count); // make the count negative
1615 __ jmp(L_copy_bytes);
1616
1617 // Copy trailing qwords
1618 __ BIND(L_copy_8_bytes);
1619 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1620 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1621 __ increment(qword_count);
1622 __ jcc(Assembler::notZero, L_copy_8_bytes);
1623
1624 // Check for and copy trailing dword
1625 __ BIND(L_copy_4_bytes);
1626 __ testl(byte_count, 4);
1627 __ jccb(Assembler::zero, L_copy_2_bytes);
1628 __ movl(rax, Address(end_from, 8));
1629 __ movl(Address(end_to, 8), rax);
1630
1631 __ addptr(end_from, 4);
1632 __ addptr(end_to, 4);
1633
1634 // Check for and copy trailing word
1635 __ BIND(L_copy_2_bytes);
1636 __ testl(byte_count, 2);
1637 __ jccb(Assembler::zero, L_copy_byte);
1638 __ movw(rax, Address(end_from, 8));
1639 __ movw(Address(end_to, 8), rax);
1640
1641 __ addptr(end_from, 2);
1642 __ addptr(end_to, 2);
1643
1644 // Check for and copy trailing byte
1645 __ BIND(L_copy_byte);
1646 __ testl(byte_count, 1);
1647 __ jccb(Assembler::zero, L_exit);
1648 __ movb(rax, Address(end_from, 8));
1649 __ movb(Address(end_to, 8), rax);
1650 }
1651 __ BIND(L_exit);
1652 address ucme_exit_pc = __ pc();
1653 restore_arg_regs();
1654 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1655 __ xorptr(rax, rax); // return 0
1656 __ vzeroupper();
1657 __ leave(); // required for proper stackwalking of RuntimeStub frame
1658 __ ret(0);
1659
1660 {
1661 UnsafeCopyMemoryMark ucmm(this, !aligned, false, ucme_exit_pc);
1662 // Copy in multi-bytes chunks
1663 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1664 __ jmp(L_copy_4_bytes);
1665 }
1666 return start;
1667 }
1668
1669 // Arguments:
1670 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1671 // ignored
1672 // name - stub name string
1673 //
1674 // Inputs:
1675 // c_rarg0 - source array address
1676 // c_rarg1 - destination array address
1677 // c_rarg2 - element count, treated as ssize_t, can be zero
1678 //
1679 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1680 // we let the hardware handle it. The one to eight bytes within words,
1681 // dwords or qwords that span cache line boundaries will still be loaded
1682 // and stored atomically.
1683 //
1684 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1685 address* entry, const char *name) {
1686 __ align(CodeEntryAlignment);
1687 StubCodeMark mark(this, "StubRoutines", name);
1688 address start = __ pc();
1689
1690 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1691 const Register from = rdi; // source array address
1692 const Register to = rsi; // destination array address
1693 const Register count = rdx; // elements count
1694 const Register byte_count = rcx;
1695 const Register qword_count = count;
1696
1697 __ enter(); // required for proper stackwalking of RuntimeStub frame
1698 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1699
1700 if (entry != NULL) {
1701 *entry = __ pc();
1702 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1703 BLOCK_COMMENT("Entry:");
1704 }
1705
1706 array_overlap_test(nooverlap_target, Address::times_1);
1707 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1708 // r9 and r10 may be used to save non-volatile registers
1709
1710 {
1711 // UnsafeCopyMemory page error: continue after ucm
1712 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1713 // 'from', 'to' and 'count' are now valid
1714 __ movptr(byte_count, count);
1715 __ shrptr(count, 3); // count => qword_count
1716
1717 // Copy from high to low addresses.
1718
1719 // Check for and copy trailing byte
1720 __ testl(byte_count, 1);
1721 __ jcc(Assembler::zero, L_copy_2_bytes);
1722 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1723 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1724 __ decrement(byte_count); // Adjust for possible trailing word
1725
1726 // Check for and copy trailing word
1727 __ BIND(L_copy_2_bytes);
1728 __ testl(byte_count, 2);
1729 __ jcc(Assembler::zero, L_copy_4_bytes);
1730 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1731 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1732
1733 // Check for and copy trailing dword
1734 __ BIND(L_copy_4_bytes);
1735 __ testl(byte_count, 4);
1736 __ jcc(Assembler::zero, L_copy_bytes);
1737 __ movl(rax, Address(from, qword_count, Address::times_8));
1738 __ movl(Address(to, qword_count, Address::times_8), rax);
1739 __ jmp(L_copy_bytes);
1740
1741 // Copy trailing qwords
1742 __ BIND(L_copy_8_bytes);
1743 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1744 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1745 __ decrement(qword_count);
1746 __ jcc(Assembler::notZero, L_copy_8_bytes);
1747 }
1748 restore_arg_regs();
1749 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1750 __ xorptr(rax, rax); // return 0
1751 __ vzeroupper();
1752 __ leave(); // required for proper stackwalking of RuntimeStub frame
1753 __ ret(0);
1754
1755 {
1756 // UnsafeCopyMemory page error: continue after ucm
1757 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1758 // Copy in multi-bytes chunks
1759 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1760 }
1761 restore_arg_regs();
1762 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1763 __ xorptr(rax, rax); // return 0
1764 __ vzeroupper();
1765 __ leave(); // required for proper stackwalking of RuntimeStub frame
1766 __ ret(0);
1767
1768 return start;
1769 }
1770
1771 // Arguments:
1772 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1773 // ignored
1774 // name - stub name string
1775 //
1776 // Inputs:
1777 // c_rarg0 - source array address
1778 // c_rarg1 - destination array address
1779 // c_rarg2 - element count, treated as ssize_t, can be zero
1780 //
1781 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1782 // let the hardware handle it. The two or four words within dwords
1783 // or qwords that span cache line boundaries will still be loaded
1784 // and stored atomically.
1785 //
1786 // Side Effects:
1787 // disjoint_short_copy_entry is set to the no-overlap entry point
1788 // used by generate_conjoint_short_copy().
1789 //
1790 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1791 __ align(CodeEntryAlignment);
1792 StubCodeMark mark(this, "StubRoutines", name);
1793 address start = __ pc();
1794
1795 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1796 const Register from = rdi; // source array address
1797 const Register to = rsi; // destination array address
1798 const Register count = rdx; // elements count
1799 const Register word_count = rcx;
1800 const Register qword_count = count;
1801 const Register end_from = from; // source array end address
1802 const Register end_to = to; // destination array end address
1803 // End pointers are inclusive, and if count is not zero they point
1804 // to the last unit copied: end_to[0] := end_from[0]
1805
1806 __ enter(); // required for proper stackwalking of RuntimeStub frame
1807 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1808
1809 if (entry != NULL) {
1810 *entry = __ pc();
1811 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1812 BLOCK_COMMENT("Entry:");
1813 }
1814
1815 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1816 // r9 and r10 may be used to save non-volatile registers
1817
1818 {
1819 // UnsafeCopyMemory page error: continue after ucm
1820 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1821 // 'from', 'to' and 'count' are now valid
1822 __ movptr(word_count, count);
1823 __ shrptr(count, 2); // count => qword_count
1824
1825 // Copy from low to high addresses. Use 'to' as scratch.
1826 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1827 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1828 __ negptr(qword_count);
1829 __ jmp(L_copy_bytes);
1830
1831 // Copy trailing qwords
1832 __ BIND(L_copy_8_bytes);
1833 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1834 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1835 __ increment(qword_count);
1836 __ jcc(Assembler::notZero, L_copy_8_bytes);
1837
1838 // Original 'dest' is trashed, so we can't use it as a
1839 // base register for a possible trailing word copy
1840
1841 // Check for and copy trailing dword
1842 __ BIND(L_copy_4_bytes);
1843 __ testl(word_count, 2);
1844 __ jccb(Assembler::zero, L_copy_2_bytes);
1845 __ movl(rax, Address(end_from, 8));
1846 __ movl(Address(end_to, 8), rax);
1847
1848 __ addptr(end_from, 4);
1849 __ addptr(end_to, 4);
1850
1851 // Check for and copy trailing word
1852 __ BIND(L_copy_2_bytes);
1853 __ testl(word_count, 1);
1854 __ jccb(Assembler::zero, L_exit);
1855 __ movw(rax, Address(end_from, 8));
1856 __ movw(Address(end_to, 8), rax);
1857 }
1858 __ BIND(L_exit);
1859 address ucme_exit_pc = __ pc();
1860 restore_arg_regs();
1861 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1862 __ xorptr(rax, rax); // return 0
1863 __ vzeroupper();
1864 __ leave(); // required for proper stackwalking of RuntimeStub frame
1865 __ ret(0);
1866
1867 {
1868 UnsafeCopyMemoryMark ucmm(this, !aligned, false, ucme_exit_pc);
1869 // Copy in multi-bytes chunks
1870 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1871 __ jmp(L_copy_4_bytes);
1872 }
1873
1874 return start;
1875 }
1876
1877 address generate_fill(BasicType t, bool aligned, const char *name) {
1878 __ align(CodeEntryAlignment);
1879 StubCodeMark mark(this, "StubRoutines", name);
1880 address start = __ pc();
1881
1882 BLOCK_COMMENT("Entry:");
1883
1884 const Register to = c_rarg0; // source array address
1885 const Register value = c_rarg1; // value
1886 const Register count = c_rarg2; // elements count
1887
1888 __ enter(); // required for proper stackwalking of RuntimeStub frame
1889
1890 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1891
1892 __ vzeroupper();
1893 __ leave(); // required for proper stackwalking of RuntimeStub frame
1894 __ ret(0);
1895 return start;
1896 }
1897
1898 // Arguments:
1899 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1900 // ignored
1901 // name - stub name string
1902 //
1903 // Inputs:
1904 // c_rarg0 - source array address
1905 // c_rarg1 - destination array address
1906 // c_rarg2 - element count, treated as ssize_t, can be zero
1907 //
1908 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1909 // let the hardware handle it. The two or four words within dwords
1910 // or qwords that span cache line boundaries will still be loaded
1911 // and stored atomically.
1912 //
1913 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1914 address *entry, const char *name) {
1915 __ align(CodeEntryAlignment);
1916 StubCodeMark mark(this, "StubRoutines", name);
1917 address start = __ pc();
1918
1919 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1920 const Register from = rdi; // source array address
1921 const Register to = rsi; // destination array address
1922 const Register count = rdx; // elements count
1923 const Register word_count = rcx;
1924 const Register qword_count = count;
1925
1926 __ enter(); // required for proper stackwalking of RuntimeStub frame
1927 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1928
1929 if (entry != NULL) {
1930 *entry = __ pc();
1931 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1932 BLOCK_COMMENT("Entry:");
1933 }
1934
1935 array_overlap_test(nooverlap_target, Address::times_2);
1936 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1937 // r9 and r10 may be used to save non-volatile registers
1938
1939 {
1940 // UnsafeCopyMemory page error: continue after ucm
1941 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1942 // 'from', 'to' and 'count' are now valid
1943 __ movptr(word_count, count);
1944 __ shrptr(count, 2); // count => qword_count
1945
1946 // Copy from high to low addresses. Use 'to' as scratch.
1947
1948 // Check for and copy trailing word
1949 __ testl(word_count, 1);
1950 __ jccb(Assembler::zero, L_copy_4_bytes);
1951 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1952 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1953
1954 // Check for and copy trailing dword
1955 __ BIND(L_copy_4_bytes);
1956 __ testl(word_count, 2);
1957 __ jcc(Assembler::zero, L_copy_bytes);
1958 __ movl(rax, Address(from, qword_count, Address::times_8));
1959 __ movl(Address(to, qword_count, Address::times_8), rax);
1960 __ jmp(L_copy_bytes);
1961
1962 // Copy trailing qwords
1963 __ BIND(L_copy_8_bytes);
1964 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1965 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1966 __ decrement(qword_count);
1967 __ jcc(Assembler::notZero, L_copy_8_bytes);
1968 }
1969 restore_arg_regs();
1970 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1971 __ xorptr(rax, rax); // return 0
1972 __ vzeroupper();
1973 __ leave(); // required for proper stackwalking of RuntimeStub frame
1974 __ ret(0);
1975
1976 {
1977 // UnsafeCopyMemory page error: continue after ucm
1978 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1979 // Copy in multi-bytes chunks
1980 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1981 }
1982 restore_arg_regs();
1983 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1984 __ xorptr(rax, rax); // return 0
1985 __ vzeroupper();
1986 __ leave(); // required for proper stackwalking of RuntimeStub frame
1987 __ ret(0);
1988
1989 return start;
1990 }
1991
1992 // Arguments:
1993 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1994 // ignored
1995 // is_oop - true => oop array, so generate store check code
1996 // name - stub name string
1997 //
1998 // Inputs:
1999 // c_rarg0 - source array address
2000 // c_rarg1 - destination array address
2001 // c_rarg2 - element count, treated as ssize_t, can be zero
2002 //
2003 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
2004 // the hardware handle it. The two dwords within qwords that span
2005 // cache line boundaries will still be loaded and stored atomicly.
2006 //
2007 // Side Effects:
2008 // disjoint_int_copy_entry is set to the no-overlap entry point
2009 // used by generate_conjoint_int_oop_copy().
2010 //
2011 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
2012 const char *name, bool dest_uninitialized = false) {
2013 __ align(CodeEntryAlignment);
2014 StubCodeMark mark(this, "StubRoutines", name);
2015 address start = __ pc();
2016
2017 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
2018 const Register from = rdi; // source array address
2019 const Register to = rsi; // destination array address
2020 const Register count = rdx; // elements count
2021 const Register dword_count = rcx;
2022 const Register qword_count = count;
2023 const Register end_from = from; // source array end address
2024 const Register end_to = to; // destination array end address
2025 // End pointers are inclusive, and if count is not zero they point
2026 // to the last unit copied: end_to[0] := end_from[0]
2027
2028 __ enter(); // required for proper stackwalking of RuntimeStub frame
2029 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2030
2031 if (entry != NULL) {
2032 *entry = __ pc();
2033 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2034 BLOCK_COMMENT("Entry:");
2035 }
2036
2037 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2038 // r9 is used to save r15_thread
2039
2040 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2041 if (dest_uninitialized) {
2042 decorators |= IS_DEST_UNINITIALIZED;
2043 }
2044 if (aligned) {
2045 decorators |= ARRAYCOPY_ALIGNED;
2046 }
2047
2048 BasicType type = is_oop ? T_OBJECT : T_INT;
2049 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2050 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2051
2052 {
2053 // UnsafeCopyMemory page error: continue after ucm
2054 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2055 // 'from', 'to' and 'count' are now valid
2056 __ movptr(dword_count, count);
2057 __ shrptr(count, 1); // count => qword_count
2058
2059 // Copy from low to high addresses. Use 'to' as scratch.
2060 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2061 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2062 __ negptr(qword_count);
2063 __ jmp(L_copy_bytes);
2064
2065 // Copy trailing qwords
2066 __ BIND(L_copy_8_bytes);
2067 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2068 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2069 __ increment(qword_count);
2070 __ jcc(Assembler::notZero, L_copy_8_bytes);
2071
2072 // Check for and copy trailing dword
2073 __ BIND(L_copy_4_bytes);
2074 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
2075 __ jccb(Assembler::zero, L_exit);
2076 __ movl(rax, Address(end_from, 8));
2077 __ movl(Address(end_to, 8), rax);
2078 }
2079 __ BIND(L_exit);
2080 address ucme_exit_pc = __ pc();
2081 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2082 restore_arg_regs_using_thread();
2083 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2084 __ vzeroupper();
2085 __ xorptr(rax, rax); // return 0
2086 __ leave(); // required for proper stackwalking of RuntimeStub frame
2087 __ ret(0);
2088
2089 {
2090 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, false, ucme_exit_pc);
2091 // Copy in multi-bytes chunks
2092 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2093 __ jmp(L_copy_4_bytes);
2094 }
2095
2096 return start;
2097 }
2098
2099 // Arguments:
2100 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
2101 // ignored
2102 // is_oop - true => oop array, so generate store check code
2103 // name - stub name string
2104 //
2105 // Inputs:
2106 // c_rarg0 - source array address
2107 // c_rarg1 - destination array address
2108 // c_rarg2 - element count, treated as ssize_t, can be zero
2109 //
2110 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
2111 // the hardware handle it. The two dwords within qwords that span
2112 // cache line boundaries will still be loaded and stored atomicly.
2113 //
2114 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
2115 address *entry, const char *name,
2116 bool dest_uninitialized = false) {
2117 __ align(CodeEntryAlignment);
2118 StubCodeMark mark(this, "StubRoutines", name);
2119 address start = __ pc();
2120
2121 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2122 const Register from = rdi; // source array address
2123 const Register to = rsi; // destination array address
2124 const Register count = rdx; // elements count
2125 const Register dword_count = rcx;
2126 const Register qword_count = count;
2127
2128 __ enter(); // required for proper stackwalking of RuntimeStub frame
2129 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2130
2131 if (entry != NULL) {
2132 *entry = __ pc();
2133 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2134 BLOCK_COMMENT("Entry:");
2135 }
2136
2137 array_overlap_test(nooverlap_target, Address::times_4);
2138 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2139 // r9 is used to save r15_thread
2140
2141 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2142 if (dest_uninitialized) {
2143 decorators |= IS_DEST_UNINITIALIZED;
2144 }
2145 if (aligned) {
2146 decorators |= ARRAYCOPY_ALIGNED;
2147 }
2148
2149 BasicType type = is_oop ? T_OBJECT : T_INT;
2150 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2151 // no registers are destroyed by this call
2152 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2153
2154 assert_clean_int(count, rax); // Make sure 'count' is clean int.
2155 {
2156 // UnsafeCopyMemory page error: continue after ucm
2157 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2158 // 'from', 'to' and 'count' are now valid
2159 __ movptr(dword_count, count);
2160 __ shrptr(count, 1); // count => qword_count
2161
2162 // Copy from high to low addresses. Use 'to' as scratch.
2163
2164 // Check for and copy trailing dword
2165 __ testl(dword_count, 1);
2166 __ jcc(Assembler::zero, L_copy_bytes);
2167 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
2168 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
2169 __ jmp(L_copy_bytes);
2170
2171 // Copy trailing qwords
2172 __ BIND(L_copy_8_bytes);
2173 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2174 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2175 __ decrement(qword_count);
2176 __ jcc(Assembler::notZero, L_copy_8_bytes);
2177 }
2178 if (is_oop) {
2179 __ jmp(L_exit);
2180 }
2181 restore_arg_regs_using_thread();
2182 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2183 __ xorptr(rax, rax); // return 0
2184 __ vzeroupper();
2185 __ leave(); // required for proper stackwalking of RuntimeStub frame
2186 __ ret(0);
2187
2188 {
2189 // UnsafeCopyMemory page error: continue after ucm
2190 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2191 // Copy in multi-bytes chunks
2192 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2193 }
2194
2195 __ BIND(L_exit);
2196 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2197 restore_arg_regs_using_thread();
2198 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2199 __ xorptr(rax, rax); // return 0
2200 __ vzeroupper();
2201 __ leave(); // required for proper stackwalking of RuntimeStub frame
2202 __ ret(0);
2203
2204 return start;
2205 }
2206
2207 // Arguments:
2208 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2209 // ignored
2210 // is_oop - true => oop array, so generate store check code
2211 // name - stub name string
2212 //
2213 // Inputs:
2214 // c_rarg0 - source array address
2215 // c_rarg1 - destination array address
2216 // c_rarg2 - element count, treated as ssize_t, can be zero
2217 //
2218 // Side Effects:
2219 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2220 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2221 //
2222 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2223 const char *name, bool dest_uninitialized = false) {
2224 __ align(CodeEntryAlignment);
2225 StubCodeMark mark(this, "StubRoutines", name);
2226 address start = __ pc();
2227
2228 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2229 const Register from = rdi; // source array address
2230 const Register to = rsi; // destination array address
2231 const Register qword_count = rdx; // elements count
2232 const Register end_from = from; // source array end address
2233 const Register end_to = rcx; // destination array end address
2234 const Register saved_count = r11;
2235 // End pointers are inclusive, and if count is not zero they point
2236 // to the last unit copied: end_to[0] := end_from[0]
2237
2238 __ enter(); // required for proper stackwalking of RuntimeStub frame
2239 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2240 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2241
2242 if (entry != NULL) {
2243 *entry = __ pc();
2244 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2245 BLOCK_COMMENT("Entry:");
2246 }
2247
2248 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2249 // r9 is used to save r15_thread
2250 // 'from', 'to' and 'qword_count' are now valid
2251
2252 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2253 if (dest_uninitialized) {
2254 decorators |= IS_DEST_UNINITIALIZED;
2255 }
2256 if (aligned) {
2257 decorators |= ARRAYCOPY_ALIGNED;
2258 }
2259
2260 BasicType type = is_oop ? T_OBJECT : T_LONG;
2261 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2262 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2263 {
2264 // UnsafeCopyMemory page error: continue after ucm
2265 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2266
2267 // Copy from low to high addresses. Use 'to' as scratch.
2268 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2269 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2270 __ negptr(qword_count);
2271 __ jmp(L_copy_bytes);
2272
2273 // Copy trailing qwords
2274 __ BIND(L_copy_8_bytes);
2275 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2276 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2277 __ increment(qword_count);
2278 __ jcc(Assembler::notZero, L_copy_8_bytes);
2279 }
2280 if (is_oop) {
2281 __ jmp(L_exit);
2282 } else {
2283 restore_arg_regs_using_thread();
2284 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2285 __ xorptr(rax, rax); // return 0
2286 __ vzeroupper();
2287 __ leave(); // required for proper stackwalking of RuntimeStub frame
2288 __ ret(0);
2289 }
2290
2291 {
2292 // UnsafeCopyMemory page error: continue after ucm
2293 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2294 // Copy in multi-bytes chunks
2295 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2296 }
2297
2298 __ BIND(L_exit);
2299 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2300 restore_arg_regs_using_thread();
2301 if (is_oop) {
2302 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2303 } else {
2304 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2305 }
2306 __ vzeroupper();
2307 __ xorptr(rax, rax); // return 0
2308 __ leave(); // required for proper stackwalking of RuntimeStub frame
2309 __ ret(0);
2310
2311 return start;
2312 }
2313
2314 // Arguments:
2315 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2316 // ignored
2317 // is_oop - true => oop array, so generate store check code
2318 // name - stub name string
2319 //
2320 // Inputs:
2321 // c_rarg0 - source array address
2322 // c_rarg1 - destination array address
2323 // c_rarg2 - element count, treated as ssize_t, can be zero
2324 //
2325 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2326 address nooverlap_target, address *entry,
2327 const char *name, bool dest_uninitialized = false) {
2328 __ align(CodeEntryAlignment);
2329 StubCodeMark mark(this, "StubRoutines", name);
2330 address start = __ pc();
2331
2332 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2333 const Register from = rdi; // source array address
2334 const Register to = rsi; // destination array address
2335 const Register qword_count = rdx; // elements count
2336 const Register saved_count = rcx;
2337
2338 __ enter(); // required for proper stackwalking of RuntimeStub frame
2339 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2340
2341 if (entry != NULL) {
2342 *entry = __ pc();
2343 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2344 BLOCK_COMMENT("Entry:");
2345 }
2346
2347 array_overlap_test(nooverlap_target, Address::times_8);
2348 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2349 // r9 is used to save r15_thread
2350 // 'from', 'to' and 'qword_count' are now valid
2351
2352 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2353 if (dest_uninitialized) {
2354 decorators |= IS_DEST_UNINITIALIZED;
2355 }
2356 if (aligned) {
2357 decorators |= ARRAYCOPY_ALIGNED;
2358 }
2359
2360 BasicType type = is_oop ? T_OBJECT : T_LONG;
2361 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2362 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2363 {
2364 // UnsafeCopyMemory page error: continue after ucm
2365 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2366
2367 __ jmp(L_copy_bytes);
2368
2369 // Copy trailing qwords
2370 __ BIND(L_copy_8_bytes);
2371 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2372 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2373 __ decrement(qword_count);
2374 __ jcc(Assembler::notZero, L_copy_8_bytes);
2375 }
2376 if (is_oop) {
2377 __ jmp(L_exit);
2378 } else {
2379 restore_arg_regs_using_thread();
2380 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2381 __ xorptr(rax, rax); // return 0
2382 __ vzeroupper();
2383 __ leave(); // required for proper stackwalking of RuntimeStub frame
2384 __ ret(0);
2385 }
2386 {
2387 // UnsafeCopyMemory page error: continue after ucm
2388 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2389
2390 // Copy in multi-bytes chunks
2391 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2392 }
2393 __ BIND(L_exit);
2394 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2395 restore_arg_regs_using_thread();
2396 if (is_oop) {
2397 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2398 } else {
2399 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2400 }
2401 __ vzeroupper();
2402 __ xorptr(rax, rax); // return 0
2403 __ leave(); // required for proper stackwalking of RuntimeStub frame
2404 __ ret(0);
2405
2406 return start;
2407 }
2408
2409
2410 // Helper for generating a dynamic type check.
2411 // Smashes no registers.
2412 void generate_type_check(Register sub_klass,
2413 Register super_check_offset,
2414 Register super_klass,
2415 Label& L_success) {
2416 assert_different_registers(sub_klass, super_check_offset, super_klass);
2417
2418 BLOCK_COMMENT("type_check:");
2419
2420 Label L_miss;
2421
2422 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2423 super_check_offset);
2424 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2425
2426 // Fall through on failure!
2427 __ BIND(L_miss);
2428 }
2429
2430 //
2431 // Generate checkcasting array copy stub
2432 //
2433 // Input:
2434 // c_rarg0 - source array address
2435 // c_rarg1 - destination array address
2436 // c_rarg2 - element count, treated as ssize_t, can be zero
2437 // c_rarg3 - size_t ckoff (super_check_offset)
2438 // not Win64
2439 // c_rarg4 - oop ckval (super_klass)
2440 // Win64
2441 // rsp+40 - oop ckval (super_klass)
2442 //
2443 // Output:
2444 // rax == 0 - success
2445 // rax == -1^K - failure, where K is partial transfer count
2446 //
2447 address generate_checkcast_copy(const char *name, address *entry,
2448 bool dest_uninitialized = false) {
2449
2450 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2451
2452 // Input registers (after setup_arg_regs)
2453 const Register from = rdi; // source array address
2454 const Register to = rsi; // destination array address
2455 const Register length = rdx; // elements count
2456 const Register ckoff = rcx; // super_check_offset
2457 const Register ckval = r8; // super_klass
2458
2459 // Registers used as temps (r13, r14 are save-on-entry)
2460 const Register end_from = from; // source array end address
2461 const Register end_to = r13; // destination array end address
2462 const Register count = rdx; // -(count_remaining)
2463 const Register r14_length = r14; // saved copy of length
2464 // End pointers are inclusive, and if length is not zero they point
2465 // to the last unit copied: end_to[0] := end_from[0]
2466
2467 const Register rax_oop = rax; // actual oop copied
2468 const Register r11_klass = r11; // oop._klass
2469
2470 //---------------------------------------------------------------
2471 // Assembler stub will be used for this call to arraycopy
2472 // if the two arrays are subtypes of Object[] but the
2473 // destination array type is not equal to or a supertype
2474 // of the source type. Each element must be separately
2475 // checked.
2476
2477 __ align(CodeEntryAlignment);
2478 StubCodeMark mark(this, "StubRoutines", name);
2479 address start = __ pc();
2480
2481 __ enter(); // required for proper stackwalking of RuntimeStub frame
2482
2483 #ifdef ASSERT
2484 // caller guarantees that the arrays really are different
2485 // otherwise, we would have to make conjoint checks
2486 { Label L;
2487 array_overlap_test(L, TIMES_OOP);
2488 __ stop("checkcast_copy within a single array");
2489 __ bind(L);
2490 }
2491 #endif //ASSERT
2492
2493 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2494 // ckoff => rcx, ckval => r8
2495 // r9 and r10 may be used to save non-volatile registers
2496 #ifdef _WIN64
2497 // last argument (#4) is on stack on Win64
2498 __ movptr(ckval, Address(rsp, 6 * wordSize));
2499 #endif
2500
2501 // Caller of this entry point must set up the argument registers.
2502 if (entry != NULL) {
2503 *entry = __ pc();
2504 BLOCK_COMMENT("Entry:");
2505 }
2506
2507 // allocate spill slots for r13, r14
2508 enum {
2509 saved_r13_offset,
2510 saved_r14_offset,
2511 saved_r10_offset,
2512 saved_rbp_offset
2513 };
2514 __ subptr(rsp, saved_rbp_offset * wordSize);
2515 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2516 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2517 __ movptr(Address(rsp, saved_r10_offset * wordSize), r10);
2518
2519 #ifdef ASSERT
2520 Label L2;
2521 __ get_thread(r14);
2522 __ cmpptr(r15_thread, r14);
2523 __ jcc(Assembler::equal, L2);
2524 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
2525 __ bind(L2);
2526 #endif // ASSERT
2527
2528 // check that int operands are properly extended to size_t
2529 assert_clean_int(length, rax);
2530 assert_clean_int(ckoff, rax);
2531
2532 #ifdef ASSERT
2533 BLOCK_COMMENT("assert consistent ckoff/ckval");
2534 // The ckoff and ckval must be mutually consistent,
2535 // even though caller generates both.
2536 { Label L;
2537 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2538 __ cmpl(ckoff, Address(ckval, sco_offset));
2539 __ jcc(Assembler::equal, L);
2540 __ stop("super_check_offset inconsistent");
2541 __ bind(L);
2542 }
2543 #endif //ASSERT
2544
2545 // Loop-invariant addresses. They are exclusive end pointers.
2546 Address end_from_addr(from, length, TIMES_OOP, 0);
2547 Address end_to_addr(to, length, TIMES_OOP, 0);
2548 // Loop-variant addresses. They assume post-incremented count < 0.
2549 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2550 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2551
2552 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST | ARRAYCOPY_DISJOINT;
2553 if (dest_uninitialized) {
2554 decorators |= IS_DEST_UNINITIALIZED;
2555 }
2556
2557 BasicType type = T_OBJECT;
2558 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2559 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2560
2561 // Copy from low to high addresses, indexed from the end of each array.
2562 __ lea(end_from, end_from_addr);
2563 __ lea(end_to, end_to_addr);
2564 __ movptr(r14_length, length); // save a copy of the length
2565 assert(length == count, ""); // else fix next line:
2566 __ negptr(count); // negate and test the length
2567 __ jcc(Assembler::notZero, L_load_element);
2568
2569 // Empty array: Nothing to do.
2570 __ xorptr(rax, rax); // return 0 on (trivial) success
2571 __ jmp(L_done);
2572
2573 // ======== begin loop ========
2574 // (Loop is rotated; its entry is L_load_element.)
2575 // Loop control:
2576 // for (count = -count; count != 0; count++)
2577 // Base pointers src, dst are biased by 8*(count-1),to last element.
2578 __ align(OptoLoopAlignment);
2579
2580 __ BIND(L_store_element);
2581 __ store_heap_oop(to_element_addr, rax_oop, noreg, noreg, AS_RAW); // store the oop
2582 __ increment(count); // increment the count toward zero
2583 __ jcc(Assembler::zero, L_do_card_marks);
2584
2585 // ======== loop entry is here ========
2586 __ BIND(L_load_element);
2587 __ load_heap_oop(rax_oop, from_element_addr, noreg, noreg, AS_RAW); // load the oop
2588 __ testptr(rax_oop, rax_oop);
2589 __ jcc(Assembler::zero, L_store_element);
2590
2591 __ load_klass(r11_klass, rax_oop, rscratch1);// query the object klass
2592 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2593 // ======== end loop ========
2594
2595 // It was a real error; we must depend on the caller to finish the job.
2596 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2597 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2598 // and report their number to the caller.
2599 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2600 Label L_post_barrier;
2601 __ addptr(r14_length, count); // K = (original - remaining) oops
2602 __ movptr(rax, r14_length); // save the value
2603 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2604 __ jccb(Assembler::notZero, L_post_barrier);
2605 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2606
2607 // Come here on success only.
2608 __ BIND(L_do_card_marks);
2609 __ xorptr(rax, rax); // return 0 on success
2610
2611 __ BIND(L_post_barrier);
2612 bs->arraycopy_epilogue(_masm, decorators, type, from, to, r14_length);
2613
2614 // Common exit point (success or failure).
2615 __ BIND(L_done);
2616 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2617 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2618 __ movptr(r10, Address(rsp, saved_r10_offset * wordSize));
2619 restore_arg_regs();
2620 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2621 __ leave(); // required for proper stackwalking of RuntimeStub frame
2622 __ ret(0);
2623
2624 return start;
2625 }
2626
2627 //
2628 // Generate 'unsafe' array copy stub
2629 // Though just as safe as the other stubs, it takes an unscaled
2630 // size_t argument instead of an element count.
2631 //
2632 // Input:
2633 // c_rarg0 - source array address
2634 // c_rarg1 - destination array address
2635 // c_rarg2 - byte count, treated as ssize_t, can be zero
2636 //
2637 // Examines the alignment of the operands and dispatches
2638 // to a long, int, short, or byte copy loop.
2639 //
2640 address generate_unsafe_copy(const char *name,
2641 address byte_copy_entry, address short_copy_entry,
2642 address int_copy_entry, address long_copy_entry) {
2643
2644 Label L_long_aligned, L_int_aligned, L_short_aligned;
2645
2646 // Input registers (before setup_arg_regs)
2647 const Register from = c_rarg0; // source array address
2648 const Register to = c_rarg1; // destination array address
2649 const Register size = c_rarg2; // byte count (size_t)
2650
2651 // Register used as a temp
2652 const Register bits = rax; // test copy of low bits
2653
2654 __ align(CodeEntryAlignment);
2655 StubCodeMark mark(this, "StubRoutines", name);
2656 address start = __ pc();
2657
2658 __ enter(); // required for proper stackwalking of RuntimeStub frame
2659
2660 // bump this on entry, not on exit:
2661 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2662
2663 __ mov(bits, from);
2664 __ orptr(bits, to);
2665 __ orptr(bits, size);
2666
2667 __ testb(bits, BytesPerLong-1);
2668 __ jccb(Assembler::zero, L_long_aligned);
2669
2670 __ testb(bits, BytesPerInt-1);
2671 __ jccb(Assembler::zero, L_int_aligned);
2672
2673 __ testb(bits, BytesPerShort-1);
2674 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2675
2676 __ BIND(L_short_aligned);
2677 __ shrptr(size, LogBytesPerShort); // size => short_count
2678 __ jump(RuntimeAddress(short_copy_entry));
2679
2680 __ BIND(L_int_aligned);
2681 __ shrptr(size, LogBytesPerInt); // size => int_count
2682 __ jump(RuntimeAddress(int_copy_entry));
2683
2684 __ BIND(L_long_aligned);
2685 __ shrptr(size, LogBytesPerLong); // size => qword_count
2686 __ jump(RuntimeAddress(long_copy_entry));
2687
2688 return start;
2689 }
2690
2691 // Perform range checks on the proposed arraycopy.
2692 // Kills temp, but nothing else.
2693 // Also, clean the sign bits of src_pos and dst_pos.
2694 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2695 Register src_pos, // source position (c_rarg1)
2696 Register dst, // destination array oo (c_rarg2)
2697 Register dst_pos, // destination position (c_rarg3)
2698 Register length,
2699 Register temp,
2700 Label& L_failed) {
2701 BLOCK_COMMENT("arraycopy_range_checks:");
2702
2703 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2704 __ movl(temp, length);
2705 __ addl(temp, src_pos); // src_pos + length
2706 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2707 __ jcc(Assembler::above, L_failed);
2708
2709 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2710 __ movl(temp, length);
2711 __ addl(temp, dst_pos); // dst_pos + length
2712 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2713 __ jcc(Assembler::above, L_failed);
2714
2715 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2716 // Move with sign extension can be used since they are positive.
2717 __ movslq(src_pos, src_pos);
2718 __ movslq(dst_pos, dst_pos);
2719
2720 BLOCK_COMMENT("arraycopy_range_checks done");
2721 }
2722
2723 //
2724 // Generate generic array copy stubs
2725 //
2726 // Input:
2727 // c_rarg0 - src oop
2728 // c_rarg1 - src_pos (32-bits)
2729 // c_rarg2 - dst oop
2730 // c_rarg3 - dst_pos (32-bits)
2731 // not Win64
2732 // c_rarg4 - element count (32-bits)
2733 // Win64
2734 // rsp+40 - element count (32-bits)
2735 //
2736 // Output:
2737 // rax == 0 - success
2738 // rax == -1^K - failure, where K is partial transfer count
2739 //
2740 address generate_generic_copy(const char *name,
2741 address byte_copy_entry, address short_copy_entry,
2742 address int_copy_entry, address oop_copy_entry,
2743 address long_copy_entry, address checkcast_copy_entry) {
2744
2745 Label L_failed, L_failed_0, L_objArray;
2746 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2747
2748 // Input registers
2749 const Register src = c_rarg0; // source array oop
2750 const Register src_pos = c_rarg1; // source position
2751 const Register dst = c_rarg2; // destination array oop
2752 const Register dst_pos = c_rarg3; // destination position
2753 #ifndef _WIN64
2754 const Register length = c_rarg4;
2755 const Register rklass_tmp = r9; // load_klass
2756 #else
2757 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2758 const Register rklass_tmp = rdi; // load_klass
2759 #endif
2760
2761 { int modulus = CodeEntryAlignment;
2762 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2763 int advance = target - (__ offset() % modulus);
2764 if (advance < 0) advance += modulus;
2765 if (advance > 0) __ nop(advance);
2766 }
2767 StubCodeMark mark(this, "StubRoutines", name);
2768
2769 // Short-hop target to L_failed. Makes for denser prologue code.
2770 __ BIND(L_failed_0);
2771 __ jmp(L_failed);
2772 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2773
2774 __ align(CodeEntryAlignment);
2775 address start = __ pc();
2776
2777 __ enter(); // required for proper stackwalking of RuntimeStub frame
2778
2779 // bump this on entry, not on exit:
2780 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2781
2782 //-----------------------------------------------------------------------
2783 // Assembler stub will be used for this call to arraycopy
2784 // if the following conditions are met:
2785 //
2786 // (1) src and dst must not be null.
2787 // (2) src_pos must not be negative.
2788 // (3) dst_pos must not be negative.
2789 // (4) length must not be negative.
2790 // (5) src klass and dst klass should be the same and not NULL.
2791 // (6) src and dst should be arrays.
2792 // (7) src_pos + length must not exceed length of src.
2793 // (8) dst_pos + length must not exceed length of dst.
2794 //
2795
2796 // if (src == NULL) return -1;
2797 __ testptr(src, src); // src oop
2798 size_t j1off = __ offset();
2799 __ jccb(Assembler::zero, L_failed_0);
2800
2801 // if (src_pos < 0) return -1;
2802 __ testl(src_pos, src_pos); // src_pos (32-bits)
2803 __ jccb(Assembler::negative, L_failed_0);
2804
2805 // if (dst == NULL) return -1;
2806 __ testptr(dst, dst); // dst oop
2807 __ jccb(Assembler::zero, L_failed_0);
2808
2809 // if (dst_pos < 0) return -1;
2810 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2811 size_t j4off = __ offset();
2812 __ jccb(Assembler::negative, L_failed_0);
2813
2814 // The first four tests are very dense code,
2815 // but not quite dense enough to put four
2816 // jumps in a 16-byte instruction fetch buffer.
2817 // That's good, because some branch predicters
2818 // do not like jumps so close together.
2819 // Make sure of this.
2820 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2821
2822 // registers used as temp
2823 const Register r11_length = r11; // elements count to copy
2824 const Register r10_src_klass = r10; // array klass
2825
2826 // if (length < 0) return -1;
2827 __ movl(r11_length, length); // length (elements count, 32-bits value)
2828 __ testl(r11_length, r11_length);
2829 __ jccb(Assembler::negative, L_failed_0);
2830
2831 __ load_klass(r10_src_klass, src, rklass_tmp);
2832 #ifdef ASSERT
2833 // assert(src->klass() != NULL);
2834 {
2835 BLOCK_COMMENT("assert klasses not null {");
2836 Label L1, L2;
2837 __ testptr(r10_src_klass, r10_src_klass);
2838 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2839 __ bind(L1);
2840 __ stop("broken null klass");
2841 __ bind(L2);
2842 __ load_klass(rax, dst, rklass_tmp);
2843 __ cmpq(rax, 0);
2844 __ jcc(Assembler::equal, L1); // this would be broken also
2845 BLOCK_COMMENT("} assert klasses not null done");
2846 }
2847 #endif
2848
2849 // Load layout helper (32-bits)
2850 //
2851 // |array_tag| | header_size | element_type | |log2_element_size|
2852 // 32 30 24 16 8 2 0
2853 //
2854 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2855 //
2856
2857 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2858
2859 // Handle objArrays completely differently...
2860 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2861 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2862 __ jcc(Assembler::equal, L_objArray);
2863
2864 // if (src->klass() != dst->klass()) return -1;
2865 __ load_klass(rax, dst, rklass_tmp);
2866 __ cmpq(r10_src_klass, rax);
2867 __ jcc(Assembler::notEqual, L_failed);
2868
2869 const Register rax_lh = rax; // layout helper
2870 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2871
2872 // if (!src->is_Array()) return -1;
2873 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2874 __ jcc(Assembler::greaterEqual, L_failed);
2875
2876 // At this point, it is known to be a typeArray (array_tag 0x3).
2877 #ifdef ASSERT
2878 {
2879 BLOCK_COMMENT("assert primitive array {");
2880 Label L;
2881 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2882 __ jcc(Assembler::greaterEqual, L);
2883 __ stop("must be a primitive array");
2884 __ bind(L);
2885 BLOCK_COMMENT("} assert primitive array done");
2886 }
2887 #endif
2888
2889 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2890 r10, L_failed);
2891
2892 // TypeArrayKlass
2893 //
2894 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2895 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2896 //
2897
2898 const Register r10_offset = r10; // array offset
2899 const Register rax_elsize = rax_lh; // element size
2900
2901 __ movl(r10_offset, rax_lh);
2902 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2903 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2904 __ addptr(src, r10_offset); // src array offset
2905 __ addptr(dst, r10_offset); // dst array offset
2906 BLOCK_COMMENT("choose copy loop based on element size");
2907 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2908
2909 // next registers should be set before the jump to corresponding stub
2910 const Register from = c_rarg0; // source array address
2911 const Register to = c_rarg1; // destination array address
2912 const Register count = c_rarg2; // elements count
2913
2914 // 'from', 'to', 'count' registers should be set in such order
2915 // since they are the same as 'src', 'src_pos', 'dst'.
2916
2917 __ BIND(L_copy_bytes);
2918 __ cmpl(rax_elsize, 0);
2919 __ jccb(Assembler::notEqual, L_copy_shorts);
2920 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2921 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2922 __ movl2ptr(count, r11_length); // length
2923 __ jump(RuntimeAddress(byte_copy_entry));
2924
2925 __ BIND(L_copy_shorts);
2926 __ cmpl(rax_elsize, LogBytesPerShort);
2927 __ jccb(Assembler::notEqual, L_copy_ints);
2928 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2929 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2930 __ movl2ptr(count, r11_length); // length
2931 __ jump(RuntimeAddress(short_copy_entry));
2932
2933 __ BIND(L_copy_ints);
2934 __ cmpl(rax_elsize, LogBytesPerInt);
2935 __ jccb(Assembler::notEqual, L_copy_longs);
2936 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2937 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2938 __ movl2ptr(count, r11_length); // length
2939 __ jump(RuntimeAddress(int_copy_entry));
2940
2941 __ BIND(L_copy_longs);
2942 #ifdef ASSERT
2943 {
2944 BLOCK_COMMENT("assert long copy {");
2945 Label L;
2946 __ cmpl(rax_elsize, LogBytesPerLong);
2947 __ jcc(Assembler::equal, L);
2948 __ stop("must be long copy, but elsize is wrong");
2949 __ bind(L);
2950 BLOCK_COMMENT("} assert long copy done");
2951 }
2952 #endif
2953 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2954 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2955 __ movl2ptr(count, r11_length); // length
2956 __ jump(RuntimeAddress(long_copy_entry));
2957
2958 // ObjArrayKlass
2959 __ BIND(L_objArray);
2960 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2961
2962 Label L_plain_copy, L_checkcast_copy;
2963 // test array classes for subtyping
2964 __ load_klass(rax, dst, rklass_tmp);
2965 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2966 __ jcc(Assembler::notEqual, L_checkcast_copy);
2967
2968 // Identically typed arrays can be copied without element-wise checks.
2969 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2970 r10, L_failed);
2971
2972 __ lea(from, Address(src, src_pos, TIMES_OOP,
2973 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2974 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2975 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2976 __ movl2ptr(count, r11_length); // length
2977 __ BIND(L_plain_copy);
2978 __ jump(RuntimeAddress(oop_copy_entry));
2979
2980 __ BIND(L_checkcast_copy);
2981 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2982 {
2983 // Before looking at dst.length, make sure dst is also an objArray.
2984 __ cmpl(Address(rax, lh_offset), objArray_lh);
2985 __ jcc(Assembler::notEqual, L_failed);
2986
2987 // It is safe to examine both src.length and dst.length.
2988 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2989 rax, L_failed);
2990
2991 const Register r11_dst_klass = r11;
2992 __ load_klass(r11_dst_klass, dst, rklass_tmp); // reload
2993
2994 // Marshal the base address arguments now, freeing registers.
2995 __ lea(from, Address(src, src_pos, TIMES_OOP,
2996 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2997 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2998 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2999 __ movl(count, length); // length (reloaded)
3000 Register sco_temp = c_rarg3; // this register is free now
3001 assert_different_registers(from, to, count, sco_temp,
3002 r11_dst_klass, r10_src_klass);
3003 assert_clean_int(count, sco_temp);
3004
3005 // Generate the type check.
3006 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
3007 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
3008 assert_clean_int(sco_temp, rax);
3009 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
3010
3011 // Fetch destination element klass from the ObjArrayKlass header.
3012 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
3013 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
3014 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
3015 assert_clean_int(sco_temp, rax);
3016
3017 // the checkcast_copy loop needs two extra arguments:
3018 assert(c_rarg3 == sco_temp, "#3 already in place");
3019 // Set up arguments for checkcast_copy_entry.
3020 setup_arg_regs(4);
3021 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
3022 __ jump(RuntimeAddress(checkcast_copy_entry));
3023 }
3024
3025 __ BIND(L_failed);
3026 __ xorptr(rax, rax);
3027 __ notptr(rax); // return -1
3028 __ leave(); // required for proper stackwalking of RuntimeStub frame
3029 __ ret(0);
3030
3031 return start;
3032 }
3033
3034 address generate_data_cache_writeback() {
3035 const Register src = c_rarg0; // source address
3036
3037 __ align(CodeEntryAlignment);
3038
3039 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback");
3040
3041 address start = __ pc();
3042 __ enter();
3043 __ cache_wb(Address(src, 0));
3044 __ leave();
3045 __ ret(0);
3046
3047 return start;
3048 }
3049
3050 address generate_data_cache_writeback_sync() {
3051 const Register is_pre = c_rarg0; // pre or post sync
3052
3053 __ align(CodeEntryAlignment);
3054
3055 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback_sync");
3056
3057 // pre wbsync is a no-op
3058 // post wbsync translates to an sfence
3059
3060 Label skip;
3061 address start = __ pc();
3062 __ enter();
3063 __ cmpl(is_pre, 0);
3064 __ jcc(Assembler::notEqual, skip);
3065 __ cache_wbsync(false);
3066 __ bind(skip);
3067 __ leave();
3068 __ ret(0);
3069
3070 return start;
3071 }
3072
3073 void generate_arraycopy_stubs() {
3074 address entry;
3075 address entry_jbyte_arraycopy;
3076 address entry_jshort_arraycopy;
3077 address entry_jint_arraycopy;
3078 address entry_oop_arraycopy;
3079 address entry_jlong_arraycopy;
3080 address entry_checkcast_arraycopy;
3081
3082 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
3083 "jbyte_disjoint_arraycopy");
3084 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
3085 "jbyte_arraycopy");
3086
3087 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
3088 "jshort_disjoint_arraycopy");
3089 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
3090 "jshort_arraycopy");
3091
3092 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
3093 "jint_disjoint_arraycopy");
3094 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
3095 &entry_jint_arraycopy, "jint_arraycopy");
3096
3097 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
3098 "jlong_disjoint_arraycopy");
3099 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
3100 &entry_jlong_arraycopy, "jlong_arraycopy");
3101
3102
3103 if (UseCompressedOops) {
3104 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
3105 "oop_disjoint_arraycopy");
3106 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
3107 &entry_oop_arraycopy, "oop_arraycopy");
3108 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
3109 "oop_disjoint_arraycopy_uninit",
3110 /*dest_uninitialized*/true);
3111 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
3112 NULL, "oop_arraycopy_uninit",
3113 /*dest_uninitialized*/true);
3114 } else {
3115 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
3116 "oop_disjoint_arraycopy");
3117 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
3118 &entry_oop_arraycopy, "oop_arraycopy");
3119 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
3120 "oop_disjoint_arraycopy_uninit",
3121 /*dest_uninitialized*/true);
3122 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
3123 NULL, "oop_arraycopy_uninit",
3124 /*dest_uninitialized*/true);
3125 }
3126
3127 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
3128 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
3129 /*dest_uninitialized*/true);
3130
3131 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
3132 entry_jbyte_arraycopy,
3133 entry_jshort_arraycopy,
3134 entry_jint_arraycopy,
3135 entry_jlong_arraycopy);
3136 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
3137 entry_jbyte_arraycopy,
3138 entry_jshort_arraycopy,
3139 entry_jint_arraycopy,
3140 entry_oop_arraycopy,
3141 entry_jlong_arraycopy,
3142 entry_checkcast_arraycopy);
3143
3144 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
3145 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
3146 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
3147 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
3148 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
3149 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
3150
3151 // We don't generate specialized code for HeapWord-aligned source
3152 // arrays, so just use the code we've already generated
3153 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
3154 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
3155
3156 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
3157 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
3158
3159 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
3160 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
3161
3162 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
3163 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
3164
3165 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
3166 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
3167
3168 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
3169 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
3170 }
3171
3172 // AES intrinsic stubs
3173 enum {AESBlockSize = 16};
3174
3175 address generate_key_shuffle_mask() {
3176 __ align(16);
3177 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
3178 address start = __ pc();
3179 __ emit_data64( 0x0405060700010203, relocInfo::none );
3180 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
3181 return start;
3182 }
3183
3184 address generate_counter_shuffle_mask() {
3185 __ align(16);
3186 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
3187 address start = __ pc();
3188 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3189 __ emit_data64(0x0001020304050607, relocInfo::none);
3190 return start;
3191 }
3192
3193 // Utility routine for loading a 128-bit key word in little endian format
3194 // can optionally specify that the shuffle mask is already in an xmmregister
3195 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
3196 __ movdqu(xmmdst, Address(key, offset));
3197 if (xmm_shuf_mask != NULL) {
3198 __ pshufb(xmmdst, xmm_shuf_mask);
3199 } else {
3200 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3201 }
3202 }
3203
3204 // Utility routine for increase 128bit counter (iv in CTR mode)
3205 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
3206 __ pextrq(reg, xmmdst, 0x0);
3207 __ addq(reg, inc_delta);
3208 __ pinsrq(xmmdst, reg, 0x0);
3209 __ jcc(Assembler::carryClear, next_block); // jump if no carry
3210 __ pextrq(reg, xmmdst, 0x01); // Carry
3211 __ addq(reg, 0x01);
3212 __ pinsrq(xmmdst, reg, 0x01); //Carry end
3213 __ BIND(next_block); // next instruction
3214 }
3215
3216 // Arguments:
3217 //
3218 // Inputs:
3219 // c_rarg0 - source byte array address
3220 // c_rarg1 - destination byte array address
3221 // c_rarg2 - K (key) in little endian int array
3222 //
3223 address generate_aescrypt_encryptBlock() {
3224 assert(UseAES, "need AES instructions and misaligned SSE support");
3225 __ align(CodeEntryAlignment);
3226 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
3227 Label L_doLast;
3228 address start = __ pc();
3229
3230 const Register from = c_rarg0; // source array address
3231 const Register to = c_rarg1; // destination array address
3232 const Register key = c_rarg2; // key array address
3233 const Register keylen = rax;
3234
3235 const XMMRegister xmm_result = xmm0;
3236 const XMMRegister xmm_key_shuf_mask = xmm1;
3237 // On win64 xmm6-xmm15 must be preserved so don't use them.
3238 const XMMRegister xmm_temp1 = xmm2;
3239 const XMMRegister xmm_temp2 = xmm3;
3240 const XMMRegister xmm_temp3 = xmm4;
3241 const XMMRegister xmm_temp4 = xmm5;
3242
3243 __ enter(); // required for proper stackwalking of RuntimeStub frame
3244
3245 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3246 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3247
3248 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3249 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
3250
3251 // For encryption, the java expanded key ordering is just what we need
3252 // we don't know if the key is aligned, hence not using load-execute form
3253
3254 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3255 __ pxor(xmm_result, xmm_temp1);
3256
3257 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3258 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3259 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3260 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3261
3262 __ aesenc(xmm_result, xmm_temp1);
3263 __ aesenc(xmm_result, xmm_temp2);
3264 __ aesenc(xmm_result, xmm_temp3);
3265 __ aesenc(xmm_result, xmm_temp4);
3266
3267 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3268 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3269 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3270 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3271
3272 __ aesenc(xmm_result, xmm_temp1);
3273 __ aesenc(xmm_result, xmm_temp2);
3274 __ aesenc(xmm_result, xmm_temp3);
3275 __ aesenc(xmm_result, xmm_temp4);
3276
3277 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3278 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3279
3280 __ cmpl(keylen, 44);
3281 __ jccb(Assembler::equal, L_doLast);
3282
3283 __ aesenc(xmm_result, xmm_temp1);
3284 __ aesenc(xmm_result, xmm_temp2);
3285
3286 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3287 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3288
3289 __ cmpl(keylen, 52);
3290 __ jccb(Assembler::equal, L_doLast);
3291
3292 __ aesenc(xmm_result, xmm_temp1);
3293 __ aesenc(xmm_result, xmm_temp2);
3294
3295 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3296 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3297
3298 __ BIND(L_doLast);
3299 __ aesenc(xmm_result, xmm_temp1);
3300 __ aesenclast(xmm_result, xmm_temp2);
3301 __ movdqu(Address(to, 0), xmm_result); // store the result
3302 __ xorptr(rax, rax); // return 0
3303 __ leave(); // required for proper stackwalking of RuntimeStub frame
3304 __ ret(0);
3305
3306 return start;
3307 }
3308
3309
3310 // Arguments:
3311 //
3312 // Inputs:
3313 // c_rarg0 - source byte array address
3314 // c_rarg1 - destination byte array address
3315 // c_rarg2 - K (key) in little endian int array
3316 //
3317 address generate_aescrypt_decryptBlock() {
3318 assert(UseAES, "need AES instructions and misaligned SSE support");
3319 __ align(CodeEntryAlignment);
3320 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3321 Label L_doLast;
3322 address start = __ pc();
3323
3324 const Register from = c_rarg0; // source array address
3325 const Register to = c_rarg1; // destination array address
3326 const Register key = c_rarg2; // key array address
3327 const Register keylen = rax;
3328
3329 const XMMRegister xmm_result = xmm0;
3330 const XMMRegister xmm_key_shuf_mask = xmm1;
3331 // On win64 xmm6-xmm15 must be preserved so don't use them.
3332 const XMMRegister xmm_temp1 = xmm2;
3333 const XMMRegister xmm_temp2 = xmm3;
3334 const XMMRegister xmm_temp3 = xmm4;
3335 const XMMRegister xmm_temp4 = xmm5;
3336
3337 __ enter(); // required for proper stackwalking of RuntimeStub frame
3338
3339 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3340 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3341
3342 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3343 __ movdqu(xmm_result, Address(from, 0));
3344
3345 // for decryption java expanded key ordering is rotated one position from what we want
3346 // so we start from 0x10 here and hit 0x00 last
3347 // we don't know if the key is aligned, hence not using load-execute form
3348 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3349 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3350 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3351 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3352
3353 __ pxor (xmm_result, xmm_temp1);
3354 __ aesdec(xmm_result, xmm_temp2);
3355 __ aesdec(xmm_result, xmm_temp3);
3356 __ aesdec(xmm_result, xmm_temp4);
3357
3358 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3359 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3360 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3361 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3362
3363 __ aesdec(xmm_result, xmm_temp1);
3364 __ aesdec(xmm_result, xmm_temp2);
3365 __ aesdec(xmm_result, xmm_temp3);
3366 __ aesdec(xmm_result, xmm_temp4);
3367
3368 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3369 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3370 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3371
3372 __ cmpl(keylen, 44);
3373 __ jccb(Assembler::equal, L_doLast);
3374
3375 __ aesdec(xmm_result, xmm_temp1);
3376 __ aesdec(xmm_result, xmm_temp2);
3377
3378 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3379 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3380
3381 __ cmpl(keylen, 52);
3382 __ jccb(Assembler::equal, L_doLast);
3383
3384 __ aesdec(xmm_result, xmm_temp1);
3385 __ aesdec(xmm_result, xmm_temp2);
3386
3387 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3388 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3389
3390 __ BIND(L_doLast);
3391 __ aesdec(xmm_result, xmm_temp1);
3392 __ aesdec(xmm_result, xmm_temp2);
3393
3394 // for decryption the aesdeclast operation is always on key+0x00
3395 __ aesdeclast(xmm_result, xmm_temp3);
3396 __ movdqu(Address(to, 0), xmm_result); // store the result
3397 __ xorptr(rax, rax); // return 0
3398 __ leave(); // required for proper stackwalking of RuntimeStub frame
3399 __ ret(0);
3400
3401 return start;
3402 }
3403
3404
3405 // Arguments:
3406 //
3407 // Inputs:
3408 // c_rarg0 - source byte array address
3409 // c_rarg1 - destination byte array address
3410 // c_rarg2 - K (key) in little endian int array
3411 // c_rarg3 - r vector byte array address
3412 // c_rarg4 - input length
3413 //
3414 // Output:
3415 // rax - input length
3416 //
3417 address generate_cipherBlockChaining_encryptAESCrypt() {
3418 assert(UseAES, "need AES instructions and misaligned SSE support");
3419 __ align(CodeEntryAlignment);
3420 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3421 address start = __ pc();
3422
3423 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3424 const Register from = c_rarg0; // source array address
3425 const Register to = c_rarg1; // destination array address
3426 const Register key = c_rarg2; // key array address
3427 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3428 // and left with the results of the last encryption block
3429 #ifndef _WIN64
3430 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3431 #else
3432 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3433 const Register len_reg = r11; // pick the volatile windows register
3434 #endif
3435 const Register pos = rax;
3436
3437 // xmm register assignments for the loops below
3438 const XMMRegister xmm_result = xmm0;
3439 const XMMRegister xmm_temp = xmm1;
3440 // keys 0-10 preloaded into xmm2-xmm12
3441 const int XMM_REG_NUM_KEY_FIRST = 2;
3442 const int XMM_REG_NUM_KEY_LAST = 15;
3443 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3444 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3445 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3446 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3447 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3448
3449 __ enter(); // required for proper stackwalking of RuntimeStub frame
3450
3451 #ifdef _WIN64
3452 // on win64, fill len_reg from stack position
3453 __ movl(len_reg, len_mem);
3454 #else
3455 __ push(len_reg); // Save
3456 #endif
3457
3458 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3459 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3460 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3461 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3462 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3463 offset += 0x10;
3464 }
3465 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3466
3467 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3468 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3469 __ cmpl(rax, 44);
3470 __ jcc(Assembler::notEqual, L_key_192_256);
3471
3472 // 128 bit code follows here
3473 __ movptr(pos, 0);
3474 __ align(OptoLoopAlignment);
3475
3476 __ BIND(L_loopTop_128);
3477 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3478 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3479 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3480 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3481 __ aesenc(xmm_result, as_XMMRegister(rnum));
3482 }
3483 __ aesenclast(xmm_result, xmm_key10);
3484 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3485 // no need to store r to memory until we exit
3486 __ addptr(pos, AESBlockSize);
3487 __ subptr(len_reg, AESBlockSize);
3488 __ jcc(Assembler::notEqual, L_loopTop_128);
3489
3490 __ BIND(L_exit);
3491 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3492
3493 #ifdef _WIN64
3494 __ movl(rax, len_mem);
3495 #else
3496 __ pop(rax); // return length
3497 #endif
3498 __ leave(); // required for proper stackwalking of RuntimeStub frame
3499 __ ret(0);
3500
3501 __ BIND(L_key_192_256);
3502 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3503 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3504 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3505 __ cmpl(rax, 52);
3506 __ jcc(Assembler::notEqual, L_key_256);
3507
3508 // 192-bit code follows here (could be changed to use more xmm registers)
3509 __ movptr(pos, 0);
3510 __ align(OptoLoopAlignment);
3511
3512 __ BIND(L_loopTop_192);
3513 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3514 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3515 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3516 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3517 __ aesenc(xmm_result, as_XMMRegister(rnum));
3518 }
3519 __ aesenclast(xmm_result, xmm_key12);
3520 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3521 // no need to store r to memory until we exit
3522 __ addptr(pos, AESBlockSize);
3523 __ subptr(len_reg, AESBlockSize);
3524 __ jcc(Assembler::notEqual, L_loopTop_192);
3525 __ jmp(L_exit);
3526
3527 __ BIND(L_key_256);
3528 // 256-bit code follows here (could be changed to use more xmm registers)
3529 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3530 __ movptr(pos, 0);
3531 __ align(OptoLoopAlignment);
3532
3533 __ BIND(L_loopTop_256);
3534 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3535 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3536 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3537 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3538 __ aesenc(xmm_result, as_XMMRegister(rnum));
3539 }
3540 load_key(xmm_temp, key, 0xe0);
3541 __ aesenclast(xmm_result, xmm_temp);
3542 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3543 // no need to store r to memory until we exit
3544 __ addptr(pos, AESBlockSize);
3545 __ subptr(len_reg, AESBlockSize);
3546 __ jcc(Assembler::notEqual, L_loopTop_256);
3547 __ jmp(L_exit);
3548
3549 return start;
3550 }
3551
3552 // Safefetch stubs.
3553 void generate_safefetch(const char* name, int size, address* entry,
3554 address* fault_pc, address* continuation_pc) {
3555 // safefetch signatures:
3556 // int SafeFetch32(int* adr, int errValue);
3557 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3558 //
3559 // arguments:
3560 // c_rarg0 = adr
3561 // c_rarg1 = errValue
3562 //
3563 // result:
3564 // PPC_RET = *adr or errValue
3565
3566 StubCodeMark mark(this, "StubRoutines", name);
3567
3568 // Entry point, pc or function descriptor.
3569 *entry = __ pc();
3570
3571 // Load *adr into c_rarg1, may fault.
3572 *fault_pc = __ pc();
3573 switch (size) {
3574 case 4:
3575 // int32_t
3576 __ movl(c_rarg1, Address(c_rarg0, 0));
3577 break;
3578 case 8:
3579 // int64_t
3580 __ movq(c_rarg1, Address(c_rarg0, 0));
3581 break;
3582 default:
3583 ShouldNotReachHere();
3584 }
3585
3586 // return errValue or *adr
3587 *continuation_pc = __ pc();
3588 __ movq(rax, c_rarg1);
3589 __ ret(0);
3590 }
3591
3592 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3593 // to hide instruction latency
3594 //
3595 // Arguments:
3596 //
3597 // Inputs:
3598 // c_rarg0 - source byte array address
3599 // c_rarg1 - destination byte array address
3600 // c_rarg2 - K (key) in little endian int array
3601 // c_rarg3 - r vector byte array address
3602 // c_rarg4 - input length
3603 //
3604 // Output:
3605 // rax - input length
3606 //
3607 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3608 assert(UseAES, "need AES instructions and misaligned SSE support");
3609 __ align(CodeEntryAlignment);
3610 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3611 address start = __ pc();
3612
3613 const Register from = c_rarg0; // source array address
3614 const Register to = c_rarg1; // destination array address
3615 const Register key = c_rarg2; // key array address
3616 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3617 // and left with the results of the last encryption block
3618 #ifndef _WIN64
3619 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3620 #else
3621 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3622 const Register len_reg = r11; // pick the volatile windows register
3623 #endif
3624 const Register pos = rax;
3625
3626 const int PARALLEL_FACTOR = 4;
3627 const int ROUNDS[3] = { 10, 12, 14 }; // aes rounds for key128, key192, key256
3628
3629 Label L_exit;
3630 Label L_singleBlock_loopTopHead[3]; // 128, 192, 256
3631 Label L_singleBlock_loopTopHead2[3]; // 128, 192, 256
3632 Label L_singleBlock_loopTop[3]; // 128, 192, 256
3633 Label L_multiBlock_loopTopHead[3]; // 128, 192, 256
3634 Label L_multiBlock_loopTop[3]; // 128, 192, 256
3635
3636 // keys 0-10 preloaded into xmm5-xmm15
3637 const int XMM_REG_NUM_KEY_FIRST = 5;
3638 const int XMM_REG_NUM_KEY_LAST = 15;
3639 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3640 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3641
3642 __ enter(); // required for proper stackwalking of RuntimeStub frame
3643
3644 #ifdef _WIN64
3645 // on win64, fill len_reg from stack position
3646 __ movl(len_reg, len_mem);
3647 #else
3648 __ push(len_reg); // Save
3649 #endif
3650 __ push(rbx);
3651 // the java expanded key ordering is rotated one position from what we want
3652 // so we start from 0x10 here and hit 0x00 last
3653 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3654 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3655 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3656 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3657 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3658 offset += 0x10;
3659 }
3660 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3661
3662 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3663
3664 // registers holding the four results in the parallelized loop
3665 const XMMRegister xmm_result0 = xmm0;
3666 const XMMRegister xmm_result1 = xmm2;
3667 const XMMRegister xmm_result2 = xmm3;
3668 const XMMRegister xmm_result3 = xmm4;
3669
3670 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3671
3672 __ xorptr(pos, pos);
3673
3674 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3675 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3676 __ cmpl(rbx, 52);
3677 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[1]);
3678 __ cmpl(rbx, 60);
3679 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[2]);
3680
3681 #define DoFour(opc, src_reg) \
3682 __ opc(xmm_result0, src_reg); \
3683 __ opc(xmm_result1, src_reg); \
3684 __ opc(xmm_result2, src_reg); \
3685 __ opc(xmm_result3, src_reg); \
3686
3687 for (int k = 0; k < 3; ++k) {
3688 __ BIND(L_multiBlock_loopTopHead[k]);
3689 if (k != 0) {
3690 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3691 __ jcc(Assembler::less, L_singleBlock_loopTopHead2[k]);
3692 }
3693 if (k == 1) {
3694 __ subptr(rsp, 6 * wordSize);
3695 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3696 load_key(xmm15, key, 0xb0); // 0xb0; 192-bit key goes up to 0xc0
3697 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3698 load_key(xmm1, key, 0xc0); // 0xc0;
3699 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3700 } else if (k == 2) {
3701 __ subptr(rsp, 10 * wordSize);
3702 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3703 load_key(xmm15, key, 0xd0); // 0xd0; 256-bit key goes upto 0xe0
3704 __ movdqu(Address(rsp, 6 * wordSize), xmm15);
3705 load_key(xmm1, key, 0xe0); // 0xe0;
3706 __ movdqu(Address(rsp, 8 * wordSize), xmm1);
3707 load_key(xmm15, key, 0xb0); // 0xb0;
3708 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3709 load_key(xmm1, key, 0xc0); // 0xc0;
3710 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3711 }
3712 __ align(OptoLoopAlignment);
3713 __ BIND(L_multiBlock_loopTop[k]);
3714 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3715 __ jcc(Assembler::less, L_singleBlock_loopTopHead[k]);
3716
3717 if (k != 0) {
3718 __ movdqu(xmm15, Address(rsp, 2 * wordSize));
3719 __ movdqu(xmm1, Address(rsp, 4 * wordSize));
3720 }
3721
3722 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
3723 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3724 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3725 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3726
3727 DoFour(pxor, xmm_key_first);
3728 if (k == 0) {
3729 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
3730 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3731 }
3732 DoFour(aesdeclast, xmm_key_last);
3733 } else if (k == 1) {
3734 for (int rnum = 1; rnum <= ROUNDS[k]-2; rnum++) {
3735 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3736 }
3737 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3738 DoFour(aesdec, xmm1); // key : 0xc0
3739 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3740 DoFour(aesdeclast, xmm_key_last);
3741 } else if (k == 2) {
3742 for (int rnum = 1; rnum <= ROUNDS[k] - 4; rnum++) {
3743 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3744 }
3745 DoFour(aesdec, xmm1); // key : 0xc0
3746 __ movdqu(xmm15, Address(rsp, 6 * wordSize));
3747 __ movdqu(xmm1, Address(rsp, 8 * wordSize));
3748 DoFour(aesdec, xmm15); // key : 0xd0
3749 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3750 DoFour(aesdec, xmm1); // key : 0xe0
3751 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3752 DoFour(aesdeclast, xmm_key_last);
3753 }
3754
3755 // for each result, xor with the r vector of previous cipher block
3756 __ pxor(xmm_result0, xmm_prev_block_cipher);
3757 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3758 __ pxor(xmm_result1, xmm_prev_block_cipher);
3759 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3760 __ pxor(xmm_result2, xmm_prev_block_cipher);
3761 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3762 __ pxor(xmm_result3, xmm_prev_block_cipher);
3763 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
3764 if (k != 0) {
3765 __ movdqu(Address(rvec, 0x00), xmm_prev_block_cipher);
3766 }
3767
3768 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3769 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3770 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3771 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3772
3773 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize);
3774 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize);
3775 __ jmp(L_multiBlock_loopTop[k]);
3776
3777 // registers used in the non-parallelized loops
3778 // xmm register assignments for the loops below
3779 const XMMRegister xmm_result = xmm0;
3780 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3781 const XMMRegister xmm_key11 = xmm3;
3782 const XMMRegister xmm_key12 = xmm4;
3783 const XMMRegister key_tmp = xmm4;
3784
3785 __ BIND(L_singleBlock_loopTopHead[k]);
3786 if (k == 1) {
3787 __ addptr(rsp, 6 * wordSize);
3788 } else if (k == 2) {
3789 __ addptr(rsp, 10 * wordSize);
3790 }
3791 __ cmpptr(len_reg, 0); // any blocks left??
3792 __ jcc(Assembler::equal, L_exit);
3793 __ BIND(L_singleBlock_loopTopHead2[k]);
3794 if (k == 1) {
3795 load_key(xmm_key11, key, 0xb0); // 0xb0; 192-bit key goes upto 0xc0
3796 load_key(xmm_key12, key, 0xc0); // 0xc0; 192-bit key goes upto 0xc0
3797 }
3798 if (k == 2) {
3799 load_key(xmm_key11, key, 0xb0); // 0xb0; 256-bit key goes upto 0xe0
3800 }
3801 __ align(OptoLoopAlignment);
3802 __ BIND(L_singleBlock_loopTop[k]);
3803 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3804 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3805 __ pxor(xmm_result, xmm_key_first); // do the aes dec rounds
3806 for (int rnum = 1; rnum <= 9 ; rnum++) {
3807 __ aesdec(xmm_result, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3808 }
3809 if (k == 1) {
3810 __ aesdec(xmm_result, xmm_key11);
3811 __ aesdec(xmm_result, xmm_key12);
3812 }
3813 if (k == 2) {
3814 __ aesdec(xmm_result, xmm_key11);
3815 load_key(key_tmp, key, 0xc0);
3816 __ aesdec(xmm_result, key_tmp);
3817 load_key(key_tmp, key, 0xd0);
3818 __ aesdec(xmm_result, key_tmp);
3819 load_key(key_tmp, key, 0xe0);
3820 __ aesdec(xmm_result, key_tmp);
3821 }
3822
3823 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3824 __ pxor(xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3825 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3826 // no need to store r to memory until we exit
3827 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3828 __ addptr(pos, AESBlockSize);
3829 __ subptr(len_reg, AESBlockSize);
3830 __ jcc(Assembler::notEqual, L_singleBlock_loopTop[k]);
3831 if (k != 2) {
3832 __ jmp(L_exit);
3833 }
3834 } //for 128/192/256
3835
3836 __ BIND(L_exit);
3837 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3838 __ pop(rbx);
3839 #ifdef _WIN64
3840 __ movl(rax, len_mem);
3841 #else
3842 __ pop(rax); // return length
3843 #endif
3844 __ leave(); // required for proper stackwalking of RuntimeStub frame
3845 __ ret(0);
3846 return start;
3847 }
3848
3849 address generate_electronicCodeBook_encryptAESCrypt() {
3850 __ align(CodeEntryAlignment);
3851 StubCodeMark mark(this, "StubRoutines", "electronicCodeBook_encryptAESCrypt");
3852 address start = __ pc();
3853 const Register from = c_rarg0; // source array address
3854 const Register to = c_rarg1; // destination array address
3855 const Register key = c_rarg2; // key array address
3856 const Register len = c_rarg3; // src len (must be multiple of blocksize 16)
3857 __ enter(); // required for proper stackwalking of RuntimeStub frame
3858 __ aesecb_encrypt(from, to, key, len);
3859 __ leave(); // required for proper stackwalking of RuntimeStub frame
3860 __ ret(0);
3861 return start;
3862 }
3863
3864 address generate_electronicCodeBook_decryptAESCrypt() {
3865 __ align(CodeEntryAlignment);
3866 StubCodeMark mark(this, "StubRoutines", "electronicCodeBook_decryptAESCrypt");
3867 address start = __ pc();
3868 const Register from = c_rarg0; // source array address
3869 const Register to = c_rarg1; // destination array address
3870 const Register key = c_rarg2; // key array address
3871 const Register len = c_rarg3; // src len (must be multiple of blocksize 16)
3872 __ enter(); // required for proper stackwalking of RuntimeStub frame
3873 __ aesecb_decrypt(from, to, key, len);
3874 __ leave(); // required for proper stackwalking of RuntimeStub frame
3875 __ ret(0);
3876 return start;
3877 }
3878
3879 address generate_upper_word_mask() {
3880 __ align(64);
3881 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3882 address start = __ pc();
3883 __ emit_data64(0x0000000000000000, relocInfo::none);
3884 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
3885 return start;
3886 }
3887
3888 address generate_shuffle_byte_flip_mask() {
3889 __ align(64);
3890 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3891 address start = __ pc();
3892 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3893 __ emit_data64(0x0001020304050607, relocInfo::none);
3894 return start;
3895 }
3896
3897 // ofs and limit are use for multi-block byte array.
3898 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3899 address generate_sha1_implCompress(bool multi_block, const char *name) {
3900 __ align(CodeEntryAlignment);
3901 StubCodeMark mark(this, "StubRoutines", name);
3902 address start = __ pc();
3903
3904 Register buf = c_rarg0;
3905 Register state = c_rarg1;
3906 Register ofs = c_rarg2;
3907 Register limit = c_rarg3;
3908
3909 const XMMRegister abcd = xmm0;
3910 const XMMRegister e0 = xmm1;
3911 const XMMRegister e1 = xmm2;
3912 const XMMRegister msg0 = xmm3;
3913
3914 const XMMRegister msg1 = xmm4;
3915 const XMMRegister msg2 = xmm5;
3916 const XMMRegister msg3 = xmm6;
3917 const XMMRegister shuf_mask = xmm7;
3918
3919 __ enter();
3920
3921 __ subptr(rsp, 4 * wordSize);
3922
3923 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3924 buf, state, ofs, limit, rsp, multi_block);
3925
3926 __ addptr(rsp, 4 * wordSize);
3927
3928 __ leave();
3929 __ ret(0);
3930 return start;
3931 }
3932
3933 address generate_pshuffle_byte_flip_mask() {
3934 __ align(64);
3935 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3936 address start = __ pc();
3937 __ emit_data64(0x0405060700010203, relocInfo::none);
3938 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3939
3940 if (VM_Version::supports_avx2()) {
3941 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
3942 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3943 // _SHUF_00BA
3944 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3945 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3946 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3947 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3948 // _SHUF_DC00
3949 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3950 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3951 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3952 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3953 }
3954
3955 return start;
3956 }
3957
3958 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
3959 address generate_pshuffle_byte_flip_mask_sha512() {
3960 __ align(32);
3961 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
3962 address start = __ pc();
3963 if (VM_Version::supports_avx2()) {
3964 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
3965 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3966 __ emit_data64(0x1011121314151617, relocInfo::none);
3967 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
3968 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
3969 __ emit_data64(0x0000000000000000, relocInfo::none);
3970 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3971 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3972 }
3973
3974 return start;
3975 }
3976
3977 // ofs and limit are use for multi-block byte array.
3978 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3979 address generate_sha256_implCompress(bool multi_block, const char *name) {
3980 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
3981 __ align(CodeEntryAlignment);
3982 StubCodeMark mark(this, "StubRoutines", name);
3983 address start = __ pc();
3984
3985 Register buf = c_rarg0;
3986 Register state = c_rarg1;
3987 Register ofs = c_rarg2;
3988 Register limit = c_rarg3;
3989
3990 const XMMRegister msg = xmm0;
3991 const XMMRegister state0 = xmm1;
3992 const XMMRegister state1 = xmm2;
3993 const XMMRegister msgtmp0 = xmm3;
3994
3995 const XMMRegister msgtmp1 = xmm4;
3996 const XMMRegister msgtmp2 = xmm5;
3997 const XMMRegister msgtmp3 = xmm6;
3998 const XMMRegister msgtmp4 = xmm7;
3999
4000 const XMMRegister shuf_mask = xmm8;
4001
4002 __ enter();
4003
4004 __ subptr(rsp, 4 * wordSize);
4005
4006 if (VM_Version::supports_sha()) {
4007 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
4008 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
4009 } else if (VM_Version::supports_avx2()) {
4010 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
4011 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
4012 }
4013 __ addptr(rsp, 4 * wordSize);
4014 __ vzeroupper();
4015 __ leave();
4016 __ ret(0);
4017 return start;
4018 }
4019
4020 address generate_sha512_implCompress(bool multi_block, const char *name) {
4021 assert(VM_Version::supports_avx2(), "");
4022 assert(VM_Version::supports_bmi2(), "");
4023 __ align(CodeEntryAlignment);
4024 StubCodeMark mark(this, "StubRoutines", name);
4025 address start = __ pc();
4026
4027 Register buf = c_rarg0;
4028 Register state = c_rarg1;
4029 Register ofs = c_rarg2;
4030 Register limit = c_rarg3;
4031
4032 const XMMRegister msg = xmm0;
4033 const XMMRegister state0 = xmm1;
4034 const XMMRegister state1 = xmm2;
4035 const XMMRegister msgtmp0 = xmm3;
4036 const XMMRegister msgtmp1 = xmm4;
4037 const XMMRegister msgtmp2 = xmm5;
4038 const XMMRegister msgtmp3 = xmm6;
4039 const XMMRegister msgtmp4 = xmm7;
4040
4041 const XMMRegister shuf_mask = xmm8;
4042
4043 __ enter();
4044
4045 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
4046 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
4047
4048 __ vzeroupper();
4049 __ leave();
4050 __ ret(0);
4051 return start;
4052 }
4053
4054 // This mask is used for incrementing counter value(linc0, linc4, etc.)
4055 address counter_mask_addr() {
4056 __ align(64);
4057 StubCodeMark mark(this, "StubRoutines", "counter_mask_addr");
4058 address start = __ pc();
4059 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);//lbswapmask
4060 __ emit_data64(0x0001020304050607, relocInfo::none);
4061 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
4062 __ emit_data64(0x0001020304050607, relocInfo::none);
4063 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
4064 __ emit_data64(0x0001020304050607, relocInfo::none);
4065 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
4066 __ emit_data64(0x0001020304050607, relocInfo::none);
4067 __ emit_data64(0x0000000000000000, relocInfo::none);//linc0 = counter_mask_addr+64
4068 __ emit_data64(0x0000000000000000, relocInfo::none);
4069 __ emit_data64(0x0000000000000001, relocInfo::none);//counter_mask_addr() + 80
4070 __ emit_data64(0x0000000000000000, relocInfo::none);
4071 __ emit_data64(0x0000000000000002, relocInfo::none);
4072 __ emit_data64(0x0000000000000000, relocInfo::none);
4073 __ emit_data64(0x0000000000000003, relocInfo::none);
4074 __ emit_data64(0x0000000000000000, relocInfo::none);
4075 __ emit_data64(0x0000000000000004, relocInfo::none);//linc4 = counter_mask_addr() + 128
4076 __ emit_data64(0x0000000000000000, relocInfo::none);
4077 __ emit_data64(0x0000000000000004, relocInfo::none);
4078 __ emit_data64(0x0000000000000000, relocInfo::none);
4079 __ emit_data64(0x0000000000000004, relocInfo::none);
4080 __ emit_data64(0x0000000000000000, relocInfo::none);
4081 __ emit_data64(0x0000000000000004, relocInfo::none);
4082 __ emit_data64(0x0000000000000000, relocInfo::none);
4083 __ emit_data64(0x0000000000000008, relocInfo::none);//linc8 = counter_mask_addr() + 192
4084 __ emit_data64(0x0000000000000000, relocInfo::none);
4085 __ emit_data64(0x0000000000000008, relocInfo::none);
4086 __ emit_data64(0x0000000000000000, relocInfo::none);
4087 __ emit_data64(0x0000000000000008, relocInfo::none);
4088 __ emit_data64(0x0000000000000000, relocInfo::none);
4089 __ emit_data64(0x0000000000000008, relocInfo::none);
4090 __ emit_data64(0x0000000000000000, relocInfo::none);
4091 __ emit_data64(0x0000000000000020, relocInfo::none);//linc32 = counter_mask_addr() + 256
4092 __ emit_data64(0x0000000000000000, relocInfo::none);
4093 __ emit_data64(0x0000000000000020, relocInfo::none);
4094 __ emit_data64(0x0000000000000000, relocInfo::none);
4095 __ emit_data64(0x0000000000000020, relocInfo::none);
4096 __ emit_data64(0x0000000000000000, relocInfo::none);
4097 __ emit_data64(0x0000000000000020, relocInfo::none);
4098 __ emit_data64(0x0000000000000000, relocInfo::none);
4099 __ emit_data64(0x0000000000000010, relocInfo::none);//linc16 = counter_mask_addr() + 320
4100 __ emit_data64(0x0000000000000000, relocInfo::none);
4101 __ emit_data64(0x0000000000000010, relocInfo::none);
4102 __ emit_data64(0x0000000000000000, relocInfo::none);
4103 __ emit_data64(0x0000000000000010, relocInfo::none);
4104 __ emit_data64(0x0000000000000000, relocInfo::none);
4105 __ emit_data64(0x0000000000000010, relocInfo::none);
4106 __ emit_data64(0x0000000000000000, relocInfo::none);
4107 return start;
4108 }
4109
4110 // Vector AES Counter implementation
4111 address generate_counterMode_VectorAESCrypt() {
4112 __ align(CodeEntryAlignment);
4113 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
4114 address start = __ pc();
4115 const Register from = c_rarg0; // source array address
4116 const Register to = c_rarg1; // destination array address
4117 const Register key = c_rarg2; // key array address r8
4118 const Register counter = c_rarg3; // counter byte array initialized from counter array address
4119 // and updated with the incremented counter in the end
4120 #ifndef _WIN64
4121 const Register len_reg = c_rarg4;
4122 const Register saved_encCounter_start = c_rarg5;
4123 const Register used_addr = r10;
4124 const Address used_mem(rbp, 2 * wordSize);
4125 const Register used = r11;
4126 #else
4127 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4128 const Address saved_encCounter_mem(rbp, 7 * wordSize); // saved encrypted counter is on stack on Win64
4129 const Address used_mem(rbp, 8 * wordSize); // used length is on stack on Win64
4130 const Register len_reg = r10; // pick the first volatile windows register
4131 const Register saved_encCounter_start = r11;
4132 const Register used_addr = r13;
4133 const Register used = r14;
4134 #endif
4135 __ enter();
4136 // Save state before entering routine
4137 __ push(r12);
4138 __ push(r13);
4139 __ push(r14);
4140 __ push(r15);
4141 #ifdef _WIN64
4142 // on win64, fill len_reg from stack position
4143 __ movl(len_reg, len_mem);
4144 __ movptr(saved_encCounter_start, saved_encCounter_mem);
4145 __ movptr(used_addr, used_mem);
4146 __ movl(used, Address(used_addr, 0));
4147 #else
4148 __ push(len_reg); // Save
4149 __ movptr(used_addr, used_mem);
4150 __ movl(used, Address(used_addr, 0));
4151 #endif
4152 __ push(rbx);
4153 __ aesctr_encrypt(from, to, key, counter, len_reg, used, used_addr, saved_encCounter_start);
4154 // Restore state before leaving routine
4155 __ pop(rbx);
4156 #ifdef _WIN64
4157 __ movl(rax, len_mem); // return length
4158 #else
4159 __ pop(rax); // return length
4160 #endif
4161 __ pop(r15);
4162 __ pop(r14);
4163 __ pop(r13);
4164 __ pop(r12);
4165
4166 __ leave(); // required for proper stackwalking of RuntimeStub frame
4167 __ ret(0);
4168 return start;
4169 }
4170
4171 // This is a version of CTR/AES crypt which does 6 blocks in a loop at a time
4172 // to hide instruction latency
4173 //
4174 // Arguments:
4175 //
4176 // Inputs:
4177 // c_rarg0 - source byte array address
4178 // c_rarg1 - destination byte array address
4179 // c_rarg2 - K (key) in little endian int array
4180 // c_rarg3 - counter vector byte array address
4181 // Linux
4182 // c_rarg4 - input length
4183 // c_rarg5 - saved encryptedCounter start
4184 // rbp + 6 * wordSize - saved used length
4185 // Windows
4186 // rbp + 6 * wordSize - input length
4187 // rbp + 7 * wordSize - saved encryptedCounter start
4188 // rbp + 8 * wordSize - saved used length
4189 //
4190 // Output:
4191 // rax - input length
4192 //
4193 address generate_counterMode_AESCrypt_Parallel() {
4194 assert(UseAES, "need AES instructions and misaligned SSE support");
4195 __ align(CodeEntryAlignment);
4196 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
4197 address start = __ pc();
4198 const Register from = c_rarg0; // source array address
4199 const Register to = c_rarg1; // destination array address
4200 const Register key = c_rarg2; // key array address
4201 const Register counter = c_rarg3; // counter byte array initialized from counter array address
4202 // and updated with the incremented counter in the end
4203 #ifndef _WIN64
4204 const Register len_reg = c_rarg4;
4205 const Register saved_encCounter_start = c_rarg5;
4206 const Register used_addr = r10;
4207 const Address used_mem(rbp, 2 * wordSize);
4208 const Register used = r11;
4209 #else
4210 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4211 const Address saved_encCounter_mem(rbp, 7 * wordSize); // length is on stack on Win64
4212 const Address used_mem(rbp, 8 * wordSize); // length is on stack on Win64
4213 const Register len_reg = r10; // pick the first volatile windows register
4214 const Register saved_encCounter_start = r11;
4215 const Register used_addr = r13;
4216 const Register used = r14;
4217 #endif
4218 const Register pos = rax;
4219
4220 const int PARALLEL_FACTOR = 6;
4221 const XMMRegister xmm_counter_shuf_mask = xmm0;
4222 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
4223 const XMMRegister xmm_curr_counter = xmm2;
4224
4225 const XMMRegister xmm_key_tmp0 = xmm3;
4226 const XMMRegister xmm_key_tmp1 = xmm4;
4227
4228 // registers holding the four results in the parallelized loop
4229 const XMMRegister xmm_result0 = xmm5;
4230 const XMMRegister xmm_result1 = xmm6;
4231 const XMMRegister xmm_result2 = xmm7;
4232 const XMMRegister xmm_result3 = xmm8;
4233 const XMMRegister xmm_result4 = xmm9;
4234 const XMMRegister xmm_result5 = xmm10;
4235
4236 const XMMRegister xmm_from0 = xmm11;
4237 const XMMRegister xmm_from1 = xmm12;
4238 const XMMRegister xmm_from2 = xmm13;
4239 const XMMRegister xmm_from3 = xmm14; //the last one is xmm14. we have to preserve it on WIN64.
4240 const XMMRegister xmm_from4 = xmm3; //reuse xmm3~4. Because xmm_key_tmp0~1 are useless when loading input text
4241 const XMMRegister xmm_from5 = xmm4;
4242
4243 //for key_128, key_192, key_256
4244 const int rounds[3] = {10, 12, 14};
4245 Label L_exit_preLoop, L_preLoop_start;
4246 Label L_multiBlock_loopTop[3];
4247 Label L_singleBlockLoopTop[3];
4248 Label L__incCounter[3][6]; //for 6 blocks
4249 Label L__incCounter_single[3]; //for single block, key128, key192, key256
4250 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];
4251 Label L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
4252
4253 Label L_exit;
4254
4255 __ enter(); // required for proper stackwalking of RuntimeStub frame
4256
4257 #ifdef _WIN64
4258 // allocate spill slots for r13, r14
4259 enum {
4260 saved_r13_offset,
4261 saved_r14_offset
4262 };
4263 __ subptr(rsp, 2 * wordSize);
4264 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
4265 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
4266
4267 // on win64, fill len_reg from stack position
4268 __ movl(len_reg, len_mem);
4269 __ movptr(saved_encCounter_start, saved_encCounter_mem);
4270 __ movptr(used_addr, used_mem);
4271 __ movl(used, Address(used_addr, 0));
4272 #else
4273 __ push(len_reg); // Save
4274 __ movptr(used_addr, used_mem);
4275 __ movl(used, Address(used_addr, 0));
4276 #endif
4277
4278 __ push(rbx); // Save RBX
4279 __ movdqu(xmm_curr_counter, Address(counter, 0x00)); // initialize counter with initial counter
4280 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()), pos); // pos as scratch
4281 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled
4282 __ movptr(pos, 0);
4283
4284 // Use the partially used encrpyted counter from last invocation
4285 __ BIND(L_preLoop_start);
4286 __ cmpptr(used, 16);
4287 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
4288 __ cmpptr(len_reg, 0);
4289 __ jcc(Assembler::lessEqual, L_exit_preLoop);
4290 __ movb(rbx, Address(saved_encCounter_start, used));
4291 __ xorb(rbx, Address(from, pos));
4292 __ movb(Address(to, pos), rbx);
4293 __ addptr(pos, 1);
4294 __ addptr(used, 1);
4295 __ subptr(len_reg, 1);
4296
4297 __ jmp(L_preLoop_start);
4298
4299 __ BIND(L_exit_preLoop);
4300 __ movl(Address(used_addr, 0), used);
4301
4302 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
4303 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()), rbx); // rbx as scratch
4304 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4305 __ cmpl(rbx, 52);
4306 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
4307 __ cmpl(rbx, 60);
4308 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
4309
4310 #define CTR_DoSix(opc, src_reg) \
4311 __ opc(xmm_result0, src_reg); \
4312 __ opc(xmm_result1, src_reg); \
4313 __ opc(xmm_result2, src_reg); \
4314 __ opc(xmm_result3, src_reg); \
4315 __ opc(xmm_result4, src_reg); \
4316 __ opc(xmm_result5, src_reg);
4317
4318 // k == 0 : generate code for key_128
4319 // k == 1 : generate code for key_192
4320 // k == 2 : generate code for key_256
4321 for (int k = 0; k < 3; ++k) {
4322 //multi blocks starts here
4323 __ align(OptoLoopAlignment);
4324 __ BIND(L_multiBlock_loopTop[k]);
4325 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
4326 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
4327 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4328
4329 //load, then increase counters
4330 CTR_DoSix(movdqa, xmm_curr_counter);
4331 inc_counter(rbx, xmm_result1, 0x01, L__incCounter[k][0]);
4332 inc_counter(rbx, xmm_result2, 0x02, L__incCounter[k][1]);
4333 inc_counter(rbx, xmm_result3, 0x03, L__incCounter[k][2]);
4334 inc_counter(rbx, xmm_result4, 0x04, L__incCounter[k][3]);
4335 inc_counter(rbx, xmm_result5, 0x05, L__incCounter[k][4]);
4336 inc_counter(rbx, xmm_curr_counter, 0x06, L__incCounter[k][5]);
4337 CTR_DoSix(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
4338 CTR_DoSix(pxor, xmm_key_tmp0); //PXOR with Round 0 key
4339
4340 //load two ROUND_KEYs at a time
4341 for (int i = 1; i < rounds[k]; ) {
4342 load_key(xmm_key_tmp1, key, (0x10 * i), xmm_key_shuf_mask);
4343 load_key(xmm_key_tmp0, key, (0x10 * (i+1)), xmm_key_shuf_mask);
4344 CTR_DoSix(aesenc, xmm_key_tmp1);
4345 i++;
4346 if (i != rounds[k]) {
4347 CTR_DoSix(aesenc, xmm_key_tmp0);
4348 } else {
4349 CTR_DoSix(aesenclast, xmm_key_tmp0);
4350 }
4351 i++;
4352 }
4353
4354 // get next PARALLEL_FACTOR blocks into xmm_result registers
4355 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4356 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
4357 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
4358 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
4359 __ movdqu(xmm_from4, Address(from, pos, Address::times_1, 4 * AESBlockSize));
4360 __ movdqu(xmm_from5, Address(from, pos, Address::times_1, 5 * AESBlockSize));
4361
4362 __ pxor(xmm_result0, xmm_from0);
4363 __ pxor(xmm_result1, xmm_from1);
4364 __ pxor(xmm_result2, xmm_from2);
4365 __ pxor(xmm_result3, xmm_from3);
4366 __ pxor(xmm_result4, xmm_from4);
4367 __ pxor(xmm_result5, xmm_from5);
4368
4369 // store 6 results into the next 64 bytes of output
4370 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4371 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
4372 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
4373 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
4374 __ movdqu(Address(to, pos, Address::times_1, 4 * AESBlockSize), xmm_result4);
4375 __ movdqu(Address(to, pos, Address::times_1, 5 * AESBlockSize), xmm_result5);
4376
4377 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
4378 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
4379 __ jmp(L_multiBlock_loopTop[k]);
4380
4381 // singleBlock starts here
4382 __ align(OptoLoopAlignment);
4383 __ BIND(L_singleBlockLoopTop[k]);
4384 __ cmpptr(len_reg, 0);
4385 __ jcc(Assembler::lessEqual, L_exit);
4386 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4387 __ movdqa(xmm_result0, xmm_curr_counter);
4388 inc_counter(rbx, xmm_curr_counter, 0x01, L__incCounter_single[k]);
4389 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
4390 __ pxor(xmm_result0, xmm_key_tmp0);
4391 for (int i = 1; i < rounds[k]; i++) {
4392 load_key(xmm_key_tmp0, key, (0x10 * i), xmm_key_shuf_mask);
4393 __ aesenc(xmm_result0, xmm_key_tmp0);
4394 }
4395 load_key(xmm_key_tmp0, key, (rounds[k] * 0x10), xmm_key_shuf_mask);
4396 __ aesenclast(xmm_result0, xmm_key_tmp0);
4397 __ cmpptr(len_reg, AESBlockSize);
4398 __ jcc(Assembler::less, L_processTail_insr[k]);
4399 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4400 __ pxor(xmm_result0, xmm_from0);
4401 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4402 __ addptr(pos, AESBlockSize);
4403 __ subptr(len_reg, AESBlockSize);
4404 __ jmp(L_singleBlockLoopTop[k]);
4405 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
4406 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
4407 __ testptr(len_reg, 8);
4408 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
4409 __ subptr(pos,8);
4410 __ pinsrq(xmm_from0, Address(from, pos), 0);
4411 __ BIND(L_processTail_4_insr[k]);
4412 __ testptr(len_reg, 4);
4413 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
4414 __ subptr(pos,4);
4415 __ pslldq(xmm_from0, 4);
4416 __ pinsrd(xmm_from0, Address(from, pos), 0);
4417 __ BIND(L_processTail_2_insr[k]);
4418 __ testptr(len_reg, 2);
4419 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
4420 __ subptr(pos, 2);
4421 __ pslldq(xmm_from0, 2);
4422 __ pinsrw(xmm_from0, Address(from, pos), 0);
4423 __ BIND(L_processTail_1_insr[k]);
4424 __ testptr(len_reg, 1);
4425 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
4426 __ subptr(pos, 1);
4427 __ pslldq(xmm_from0, 1);
4428 __ pinsrb(xmm_from0, Address(from, pos), 0);
4429 __ BIND(L_processTail_exit_insr[k]);
4430
4431 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
4432 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
4433
4434 __ testptr(len_reg, 8);
4435 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
4436 __ pextrq(Address(to, pos), xmm_result0, 0);
4437 __ psrldq(xmm_result0, 8);
4438 __ addptr(pos, 8);
4439 __ BIND(L_processTail_4_extr[k]);
4440 __ testptr(len_reg, 4);
4441 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
4442 __ pextrd(Address(to, pos), xmm_result0, 0);
4443 __ psrldq(xmm_result0, 4);
4444 __ addptr(pos, 4);
4445 __ BIND(L_processTail_2_extr[k]);
4446 __ testptr(len_reg, 2);
4447 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
4448 __ pextrw(Address(to, pos), xmm_result0, 0);
4449 __ psrldq(xmm_result0, 2);
4450 __ addptr(pos, 2);
4451 __ BIND(L_processTail_1_extr[k]);
4452 __ testptr(len_reg, 1);
4453 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
4454 __ pextrb(Address(to, pos), xmm_result0, 0);
4455
4456 __ BIND(L_processTail_exit_extr[k]);
4457 __ movl(Address(used_addr, 0), len_reg);
4458 __ jmp(L_exit);
4459
4460 }
4461
4462 __ BIND(L_exit);
4463 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
4464 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
4465 __ pop(rbx); // pop the saved RBX.
4466 #ifdef _WIN64
4467 __ movl(rax, len_mem);
4468 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
4469 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
4470 __ addptr(rsp, 2 * wordSize);
4471 #else
4472 __ pop(rax); // return 'len'
4473 #endif
4474 __ leave(); // required for proper stackwalking of RuntimeStub frame
4475 __ ret(0);
4476 return start;
4477 }
4478
4479 void roundDec(XMMRegister xmm_reg) {
4480 __ vaesdec(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4481 __ vaesdec(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4482 __ vaesdec(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4483 __ vaesdec(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4484 __ vaesdec(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4485 __ vaesdec(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4486 __ vaesdec(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4487 __ vaesdec(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4488 }
4489
4490 void roundDeclast(XMMRegister xmm_reg) {
4491 __ vaesdeclast(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4492 __ vaesdeclast(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4493 __ vaesdeclast(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4494 __ vaesdeclast(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4495 __ vaesdeclast(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4496 __ vaesdeclast(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4497 __ vaesdeclast(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4498 __ vaesdeclast(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4499 }
4500
4501 void ev_load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask = NULL) {
4502 __ movdqu(xmmdst, Address(key, offset));
4503 if (xmm_shuf_mask != NULL) {
4504 __ pshufb(xmmdst, xmm_shuf_mask);
4505 } else {
4506 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4507 }
4508 __ evshufi64x2(xmmdst, xmmdst, xmmdst, 0x0, Assembler::AVX_512bit);
4509
4510 }
4511
4512 address generate_cipherBlockChaining_decryptVectorAESCrypt() {
4513 assert(VM_Version::supports_avx512_vaes(), "need AES instructions and misaligned SSE support");
4514 __ align(CodeEntryAlignment);
4515 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
4516 address start = __ pc();
4517
4518 const Register from = c_rarg0; // source array address
4519 const Register to = c_rarg1; // destination array address
4520 const Register key = c_rarg2; // key array address
4521 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
4522 // and left with the results of the last encryption block
4523 #ifndef _WIN64
4524 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
4525 #else
4526 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4527 const Register len_reg = r11; // pick the volatile windows register
4528 #endif
4529
4530 Label Loop, Loop1, L_128, L_256, L_192, KEY_192, KEY_256, Loop2, Lcbc_dec_rem_loop,
4531 Lcbc_dec_rem_last, Lcbc_dec_ret, Lcbc_dec_rem, Lcbc_exit;
4532
4533 __ enter();
4534
4535 #ifdef _WIN64
4536 // on win64, fill len_reg from stack position
4537 __ movl(len_reg, len_mem);
4538 #else
4539 __ push(len_reg); // Save
4540 #endif
4541 __ push(rbx);
4542 __ vzeroupper();
4543
4544 // Temporary variable declaration for swapping key bytes
4545 const XMMRegister xmm_key_shuf_mask = xmm1;
4546 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4547
4548 // Calculate number of rounds from key size: 44 for 10-rounds, 52 for 12-rounds, 60 for 14-rounds
4549 const Register rounds = rbx;
4550 __ movl(rounds, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4551
4552 const XMMRegister IV = xmm0;
4553 // Load IV and broadcast value to 512-bits
4554 __ evbroadcasti64x2(IV, Address(rvec, 0), Assembler::AVX_512bit);
4555
4556 // Temporary variables for storing round keys
4557 const XMMRegister RK0 = xmm30;
4558 const XMMRegister RK1 = xmm9;
4559 const XMMRegister RK2 = xmm18;
4560 const XMMRegister RK3 = xmm19;
4561 const XMMRegister RK4 = xmm20;
4562 const XMMRegister RK5 = xmm21;
4563 const XMMRegister RK6 = xmm22;
4564 const XMMRegister RK7 = xmm23;
4565 const XMMRegister RK8 = xmm24;
4566 const XMMRegister RK9 = xmm25;
4567 const XMMRegister RK10 = xmm26;
4568
4569 // Load and shuffle key
4570 // the java expanded key ordering is rotated one position from what we want
4571 // so we start from 1*16 here and hit 0*16 last
4572 ev_load_key(RK1, key, 1 * 16, xmm_key_shuf_mask);
4573 ev_load_key(RK2, key, 2 * 16, xmm_key_shuf_mask);
4574 ev_load_key(RK3, key, 3 * 16, xmm_key_shuf_mask);
4575 ev_load_key(RK4, key, 4 * 16, xmm_key_shuf_mask);
4576 ev_load_key(RK5, key, 5 * 16, xmm_key_shuf_mask);
4577 ev_load_key(RK6, key, 6 * 16, xmm_key_shuf_mask);
4578 ev_load_key(RK7, key, 7 * 16, xmm_key_shuf_mask);
4579 ev_load_key(RK8, key, 8 * 16, xmm_key_shuf_mask);
4580 ev_load_key(RK9, key, 9 * 16, xmm_key_shuf_mask);
4581 ev_load_key(RK10, key, 10 * 16, xmm_key_shuf_mask);
4582 ev_load_key(RK0, key, 0*16, xmm_key_shuf_mask);
4583
4584 // Variables for storing source cipher text
4585 const XMMRegister S0 = xmm10;
4586 const XMMRegister S1 = xmm11;
4587 const XMMRegister S2 = xmm12;
4588 const XMMRegister S3 = xmm13;
4589 const XMMRegister S4 = xmm14;
4590 const XMMRegister S5 = xmm15;
4591 const XMMRegister S6 = xmm16;
4592 const XMMRegister S7 = xmm17;
4593
4594 // Variables for storing decrypted text
4595 const XMMRegister B0 = xmm1;
4596 const XMMRegister B1 = xmm2;
4597 const XMMRegister B2 = xmm3;
4598 const XMMRegister B3 = xmm4;
4599 const XMMRegister B4 = xmm5;
4600 const XMMRegister B5 = xmm6;
4601 const XMMRegister B6 = xmm7;
4602 const XMMRegister B7 = xmm8;
4603
4604 __ cmpl(rounds, 44);
4605 __ jcc(Assembler::greater, KEY_192);
4606 __ jmp(Loop);
4607
4608 __ BIND(KEY_192);
4609 const XMMRegister RK11 = xmm27;
4610 const XMMRegister RK12 = xmm28;
4611 ev_load_key(RK11, key, 11*16, xmm_key_shuf_mask);
4612 ev_load_key(RK12, key, 12*16, xmm_key_shuf_mask);
4613
4614 __ cmpl(rounds, 52);
4615 __ jcc(Assembler::greater, KEY_256);
4616 __ jmp(Loop);
4617
4618 __ BIND(KEY_256);
4619 const XMMRegister RK13 = xmm29;
4620 const XMMRegister RK14 = xmm31;
4621 ev_load_key(RK13, key, 13*16, xmm_key_shuf_mask);
4622 ev_load_key(RK14, key, 14*16, xmm_key_shuf_mask);
4623
4624 __ BIND(Loop);
4625 __ cmpl(len_reg, 512);
4626 __ jcc(Assembler::below, Lcbc_dec_rem);
4627 __ BIND(Loop1);
4628 __ subl(len_reg, 512);
4629 __ evmovdquq(S0, Address(from, 0 * 64), Assembler::AVX_512bit);
4630 __ evmovdquq(S1, Address(from, 1 * 64), Assembler::AVX_512bit);
4631 __ evmovdquq(S2, Address(from, 2 * 64), Assembler::AVX_512bit);
4632 __ evmovdquq(S3, Address(from, 3 * 64), Assembler::AVX_512bit);
4633 __ evmovdquq(S4, Address(from, 4 * 64), Assembler::AVX_512bit);
4634 __ evmovdquq(S5, Address(from, 5 * 64), Assembler::AVX_512bit);
4635 __ evmovdquq(S6, Address(from, 6 * 64), Assembler::AVX_512bit);
4636 __ evmovdquq(S7, Address(from, 7 * 64), Assembler::AVX_512bit);
4637 __ leaq(from, Address(from, 8 * 64));
4638
4639 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4640 __ evpxorq(B1, S1, RK1, Assembler::AVX_512bit);
4641 __ evpxorq(B2, S2, RK1, Assembler::AVX_512bit);
4642 __ evpxorq(B3, S3, RK1, Assembler::AVX_512bit);
4643 __ evpxorq(B4, S4, RK1, Assembler::AVX_512bit);
4644 __ evpxorq(B5, S5, RK1, Assembler::AVX_512bit);
4645 __ evpxorq(B6, S6, RK1, Assembler::AVX_512bit);
4646 __ evpxorq(B7, S7, RK1, Assembler::AVX_512bit);
4647
4648 __ evalignq(IV, S0, IV, 0x06);
4649 __ evalignq(S0, S1, S0, 0x06);
4650 __ evalignq(S1, S2, S1, 0x06);
4651 __ evalignq(S2, S3, S2, 0x06);
4652 __ evalignq(S3, S4, S3, 0x06);
4653 __ evalignq(S4, S5, S4, 0x06);
4654 __ evalignq(S5, S6, S5, 0x06);
4655 __ evalignq(S6, S7, S6, 0x06);
4656
4657 roundDec(RK2);
4658 roundDec(RK3);
4659 roundDec(RK4);
4660 roundDec(RK5);
4661 roundDec(RK6);
4662 roundDec(RK7);
4663 roundDec(RK8);
4664 roundDec(RK9);
4665 roundDec(RK10);
4666
4667 __ cmpl(rounds, 44);
4668 __ jcc(Assembler::belowEqual, L_128);
4669 roundDec(RK11);
4670 roundDec(RK12);
4671
4672 __ cmpl(rounds, 52);
4673 __ jcc(Assembler::belowEqual, L_192);
4674 roundDec(RK13);
4675 roundDec(RK14);
4676
4677 __ BIND(L_256);
4678 roundDeclast(RK0);
4679 __ jmp(Loop2);
4680
4681 __ BIND(L_128);
4682 roundDeclast(RK0);
4683 __ jmp(Loop2);
4684
4685 __ BIND(L_192);
4686 roundDeclast(RK0);
4687
4688 __ BIND(Loop2);
4689 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4690 __ evpxorq(B1, B1, S0, Assembler::AVX_512bit);
4691 __ evpxorq(B2, B2, S1, Assembler::AVX_512bit);
4692 __ evpxorq(B3, B3, S2, Assembler::AVX_512bit);
4693 __ evpxorq(B4, B4, S3, Assembler::AVX_512bit);
4694 __ evpxorq(B5, B5, S4, Assembler::AVX_512bit);
4695 __ evpxorq(B6, B6, S5, Assembler::AVX_512bit);
4696 __ evpxorq(B7, B7, S6, Assembler::AVX_512bit);
4697 __ evmovdquq(IV, S7, Assembler::AVX_512bit);
4698
4699 __ evmovdquq(Address(to, 0 * 64), B0, Assembler::AVX_512bit);
4700 __ evmovdquq(Address(to, 1 * 64), B1, Assembler::AVX_512bit);
4701 __ evmovdquq(Address(to, 2 * 64), B2, Assembler::AVX_512bit);
4702 __ evmovdquq(Address(to, 3 * 64), B3, Assembler::AVX_512bit);
4703 __ evmovdquq(Address(to, 4 * 64), B4, Assembler::AVX_512bit);
4704 __ evmovdquq(Address(to, 5 * 64), B5, Assembler::AVX_512bit);
4705 __ evmovdquq(Address(to, 6 * 64), B6, Assembler::AVX_512bit);
4706 __ evmovdquq(Address(to, 7 * 64), B7, Assembler::AVX_512bit);
4707 __ leaq(to, Address(to, 8 * 64));
4708 __ jmp(Loop);
4709
4710 __ BIND(Lcbc_dec_rem);
4711 __ evshufi64x2(IV, IV, IV, 0x03, Assembler::AVX_512bit);
4712
4713 __ BIND(Lcbc_dec_rem_loop);
4714 __ subl(len_reg, 16);
4715 __ jcc(Assembler::carrySet, Lcbc_dec_ret);
4716
4717 __ movdqu(S0, Address(from, 0));
4718 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4719 __ vaesdec(B0, B0, RK2, Assembler::AVX_512bit);
4720 __ vaesdec(B0, B0, RK3, Assembler::AVX_512bit);
4721 __ vaesdec(B0, B0, RK4, Assembler::AVX_512bit);
4722 __ vaesdec(B0, B0, RK5, Assembler::AVX_512bit);
4723 __ vaesdec(B0, B0, RK6, Assembler::AVX_512bit);
4724 __ vaesdec(B0, B0, RK7, Assembler::AVX_512bit);
4725 __ vaesdec(B0, B0, RK8, Assembler::AVX_512bit);
4726 __ vaesdec(B0, B0, RK9, Assembler::AVX_512bit);
4727 __ vaesdec(B0, B0, RK10, Assembler::AVX_512bit);
4728 __ cmpl(rounds, 44);
4729 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4730
4731 __ vaesdec(B0, B0, RK11, Assembler::AVX_512bit);
4732 __ vaesdec(B0, B0, RK12, Assembler::AVX_512bit);
4733 __ cmpl(rounds, 52);
4734 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4735
4736 __ vaesdec(B0, B0, RK13, Assembler::AVX_512bit);
4737 __ vaesdec(B0, B0, RK14, Assembler::AVX_512bit);
4738
4739 __ BIND(Lcbc_dec_rem_last);
4740 __ vaesdeclast(B0, B0, RK0, Assembler::AVX_512bit);
4741
4742 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4743 __ evmovdquq(IV, S0, Assembler::AVX_512bit);
4744 __ movdqu(Address(to, 0), B0);
4745 __ leaq(from, Address(from, 16));
4746 __ leaq(to, Address(to, 16));
4747 __ jmp(Lcbc_dec_rem_loop);
4748
4749 __ BIND(Lcbc_dec_ret);
4750 __ movdqu(Address(rvec, 0), IV);
4751
4752 // Zero out the round keys
4753 __ evpxorq(RK0, RK0, RK0, Assembler::AVX_512bit);
4754 __ evpxorq(RK1, RK1, RK1, Assembler::AVX_512bit);
4755 __ evpxorq(RK2, RK2, RK2, Assembler::AVX_512bit);
4756 __ evpxorq(RK3, RK3, RK3, Assembler::AVX_512bit);
4757 __ evpxorq(RK4, RK4, RK4, Assembler::AVX_512bit);
4758 __ evpxorq(RK5, RK5, RK5, Assembler::AVX_512bit);
4759 __ evpxorq(RK6, RK6, RK6, Assembler::AVX_512bit);
4760 __ evpxorq(RK7, RK7, RK7, Assembler::AVX_512bit);
4761 __ evpxorq(RK8, RK8, RK8, Assembler::AVX_512bit);
4762 __ evpxorq(RK9, RK9, RK9, Assembler::AVX_512bit);
4763 __ evpxorq(RK10, RK10, RK10, Assembler::AVX_512bit);
4764 __ cmpl(rounds, 44);
4765 __ jcc(Assembler::belowEqual, Lcbc_exit);
4766 __ evpxorq(RK11, RK11, RK11, Assembler::AVX_512bit);
4767 __ evpxorq(RK12, RK12, RK12, Assembler::AVX_512bit);
4768 __ cmpl(rounds, 52);
4769 __ jcc(Assembler::belowEqual, Lcbc_exit);
4770 __ evpxorq(RK13, RK13, RK13, Assembler::AVX_512bit);
4771 __ evpxorq(RK14, RK14, RK14, Assembler::AVX_512bit);
4772
4773 __ BIND(Lcbc_exit);
4774 __ pop(rbx);
4775 #ifdef _WIN64
4776 __ movl(rax, len_mem);
4777 #else
4778 __ pop(rax); // return length
4779 #endif
4780 __ leave(); // required for proper stackwalking of RuntimeStub frame
4781 __ ret(0);
4782 return start;
4783 }
4784
4785 // Polynomial x^128+x^127+x^126+x^121+1
4786 address ghash_polynomial_addr() {
4787 __ align(CodeEntryAlignment);
4788 StubCodeMark mark(this, "StubRoutines", "_ghash_poly_addr");
4789 address start = __ pc();
4790 __ emit_data64(0x0000000000000001, relocInfo::none);
4791 __ emit_data64(0xc200000000000000, relocInfo::none);
4792 return start;
4793 }
4794
4795 address ghash_shufflemask_addr() {
4796 __ align(CodeEntryAlignment);
4797 StubCodeMark mark(this, "StubRoutines", "_ghash_shuffmask_addr");
4798 address start = __ pc();
4799 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4800 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4801 return start;
4802 }
4803
4804 // Ghash single and multi block operations using AVX instructions
4805 address generate_avx_ghash_processBlocks() {
4806 __ align(CodeEntryAlignment);
4807
4808 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4809 address start = __ pc();
4810
4811 // arguments
4812 const Register state = c_rarg0;
4813 const Register htbl = c_rarg1;
4814 const Register data = c_rarg2;
4815 const Register blocks = c_rarg3;
4816 __ enter();
4817 // Save state before entering routine
4818 __ avx_ghash(state, htbl, data, blocks);
4819 __ leave(); // required for proper stackwalking of RuntimeStub frame
4820 __ ret(0);
4821 return start;
4822 }
4823
4824 // byte swap x86 long
4825 address generate_ghash_long_swap_mask() {
4826 __ align(CodeEntryAlignment);
4827 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
4828 address start = __ pc();
4829 __ emit_data64(0x0f0e0d0c0b0a0908, relocInfo::none );
4830 __ emit_data64(0x0706050403020100, relocInfo::none );
4831 return start;
4832 }
4833
4834 // byte swap x86 byte array
4835 address generate_ghash_byte_swap_mask() {
4836 __ align(CodeEntryAlignment);
4837 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
4838 address start = __ pc();
4839 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none );
4840 __ emit_data64(0x0001020304050607, relocInfo::none );
4841 return start;
4842 }
4843
4844 /* Single and multi-block ghash operations */
4845 address generate_ghash_processBlocks() {
4846 __ align(CodeEntryAlignment);
4847 Label L_ghash_loop, L_exit;
4848 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4849 address start = __ pc();
4850
4851 const Register state = c_rarg0;
4852 const Register subkeyH = c_rarg1;
4853 const Register data = c_rarg2;
4854 const Register blocks = c_rarg3;
4855
4856 const XMMRegister xmm_temp0 = xmm0;
4857 const XMMRegister xmm_temp1 = xmm1;
4858 const XMMRegister xmm_temp2 = xmm2;
4859 const XMMRegister xmm_temp3 = xmm3;
4860 const XMMRegister xmm_temp4 = xmm4;
4861 const XMMRegister xmm_temp5 = xmm5;
4862 const XMMRegister xmm_temp6 = xmm6;
4863 const XMMRegister xmm_temp7 = xmm7;
4864 const XMMRegister xmm_temp8 = xmm8;
4865 const XMMRegister xmm_temp9 = xmm9;
4866 const XMMRegister xmm_temp10 = xmm10;
4867
4868 __ enter();
4869
4870 __ movdqu(xmm_temp10, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
4871
4872 __ movdqu(xmm_temp0, Address(state, 0));
4873 __ pshufb(xmm_temp0, xmm_temp10);
4874
4875
4876 __ BIND(L_ghash_loop);
4877 __ movdqu(xmm_temp2, Address(data, 0));
4878 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
4879
4880 __ movdqu(xmm_temp1, Address(subkeyH, 0));
4881 __ pshufb(xmm_temp1, xmm_temp10);
4882
4883 __ pxor(xmm_temp0, xmm_temp2);
4884
4885 //
4886 // Multiply with the hash key
4887 //
4888 __ movdqu(xmm_temp3, xmm_temp0);
4889 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
4890 __ movdqu(xmm_temp4, xmm_temp0);
4891 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
4892
4893 __ movdqu(xmm_temp5, xmm_temp0);
4894 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
4895 __ movdqu(xmm_temp6, xmm_temp0);
4896 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
4897
4898 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
4899
4900 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
4901 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
4902 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
4903 __ pxor(xmm_temp3, xmm_temp5);
4904 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
4905 // of the carry-less multiplication of
4906 // xmm0 by xmm1.
4907
4908 // We shift the result of the multiplication by one bit position
4909 // to the left to cope for the fact that the bits are reversed.
4910 __ movdqu(xmm_temp7, xmm_temp3);
4911 __ movdqu(xmm_temp8, xmm_temp6);
4912 __ pslld(xmm_temp3, 1);
4913 __ pslld(xmm_temp6, 1);
4914 __ psrld(xmm_temp7, 31);
4915 __ psrld(xmm_temp8, 31);
4916 __ movdqu(xmm_temp9, xmm_temp7);
4917 __ pslldq(xmm_temp8, 4);
4918 __ pslldq(xmm_temp7, 4);
4919 __ psrldq(xmm_temp9, 12);
4920 __ por(xmm_temp3, xmm_temp7);
4921 __ por(xmm_temp6, xmm_temp8);
4922 __ por(xmm_temp6, xmm_temp9);
4923
4924 //
4925 // First phase of the reduction
4926 //
4927 // Move xmm3 into xmm7, xmm8, xmm9 in order to perform the shifts
4928 // independently.
4929 __ movdqu(xmm_temp7, xmm_temp3);
4930 __ movdqu(xmm_temp8, xmm_temp3);
4931 __ movdqu(xmm_temp9, xmm_temp3);
4932 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
4933 __ pslld(xmm_temp8, 30); // packed right shift shifting << 30
4934 __ pslld(xmm_temp9, 25); // packed right shift shifting << 25
4935 __ pxor(xmm_temp7, xmm_temp8); // xor the shifted versions
4936 __ pxor(xmm_temp7, xmm_temp9);
4937 __ movdqu(xmm_temp8, xmm_temp7);
4938 __ pslldq(xmm_temp7, 12);
4939 __ psrldq(xmm_temp8, 4);
4940 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
4941
4942 //
4943 // Second phase of the reduction
4944 //
4945 // Make 3 copies of xmm3 in xmm2, xmm4, xmm5 for doing these
4946 // shift operations.
4947 __ movdqu(xmm_temp2, xmm_temp3);
4948 __ movdqu(xmm_temp4, xmm_temp3);
4949 __ movdqu(xmm_temp5, xmm_temp3);
4950 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
4951 __ psrld(xmm_temp4, 2); // packed left shifting >> 2
4952 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
4953 __ pxor(xmm_temp2, xmm_temp4); // xor the shifted versions
4954 __ pxor(xmm_temp2, xmm_temp5);
4955 __ pxor(xmm_temp2, xmm_temp8);
4956 __ pxor(xmm_temp3, xmm_temp2);
4957 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
4958
4959 __ decrement(blocks);
4960 __ jcc(Assembler::zero, L_exit);
4961 __ movdqu(xmm_temp0, xmm_temp6);
4962 __ addptr(data, 16);
4963 __ jmp(L_ghash_loop);
4964
4965 __ BIND(L_exit);
4966 __ pshufb(xmm_temp6, xmm_temp10); // Byte swap 16-byte result
4967 __ movdqu(Address(state, 0), xmm_temp6); // store the result
4968 __ leave();
4969 __ ret(0);
4970 return start;
4971 }
4972
4973 //base64 character set
4974 address base64_charset_addr() {
4975 __ align(CodeEntryAlignment);
4976 StubCodeMark mark(this, "StubRoutines", "base64_charset");
4977 address start = __ pc();
4978 __ emit_data64(0x0000004200000041, relocInfo::none);
4979 __ emit_data64(0x0000004400000043, relocInfo::none);
4980 __ emit_data64(0x0000004600000045, relocInfo::none);
4981 __ emit_data64(0x0000004800000047, relocInfo::none);
4982 __ emit_data64(0x0000004a00000049, relocInfo::none);
4983 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4984 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4985 __ emit_data64(0x000000500000004f, relocInfo::none);
4986 __ emit_data64(0x0000005200000051, relocInfo::none);
4987 __ emit_data64(0x0000005400000053, relocInfo::none);
4988 __ emit_data64(0x0000005600000055, relocInfo::none);
4989 __ emit_data64(0x0000005800000057, relocInfo::none);
4990 __ emit_data64(0x0000005a00000059, relocInfo::none);
4991 __ emit_data64(0x0000006200000061, relocInfo::none);
4992 __ emit_data64(0x0000006400000063, relocInfo::none);
4993 __ emit_data64(0x0000006600000065, relocInfo::none);
4994 __ emit_data64(0x0000006800000067, relocInfo::none);
4995 __ emit_data64(0x0000006a00000069, relocInfo::none);
4996 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4997 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4998 __ emit_data64(0x000000700000006f, relocInfo::none);
4999 __ emit_data64(0x0000007200000071, relocInfo::none);
5000 __ emit_data64(0x0000007400000073, relocInfo::none);
5001 __ emit_data64(0x0000007600000075, relocInfo::none);
5002 __ emit_data64(0x0000007800000077, relocInfo::none);
5003 __ emit_data64(0x0000007a00000079, relocInfo::none);
5004 __ emit_data64(0x0000003100000030, relocInfo::none);
5005 __ emit_data64(0x0000003300000032, relocInfo::none);
5006 __ emit_data64(0x0000003500000034, relocInfo::none);
5007 __ emit_data64(0x0000003700000036, relocInfo::none);
5008 __ emit_data64(0x0000003900000038, relocInfo::none);
5009 __ emit_data64(0x0000002f0000002b, relocInfo::none);
5010 return start;
5011 }
5012
5013 //base64 url character set
5014 address base64url_charset_addr() {
5015 __ align(CodeEntryAlignment);
5016 StubCodeMark mark(this, "StubRoutines", "base64url_charset");
5017 address start = __ pc();
5018 __ emit_data64(0x0000004200000041, relocInfo::none);
5019 __ emit_data64(0x0000004400000043, relocInfo::none);
5020 __ emit_data64(0x0000004600000045, relocInfo::none);
5021 __ emit_data64(0x0000004800000047, relocInfo::none);
5022 __ emit_data64(0x0000004a00000049, relocInfo::none);
5023 __ emit_data64(0x0000004c0000004b, relocInfo::none);
5024 __ emit_data64(0x0000004e0000004d, relocInfo::none);
5025 __ emit_data64(0x000000500000004f, relocInfo::none);
5026 __ emit_data64(0x0000005200000051, relocInfo::none);
5027 __ emit_data64(0x0000005400000053, relocInfo::none);
5028 __ emit_data64(0x0000005600000055, relocInfo::none);
5029 __ emit_data64(0x0000005800000057, relocInfo::none);
5030 __ emit_data64(0x0000005a00000059, relocInfo::none);
5031 __ emit_data64(0x0000006200000061, relocInfo::none);
5032 __ emit_data64(0x0000006400000063, relocInfo::none);
5033 __ emit_data64(0x0000006600000065, relocInfo::none);
5034 __ emit_data64(0x0000006800000067, relocInfo::none);
5035 __ emit_data64(0x0000006a00000069, relocInfo::none);
5036 __ emit_data64(0x0000006c0000006b, relocInfo::none);
5037 __ emit_data64(0x0000006e0000006d, relocInfo::none);
5038 __ emit_data64(0x000000700000006f, relocInfo::none);
5039 __ emit_data64(0x0000007200000071, relocInfo::none);
5040 __ emit_data64(0x0000007400000073, relocInfo::none);
5041 __ emit_data64(0x0000007600000075, relocInfo::none);
5042 __ emit_data64(0x0000007800000077, relocInfo::none);
5043 __ emit_data64(0x0000007a00000079, relocInfo::none);
5044 __ emit_data64(0x0000003100000030, relocInfo::none);
5045 __ emit_data64(0x0000003300000032, relocInfo::none);
5046 __ emit_data64(0x0000003500000034, relocInfo::none);
5047 __ emit_data64(0x0000003700000036, relocInfo::none);
5048 __ emit_data64(0x0000003900000038, relocInfo::none);
5049 __ emit_data64(0x0000005f0000002d, relocInfo::none);
5050
5051 return start;
5052 }
5053
5054 address base64_bswap_mask_addr() {
5055 __ align(CodeEntryAlignment);
5056 StubCodeMark mark(this, "StubRoutines", "bswap_mask_base64");
5057 address start = __ pc();
5058 __ emit_data64(0x0504038002010080, relocInfo::none);
5059 __ emit_data64(0x0b0a098008070680, relocInfo::none);
5060 __ emit_data64(0x0908078006050480, relocInfo::none);
5061 __ emit_data64(0x0f0e0d800c0b0a80, relocInfo::none);
5062 __ emit_data64(0x0605048003020180, relocInfo::none);
5063 __ emit_data64(0x0c0b0a8009080780, relocInfo::none);
5064 __ emit_data64(0x0504038002010080, relocInfo::none);
5065 __ emit_data64(0x0b0a098008070680, relocInfo::none);
5066
5067 return start;
5068 }
5069
5070 address base64_right_shift_mask_addr() {
5071 __ align(CodeEntryAlignment);
5072 StubCodeMark mark(this, "StubRoutines", "right_shift_mask");
5073 address start = __ pc();
5074 __ emit_data64(0x0006000400020000, relocInfo::none);
5075 __ emit_data64(0x0006000400020000, relocInfo::none);
5076 __ emit_data64(0x0006000400020000, relocInfo::none);
5077 __ emit_data64(0x0006000400020000, relocInfo::none);
5078 __ emit_data64(0x0006000400020000, relocInfo::none);
5079 __ emit_data64(0x0006000400020000, relocInfo::none);
5080 __ emit_data64(0x0006000400020000, relocInfo::none);
5081 __ emit_data64(0x0006000400020000, relocInfo::none);
5082
5083 return start;
5084 }
5085
5086 address base64_left_shift_mask_addr() {
5087 __ align(CodeEntryAlignment);
5088 StubCodeMark mark(this, "StubRoutines", "left_shift_mask");
5089 address start = __ pc();
5090 __ emit_data64(0x0000000200040000, relocInfo::none);
5091 __ emit_data64(0x0000000200040000, relocInfo::none);
5092 __ emit_data64(0x0000000200040000, relocInfo::none);
5093 __ emit_data64(0x0000000200040000, relocInfo::none);
5094 __ emit_data64(0x0000000200040000, relocInfo::none);
5095 __ emit_data64(0x0000000200040000, relocInfo::none);
5096 __ emit_data64(0x0000000200040000, relocInfo::none);
5097 __ emit_data64(0x0000000200040000, relocInfo::none);
5098
5099 return start;
5100 }
5101
5102 address base64_and_mask_addr() {
5103 __ align(CodeEntryAlignment);
5104 StubCodeMark mark(this, "StubRoutines", "and_mask");
5105 address start = __ pc();
5106 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5107 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5108 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5109 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5110 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5111 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5112 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5113 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5114 return start;
5115 }
5116
5117 address base64_gather_mask_addr() {
5118 __ align(CodeEntryAlignment);
5119 StubCodeMark mark(this, "StubRoutines", "gather_mask");
5120 address start = __ pc();
5121 __ emit_data64(0xffffffffffffffff, relocInfo::none);
5122 return start;
5123 }
5124
5125 // Code for generating Base64 encoding.
5126 // Intrinsic function prototype in Base64.java:
5127 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL) {
5128 address generate_base64_encodeBlock() {
5129 __ align(CodeEntryAlignment);
5130 StubCodeMark mark(this, "StubRoutines", "implEncode");
5131 address start = __ pc();
5132 __ enter();
5133
5134 // Save callee-saved registers before using them
5135 __ push(r12);
5136 __ push(r13);
5137 __ push(r14);
5138 __ push(r15);
5139
5140 // arguments
5141 const Register source = c_rarg0; // Source Array
5142 const Register start_offset = c_rarg1; // start offset
5143 const Register end_offset = c_rarg2; // end offset
5144 const Register dest = c_rarg3; // destination array
5145
5146 #ifndef _WIN64
5147 const Register dp = c_rarg4; // Position for writing to dest array
5148 const Register isURL = c_rarg5;// Base64 or URL character set
5149 #else
5150 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
5151 const Address isURL_mem(rbp, 7 * wordSize);
5152 const Register isURL = r10; // pick the volatile windows register
5153 const Register dp = r12;
5154 __ movl(dp, dp_mem);
5155 __ movl(isURL, isURL_mem);
5156 #endif
5157
5158 const Register length = r14;
5159 Label L_process80, L_process32, L_process3, L_exit, L_processdata;
5160
5161 // calculate length from offsets
5162 __ movl(length, end_offset);
5163 __ subl(length, start_offset);
5164 __ cmpl(length, 0);
5165 __ jcc(Assembler::lessEqual, L_exit);
5166
5167 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_charset_addr()));
5168 // check if base64 charset(isURL=0) or base64 url charset(isURL=1) needs to be loaded
5169 __ cmpl(isURL, 0);
5170 __ jcc(Assembler::equal, L_processdata);
5171 __ lea(r11, ExternalAddress(StubRoutines::x86::base64url_charset_addr()));
5172
5173 // load masks required for encoding data
5174 __ BIND(L_processdata);
5175 __ movdqu(xmm16, ExternalAddress(StubRoutines::x86::base64_gather_mask_addr()));
5176 // Set 64 bits of K register.
5177 __ evpcmpeqb(k3, xmm16, xmm16, Assembler::AVX_512bit);
5178 __ evmovdquq(xmm12, ExternalAddress(StubRoutines::x86::base64_bswap_mask_addr()), Assembler::AVX_256bit, r13);
5179 __ evmovdquq(xmm13, ExternalAddress(StubRoutines::x86::base64_right_shift_mask_addr()), Assembler::AVX_512bit, r13);
5180 __ evmovdquq(xmm14, ExternalAddress(StubRoutines::x86::base64_left_shift_mask_addr()), Assembler::AVX_512bit, r13);
5181 __ evmovdquq(xmm15, ExternalAddress(StubRoutines::x86::base64_and_mask_addr()), Assembler::AVX_512bit, r13);
5182
5183 // Vector Base64 implementation, producing 96 bytes of encoded data
5184 __ BIND(L_process80);
5185 __ cmpl(length, 80);
5186 __ jcc(Assembler::below, L_process32);
5187 __ evmovdquq(xmm0, Address(source, start_offset, Address::times_1, 0), Assembler::AVX_256bit);
5188 __ evmovdquq(xmm1, Address(source, start_offset, Address::times_1, 24), Assembler::AVX_256bit);
5189 __ evmovdquq(xmm2, Address(source, start_offset, Address::times_1, 48), Assembler::AVX_256bit);
5190
5191 //permute the input data in such a manner that we have continuity of the source
5192 __ vpermq(xmm3, xmm0, 148, Assembler::AVX_256bit);
5193 __ vpermq(xmm4, xmm1, 148, Assembler::AVX_256bit);
5194 __ vpermq(xmm5, xmm2, 148, Assembler::AVX_256bit);
5195
5196 //shuffle input and group 3 bytes of data and to it add 0 as the 4th byte.
5197 //we can deal with 12 bytes at a time in a 128 bit register
5198 __ vpshufb(xmm3, xmm3, xmm12, Assembler::AVX_256bit);
5199 __ vpshufb(xmm4, xmm4, xmm12, Assembler::AVX_256bit);
5200 __ vpshufb(xmm5, xmm5, xmm12, Assembler::AVX_256bit);
5201
5202 //convert byte to word. Each 128 bit register will have 6 bytes for processing
5203 __ vpmovzxbw(xmm3, xmm3, Assembler::AVX_512bit);
5204 __ vpmovzxbw(xmm4, xmm4, Assembler::AVX_512bit);
5205 __ vpmovzxbw(xmm5, xmm5, Assembler::AVX_512bit);
5206
5207 // Extract bits in the following pattern 6, 4+2, 2+4, 6 to convert 3, 8 bit numbers to 4, 6 bit numbers
5208 __ evpsrlvw(xmm0, xmm3, xmm13, Assembler::AVX_512bit);
5209 __ evpsrlvw(xmm1, xmm4, xmm13, Assembler::AVX_512bit);
5210 __ evpsrlvw(xmm2, xmm5, xmm13, Assembler::AVX_512bit);
5211
5212 __ evpsllvw(xmm3, xmm3, xmm14, Assembler::AVX_512bit);
5213 __ evpsllvw(xmm4, xmm4, xmm14, Assembler::AVX_512bit);
5214 __ evpsllvw(xmm5, xmm5, xmm14, Assembler::AVX_512bit);
5215
5216 __ vpsrlq(xmm0, xmm0, 8, Assembler::AVX_512bit);
5217 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
5218 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
5219
5220 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5221 __ vpsllq(xmm4, xmm4, 8, Assembler::AVX_512bit);
5222 __ vpsllq(xmm5, xmm5, 8, Assembler::AVX_512bit);
5223
5224 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
5225 __ vpandq(xmm4, xmm4, xmm15, Assembler::AVX_512bit);
5226 __ vpandq(xmm5, xmm5, xmm15, Assembler::AVX_512bit);
5227
5228 // Get the final 4*6 bits base64 encoding
5229 __ vporq(xmm3, xmm3, xmm0, Assembler::AVX_512bit);
5230 __ vporq(xmm4, xmm4, xmm1, Assembler::AVX_512bit);
5231 __ vporq(xmm5, xmm5, xmm2, Assembler::AVX_512bit);
5232
5233 // Shift
5234 __ vpsrlq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5235 __ vpsrlq(xmm4, xmm4, 8, Assembler::AVX_512bit);
5236 __ vpsrlq(xmm5, xmm5, 8, Assembler::AVX_512bit);
5237
5238 // look up 6 bits in the base64 character set to fetch the encoding
5239 // we are converting word to dword as gather instructions need dword indices for looking up encoding
5240 __ vextracti64x4(xmm6, xmm3, 0);
5241 __ vpmovzxwd(xmm0, xmm6, Assembler::AVX_512bit);
5242 __ vextracti64x4(xmm6, xmm3, 1);
5243 __ vpmovzxwd(xmm1, xmm6, Assembler::AVX_512bit);
5244
5245 __ vextracti64x4(xmm6, xmm4, 0);
5246 __ vpmovzxwd(xmm2, xmm6, Assembler::AVX_512bit);
5247 __ vextracti64x4(xmm6, xmm4, 1);
5248 __ vpmovzxwd(xmm3, xmm6, Assembler::AVX_512bit);
5249
5250 __ vextracti64x4(xmm4, xmm5, 0);
5251 __ vpmovzxwd(xmm6, xmm4, Assembler::AVX_512bit);
5252
5253 __ vextracti64x4(xmm4, xmm5, 1);
5254 __ vpmovzxwd(xmm7, xmm4, Assembler::AVX_512bit);
5255
5256 __ kmovql(k2, k3);
5257 __ evpgatherdd(xmm4, k2, Address(r11, xmm0, Address::times_4, 0), Assembler::AVX_512bit);
5258 __ kmovql(k2, k3);
5259 __ evpgatherdd(xmm5, k2, Address(r11, xmm1, Address::times_4, 0), Assembler::AVX_512bit);
5260 __ kmovql(k2, k3);
5261 __ evpgatherdd(xmm8, k2, Address(r11, xmm2, Address::times_4, 0), Assembler::AVX_512bit);
5262 __ kmovql(k2, k3);
5263 __ evpgatherdd(xmm9, k2, Address(r11, xmm3, Address::times_4, 0), Assembler::AVX_512bit);
5264 __ kmovql(k2, k3);
5265 __ evpgatherdd(xmm10, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
5266 __ kmovql(k2, k3);
5267 __ evpgatherdd(xmm11, k2, Address(r11, xmm7, Address::times_4, 0), Assembler::AVX_512bit);
5268
5269 //Down convert dword to byte. Final output is 16*6 = 96 bytes long
5270 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm4, Assembler::AVX_512bit);
5271 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm5, Assembler::AVX_512bit);
5272 __ evpmovdb(Address(dest, dp, Address::times_1, 32), xmm8, Assembler::AVX_512bit);
5273 __ evpmovdb(Address(dest, dp, Address::times_1, 48), xmm9, Assembler::AVX_512bit);
5274 __ evpmovdb(Address(dest, dp, Address::times_1, 64), xmm10, Assembler::AVX_512bit);
5275 __ evpmovdb(Address(dest, dp, Address::times_1, 80), xmm11, Assembler::AVX_512bit);
5276
5277 __ addq(dest, 96);
5278 __ addq(source, 72);
5279 __ subq(length, 72);
5280 __ jmp(L_process80);
5281
5282 // Vector Base64 implementation generating 32 bytes of encoded data
5283 __ BIND(L_process32);
5284 __ cmpl(length, 32);
5285 __ jcc(Assembler::below, L_process3);
5286 __ evmovdquq(xmm0, Address(source, start_offset), Assembler::AVX_256bit);
5287 __ vpermq(xmm0, xmm0, 148, Assembler::AVX_256bit);
5288 __ vpshufb(xmm6, xmm0, xmm12, Assembler::AVX_256bit);
5289 __ vpmovzxbw(xmm6, xmm6, Assembler::AVX_512bit);
5290 __ evpsrlvw(xmm2, xmm6, xmm13, Assembler::AVX_512bit);
5291 __ evpsllvw(xmm3, xmm6, xmm14, Assembler::AVX_512bit);
5292
5293 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
5294 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5295 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
5296 __ vporq(xmm1, xmm2, xmm3, Assembler::AVX_512bit);
5297 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
5298 __ vextracti64x4(xmm9, xmm1, 0);
5299 __ vpmovzxwd(xmm6, xmm9, Assembler::AVX_512bit);
5300 __ vextracti64x4(xmm9, xmm1, 1);
5301 __ vpmovzxwd(xmm5, xmm9, Assembler::AVX_512bit);
5302 __ kmovql(k2, k3);
5303 __ evpgatherdd(xmm8, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
5304 __ kmovql(k2, k3);
5305 __ evpgatherdd(xmm10, k2, Address(r11, xmm5, Address::times_4, 0), Assembler::AVX_512bit);
5306 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm8, Assembler::AVX_512bit);
5307 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm10, Assembler::AVX_512bit);
5308 __ subq(length, 24);
5309 __ addq(dest, 32);
5310 __ addq(source, 24);
5311 __ jmp(L_process32);
5312
5313 // Scalar data processing takes 3 bytes at a time and produces 4 bytes of encoded data
5314 /* This code corresponds to the scalar version of the following snippet in Base64.java
5315 ** int bits = (src[sp0++] & 0xff) << 16 |(src[sp0++] & 0xff) << 8 |(src[sp0++] & 0xff);
5316 ** dst[dp0++] = (byte)base64[(bits >> > 18) & 0x3f];
5317 ** dst[dp0++] = (byte)base64[(bits >> > 12) & 0x3f];
5318 ** dst[dp0++] = (byte)base64[(bits >> > 6) & 0x3f];
5319 ** dst[dp0++] = (byte)base64[bits & 0x3f];*/
5320 __ BIND(L_process3);
5321 __ cmpl(length, 3);
5322 __ jcc(Assembler::below, L_exit);
5323 // Read 1 byte at a time
5324 __ movzbl(rax, Address(source, start_offset));
5325 __ shll(rax, 0x10);
5326 __ movl(r15, rax);
5327 __ movzbl(rax, Address(source, start_offset, Address::times_1, 1));
5328 __ shll(rax, 0x8);
5329 __ movzwl(rax, rax);
5330 __ orl(r15, rax);
5331 __ movzbl(rax, Address(source, start_offset, Address::times_1, 2));
5332 __ orl(rax, r15);
5333 // Save 3 bytes read in r15
5334 __ movl(r15, rax);
5335 __ shrl(rax, 0x12);
5336 __ andl(rax, 0x3f);
5337 // rax contains the index, r11 contains base64 lookup table
5338 __ movb(rax, Address(r11, rax, Address::times_4));
5339 // Write the encoded byte to destination
5340 __ movb(Address(dest, dp, Address::times_1, 0), rax);
5341 __ movl(rax, r15);
5342 __ shrl(rax, 0xc);
5343 __ andl(rax, 0x3f);
5344 __ movb(rax, Address(r11, rax, Address::times_4));
5345 __ movb(Address(dest, dp, Address::times_1, 1), rax);
5346 __ movl(rax, r15);
5347 __ shrl(rax, 0x6);
5348 __ andl(rax, 0x3f);
5349 __ movb(rax, Address(r11, rax, Address::times_4));
5350 __ movb(Address(dest, dp, Address::times_1, 2), rax);
5351 __ movl(rax, r15);
5352 __ andl(rax, 0x3f);
5353 __ movb(rax, Address(r11, rax, Address::times_4));
5354 __ movb(Address(dest, dp, Address::times_1, 3), rax);
5355 __ subl(length, 3);
5356 __ addq(dest, 4);
5357 __ addq(source, 3);
5358 __ jmp(L_process3);
5359 __ BIND(L_exit);
5360 __ pop(r15);
5361 __ pop(r14);
5362 __ pop(r13);
5363 __ pop(r12);
5364 __ leave();
5365 __ ret(0);
5366 return start;
5367 }
5368
5369 /**
5370 * Arguments:
5371 *
5372 * Inputs:
5373 * c_rarg0 - int crc
5374 * c_rarg1 - byte* buf
5375 * c_rarg2 - int length
5376 *
5377 * Ouput:
5378 * rax - int crc result
5379 */
5380 address generate_updateBytesCRC32() {
5381 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
5382
5383 __ align(CodeEntryAlignment);
5384 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
5385
5386 address start = __ pc();
5387 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5388 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5389 // rscratch1: r10
5390 const Register crc = c_rarg0; // crc
5391 const Register buf = c_rarg1; // source java byte array address
5392 const Register len = c_rarg2; // length
5393 const Register table = c_rarg3; // crc_table address (reuse register)
5394 const Register tmp1 = r11;
5395 const Register tmp2 = r10;
5396 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax);
5397
5398 BLOCK_COMMENT("Entry:");
5399 __ enter(); // required for proper stackwalking of RuntimeStub frame
5400
5401 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
5402 VM_Version::supports_avx512bw() &&
5403 VM_Version::supports_avx512vl()) {
5404 __ kernel_crc32_avx512(crc, buf, len, table, tmp1, tmp2);
5405 } else {
5406 __ kernel_crc32(crc, buf, len, table, tmp1);
5407 }
5408
5409 __ movl(rax, crc);
5410 __ vzeroupper();
5411 __ leave(); // required for proper stackwalking of RuntimeStub frame
5412 __ ret(0);
5413
5414 return start;
5415 }
5416
5417 /**
5418 * Arguments:
5419 *
5420 * Inputs:
5421 * c_rarg0 - int crc
5422 * c_rarg1 - byte* buf
5423 * c_rarg2 - long length
5424 * c_rarg3 - table_start - optional (present only when doing a library_call,
5425 * not used by x86 algorithm)
5426 *
5427 * Ouput:
5428 * rax - int crc result
5429 */
5430 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
5431 assert(UseCRC32CIntrinsics, "need SSE4_2");
5432 __ align(CodeEntryAlignment);
5433 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
5434 address start = __ pc();
5435 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
5436 //Windows RCX RDX R8 R9 none none XMM0..XMM3
5437 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
5438 const Register crc = c_rarg0; // crc
5439 const Register buf = c_rarg1; // source java byte array address
5440 const Register len = c_rarg2; // length
5441 const Register a = rax;
5442 const Register j = r9;
5443 const Register k = r10;
5444 const Register l = r11;
5445 #ifdef _WIN64
5446 const Register y = rdi;
5447 const Register z = rsi;
5448 #else
5449 const Register y = rcx;
5450 const Register z = r8;
5451 #endif
5452 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
5453
5454 BLOCK_COMMENT("Entry:");
5455 __ enter(); // required for proper stackwalking of RuntimeStub frame
5456 #ifdef _WIN64
5457 __ push(y);
5458 __ push(z);
5459 #endif
5460 __ crc32c_ipl_alg2_alt2(crc, buf, len,
5461 a, j, k,
5462 l, y, z,
5463 c_farg0, c_farg1, c_farg2,
5464 is_pclmulqdq_supported);
5465 __ movl(rax, crc);
5466 #ifdef _WIN64
5467 __ pop(z);
5468 __ pop(y);
5469 #endif
5470 __ vzeroupper();
5471 __ leave(); // required for proper stackwalking of RuntimeStub frame
5472 __ ret(0);
5473
5474 return start;
5475 }
5476
5477 /**
5478 * Arguments:
5479 *
5480 * Input:
5481 * c_rarg0 - x address
5482 * c_rarg1 - x length
5483 * c_rarg2 - y address
5484 * c_rarg3 - y length
5485 * not Win64
5486 * c_rarg4 - z address
5487 * c_rarg5 - z length
5488 * Win64
5489 * rsp+40 - z address
5490 * rsp+48 - z length
5491 */
5492 address generate_multiplyToLen() {
5493 __ align(CodeEntryAlignment);
5494 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
5495
5496 address start = __ pc();
5497 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5498 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5499 const Register x = rdi;
5500 const Register xlen = rax;
5501 const Register y = rsi;
5502 const Register ylen = rcx;
5503 const Register z = r8;
5504 const Register zlen = r11;
5505
5506 // Next registers will be saved on stack in multiply_to_len().
5507 const Register tmp1 = r12;
5508 const Register tmp2 = r13;
5509 const Register tmp3 = r14;
5510 const Register tmp4 = r15;
5511 const Register tmp5 = rbx;
5512
5513 BLOCK_COMMENT("Entry:");
5514 __ enter(); // required for proper stackwalking of RuntimeStub frame
5515
5516 #ifndef _WIN64
5517 __ movptr(zlen, r9); // Save r9 in r11 - zlen
5518 #endif
5519 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
5520 // ylen => rcx, z => r8, zlen => r11
5521 // r9 and r10 may be used to save non-volatile registers
5522 #ifdef _WIN64
5523 // last 2 arguments (#4, #5) are on stack on Win64
5524 __ movptr(z, Address(rsp, 6 * wordSize));
5525 __ movptr(zlen, Address(rsp, 7 * wordSize));
5526 #endif
5527
5528 __ movptr(xlen, rsi);
5529 __ movptr(y, rdx);
5530 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
5531
5532 restore_arg_regs();
5533
5534 __ leave(); // required for proper stackwalking of RuntimeStub frame
5535 __ ret(0);
5536
5537 return start;
5538 }
5539
5540 /**
5541 * Arguments:
5542 *
5543 * Input:
5544 * c_rarg0 - obja address
5545 * c_rarg1 - objb address
5546 * c_rarg3 - length length
5547 * c_rarg4 - scale log2_array_indxscale
5548 *
5549 * Output:
5550 * rax - int >= mismatched index, < 0 bitwise complement of tail
5551 */
5552 address generate_vectorizedMismatch() {
5553 __ align(CodeEntryAlignment);
5554 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
5555 address start = __ pc();
5556
5557 BLOCK_COMMENT("Entry:");
5558 __ enter();
5559
5560 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5561 const Register scale = c_rarg0; //rcx, will exchange with r9
5562 const Register objb = c_rarg1; //rdx
5563 const Register length = c_rarg2; //r8
5564 const Register obja = c_rarg3; //r9
5565 __ xchgq(obja, scale); //now obja and scale contains the correct contents
5566
5567 const Register tmp1 = r10;
5568 const Register tmp2 = r11;
5569 #endif
5570 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5571 const Register obja = c_rarg0; //U:rdi
5572 const Register objb = c_rarg1; //U:rsi
5573 const Register length = c_rarg2; //U:rdx
5574 const Register scale = c_rarg3; //U:rcx
5575 const Register tmp1 = r8;
5576 const Register tmp2 = r9;
5577 #endif
5578 const Register result = rax; //return value
5579 const XMMRegister vec0 = xmm0;
5580 const XMMRegister vec1 = xmm1;
5581 const XMMRegister vec2 = xmm2;
5582
5583 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
5584
5585 __ vzeroupper();
5586 __ leave();
5587 __ ret(0);
5588
5589 return start;
5590 }
5591
5592 /**
5593 * Arguments:
5594 *
5595 // Input:
5596 // c_rarg0 - x address
5597 // c_rarg1 - x length
5598 // c_rarg2 - z address
5599 // c_rarg3 - z lenth
5600 *
5601 */
5602 address generate_squareToLen() {
5603
5604 __ align(CodeEntryAlignment);
5605 StubCodeMark mark(this, "StubRoutines", "squareToLen");
5606
5607 address start = __ pc();
5608 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5609 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
5610 const Register x = rdi;
5611 const Register len = rsi;
5612 const Register z = r8;
5613 const Register zlen = rcx;
5614
5615 const Register tmp1 = r12;
5616 const Register tmp2 = r13;
5617 const Register tmp3 = r14;
5618 const Register tmp4 = r15;
5619 const Register tmp5 = rbx;
5620
5621 BLOCK_COMMENT("Entry:");
5622 __ enter(); // required for proper stackwalking of RuntimeStub frame
5623
5624 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
5625 // zlen => rcx
5626 // r9 and r10 may be used to save non-volatile registers
5627 __ movptr(r8, rdx);
5628 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5629
5630 restore_arg_regs();
5631
5632 __ leave(); // required for proper stackwalking of RuntimeStub frame
5633 __ ret(0);
5634
5635 return start;
5636 }
5637
5638 address generate_method_entry_barrier() {
5639 __ align(CodeEntryAlignment);
5640 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
5641
5642 Label deoptimize_label;
5643
5644 address start = __ pc();
5645
5646 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
5647
5648 BLOCK_COMMENT("Entry:");
5649 __ enter(); // save rbp
5650
5651 // save c_rarg0, because we want to use that value.
5652 // We could do without it but then we depend on the number of slots used by pusha
5653 __ push(c_rarg0);
5654
5655 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
5656
5657 __ pusha();
5658
5659 // The method may have floats as arguments, and we must spill them before calling
5660 // the VM runtime.
5661 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
5662 const int xmm_size = wordSize * 2;
5663 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
5664 __ subptr(rsp, xmm_spill_size);
5665 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
5666 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
5667 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
5668 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
5669 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
5670 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
5671 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
5672 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
5673
5674 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
5675
5676 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
5677 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
5678 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
5679 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
5680 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
5681 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
5682 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
5683 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
5684 __ addptr(rsp, xmm_spill_size);
5685
5686 __ cmpl(rax, 1); // 1 means deoptimize
5687 __ jcc(Assembler::equal, deoptimize_label);
5688
5689 __ popa();
5690 __ pop(c_rarg0);
5691
5692 __ leave();
5693
5694 __ addptr(rsp, 1 * wordSize); // cookie
5695 __ ret(0);
5696
5697
5698 __ BIND(deoptimize_label);
5699
5700 __ popa();
5701 __ pop(c_rarg0);
5702
5703 __ leave();
5704
5705 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
5706 // here while still having a correct stack is valuable
5707 __ testptr(rsp, Address(rsp, 0));
5708
5709 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
5710 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
5711
5712 return start;
5713 }
5714
5715 /**
5716 * Arguments:
5717 *
5718 * Input:
5719 * c_rarg0 - out address
5720 * c_rarg1 - in address
5721 * c_rarg2 - offset
5722 * c_rarg3 - len
5723 * not Win64
5724 * c_rarg4 - k
5725 * Win64
5726 * rsp+40 - k
5727 */
5728 address generate_mulAdd() {
5729 __ align(CodeEntryAlignment);
5730 StubCodeMark mark(this, "StubRoutines", "mulAdd");
5731
5732 address start = __ pc();
5733 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5734 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5735 const Register out = rdi;
5736 const Register in = rsi;
5737 const Register offset = r11;
5738 const Register len = rcx;
5739 const Register k = r8;
5740
5741 // Next registers will be saved on stack in mul_add().
5742 const Register tmp1 = r12;
5743 const Register tmp2 = r13;
5744 const Register tmp3 = r14;
5745 const Register tmp4 = r15;
5746 const Register tmp5 = rbx;
5747
5748 BLOCK_COMMENT("Entry:");
5749 __ enter(); // required for proper stackwalking of RuntimeStub frame
5750
5751 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
5752 // len => rcx, k => r8
5753 // r9 and r10 may be used to save non-volatile registers
5754 #ifdef _WIN64
5755 // last argument is on stack on Win64
5756 __ movl(k, Address(rsp, 6 * wordSize));
5757 #endif
5758 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
5759 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5760
5761 restore_arg_regs();
5762
5763 __ leave(); // required for proper stackwalking of RuntimeStub frame
5764 __ ret(0);
5765
5766 return start;
5767 }
5768
5769 address generate_bigIntegerRightShift() {
5770 __ align(CodeEntryAlignment);
5771 StubCodeMark mark(this, "StubRoutines", "bigIntegerRightShiftWorker");
5772
5773 address start = __ pc();
5774 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
5775 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
5776 const Register newArr = rdi;
5777 const Register oldArr = rsi;
5778 const Register newIdx = rdx;
5779 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
5780 const Register totalNumIter = r8;
5781
5782 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
5783 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
5784 const Register tmp1 = r11; // Caller save.
5785 const Register tmp2 = rax; // Caller save.
5786 const Register tmp3 = WINDOWS_ONLY(r12) NOT_WINDOWS(r9); // Windows: Callee save. Linux: Caller save.
5787 const Register tmp4 = WINDOWS_ONLY(r13) NOT_WINDOWS(r10); // Windows: Callee save. Linux: Caller save.
5788 const Register tmp5 = r14; // Callee save.
5789 const Register tmp6 = r15;
5790
5791 const XMMRegister x0 = xmm0;
5792 const XMMRegister x1 = xmm1;
5793 const XMMRegister x2 = xmm2;
5794
5795 BLOCK_COMMENT("Entry:");
5796 __ enter(); // required for proper stackwalking of RuntimeStub frame
5797
5798 #ifdef _WINDOWS
5799 setup_arg_regs(4);
5800 // For windows, since last argument is on stack, we need to move it to the appropriate register.
5801 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
5802 // Save callee save registers.
5803 __ push(tmp3);
5804 __ push(tmp4);
5805 #endif
5806 __ push(tmp5);
5807
5808 // Rename temps used throughout the code.
5809 const Register idx = tmp1;
5810 const Register nIdx = tmp2;
5811
5812 __ xorl(idx, idx);
5813
5814 // Start right shift from end of the array.
5815 // For example, if #iteration = 4 and newIdx = 1
5816 // then dest[4] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
5817 // if #iteration = 4 and newIdx = 0
5818 // then dest[3] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
5819 __ movl(idx, totalNumIter);
5820 __ movl(nIdx, idx);
5821 __ addl(nIdx, newIdx);
5822
5823 // If vectorization is enabled, check if the number of iterations is at least 64
5824 // If not, then go to ShifTwo processing 2 iterations
5825 if (VM_Version::supports_avx512_vbmi2()) {
5826 __ cmpptr(totalNumIter, (AVX3Threshold/64));
5827 __ jcc(Assembler::less, ShiftTwo);
5828
5829 if (AVX3Threshold < 16 * 64) {
5830 __ cmpl(totalNumIter, 16);
5831 __ jcc(Assembler::less, ShiftTwo);
5832 }
5833 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
5834 __ subl(idx, 16);
5835 __ subl(nIdx, 16);
5836 __ BIND(Shift512Loop);
5837 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 4), Assembler::AVX_512bit);
5838 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
5839 __ vpshrdvd(x2, x1, x0, Assembler::AVX_512bit);
5840 __ evmovdqul(Address(newArr, nIdx, Address::times_4), x2, Assembler::AVX_512bit);
5841 __ subl(nIdx, 16);
5842 __ subl(idx, 16);
5843 __ jcc(Assembler::greaterEqual, Shift512Loop);
5844 __ addl(idx, 16);
5845 __ addl(nIdx, 16);
5846 }
5847 __ BIND(ShiftTwo);
5848 __ cmpl(idx, 2);
5849 __ jcc(Assembler::less, ShiftOne);
5850 __ subl(idx, 2);
5851 __ subl(nIdx, 2);
5852 __ BIND(ShiftTwoLoop);
5853 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 8));
5854 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
5855 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
5856 __ shrdl(tmp5, tmp4);
5857 __ shrdl(tmp4, tmp3);
5858 __ movl(Address(newArr, nIdx, Address::times_4, 4), tmp5);
5859 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
5860 __ subl(nIdx, 2);
5861 __ subl(idx, 2);
5862 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
5863 __ addl(idx, 2);
5864 __ addl(nIdx, 2);
5865
5866 // Do the last iteration
5867 __ BIND(ShiftOne);
5868 __ cmpl(idx, 1);
5869 __ jcc(Assembler::less, Exit);
5870 __ subl(idx, 1);
5871 __ subl(nIdx, 1);
5872 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
5873 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
5874 __ shrdl(tmp4, tmp3);
5875 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
5876 __ BIND(Exit);
5877 // Restore callee save registers.
5878 __ pop(tmp5);
5879 #ifdef _WINDOWS
5880 __ pop(tmp4);
5881 __ pop(tmp3);
5882 restore_arg_regs();
5883 #endif
5884 __ leave(); // required for proper stackwalking of RuntimeStub frame
5885 __ ret(0);
5886 return start;
5887 }
5888
5889 /**
5890 * Arguments:
5891 *
5892 * Input:
5893 * c_rarg0 - newArr address
5894 * c_rarg1 - oldArr address
5895 * c_rarg2 - newIdx
5896 * c_rarg3 - shiftCount
5897 * not Win64
5898 * c_rarg4 - numIter
5899 * Win64
5900 * rsp40 - numIter
5901 */
5902 address generate_bigIntegerLeftShift() {
5903 __ align(CodeEntryAlignment);
5904 StubCodeMark mark(this, "StubRoutines", "bigIntegerLeftShiftWorker");
5905 address start = __ pc();
5906 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
5907 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
5908 const Register newArr = rdi;
5909 const Register oldArr = rsi;
5910 const Register newIdx = rdx;
5911 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
5912 const Register totalNumIter = r8;
5913 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
5914 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
5915 const Register tmp1 = r11; // Caller save.
5916 const Register tmp2 = rax; // Caller save.
5917 const Register tmp3 = WINDOWS_ONLY(r12) NOT_WINDOWS(r9); // Windows: Callee save. Linux: Caller save.
5918 const Register tmp4 = WINDOWS_ONLY(r13) NOT_WINDOWS(r10); // Windows: Callee save. Linux: Caller save.
5919 const Register tmp5 = r14; // Callee save.
5920
5921 const XMMRegister x0 = xmm0;
5922 const XMMRegister x1 = xmm1;
5923 const XMMRegister x2 = xmm2;
5924 BLOCK_COMMENT("Entry:");
5925 __ enter(); // required for proper stackwalking of RuntimeStub frame
5926
5927 #ifdef _WINDOWS
5928 setup_arg_regs(4);
5929 // For windows, since last argument is on stack, we need to move it to the appropriate register.
5930 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
5931 // Save callee save registers.
5932 __ push(tmp3);
5933 __ push(tmp4);
5934 #endif
5935 __ push(tmp5);
5936
5937 // Rename temps used throughout the code
5938 const Register idx = tmp1;
5939 const Register numIterTmp = tmp2;
5940
5941 // Start idx from zero.
5942 __ xorl(idx, idx);
5943 // Compute interior pointer for new array. We do this so that we can use same index for both old and new arrays.
5944 __ lea(newArr, Address(newArr, newIdx, Address::times_4));
5945 __ movl(numIterTmp, totalNumIter);
5946
5947 // If vectorization is enabled, check if the number of iterations is at least 64
5948 // If not, then go to ShiftTwo shifting two numbers at a time
5949 if (VM_Version::supports_avx512_vbmi2()) {
5950 __ cmpl(totalNumIter, (AVX3Threshold/64));
5951 __ jcc(Assembler::less, ShiftTwo);
5952
5953 if (AVX3Threshold < 16 * 64) {
5954 __ cmpl(totalNumIter, 16);
5955 __ jcc(Assembler::less, ShiftTwo);
5956 }
5957 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
5958 __ subl(numIterTmp, 16);
5959 __ BIND(Shift512Loop);
5960 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
5961 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 0x4), Assembler::AVX_512bit);
5962 __ vpshldvd(x1, x2, x0, Assembler::AVX_512bit);
5963 __ evmovdqul(Address(newArr, idx, Address::times_4), x1, Assembler::AVX_512bit);
5964 __ addl(idx, 16);
5965 __ subl(numIterTmp, 16);
5966 __ jcc(Assembler::greaterEqual, Shift512Loop);
5967 __ addl(numIterTmp, 16);
5968 }
5969 __ BIND(ShiftTwo);
5970 __ cmpl(totalNumIter, 1);
5971 __ jcc(Assembler::less, Exit);
5972 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
5973 __ subl(numIterTmp, 2);
5974 __ jcc(Assembler::less, ShiftOne);
5975
5976 __ BIND(ShiftTwoLoop);
5977 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
5978 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 0x8));
5979 __ shldl(tmp3, tmp4);
5980 __ shldl(tmp4, tmp5);
5981 __ movl(Address(newArr, idx, Address::times_4), tmp3);
5982 __ movl(Address(newArr, idx, Address::times_4, 0x4), tmp4);
5983 __ movl(tmp3, tmp5);
5984 __ addl(idx, 2);
5985 __ subl(numIterTmp, 2);
5986 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
5987
5988 // Do the last iteration
5989 __ BIND(ShiftOne);
5990 __ addl(numIterTmp, 2);
5991 __ cmpl(numIterTmp, 1);
5992 __ jcc(Assembler::less, Exit);
5993 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
5994 __ shldl(tmp3, tmp4);
5995 __ movl(Address(newArr, idx, Address::times_4), tmp3);
5996
5997 __ BIND(Exit);
5998 // Restore callee save registers.
5999 __ pop(tmp5);
6000 #ifdef _WINDOWS
6001 __ pop(tmp4);
6002 __ pop(tmp3);
6003 restore_arg_regs();
6004 #endif
6005 __ leave(); // required for proper stackwalking of RuntimeStub frame
6006 __ ret(0);
6007 return start;
6008 }
6009
6010 address generate_libmExp() {
6011 StubCodeMark mark(this, "StubRoutines", "libmExp");
6012
6013 address start = __ pc();
6014
6015 const XMMRegister x0 = xmm0;
6016 const XMMRegister x1 = xmm1;
6017 const XMMRegister x2 = xmm2;
6018 const XMMRegister x3 = xmm3;
6019
6020 const XMMRegister x4 = xmm4;
6021 const XMMRegister x5 = xmm5;
6022 const XMMRegister x6 = xmm6;
6023 const XMMRegister x7 = xmm7;
6024
6025 const Register tmp = r11;
6026
6027 BLOCK_COMMENT("Entry:");
6028 __ enter(); // required for proper stackwalking of RuntimeStub frame
6029
6030 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
6031
6032 __ leave(); // required for proper stackwalking of RuntimeStub frame
6033 __ ret(0);
6034
6035 return start;
6036
6037 }
6038
6039 address generate_libmLog() {
6040 StubCodeMark mark(this, "StubRoutines", "libmLog");
6041
6042 address start = __ pc();
6043
6044 const XMMRegister x0 = xmm0;
6045 const XMMRegister x1 = xmm1;
6046 const XMMRegister x2 = xmm2;
6047 const XMMRegister x3 = xmm3;
6048
6049 const XMMRegister x4 = xmm4;
6050 const XMMRegister x5 = xmm5;
6051 const XMMRegister x6 = xmm6;
6052 const XMMRegister x7 = xmm7;
6053
6054 const Register tmp1 = r11;
6055 const Register tmp2 = r8;
6056
6057 BLOCK_COMMENT("Entry:");
6058 __ enter(); // required for proper stackwalking of RuntimeStub frame
6059
6060 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2);
6061
6062 __ leave(); // required for proper stackwalking of RuntimeStub frame
6063 __ ret(0);
6064
6065 return start;
6066
6067 }
6068
6069 address generate_libmLog10() {
6070 StubCodeMark mark(this, "StubRoutines", "libmLog10");
6071
6072 address start = __ pc();
6073
6074 const XMMRegister x0 = xmm0;
6075 const XMMRegister x1 = xmm1;
6076 const XMMRegister x2 = xmm2;
6077 const XMMRegister x3 = xmm3;
6078
6079 const XMMRegister x4 = xmm4;
6080 const XMMRegister x5 = xmm5;
6081 const XMMRegister x6 = xmm6;
6082 const XMMRegister x7 = xmm7;
6083
6084 const Register tmp = r11;
6085
6086 BLOCK_COMMENT("Entry:");
6087 __ enter(); // required for proper stackwalking of RuntimeStub frame
6088
6089 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
6090
6091 __ leave(); // required for proper stackwalking of RuntimeStub frame
6092 __ ret(0);
6093
6094 return start;
6095
6096 }
6097
6098 address generate_libmPow() {
6099 StubCodeMark mark(this, "StubRoutines", "libmPow");
6100
6101 address start = __ pc();
6102
6103 const XMMRegister x0 = xmm0;
6104 const XMMRegister x1 = xmm1;
6105 const XMMRegister x2 = xmm2;
6106 const XMMRegister x3 = xmm3;
6107
6108 const XMMRegister x4 = xmm4;
6109 const XMMRegister x5 = xmm5;
6110 const XMMRegister x6 = xmm6;
6111 const XMMRegister x7 = xmm7;
6112
6113 const Register tmp1 = r8;
6114 const Register tmp2 = r9;
6115 const Register tmp3 = r10;
6116 const Register tmp4 = r11;
6117
6118 BLOCK_COMMENT("Entry:");
6119 __ enter(); // required for proper stackwalking of RuntimeStub frame
6120
6121 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
6122
6123 __ leave(); // required for proper stackwalking of RuntimeStub frame
6124 __ ret(0);
6125
6126 return start;
6127
6128 }
6129
6130 address generate_libmSin() {
6131 StubCodeMark mark(this, "StubRoutines", "libmSin");
6132
6133 address start = __ pc();
6134
6135 const XMMRegister x0 = xmm0;
6136 const XMMRegister x1 = xmm1;
6137 const XMMRegister x2 = xmm2;
6138 const XMMRegister x3 = xmm3;
6139
6140 const XMMRegister x4 = xmm4;
6141 const XMMRegister x5 = xmm5;
6142 const XMMRegister x6 = xmm6;
6143 const XMMRegister x7 = xmm7;
6144
6145 const Register tmp1 = r8;
6146 const Register tmp2 = r9;
6147 const Register tmp3 = r10;
6148 const Register tmp4 = r11;
6149
6150 BLOCK_COMMENT("Entry:");
6151 __ enter(); // required for proper stackwalking of RuntimeStub frame
6152
6153 #ifdef _WIN64
6154 __ push(rsi);
6155 __ push(rdi);
6156 #endif
6157 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
6158
6159 #ifdef _WIN64
6160 __ pop(rdi);
6161 __ pop(rsi);
6162 #endif
6163
6164 __ leave(); // required for proper stackwalking of RuntimeStub frame
6165 __ ret(0);
6166
6167 return start;
6168
6169 }
6170
6171 address generate_libmCos() {
6172 StubCodeMark mark(this, "StubRoutines", "libmCos");
6173
6174 address start = __ pc();
6175
6176 const XMMRegister x0 = xmm0;
6177 const XMMRegister x1 = xmm1;
6178 const XMMRegister x2 = xmm2;
6179 const XMMRegister x3 = xmm3;
6180
6181 const XMMRegister x4 = xmm4;
6182 const XMMRegister x5 = xmm5;
6183 const XMMRegister x6 = xmm6;
6184 const XMMRegister x7 = xmm7;
6185
6186 const Register tmp1 = r8;
6187 const Register tmp2 = r9;
6188 const Register tmp3 = r10;
6189 const Register tmp4 = r11;
6190
6191 BLOCK_COMMENT("Entry:");
6192 __ enter(); // required for proper stackwalking of RuntimeStub frame
6193
6194 #ifdef _WIN64
6195 __ push(rsi);
6196 __ push(rdi);
6197 #endif
6198 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
6199
6200 #ifdef _WIN64
6201 __ pop(rdi);
6202 __ pop(rsi);
6203 #endif
6204
6205 __ leave(); // required for proper stackwalking of RuntimeStub frame
6206 __ ret(0);
6207
6208 return start;
6209
6210 }
6211
6212 address generate_libmTan() {
6213 StubCodeMark mark(this, "StubRoutines", "libmTan");
6214
6215 address start = __ pc();
6216
6217 const XMMRegister x0 = xmm0;
6218 const XMMRegister x1 = xmm1;
6219 const XMMRegister x2 = xmm2;
6220 const XMMRegister x3 = xmm3;
6221
6222 const XMMRegister x4 = xmm4;
6223 const XMMRegister x5 = xmm5;
6224 const XMMRegister x6 = xmm6;
6225 const XMMRegister x7 = xmm7;
6226
6227 const Register tmp1 = r8;
6228 const Register tmp2 = r9;
6229 const Register tmp3 = r10;
6230 const Register tmp4 = r11;
6231
6232 BLOCK_COMMENT("Entry:");
6233 __ enter(); // required for proper stackwalking of RuntimeStub frame
6234
6235 #ifdef _WIN64
6236 __ push(rsi);
6237 __ push(rdi);
6238 #endif
6239 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
6240
6241 #ifdef _WIN64
6242 __ pop(rdi);
6243 __ pop(rsi);
6244 #endif
6245
6246 __ leave(); // required for proper stackwalking of RuntimeStub frame
6247 __ ret(0);
6248
6249 return start;
6250
6251 }
6252
6253 #undef __
6254 #define __ masm->
6255
6256 // Continuation point for throwing of implicit exceptions that are
6257 // not handled in the current activation. Fabricates an exception
6258 // oop and initiates normal exception dispatching in this
6259 // frame. Since we need to preserve callee-saved values (currently
6260 // only for C2, but done for C1 as well) we need a callee-saved oop
6261 // map and therefore have to make these stubs into RuntimeStubs
6262 // rather than BufferBlobs. If the compiler needs all registers to
6263 // be preserved between the fault point and the exception handler
6264 // then it must assume responsibility for that in
6265 // AbstractCompiler::continuation_for_implicit_null_exception or
6266 // continuation_for_implicit_division_by_zero_exception. All other
6267 // implicit exceptions (e.g., NullPointerException or
6268 // AbstractMethodError on entry) are either at call sites or
6269 // otherwise assume that stack unwinding will be initiated, so
6270 // caller saved registers were assumed volatile in the compiler.
6271 address generate_throw_exception(const char* name,
6272 address runtime_entry,
6273 Register arg1 = noreg,
6274 Register arg2 = noreg) {
6275 // Information about frame layout at time of blocking runtime call.
6276 // Note that we only have to preserve callee-saved registers since
6277 // the compilers are responsible for supplying a continuation point
6278 // if they expect all registers to be preserved.
6279 enum layout {
6280 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
6281 rbp_off2,
6282 return_off,
6283 return_off2,
6284 framesize // inclusive of return address
6285 };
6286
6287 int insts_size = 512;
6288 int locs_size = 64;
6289
6290 CodeBuffer code(name, insts_size, locs_size);
6291 OopMapSet* oop_maps = new OopMapSet();
6292 MacroAssembler* masm = new MacroAssembler(&code);
6293
6294 address start = __ pc();
6295
6296 // This is an inlined and slightly modified version of call_VM
6297 // which has the ability to fetch the return PC out of
6298 // thread-local storage and also sets up last_Java_sp slightly
6299 // differently than the real call_VM
6300
6301 __ enter(); // required for proper stackwalking of RuntimeStub frame
6302
6303 assert(is_even(framesize/2), "sp not 16-byte aligned");
6304
6305 // return address and rbp are already in place
6306 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
6307
6308 int frame_complete = __ pc() - start;
6309
6310 // Set up last_Java_sp and last_Java_fp
6311 address the_pc = __ pc();
6312 __ set_last_Java_frame(rsp, rbp, the_pc);
6313 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
6314
6315 // Call runtime
6316 if (arg1 != noreg) {
6317 assert(arg2 != c_rarg1, "clobbered");
6318 __ movptr(c_rarg1, arg1);
6319 }
6320 if (arg2 != noreg) {
6321 __ movptr(c_rarg2, arg2);
6322 }
6323 __ movptr(c_rarg0, r15_thread);
6324 BLOCK_COMMENT("call runtime_entry");
6325 __ call(RuntimeAddress(runtime_entry));
6326
6327 // Generate oop map
6328 OopMap* map = new OopMap(framesize, 0);
6329
6330 oop_maps->add_gc_map(the_pc - start, map);
6331
6332 __ reset_last_Java_frame(true);
6333
6334 __ leave(); // required for proper stackwalking of RuntimeStub frame
6335
6336 // check for pending exceptions
6337 #ifdef ASSERT
6338 Label L;
6339 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
6340 (int32_t) NULL_WORD);
6341 __ jcc(Assembler::notEqual, L);
6342 __ should_not_reach_here();
6343 __ bind(L);
6344 #endif // ASSERT
6345 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
6346
6347
6348 // codeBlob framesize is in words (not VMRegImpl::slot_size)
6349 RuntimeStub* stub =
6350 RuntimeStub::new_runtime_stub(name,
6351 &code,
6352 frame_complete,
6353 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
6354 oop_maps, false);
6355 return stub->entry_point();
6356 }
6357
6358 void create_control_words() {
6359 // Round to nearest, 53-bit mode, exceptions masked
6360 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
6361 // Round to zero, 53-bit mode, exception mased
6362 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
6363 // Round to nearest, 24-bit mode, exceptions masked
6364 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
6365 // Round to nearest, 64-bit mode, exceptions masked
6366 StubRoutines::_mxcsr_std = 0x1F80;
6367 // Note: the following two constants are 80-bit values
6368 // layout is critical for correct loading by FPU.
6369 // Bias for strict fp multiply/divide
6370 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
6371 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
6372 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
6373 // Un-Bias for strict fp multiply/divide
6374 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
6375 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
6376 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
6377 }
6378
6379 // Initialization
6380 void generate_initial() {
6381 // Generates all stubs and initializes the entry points
6382
6383 // This platform-specific settings are needed by generate_call_stub()
6384 create_control_words();
6385
6386 // entry points that exist in all platforms Note: This is code
6387 // that could be shared among different platforms - however the
6388 // benefit seems to be smaller than the disadvantage of having a
6389 // much more complicated generator structure. See also comment in
6390 // stubRoutines.hpp.
6391
6392 StubRoutines::_forward_exception_entry = generate_forward_exception();
6393
6394 StubRoutines::_call_stub_entry =
6395 generate_call_stub(StubRoutines::_call_stub_return_address);
6396
6397 // is referenced by megamorphic call
6398 StubRoutines::_catch_exception_entry = generate_catch_exception();
6399
6400 // atomic calls
6401 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
6402 StubRoutines::_atomic_xchg_long_entry = generate_atomic_xchg_long();
6403 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
6404 StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
6405 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
6406 StubRoutines::_atomic_add_entry = generate_atomic_add();
6407 StubRoutines::_atomic_add_long_entry = generate_atomic_add_long();
6408 StubRoutines::_fence_entry = generate_orderaccess_fence();
6409
6410 // platform dependent
6411 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
6412 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
6413
6414 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
6415
6416 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
6417 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
6418 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
6419 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
6420
6421 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6422 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
6423 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6424 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
6425
6426 // Build this early so it's available for the interpreter.
6427 StubRoutines::_throw_StackOverflowError_entry =
6428 generate_throw_exception("StackOverflowError throw_exception",
6429 CAST_FROM_FN_PTR(address,
6430 SharedRuntime::
6431 throw_StackOverflowError));
6432 StubRoutines::_throw_delayed_StackOverflowError_entry =
6433 generate_throw_exception("delayed StackOverflowError throw_exception",
6434 CAST_FROM_FN_PTR(address,
6435 SharedRuntime::
6436 throw_delayed_StackOverflowError));
6437 if (UseCRC32Intrinsics) {
6438 // set table address before stub generation which use it
6439 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
6440 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
6441 }
6442
6443 if (UseCRC32CIntrinsics) {
6444 bool supports_clmul = VM_Version::supports_clmul();
6445 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
6446 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
6447 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
6448 }
6449 if (UseLibmIntrinsic && InlineIntrinsics) {
6450 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
6451 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
6452 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6453 StubRoutines::x86::_ONEHALF_adr = (address)StubRoutines::x86::_ONEHALF;
6454 StubRoutines::x86::_P_2_adr = (address)StubRoutines::x86::_P_2;
6455 StubRoutines::x86::_SC_4_adr = (address)StubRoutines::x86::_SC_4;
6456 StubRoutines::x86::_Ctable_adr = (address)StubRoutines::x86::_Ctable;
6457 StubRoutines::x86::_SC_2_adr = (address)StubRoutines::x86::_SC_2;
6458 StubRoutines::x86::_SC_3_adr = (address)StubRoutines::x86::_SC_3;
6459 StubRoutines::x86::_SC_1_adr = (address)StubRoutines::x86::_SC_1;
6460 StubRoutines::x86::_PI_INV_TABLE_adr = (address)StubRoutines::x86::_PI_INV_TABLE;
6461 StubRoutines::x86::_PI_4_adr = (address)StubRoutines::x86::_PI_4;
6462 StubRoutines::x86::_PI32INV_adr = (address)StubRoutines::x86::_PI32INV;
6463 StubRoutines::x86::_SIGN_MASK_adr = (address)StubRoutines::x86::_SIGN_MASK;
6464 StubRoutines::x86::_P_1_adr = (address)StubRoutines::x86::_P_1;
6465 StubRoutines::x86::_P_3_adr = (address)StubRoutines::x86::_P_3;
6466 StubRoutines::x86::_NEG_ZERO_adr = (address)StubRoutines::x86::_NEG_ZERO;
6467 }
6468 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
6469 StubRoutines::_dexp = generate_libmExp();
6470 }
6471 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
6472 StubRoutines::_dlog = generate_libmLog();
6473 }
6474 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
6475 StubRoutines::_dlog10 = generate_libmLog10();
6476 }
6477 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
6478 StubRoutines::_dpow = generate_libmPow();
6479 }
6480 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
6481 StubRoutines::_dsin = generate_libmSin();
6482 }
6483 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
6484 StubRoutines::_dcos = generate_libmCos();
6485 }
6486 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6487 StubRoutines::_dtan = generate_libmTan();
6488 }
6489 }
6490
6491 // Safefetch stubs.
6492 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
6493 &StubRoutines::_safefetch32_fault_pc,
6494 &StubRoutines::_safefetch32_continuation_pc);
6495 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
6496 &StubRoutines::_safefetchN_fault_pc,
6497 &StubRoutines::_safefetchN_continuation_pc);
6498 }
6499
6500 void generate_all() {
6501 // Generates all stubs and initializes the entry points
6502
6503 // These entry points require SharedInfo::stack0 to be set up in
6504 // non-core builds and need to be relocatable, so they each
6505 // fabricate a RuntimeStub internally.
6506 StubRoutines::_throw_AbstractMethodError_entry =
6507 generate_throw_exception("AbstractMethodError throw_exception",
6508 CAST_FROM_FN_PTR(address,
6509 SharedRuntime::
6510 throw_AbstractMethodError));
6511
6512 StubRoutines::_throw_IncompatibleClassChangeError_entry =
6513 generate_throw_exception("IncompatibleClassChangeError throw_exception",
6514 CAST_FROM_FN_PTR(address,
6515 SharedRuntime::
6516 throw_IncompatibleClassChangeError));
6517
6518 StubRoutines::_throw_NullPointerException_at_call_entry =
6519 generate_throw_exception("NullPointerException at call throw_exception",
6520 CAST_FROM_FN_PTR(address,
6521 SharedRuntime::
6522 throw_NullPointerException_at_call));
6523
6524 // entry points that are platform specific
6525 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6526 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
6527 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6528 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
6529 StubRoutines::x86::_vector_all_bits_set = generate_vector_mask("vector_all_bits_set", 0xFFFFFFFFFFFFFFFF);
6530 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
6531 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
6532 StubRoutines::x86::_vector_int_to_byte_mask = generate_vector_mask("vector_int_to_byte_mask", 0x000000ff000000ff);
6533 StubRoutines::x86::_vector_int_to_short_mask = generate_vector_mask("vector_int_to_short_mask", 0x0000ffff0000ffff);
6534 StubRoutines::x86::_vector_32_bit_mask = generate_vector_custom_i32("vector_32_bit_mask", Assembler::AVX_512bit,
6535 0xFFFFFFFF, 0, 0, 0);
6536 StubRoutines::x86::_vector_64_bit_mask = generate_vector_custom_i32("vector_64_bit_mask", Assembler::AVX_512bit,
6537 0xFFFFFFFF, 0xFFFFFFFF, 0, 0);
6538 StubRoutines::x86::_vector_int_shuffle_mask = generate_vector_mask("vector_int_shuffle_mask", 0x0302010003020100);
6539 StubRoutines::x86::_vector_short_shuffle_mask = generate_vector_mask("vector_short_shuffle_mask", 0x0100010001000100);
6540 StubRoutines::x86::_vector_long_shuffle_mask = generate_vector_mask("vector_long_shuffle_mask", 0x0000000100000000);
6541 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
6542 StubRoutines::x86::_vector_iota_indices = generate_iota_indices("iota_indices");
6543
6544 // support for verify_oop (must happen after universe_init)
6545 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
6546
6547 // data cache line writeback
6548 StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
6549 StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
6550
6551 // arraycopy stubs used by compilers
6552 generate_arraycopy_stubs();
6553
6554 // don't bother generating these AES intrinsic stubs unless global flag is set
6555 if (UseAESIntrinsics) {
6556 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
6557 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
6558 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
6559 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
6560 if (VM_Version::supports_avx512_vaes() && VM_Version::supports_avx512vl() && VM_Version::supports_avx512dq() ) {
6561 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptVectorAESCrypt();
6562 StubRoutines::_electronicCodeBook_encryptAESCrypt = generate_electronicCodeBook_encryptAESCrypt();
6563 StubRoutines::_electronicCodeBook_decryptAESCrypt = generate_electronicCodeBook_decryptAESCrypt();
6564 } else {
6565 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
6566 }
6567 }
6568 if (UseAESCTRIntrinsics) {
6569 if (VM_Version::supports_avx512_vaes() && VM_Version::supports_avx512bw() && VM_Version::supports_avx512vl()) {
6570 StubRoutines::x86::_counter_mask_addr = counter_mask_addr();
6571 StubRoutines::_counterMode_AESCrypt = generate_counterMode_VectorAESCrypt();
6572 } else {
6573 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
6574 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
6575 }
6576 }
6577
6578 if (UseSHA1Intrinsics) {
6579 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
6580 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
6581 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
6582 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
6583 }
6584 if (UseSHA256Intrinsics) {
6585 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
6586 char* dst = (char*)StubRoutines::x86::_k256_W;
6587 char* src = (char*)StubRoutines::x86::_k256;
6588 for (int ii = 0; ii < 16; ++ii) {
6589 memcpy(dst + 32 * ii, src + 16 * ii, 16);
6590 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
6591 }
6592 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
6593 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
6594 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
6595 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
6596 }
6597 if (UseSHA512Intrinsics) {
6598 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
6599 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
6600 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
6601 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
6602 }
6603
6604 // Generate GHASH intrinsics code
6605 if (UseGHASHIntrinsics) {
6606 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
6607 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
6608 if (VM_Version::supports_avx()) {
6609 StubRoutines::x86::_ghash_shuffmask_addr = ghash_shufflemask_addr();
6610 StubRoutines::x86::_ghash_poly_addr = ghash_polynomial_addr();
6611 StubRoutines::_ghash_processBlocks = generate_avx_ghash_processBlocks();
6612 } else {
6613 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
6614 }
6615 }
6616
6617 if (UseBASE64Intrinsics) {
6618 StubRoutines::x86::_and_mask = base64_and_mask_addr();
6619 StubRoutines::x86::_bswap_mask = base64_bswap_mask_addr();
6620 StubRoutines::x86::_base64_charset = base64_charset_addr();
6621 StubRoutines::x86::_url_charset = base64url_charset_addr();
6622 StubRoutines::x86::_gather_mask = base64_gather_mask_addr();
6623 StubRoutines::x86::_left_shift_mask = base64_left_shift_mask_addr();
6624 StubRoutines::x86::_right_shift_mask = base64_right_shift_mask_addr();
6625 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
6626 }
6627
6628 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
6629 if (bs_nm != NULL) {
6630 StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
6631 }
6632 #ifdef COMPILER2
6633 if (UseMultiplyToLenIntrinsic) {
6634 StubRoutines::_multiplyToLen = generate_multiplyToLen();
6635 }
6636 if (UseSquareToLenIntrinsic) {
6637 StubRoutines::_squareToLen = generate_squareToLen();
6638 }
6639 if (UseMulAddIntrinsic) {
6640 StubRoutines::_mulAdd = generate_mulAdd();
6641 }
6642 if (VM_Version::supports_avx512_vbmi2()) {
6643 StubRoutines::_bigIntegerRightShiftWorker = generate_bigIntegerRightShift();
6644 StubRoutines::_bigIntegerLeftShiftWorker = generate_bigIntegerLeftShift();
6645 }
6646 if (UseMontgomeryMultiplyIntrinsic) {
6647 StubRoutines::_montgomeryMultiply
6648 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
6649 }
6650 if (UseMontgomerySquareIntrinsic) {
6651 StubRoutines::_montgomerySquare
6652 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
6653 }
6654 #endif // COMPILER2
6655
6656 if (UseVectorizedMismatchIntrinsic) {
6657 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
6658 }
6659 }
6660
6661 public:
6662 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
6663 if (all) {
6664 generate_all();
6665 } else {
6666 generate_initial();
6667 }
6668 }
6669 }; // end class declaration
6670
6671 #define UCM_TABLE_MAX_ENTRIES 16
6672 void StubGenerator_generate(CodeBuffer* code, bool all) {
6673 if (UnsafeCopyMemory::_table == NULL) {
6674 UnsafeCopyMemory::create_table(UCM_TABLE_MAX_ENTRIES);
6675 }
6676 StubGenerator g(code, all);
6677 }