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_fp_mask(const char *stub_name, int64_t mask) {
977 __ align(CodeEntryAlignment);
978 StubCodeMark mark(this, "StubRoutines", stub_name);
979 address start = __ pc();
980
981 __ emit_data64( mask, relocInfo::none );
982 __ emit_data64( mask, relocInfo::none );
983
984 return start;
985 }
986
987 address generate_vector_mask(const char *stub_name, int64_t mask) {
988 __ align(CodeEntryAlignment);
989 StubCodeMark mark(this, "StubRoutines", stub_name);
990 address start = __ pc();
991
992 __ emit_data64(mask, relocInfo::none);
993 __ emit_data64(mask, relocInfo::none);
994 __ emit_data64(mask, relocInfo::none);
995 __ emit_data64(mask, relocInfo::none);
996 __ emit_data64(mask, relocInfo::none);
997 __ emit_data64(mask, relocInfo::none);
998 __ emit_data64(mask, relocInfo::none);
999 __ emit_data64(mask, relocInfo::none);
1000
1001 return start;
1002 }
1003
1004 address generate_vector_byte_perm_mask(const char *stub_name) {
1005 __ align(CodeEntryAlignment);
1006 StubCodeMark mark(this, "StubRoutines", stub_name);
1007 address start = __ pc();
1008
1009 __ emit_data64(0x0000000000000001, relocInfo::none);
1010 __ emit_data64(0x0000000000000003, relocInfo::none);
1011 __ emit_data64(0x0000000000000005, relocInfo::none);
1012 __ emit_data64(0x0000000000000007, relocInfo::none);
1013 __ emit_data64(0x0000000000000000, relocInfo::none);
1014 __ emit_data64(0x0000000000000002, relocInfo::none);
1015 __ emit_data64(0x0000000000000004, relocInfo::none);
1016 __ emit_data64(0x0000000000000006, relocInfo::none);
1017
1018 return start;
1019 }
1020
1021 // Non-destructive plausibility checks for oops
1022 //
1023 // Arguments:
1024 // all args on stack!
1025 //
1026 // Stack after saving c_rarg3:
1027 // [tos + 0]: saved c_rarg3
1028 // [tos + 1]: saved c_rarg2
1029 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1030 // [tos + 3]: saved flags
1031 // [tos + 4]: return address
1032 // * [tos + 5]: error message (char*)
1033 // * [tos + 6]: object to verify (oop)
1034 // * [tos + 7]: saved rax - saved by caller and bashed
1035 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1036 // * = popped on exit
1037 address generate_verify_oop() {
1038 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1039 address start = __ pc();
1040
1041 Label exit, error;
1042
1043 __ pushf();
1044 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1045
1046 __ push(r12);
1047
1048 // save c_rarg2 and c_rarg3
1049 __ push(c_rarg2);
1050 __ push(c_rarg3);
1051
1052 enum {
1053 // After previous pushes.
1054 oop_to_verify = 6 * wordSize,
1055 saved_rax = 7 * wordSize,
1056 saved_r10 = 8 * wordSize,
1057
1058 // Before the call to MacroAssembler::debug(), see below.
1059 return_addr = 16 * wordSize,
1060 error_msg = 17 * wordSize
1061 };
1062
1063 // get object
1064 __ movptr(rax, Address(rsp, oop_to_verify));
1065
1066 // make sure object is 'reasonable'
1067 __ testptr(rax, rax);
1068 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1069
1070 #if INCLUDE_ZGC
1071 if (UseZGC) {
1072 // Check if metadata bits indicate a bad oop
1073 __ testptr(rax, Address(r15_thread, ZThreadLocalData::address_bad_mask_offset()));
1074 __ jcc(Assembler::notZero, error);
1075 }
1076 #endif
1077
1078 // Check if the oop is in the right area of memory
1079 __ movptr(c_rarg2, rax);
1080 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1081 __ andptr(c_rarg2, c_rarg3);
1082 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1083 __ cmpptr(c_rarg2, c_rarg3);
1084 __ jcc(Assembler::notZero, error);
1085
1086 // make sure klass is 'reasonable', which is not zero.
1087 __ load_klass(rax, rax, rscratch1); // get klass
1088 __ testptr(rax, rax);
1089 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1090
1091 // return if everything seems ok
1092 __ bind(exit);
1093 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1094 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1095 __ pop(c_rarg3); // restore c_rarg3
1096 __ pop(c_rarg2); // restore c_rarg2
1097 __ pop(r12); // restore r12
1098 __ popf(); // restore flags
1099 __ ret(4 * wordSize); // pop caller saved stuff
1100
1101 // handle errors
1102 __ bind(error);
1103 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1104 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1105 __ pop(c_rarg3); // get saved c_rarg3 back
1106 __ pop(c_rarg2); // get saved c_rarg2 back
1107 __ pop(r12); // get saved r12 back
1108 __ popf(); // get saved flags off stack --
1109 // will be ignored
1110
1111 __ pusha(); // push registers
1112 // (rip is already
1113 // already pushed)
1114 // debug(char* msg, int64_t pc, int64_t regs[])
1115 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1116 // pushed all the registers, so now the stack looks like:
1117 // [tos + 0] 16 saved registers
1118 // [tos + 16] return address
1119 // * [tos + 17] error message (char*)
1120 // * [tos + 18] object to verify (oop)
1121 // * [tos + 19] saved rax - saved by caller and bashed
1122 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1123 // * = popped on exit
1124
1125 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1126 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1127 __ movq(c_rarg2, rsp); // pass address of regs on stack
1128 __ mov(r12, rsp); // remember rsp
1129 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1130 __ andptr(rsp, -16); // align stack as required by ABI
1131 BLOCK_COMMENT("call MacroAssembler::debug");
1132 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1133 __ hlt();
1134 return start;
1135 }
1136
1137 //
1138 // Verify that a register contains clean 32-bits positive value
1139 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1140 //
1141 // Input:
1142 // Rint - 32-bits value
1143 // Rtmp - scratch
1144 //
1145 void assert_clean_int(Register Rint, Register Rtmp) {
1146 #ifdef ASSERT
1147 Label L;
1148 assert_different_registers(Rtmp, Rint);
1149 __ movslq(Rtmp, Rint);
1150 __ cmpq(Rtmp, Rint);
1151 __ jcc(Assembler::equal, L);
1152 __ stop("high 32-bits of int value are not 0");
1153 __ bind(L);
1154 #endif
1155 }
1156
1157 // Generate overlap test for array copy stubs
1158 //
1159 // Input:
1160 // c_rarg0 - from
1161 // c_rarg1 - to
1162 // c_rarg2 - element count
1163 //
1164 // Output:
1165 // rax - &from[element count - 1]
1166 //
1167 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1168 assert(no_overlap_target != NULL, "must be generated");
1169 array_overlap_test(no_overlap_target, NULL, sf);
1170 }
1171 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1172 array_overlap_test(NULL, &L_no_overlap, sf);
1173 }
1174 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1175 const Register from = c_rarg0;
1176 const Register to = c_rarg1;
1177 const Register count = c_rarg2;
1178 const Register end_from = rax;
1179
1180 __ cmpptr(to, from);
1181 __ lea(end_from, Address(from, count, sf, 0));
1182 if (NOLp == NULL) {
1183 ExternalAddress no_overlap(no_overlap_target);
1184 __ jump_cc(Assembler::belowEqual, no_overlap);
1185 __ cmpptr(to, end_from);
1186 __ jump_cc(Assembler::aboveEqual, no_overlap);
1187 } else {
1188 __ jcc(Assembler::belowEqual, (*NOLp));
1189 __ cmpptr(to, end_from);
1190 __ jcc(Assembler::aboveEqual, (*NOLp));
1191 }
1192 }
1193
1194 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1195 //
1196 // Outputs:
1197 // rdi - rcx
1198 // rsi - rdx
1199 // rdx - r8
1200 // rcx - r9
1201 //
1202 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1203 // are non-volatile. r9 and r10 should not be used by the caller.
1204 //
1205 DEBUG_ONLY(bool regs_in_thread;)
1206
1207 void setup_arg_regs(int nargs = 3) {
1208 const Register saved_rdi = r9;
1209 const Register saved_rsi = r10;
1210 assert(nargs == 3 || nargs == 4, "else fix");
1211 #ifdef _WIN64
1212 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1213 "unexpected argument registers");
1214 if (nargs >= 4)
1215 __ mov(rax, r9); // r9 is also saved_rdi
1216 __ movptr(saved_rdi, rdi);
1217 __ movptr(saved_rsi, rsi);
1218 __ mov(rdi, rcx); // c_rarg0
1219 __ mov(rsi, rdx); // c_rarg1
1220 __ mov(rdx, r8); // c_rarg2
1221 if (nargs >= 4)
1222 __ mov(rcx, rax); // c_rarg3 (via rax)
1223 #else
1224 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1225 "unexpected argument registers");
1226 #endif
1227 DEBUG_ONLY(regs_in_thread = false;)
1228 }
1229
1230 void restore_arg_regs() {
1231 assert(!regs_in_thread, "wrong call to restore_arg_regs");
1232 const Register saved_rdi = r9;
1233 const Register saved_rsi = r10;
1234 #ifdef _WIN64
1235 __ movptr(rdi, saved_rdi);
1236 __ movptr(rsi, saved_rsi);
1237 #endif
1238 }
1239
1240 // This is used in places where r10 is a scratch register, and can
1241 // be adapted if r9 is needed also.
1242 void setup_arg_regs_using_thread() {
1243 const Register saved_r15 = r9;
1244 #ifdef _WIN64
1245 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1246 __ get_thread(r15_thread);
1247 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1248 "unexpected argument registers");
1249 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1250 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1251
1252 __ mov(rdi, rcx); // c_rarg0
1253 __ mov(rsi, rdx); // c_rarg1
1254 __ mov(rdx, r8); // c_rarg2
1255 #else
1256 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1257 "unexpected argument registers");
1258 #endif
1259 DEBUG_ONLY(regs_in_thread = true;)
1260 }
1261
1262 void restore_arg_regs_using_thread() {
1263 assert(regs_in_thread, "wrong call to restore_arg_regs");
1264 const Register saved_r15 = r9;
1265 #ifdef _WIN64
1266 __ get_thread(r15_thread);
1267 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1268 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1269 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1270 #endif
1271 }
1272
1273 // Copy big chunks forward
1274 //
1275 // Inputs:
1276 // end_from - source arrays end address
1277 // end_to - destination array end address
1278 // qword_count - 64-bits element count, negative
1279 // to - scratch
1280 // L_copy_bytes - entry label
1281 // L_copy_8_bytes - exit label
1282 //
1283 void copy_bytes_forward(Register end_from, Register end_to,
1284 Register qword_count, Register to,
1285 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1286 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1287 Label L_loop;
1288 __ align(OptoLoopAlignment);
1289 if (UseUnalignedLoadStores) {
1290 Label L_end;
1291 // Copy 64-bytes per iteration
1292 if (UseAVX > 2) {
1293 Label L_loop_avx512, L_loop_avx2, L_32_byte_head, L_above_threshold, L_below_threshold;
1294
1295 __ BIND(L_copy_bytes);
1296 __ cmpptr(qword_count, (-1 * AVX3Threshold / 8));
1297 __ jccb(Assembler::less, L_above_threshold);
1298 __ jmpb(L_below_threshold);
1299
1300 __ bind(L_loop_avx512);
1301 __ evmovdqul(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1302 __ evmovdqul(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1303 __ bind(L_above_threshold);
1304 __ addptr(qword_count, 8);
1305 __ jcc(Assembler::lessEqual, L_loop_avx512);
1306 __ jmpb(L_32_byte_head);
1307
1308 __ bind(L_loop_avx2);
1309 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1310 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1311 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1312 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1313 __ bind(L_below_threshold);
1314 __ addptr(qword_count, 8);
1315 __ jcc(Assembler::lessEqual, L_loop_avx2);
1316
1317 __ bind(L_32_byte_head);
1318 __ subptr(qword_count, 4); // sub(8) and add(4)
1319 __ jccb(Assembler::greater, L_end);
1320 } else {
1321 __ BIND(L_loop);
1322 if (UseAVX == 2) {
1323 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1324 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1325 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1326 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1327 } else {
1328 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1329 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1330 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1331 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1332 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1333 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1334 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1335 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1336 }
1337
1338 __ BIND(L_copy_bytes);
1339 __ addptr(qword_count, 8);
1340 __ jcc(Assembler::lessEqual, L_loop);
1341 __ subptr(qword_count, 4); // sub(8) and add(4)
1342 __ jccb(Assembler::greater, L_end);
1343 }
1344 // Copy trailing 32 bytes
1345 if (UseAVX >= 2) {
1346 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1347 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1348 } else {
1349 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1350 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1351 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1352 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1353 }
1354 __ addptr(qword_count, 4);
1355 __ BIND(L_end);
1356 if (UseAVX >= 2) {
1357 // clean upper bits of YMM registers
1358 __ vpxor(xmm0, xmm0);
1359 __ vpxor(xmm1, xmm1);
1360 }
1361 } else {
1362 // Copy 32-bytes per iteration
1363 __ BIND(L_loop);
1364 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1365 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1366 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1367 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1368 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1369 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1370 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1371 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1372
1373 __ BIND(L_copy_bytes);
1374 __ addptr(qword_count, 4);
1375 __ jcc(Assembler::lessEqual, L_loop);
1376 }
1377 __ subptr(qword_count, 4);
1378 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1379 }
1380
1381 // Copy big chunks backward
1382 //
1383 // Inputs:
1384 // from - source arrays address
1385 // dest - destination array address
1386 // qword_count - 64-bits element count
1387 // to - scratch
1388 // L_copy_bytes - entry label
1389 // L_copy_8_bytes - exit label
1390 //
1391 void copy_bytes_backward(Register from, Register dest,
1392 Register qword_count, Register to,
1393 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1394 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1395 Label L_loop;
1396 __ align(OptoLoopAlignment);
1397 if (UseUnalignedLoadStores) {
1398 Label L_end;
1399 // Copy 64-bytes per iteration
1400 if (UseAVX > 2) {
1401 Label L_loop_avx512, L_loop_avx2, L_32_byte_head, L_above_threshold, L_below_threshold;
1402
1403 __ BIND(L_copy_bytes);
1404 __ cmpptr(qword_count, (AVX3Threshold / 8));
1405 __ jccb(Assembler::greater, L_above_threshold);
1406 __ jmpb(L_below_threshold);
1407
1408 __ BIND(L_loop_avx512);
1409 __ evmovdqul(xmm0, Address(from, qword_count, Address::times_8, 0), Assembler::AVX_512bit);
1410 __ evmovdqul(Address(dest, qword_count, Address::times_8, 0), xmm0, Assembler::AVX_512bit);
1411 __ bind(L_above_threshold);
1412 __ subptr(qword_count, 8);
1413 __ jcc(Assembler::greaterEqual, L_loop_avx512);
1414 __ jmpb(L_32_byte_head);
1415
1416 __ bind(L_loop_avx2);
1417 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1418 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1419 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1420 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1421 __ bind(L_below_threshold);
1422 __ subptr(qword_count, 8);
1423 __ jcc(Assembler::greaterEqual, L_loop_avx2);
1424
1425 __ bind(L_32_byte_head);
1426 __ addptr(qword_count, 4); // add(8) and sub(4)
1427 __ jccb(Assembler::less, L_end);
1428 } else {
1429 __ BIND(L_loop);
1430 if (UseAVX == 2) {
1431 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1432 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1433 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1434 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1435 } else {
1436 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1437 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1438 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1439 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1440 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1441 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1442 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1443 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1444 }
1445
1446 __ BIND(L_copy_bytes);
1447 __ subptr(qword_count, 8);
1448 __ jcc(Assembler::greaterEqual, L_loop);
1449
1450 __ addptr(qword_count, 4); // add(8) and sub(4)
1451 __ jccb(Assembler::less, L_end);
1452 }
1453 // Copy trailing 32 bytes
1454 if (UseAVX >= 2) {
1455 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1456 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1457 } else {
1458 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1459 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1460 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1461 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1462 }
1463 __ subptr(qword_count, 4);
1464 __ BIND(L_end);
1465 if (UseAVX >= 2) {
1466 // clean upper bits of YMM registers
1467 __ vpxor(xmm0, xmm0);
1468 __ vpxor(xmm1, xmm1);
1469 }
1470 } else {
1471 // Copy 32-bytes per iteration
1472 __ BIND(L_loop);
1473 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1474 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1475 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1476 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1477 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1478 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1479 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1480 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1481
1482 __ BIND(L_copy_bytes);
1483 __ subptr(qword_count, 4);
1484 __ jcc(Assembler::greaterEqual, L_loop);
1485 }
1486 __ addptr(qword_count, 4);
1487 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1488 }
1489
1490 // Arguments:
1491 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1492 // ignored
1493 // name - stub name string
1494 //
1495 // Inputs:
1496 // c_rarg0 - source array address
1497 // c_rarg1 - destination array address
1498 // c_rarg2 - element count, treated as ssize_t, can be zero
1499 //
1500 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1501 // we let the hardware handle it. The one to eight bytes within words,
1502 // dwords or qwords that span cache line boundaries will still be loaded
1503 // and stored atomically.
1504 //
1505 // Side Effects:
1506 // disjoint_byte_copy_entry is set to the no-overlap entry point
1507 // used by generate_conjoint_byte_copy().
1508 //
1509 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1510 __ align(CodeEntryAlignment);
1511 StubCodeMark mark(this, "StubRoutines", name);
1512 address start = __ pc();
1513
1514 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1515 Label L_copy_byte, L_exit;
1516 const Register from = rdi; // source array address
1517 const Register to = rsi; // destination array address
1518 const Register count = rdx; // elements count
1519 const Register byte_count = rcx;
1520 const Register qword_count = count;
1521 const Register end_from = from; // source array end address
1522 const Register end_to = to; // destination array end address
1523 // End pointers are inclusive, and if count is not zero they point
1524 // to the last unit copied: end_to[0] := end_from[0]
1525
1526 __ enter(); // required for proper stackwalking of RuntimeStub frame
1527 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1528
1529 if (entry != NULL) {
1530 *entry = __ pc();
1531 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1532 BLOCK_COMMENT("Entry:");
1533 }
1534
1535 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1536 // r9 and r10 may be used to save non-volatile registers
1537
1538 {
1539 // UnsafeCopyMemory page error: continue after ucm
1540 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1541 // 'from', 'to' and 'count' are now valid
1542 __ movptr(byte_count, count);
1543 __ shrptr(count, 3); // count => qword_count
1544
1545 // Copy from low to high addresses. Use 'to' as scratch.
1546 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1547 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1548 __ negptr(qword_count); // make the count negative
1549 __ jmp(L_copy_bytes);
1550
1551 // Copy trailing qwords
1552 __ BIND(L_copy_8_bytes);
1553 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1554 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1555 __ increment(qword_count);
1556 __ jcc(Assembler::notZero, L_copy_8_bytes);
1557
1558 // Check for and copy trailing dword
1559 __ BIND(L_copy_4_bytes);
1560 __ testl(byte_count, 4);
1561 __ jccb(Assembler::zero, L_copy_2_bytes);
1562 __ movl(rax, Address(end_from, 8));
1563 __ movl(Address(end_to, 8), rax);
1564
1565 __ addptr(end_from, 4);
1566 __ addptr(end_to, 4);
1567
1568 // Check for and copy trailing word
1569 __ BIND(L_copy_2_bytes);
1570 __ testl(byte_count, 2);
1571 __ jccb(Assembler::zero, L_copy_byte);
1572 __ movw(rax, Address(end_from, 8));
1573 __ movw(Address(end_to, 8), rax);
1574
1575 __ addptr(end_from, 2);
1576 __ addptr(end_to, 2);
1577
1578 // Check for and copy trailing byte
1579 __ BIND(L_copy_byte);
1580 __ testl(byte_count, 1);
1581 __ jccb(Assembler::zero, L_exit);
1582 __ movb(rax, Address(end_from, 8));
1583 __ movb(Address(end_to, 8), rax);
1584 }
1585 __ BIND(L_exit);
1586 address ucme_exit_pc = __ pc();
1587 restore_arg_regs();
1588 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1589 __ xorptr(rax, rax); // return 0
1590 __ vzeroupper();
1591 __ leave(); // required for proper stackwalking of RuntimeStub frame
1592 __ ret(0);
1593
1594 {
1595 UnsafeCopyMemoryMark ucmm(this, !aligned, false, ucme_exit_pc);
1596 // Copy in multi-bytes chunks
1597 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1598 __ jmp(L_copy_4_bytes);
1599 }
1600 return start;
1601 }
1602
1603 // Arguments:
1604 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1605 // ignored
1606 // name - stub name string
1607 //
1608 // Inputs:
1609 // c_rarg0 - source array address
1610 // c_rarg1 - destination array address
1611 // c_rarg2 - element count, treated as ssize_t, can be zero
1612 //
1613 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1614 // we let the hardware handle it. The one to eight bytes within words,
1615 // dwords or qwords that span cache line boundaries will still be loaded
1616 // and stored atomically.
1617 //
1618 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1619 address* entry, const char *name) {
1620 __ align(CodeEntryAlignment);
1621 StubCodeMark mark(this, "StubRoutines", name);
1622 address start = __ pc();
1623
1624 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1625 const Register from = rdi; // source array address
1626 const Register to = rsi; // destination array address
1627 const Register count = rdx; // elements count
1628 const Register byte_count = rcx;
1629 const Register qword_count = count;
1630
1631 __ enter(); // required for proper stackwalking of RuntimeStub frame
1632 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1633
1634 if (entry != NULL) {
1635 *entry = __ pc();
1636 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1637 BLOCK_COMMENT("Entry:");
1638 }
1639
1640 array_overlap_test(nooverlap_target, Address::times_1);
1641 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1642 // r9 and r10 may be used to save non-volatile registers
1643
1644 {
1645 // UnsafeCopyMemory page error: continue after ucm
1646 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1647 // 'from', 'to' and 'count' are now valid
1648 __ movptr(byte_count, count);
1649 __ shrptr(count, 3); // count => qword_count
1650
1651 // Copy from high to low addresses.
1652
1653 // Check for and copy trailing byte
1654 __ testl(byte_count, 1);
1655 __ jcc(Assembler::zero, L_copy_2_bytes);
1656 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1657 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1658 __ decrement(byte_count); // Adjust for possible trailing word
1659
1660 // Check for and copy trailing word
1661 __ BIND(L_copy_2_bytes);
1662 __ testl(byte_count, 2);
1663 __ jcc(Assembler::zero, L_copy_4_bytes);
1664 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1665 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1666
1667 // Check for and copy trailing dword
1668 __ BIND(L_copy_4_bytes);
1669 __ testl(byte_count, 4);
1670 __ jcc(Assembler::zero, L_copy_bytes);
1671 __ movl(rax, Address(from, qword_count, Address::times_8));
1672 __ movl(Address(to, qword_count, Address::times_8), rax);
1673 __ jmp(L_copy_bytes);
1674
1675 // Copy trailing qwords
1676 __ BIND(L_copy_8_bytes);
1677 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1678 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1679 __ decrement(qword_count);
1680 __ jcc(Assembler::notZero, L_copy_8_bytes);
1681 }
1682 restore_arg_regs();
1683 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1684 __ xorptr(rax, rax); // return 0
1685 __ vzeroupper();
1686 __ leave(); // required for proper stackwalking of RuntimeStub frame
1687 __ ret(0);
1688
1689 {
1690 // UnsafeCopyMemory page error: continue after ucm
1691 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1692 // Copy in multi-bytes chunks
1693 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1694 }
1695 restore_arg_regs();
1696 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1697 __ xorptr(rax, rax); // return 0
1698 __ vzeroupper();
1699 __ leave(); // required for proper stackwalking of RuntimeStub frame
1700 __ ret(0);
1701
1702 return start;
1703 }
1704
1705 // Arguments:
1706 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1707 // ignored
1708 // name - stub name string
1709 //
1710 // Inputs:
1711 // c_rarg0 - source array address
1712 // c_rarg1 - destination array address
1713 // c_rarg2 - element count, treated as ssize_t, can be zero
1714 //
1715 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1716 // let the hardware handle it. The two or four words within dwords
1717 // or qwords that span cache line boundaries will still be loaded
1718 // and stored atomically.
1719 //
1720 // Side Effects:
1721 // disjoint_short_copy_entry is set to the no-overlap entry point
1722 // used by generate_conjoint_short_copy().
1723 //
1724 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1725 __ align(CodeEntryAlignment);
1726 StubCodeMark mark(this, "StubRoutines", name);
1727 address start = __ pc();
1728
1729 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1730 const Register from = rdi; // source array address
1731 const Register to = rsi; // destination array address
1732 const Register count = rdx; // elements count
1733 const Register word_count = rcx;
1734 const Register qword_count = count;
1735 const Register end_from = from; // source array end address
1736 const Register end_to = to; // destination array end address
1737 // End pointers are inclusive, and if count is not zero they point
1738 // to the last unit copied: end_to[0] := end_from[0]
1739
1740 __ enter(); // required for proper stackwalking of RuntimeStub frame
1741 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1742
1743 if (entry != NULL) {
1744 *entry = __ pc();
1745 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1746 BLOCK_COMMENT("Entry:");
1747 }
1748
1749 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1750 // r9 and r10 may be used to save non-volatile registers
1751
1752 {
1753 // UnsafeCopyMemory page error: continue after ucm
1754 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1755 // 'from', 'to' and 'count' are now valid
1756 __ movptr(word_count, count);
1757 __ shrptr(count, 2); // count => qword_count
1758
1759 // Copy from low to high addresses. Use 'to' as scratch.
1760 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1761 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1762 __ negptr(qword_count);
1763 __ jmp(L_copy_bytes);
1764
1765 // Copy trailing qwords
1766 __ BIND(L_copy_8_bytes);
1767 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1768 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1769 __ increment(qword_count);
1770 __ jcc(Assembler::notZero, L_copy_8_bytes);
1771
1772 // Original 'dest' is trashed, so we can't use it as a
1773 // base register for a possible trailing word copy
1774
1775 // Check for and copy trailing dword
1776 __ BIND(L_copy_4_bytes);
1777 __ testl(word_count, 2);
1778 __ jccb(Assembler::zero, L_copy_2_bytes);
1779 __ movl(rax, Address(end_from, 8));
1780 __ movl(Address(end_to, 8), rax);
1781
1782 __ addptr(end_from, 4);
1783 __ addptr(end_to, 4);
1784
1785 // Check for and copy trailing word
1786 __ BIND(L_copy_2_bytes);
1787 __ testl(word_count, 1);
1788 __ jccb(Assembler::zero, L_exit);
1789 __ movw(rax, Address(end_from, 8));
1790 __ movw(Address(end_to, 8), rax);
1791 }
1792 __ BIND(L_exit);
1793 address ucme_exit_pc = __ pc();
1794 restore_arg_regs();
1795 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1796 __ xorptr(rax, rax); // return 0
1797 __ vzeroupper();
1798 __ leave(); // required for proper stackwalking of RuntimeStub frame
1799 __ ret(0);
1800
1801 {
1802 UnsafeCopyMemoryMark ucmm(this, !aligned, false, ucme_exit_pc);
1803 // Copy in multi-bytes chunks
1804 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1805 __ jmp(L_copy_4_bytes);
1806 }
1807
1808 return start;
1809 }
1810
1811 address generate_fill(BasicType t, bool aligned, const char *name) {
1812 __ align(CodeEntryAlignment);
1813 StubCodeMark mark(this, "StubRoutines", name);
1814 address start = __ pc();
1815
1816 BLOCK_COMMENT("Entry:");
1817
1818 const Register to = c_rarg0; // source array address
1819 const Register value = c_rarg1; // value
1820 const Register count = c_rarg2; // elements count
1821
1822 __ enter(); // required for proper stackwalking of RuntimeStub frame
1823
1824 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1825
1826 __ vzeroupper();
1827 __ leave(); // required for proper stackwalking of RuntimeStub frame
1828 __ ret(0);
1829 return start;
1830 }
1831
1832 // Arguments:
1833 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1834 // ignored
1835 // name - stub name string
1836 //
1837 // Inputs:
1838 // c_rarg0 - source array address
1839 // c_rarg1 - destination array address
1840 // c_rarg2 - element count, treated as ssize_t, can be zero
1841 //
1842 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1843 // let the hardware handle it. The two or four words within dwords
1844 // or qwords that span cache line boundaries will still be loaded
1845 // and stored atomically.
1846 //
1847 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1848 address *entry, const char *name) {
1849 __ align(CodeEntryAlignment);
1850 StubCodeMark mark(this, "StubRoutines", name);
1851 address start = __ pc();
1852
1853 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1854 const Register from = rdi; // source array address
1855 const Register to = rsi; // destination array address
1856 const Register count = rdx; // elements count
1857 const Register word_count = rcx;
1858 const Register qword_count = count;
1859
1860 __ enter(); // required for proper stackwalking of RuntimeStub frame
1861 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1862
1863 if (entry != NULL) {
1864 *entry = __ pc();
1865 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1866 BLOCK_COMMENT("Entry:");
1867 }
1868
1869 array_overlap_test(nooverlap_target, Address::times_2);
1870 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1871 // r9 and r10 may be used to save non-volatile registers
1872
1873 {
1874 // UnsafeCopyMemory page error: continue after ucm
1875 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1876 // 'from', 'to' and 'count' are now valid
1877 __ movptr(word_count, count);
1878 __ shrptr(count, 2); // count => qword_count
1879
1880 // Copy from high to low addresses. Use 'to' as scratch.
1881
1882 // Check for and copy trailing word
1883 __ testl(word_count, 1);
1884 __ jccb(Assembler::zero, L_copy_4_bytes);
1885 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1886 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1887
1888 // Check for and copy trailing dword
1889 __ BIND(L_copy_4_bytes);
1890 __ testl(word_count, 2);
1891 __ jcc(Assembler::zero, L_copy_bytes);
1892 __ movl(rax, Address(from, qword_count, Address::times_8));
1893 __ movl(Address(to, qword_count, Address::times_8), rax);
1894 __ jmp(L_copy_bytes);
1895
1896 // Copy trailing qwords
1897 __ BIND(L_copy_8_bytes);
1898 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1899 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1900 __ decrement(qword_count);
1901 __ jcc(Assembler::notZero, L_copy_8_bytes);
1902 }
1903 restore_arg_regs();
1904 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1905 __ xorptr(rax, rax); // return 0
1906 __ vzeroupper();
1907 __ leave(); // required for proper stackwalking of RuntimeStub frame
1908 __ ret(0);
1909
1910 {
1911 // UnsafeCopyMemory page error: continue after ucm
1912 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1913 // Copy in multi-bytes chunks
1914 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1915 }
1916 restore_arg_regs();
1917 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1918 __ xorptr(rax, rax); // return 0
1919 __ vzeroupper();
1920 __ leave(); // required for proper stackwalking of RuntimeStub frame
1921 __ ret(0);
1922
1923 return start;
1924 }
1925
1926 // Arguments:
1927 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1928 // ignored
1929 // is_oop - true => oop array, so generate store check code
1930 // name - stub name string
1931 //
1932 // Inputs:
1933 // c_rarg0 - source array address
1934 // c_rarg1 - destination array address
1935 // c_rarg2 - element count, treated as ssize_t, can be zero
1936 //
1937 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1938 // the hardware handle it. The two dwords within qwords that span
1939 // cache line boundaries will still be loaded and stored atomicly.
1940 //
1941 // Side Effects:
1942 // disjoint_int_copy_entry is set to the no-overlap entry point
1943 // used by generate_conjoint_int_oop_copy().
1944 //
1945 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1946 const char *name, bool dest_uninitialized = false) {
1947 __ align(CodeEntryAlignment);
1948 StubCodeMark mark(this, "StubRoutines", name);
1949 address start = __ pc();
1950
1951 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1952 const Register from = rdi; // source array address
1953 const Register to = rsi; // destination array address
1954 const Register count = rdx; // elements count
1955 const Register dword_count = rcx;
1956 const Register qword_count = count;
1957 const Register end_from = from; // source array end address
1958 const Register end_to = to; // destination array end address
1959 // End pointers are inclusive, and if count is not zero they point
1960 // to the last unit copied: end_to[0] := end_from[0]
1961
1962 __ enter(); // required for proper stackwalking of RuntimeStub frame
1963 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1964
1965 if (entry != NULL) {
1966 *entry = __ pc();
1967 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1968 BLOCK_COMMENT("Entry:");
1969 }
1970
1971 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1972 // r9 is used to save r15_thread
1973
1974 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
1975 if (dest_uninitialized) {
1976 decorators |= IS_DEST_UNINITIALIZED;
1977 }
1978 if (aligned) {
1979 decorators |= ARRAYCOPY_ALIGNED;
1980 }
1981
1982 BasicType type = is_oop ? T_OBJECT : T_INT;
1983 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1984 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1985
1986 {
1987 // UnsafeCopyMemory page error: continue after ucm
1988 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
1989 // 'from', 'to' and 'count' are now valid
1990 __ movptr(dword_count, count);
1991 __ shrptr(count, 1); // count => qword_count
1992
1993 // Copy from low to high addresses. Use 'to' as scratch.
1994 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1995 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1996 __ negptr(qword_count);
1997 __ jmp(L_copy_bytes);
1998
1999 // Copy trailing qwords
2000 __ BIND(L_copy_8_bytes);
2001 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2002 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2003 __ increment(qword_count);
2004 __ jcc(Assembler::notZero, L_copy_8_bytes);
2005
2006 // Check for and copy trailing dword
2007 __ BIND(L_copy_4_bytes);
2008 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
2009 __ jccb(Assembler::zero, L_exit);
2010 __ movl(rax, Address(end_from, 8));
2011 __ movl(Address(end_to, 8), rax);
2012 }
2013 __ BIND(L_exit);
2014 address ucme_exit_pc = __ pc();
2015 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2016 restore_arg_regs_using_thread();
2017 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2018 __ vzeroupper();
2019 __ xorptr(rax, rax); // return 0
2020 __ leave(); // required for proper stackwalking of RuntimeStub frame
2021 __ ret(0);
2022
2023 {
2024 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, false, ucme_exit_pc);
2025 // Copy in multi-bytes chunks
2026 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2027 __ jmp(L_copy_4_bytes);
2028 }
2029
2030 return start;
2031 }
2032
2033 // Arguments:
2034 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
2035 // ignored
2036 // is_oop - true => oop array, so generate store check code
2037 // name - stub name string
2038 //
2039 // Inputs:
2040 // c_rarg0 - source array address
2041 // c_rarg1 - destination array address
2042 // c_rarg2 - element count, treated as ssize_t, can be zero
2043 //
2044 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
2045 // the hardware handle it. The two dwords within qwords that span
2046 // cache line boundaries will still be loaded and stored atomicly.
2047 //
2048 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
2049 address *entry, const char *name,
2050 bool dest_uninitialized = false) {
2051 __ align(CodeEntryAlignment);
2052 StubCodeMark mark(this, "StubRoutines", name);
2053 address start = __ pc();
2054
2055 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2056 const Register from = rdi; // source array address
2057 const Register to = rsi; // destination array address
2058 const Register count = rdx; // elements count
2059 const Register dword_count = rcx;
2060 const Register qword_count = count;
2061
2062 __ enter(); // required for proper stackwalking of RuntimeStub frame
2063 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2064
2065 if (entry != NULL) {
2066 *entry = __ pc();
2067 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2068 BLOCK_COMMENT("Entry:");
2069 }
2070
2071 array_overlap_test(nooverlap_target, Address::times_4);
2072 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2073 // r9 is used to save r15_thread
2074
2075 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2076 if (dest_uninitialized) {
2077 decorators |= IS_DEST_UNINITIALIZED;
2078 }
2079 if (aligned) {
2080 decorators |= ARRAYCOPY_ALIGNED;
2081 }
2082
2083 BasicType type = is_oop ? T_OBJECT : T_INT;
2084 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2085 // no registers are destroyed by this call
2086 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2087
2088 assert_clean_int(count, rax); // Make sure 'count' is clean int.
2089 {
2090 // UnsafeCopyMemory page error: continue after ucm
2091 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2092 // 'from', 'to' and 'count' are now valid
2093 __ movptr(dword_count, count);
2094 __ shrptr(count, 1); // count => qword_count
2095
2096 // Copy from high to low addresses. Use 'to' as scratch.
2097
2098 // Check for and copy trailing dword
2099 __ testl(dword_count, 1);
2100 __ jcc(Assembler::zero, L_copy_bytes);
2101 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
2102 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
2103 __ jmp(L_copy_bytes);
2104
2105 // Copy trailing qwords
2106 __ BIND(L_copy_8_bytes);
2107 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2108 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2109 __ decrement(qword_count);
2110 __ jcc(Assembler::notZero, L_copy_8_bytes);
2111 }
2112 if (is_oop) {
2113 __ jmp(L_exit);
2114 }
2115 restore_arg_regs_using_thread();
2116 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2117 __ xorptr(rax, rax); // return 0
2118 __ vzeroupper();
2119 __ leave(); // required for proper stackwalking of RuntimeStub frame
2120 __ ret(0);
2121
2122 {
2123 // UnsafeCopyMemory page error: continue after ucm
2124 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2125 // Copy in multi-bytes chunks
2126 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2127 }
2128
2129 __ BIND(L_exit);
2130 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2131 restore_arg_regs_using_thread();
2132 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2133 __ xorptr(rax, rax); // return 0
2134 __ vzeroupper();
2135 __ leave(); // required for proper stackwalking of RuntimeStub frame
2136 __ ret(0);
2137
2138 return start;
2139 }
2140
2141 // Arguments:
2142 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2143 // ignored
2144 // is_oop - true => oop array, so generate store check code
2145 // name - stub name string
2146 //
2147 // Inputs:
2148 // c_rarg0 - source array address
2149 // c_rarg1 - destination array address
2150 // c_rarg2 - element count, treated as ssize_t, can be zero
2151 //
2152 // Side Effects:
2153 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2154 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2155 //
2156 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2157 const char *name, bool dest_uninitialized = false) {
2158 __ align(CodeEntryAlignment);
2159 StubCodeMark mark(this, "StubRoutines", name);
2160 address start = __ pc();
2161
2162 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2163 const Register from = rdi; // source array address
2164 const Register to = rsi; // destination array address
2165 const Register qword_count = rdx; // elements count
2166 const Register end_from = from; // source array end address
2167 const Register end_to = rcx; // destination array end address
2168 const Register saved_count = r11;
2169 // End pointers are inclusive, and if count is not zero they point
2170 // to the last unit copied: end_to[0] := end_from[0]
2171
2172 __ enter(); // required for proper stackwalking of RuntimeStub frame
2173 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2174 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2175
2176 if (entry != NULL) {
2177 *entry = __ pc();
2178 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2179 BLOCK_COMMENT("Entry:");
2180 }
2181
2182 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2183 // r9 is used to save r15_thread
2184 // 'from', 'to' and 'qword_count' are now valid
2185
2186 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2187 if (dest_uninitialized) {
2188 decorators |= IS_DEST_UNINITIALIZED;
2189 }
2190 if (aligned) {
2191 decorators |= ARRAYCOPY_ALIGNED;
2192 }
2193
2194 BasicType type = is_oop ? T_OBJECT : T_LONG;
2195 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2196 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2197 {
2198 // UnsafeCopyMemory page error: continue after ucm
2199 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2200
2201 // Copy from low to high addresses. Use 'to' as scratch.
2202 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2203 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2204 __ negptr(qword_count);
2205 __ jmp(L_copy_bytes);
2206
2207 // Copy trailing qwords
2208 __ BIND(L_copy_8_bytes);
2209 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2210 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2211 __ increment(qword_count);
2212 __ jcc(Assembler::notZero, L_copy_8_bytes);
2213 }
2214 if (is_oop) {
2215 __ jmp(L_exit);
2216 } else {
2217 restore_arg_regs_using_thread();
2218 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2219 __ xorptr(rax, rax); // return 0
2220 __ vzeroupper();
2221 __ leave(); // required for proper stackwalking of RuntimeStub frame
2222 __ ret(0);
2223 }
2224
2225 {
2226 // UnsafeCopyMemory page error: continue after ucm
2227 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2228 // Copy in multi-bytes chunks
2229 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2230 }
2231
2232 __ BIND(L_exit);
2233 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2234 restore_arg_regs_using_thread();
2235 if (is_oop) {
2236 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2237 } else {
2238 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2239 }
2240 __ vzeroupper();
2241 __ xorptr(rax, rax); // return 0
2242 __ leave(); // required for proper stackwalking of RuntimeStub frame
2243 __ ret(0);
2244
2245 return start;
2246 }
2247
2248 // Arguments:
2249 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2250 // ignored
2251 // is_oop - true => oop array, so generate store check code
2252 // name - stub name string
2253 //
2254 // Inputs:
2255 // c_rarg0 - source array address
2256 // c_rarg1 - destination array address
2257 // c_rarg2 - element count, treated as ssize_t, can be zero
2258 //
2259 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2260 address nooverlap_target, address *entry,
2261 const char *name, bool dest_uninitialized = false) {
2262 __ align(CodeEntryAlignment);
2263 StubCodeMark mark(this, "StubRoutines", name);
2264 address start = __ pc();
2265
2266 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2267 const Register from = rdi; // source array address
2268 const Register to = rsi; // destination array address
2269 const Register qword_count = rdx; // elements count
2270 const Register saved_count = rcx;
2271
2272 __ enter(); // required for proper stackwalking of RuntimeStub frame
2273 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2274
2275 if (entry != NULL) {
2276 *entry = __ pc();
2277 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2278 BLOCK_COMMENT("Entry:");
2279 }
2280
2281 array_overlap_test(nooverlap_target, Address::times_8);
2282 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2283 // r9 is used to save r15_thread
2284 // 'from', 'to' and 'qword_count' are now valid
2285
2286 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2287 if (dest_uninitialized) {
2288 decorators |= IS_DEST_UNINITIALIZED;
2289 }
2290 if (aligned) {
2291 decorators |= ARRAYCOPY_ALIGNED;
2292 }
2293
2294 BasicType type = is_oop ? T_OBJECT : T_LONG;
2295 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2296 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2297 {
2298 // UnsafeCopyMemory page error: continue after ucm
2299 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2300
2301 __ jmp(L_copy_bytes);
2302
2303 // Copy trailing qwords
2304 __ BIND(L_copy_8_bytes);
2305 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2306 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2307 __ decrement(qword_count);
2308 __ jcc(Assembler::notZero, L_copy_8_bytes);
2309 }
2310 if (is_oop) {
2311 __ jmp(L_exit);
2312 } else {
2313 restore_arg_regs_using_thread();
2314 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2315 __ xorptr(rax, rax); // return 0
2316 __ vzeroupper();
2317 __ leave(); // required for proper stackwalking of RuntimeStub frame
2318 __ ret(0);
2319 }
2320 {
2321 // UnsafeCopyMemory page error: continue after ucm
2322 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2323
2324 // Copy in multi-bytes chunks
2325 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2326 }
2327 __ BIND(L_exit);
2328 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2329 restore_arg_regs_using_thread();
2330 if (is_oop) {
2331 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2332 } else {
2333 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2334 }
2335 __ vzeroupper();
2336 __ xorptr(rax, rax); // return 0
2337 __ leave(); // required for proper stackwalking of RuntimeStub frame
2338 __ ret(0);
2339
2340 return start;
2341 }
2342
2343
2344 // Helper for generating a dynamic type check.
2345 // Smashes no registers.
2346 void generate_type_check(Register sub_klass,
2347 Register super_check_offset,
2348 Register super_klass,
2349 Label& L_success) {
2350 assert_different_registers(sub_klass, super_check_offset, super_klass);
2351
2352 BLOCK_COMMENT("type_check:");
2353
2354 Label L_miss;
2355
2356 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2357 super_check_offset);
2358 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2359
2360 // Fall through on failure!
2361 __ BIND(L_miss);
2362 }
2363
2364 //
2365 // Generate checkcasting array copy stub
2366 //
2367 // Input:
2368 // c_rarg0 - source array address
2369 // c_rarg1 - destination array address
2370 // c_rarg2 - element count, treated as ssize_t, can be zero
2371 // c_rarg3 - size_t ckoff (super_check_offset)
2372 // not Win64
2373 // c_rarg4 - oop ckval (super_klass)
2374 // Win64
2375 // rsp+40 - oop ckval (super_klass)
2376 //
2377 // Output:
2378 // rax == 0 - success
2379 // rax == -1^K - failure, where K is partial transfer count
2380 //
2381 address generate_checkcast_copy(const char *name, address *entry,
2382 bool dest_uninitialized = false) {
2383
2384 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2385
2386 // Input registers (after setup_arg_regs)
2387 const Register from = rdi; // source array address
2388 const Register to = rsi; // destination array address
2389 const Register length = rdx; // elements count
2390 const Register ckoff = rcx; // super_check_offset
2391 const Register ckval = r8; // super_klass
2392
2393 // Registers used as temps (r13, r14 are save-on-entry)
2394 const Register end_from = from; // source array end address
2395 const Register end_to = r13; // destination array end address
2396 const Register count = rdx; // -(count_remaining)
2397 const Register r14_length = r14; // saved copy of length
2398 // End pointers are inclusive, and if length is not zero they point
2399 // to the last unit copied: end_to[0] := end_from[0]
2400
2401 const Register rax_oop = rax; // actual oop copied
2402 const Register r11_klass = r11; // oop._klass
2403
2404 //---------------------------------------------------------------
2405 // Assembler stub will be used for this call to arraycopy
2406 // if the two arrays are subtypes of Object[] but the
2407 // destination array type is not equal to or a supertype
2408 // of the source type. Each element must be separately
2409 // checked.
2410
2411 __ align(CodeEntryAlignment);
2412 StubCodeMark mark(this, "StubRoutines", name);
2413 address start = __ pc();
2414
2415 __ enter(); // required for proper stackwalking of RuntimeStub frame
2416
2417 #ifdef ASSERT
2418 // caller guarantees that the arrays really are different
2419 // otherwise, we would have to make conjoint checks
2420 { Label L;
2421 array_overlap_test(L, TIMES_OOP);
2422 __ stop("checkcast_copy within a single array");
2423 __ bind(L);
2424 }
2425 #endif //ASSERT
2426
2427 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2428 // ckoff => rcx, ckval => r8
2429 // r9 and r10 may be used to save non-volatile registers
2430 #ifdef _WIN64
2431 // last argument (#4) is on stack on Win64
2432 __ movptr(ckval, Address(rsp, 6 * wordSize));
2433 #endif
2434
2435 // Caller of this entry point must set up the argument registers.
2436 if (entry != NULL) {
2437 *entry = __ pc();
2438 BLOCK_COMMENT("Entry:");
2439 }
2440
2441 // allocate spill slots for r13, r14
2442 enum {
2443 saved_r13_offset,
2444 saved_r14_offset,
2445 saved_r10_offset,
2446 saved_rbp_offset
2447 };
2448 __ subptr(rsp, saved_rbp_offset * wordSize);
2449 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2450 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2451 __ movptr(Address(rsp, saved_r10_offset * wordSize), r10);
2452
2453 #ifdef ASSERT
2454 Label L2;
2455 __ get_thread(r14);
2456 __ cmpptr(r15_thread, r14);
2457 __ jcc(Assembler::equal, L2);
2458 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
2459 __ bind(L2);
2460 #endif // ASSERT
2461
2462 // check that int operands are properly extended to size_t
2463 assert_clean_int(length, rax);
2464 assert_clean_int(ckoff, rax);
2465
2466 #ifdef ASSERT
2467 BLOCK_COMMENT("assert consistent ckoff/ckval");
2468 // The ckoff and ckval must be mutually consistent,
2469 // even though caller generates both.
2470 { Label L;
2471 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2472 __ cmpl(ckoff, Address(ckval, sco_offset));
2473 __ jcc(Assembler::equal, L);
2474 __ stop("super_check_offset inconsistent");
2475 __ bind(L);
2476 }
2477 #endif //ASSERT
2478
2479 // Loop-invariant addresses. They are exclusive end pointers.
2480 Address end_from_addr(from, length, TIMES_OOP, 0);
2481 Address end_to_addr(to, length, TIMES_OOP, 0);
2482 // Loop-variant addresses. They assume post-incremented count < 0.
2483 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2484 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2485
2486 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST | ARRAYCOPY_DISJOINT;
2487 if (dest_uninitialized) {
2488 decorators |= IS_DEST_UNINITIALIZED;
2489 }
2490
2491 BasicType type = T_OBJECT;
2492 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2493 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2494
2495 // Copy from low to high addresses, indexed from the end of each array.
2496 __ lea(end_from, end_from_addr);
2497 __ lea(end_to, end_to_addr);
2498 __ movptr(r14_length, length); // save a copy of the length
2499 assert(length == count, ""); // else fix next line:
2500 __ negptr(count); // negate and test the length
2501 __ jcc(Assembler::notZero, L_load_element);
2502
2503 // Empty array: Nothing to do.
2504 __ xorptr(rax, rax); // return 0 on (trivial) success
2505 __ jmp(L_done);
2506
2507 // ======== begin loop ========
2508 // (Loop is rotated; its entry is L_load_element.)
2509 // Loop control:
2510 // for (count = -count; count != 0; count++)
2511 // Base pointers src, dst are biased by 8*(count-1),to last element.
2512 __ align(OptoLoopAlignment);
2513
2514 __ BIND(L_store_element);
2515 __ store_heap_oop(to_element_addr, rax_oop, noreg, noreg, AS_RAW); // store the oop
2516 __ increment(count); // increment the count toward zero
2517 __ jcc(Assembler::zero, L_do_card_marks);
2518
2519 // ======== loop entry is here ========
2520 __ BIND(L_load_element);
2521 __ load_heap_oop(rax_oop, from_element_addr, noreg, noreg, AS_RAW); // load the oop
2522 __ testptr(rax_oop, rax_oop);
2523 __ jcc(Assembler::zero, L_store_element);
2524
2525 __ load_klass(r11_klass, rax_oop, rscratch1);// query the object klass
2526 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2527 // ======== end loop ========
2528
2529 // It was a real error; we must depend on the caller to finish the job.
2530 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2531 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2532 // and report their number to the caller.
2533 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2534 Label L_post_barrier;
2535 __ addptr(r14_length, count); // K = (original - remaining) oops
2536 __ movptr(rax, r14_length); // save the value
2537 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2538 __ jccb(Assembler::notZero, L_post_barrier);
2539 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2540
2541 // Come here on success only.
2542 __ BIND(L_do_card_marks);
2543 __ xorptr(rax, rax); // return 0 on success
2544
2545 __ BIND(L_post_barrier);
2546 bs->arraycopy_epilogue(_masm, decorators, type, from, to, r14_length);
2547
2548 // Common exit point (success or failure).
2549 __ BIND(L_done);
2550 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2551 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2552 __ movptr(r10, Address(rsp, saved_r10_offset * wordSize));
2553 restore_arg_regs();
2554 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2555 __ leave(); // required for proper stackwalking of RuntimeStub frame
2556 __ ret(0);
2557
2558 return start;
2559 }
2560
2561 //
2562 // Generate 'unsafe' array copy stub
2563 // Though just as safe as the other stubs, it takes an unscaled
2564 // size_t argument instead of an element count.
2565 //
2566 // Input:
2567 // c_rarg0 - source array address
2568 // c_rarg1 - destination array address
2569 // c_rarg2 - byte count, treated as ssize_t, can be zero
2570 //
2571 // Examines the alignment of the operands and dispatches
2572 // to a long, int, short, or byte copy loop.
2573 //
2574 address generate_unsafe_copy(const char *name,
2575 address byte_copy_entry, address short_copy_entry,
2576 address int_copy_entry, address long_copy_entry) {
2577
2578 Label L_long_aligned, L_int_aligned, L_short_aligned;
2579
2580 // Input registers (before setup_arg_regs)
2581 const Register from = c_rarg0; // source array address
2582 const Register to = c_rarg1; // destination array address
2583 const Register size = c_rarg2; // byte count (size_t)
2584
2585 // Register used as a temp
2586 const Register bits = rax; // test copy of low bits
2587
2588 __ align(CodeEntryAlignment);
2589 StubCodeMark mark(this, "StubRoutines", name);
2590 address start = __ pc();
2591
2592 __ enter(); // required for proper stackwalking of RuntimeStub frame
2593
2594 // bump this on entry, not on exit:
2595 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2596
2597 __ mov(bits, from);
2598 __ orptr(bits, to);
2599 __ orptr(bits, size);
2600
2601 __ testb(bits, BytesPerLong-1);
2602 __ jccb(Assembler::zero, L_long_aligned);
2603
2604 __ testb(bits, BytesPerInt-1);
2605 __ jccb(Assembler::zero, L_int_aligned);
2606
2607 __ testb(bits, BytesPerShort-1);
2608 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2609
2610 __ BIND(L_short_aligned);
2611 __ shrptr(size, LogBytesPerShort); // size => short_count
2612 __ jump(RuntimeAddress(short_copy_entry));
2613
2614 __ BIND(L_int_aligned);
2615 __ shrptr(size, LogBytesPerInt); // size => int_count
2616 __ jump(RuntimeAddress(int_copy_entry));
2617
2618 __ BIND(L_long_aligned);
2619 __ shrptr(size, LogBytesPerLong); // size => qword_count
2620 __ jump(RuntimeAddress(long_copy_entry));
2621
2622 return start;
2623 }
2624
2625 // Perform range checks on the proposed arraycopy.
2626 // Kills temp, but nothing else.
2627 // Also, clean the sign bits of src_pos and dst_pos.
2628 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2629 Register src_pos, // source position (c_rarg1)
2630 Register dst, // destination array oo (c_rarg2)
2631 Register dst_pos, // destination position (c_rarg3)
2632 Register length,
2633 Register temp,
2634 Label& L_failed) {
2635 BLOCK_COMMENT("arraycopy_range_checks:");
2636
2637 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2638 __ movl(temp, length);
2639 __ addl(temp, src_pos); // src_pos + length
2640 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2641 __ jcc(Assembler::above, L_failed);
2642
2643 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2644 __ movl(temp, length);
2645 __ addl(temp, dst_pos); // dst_pos + length
2646 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2647 __ jcc(Assembler::above, L_failed);
2648
2649 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2650 // Move with sign extension can be used since they are positive.
2651 __ movslq(src_pos, src_pos);
2652 __ movslq(dst_pos, dst_pos);
2653
2654 BLOCK_COMMENT("arraycopy_range_checks done");
2655 }
2656
2657 //
2658 // Generate generic array copy stubs
2659 //
2660 // Input:
2661 // c_rarg0 - src oop
2662 // c_rarg1 - src_pos (32-bits)
2663 // c_rarg2 - dst oop
2664 // c_rarg3 - dst_pos (32-bits)
2665 // not Win64
2666 // c_rarg4 - element count (32-bits)
2667 // Win64
2668 // rsp+40 - element count (32-bits)
2669 //
2670 // Output:
2671 // rax == 0 - success
2672 // rax == -1^K - failure, where K is partial transfer count
2673 //
2674 address generate_generic_copy(const char *name,
2675 address byte_copy_entry, address short_copy_entry,
2676 address int_copy_entry, address oop_copy_entry,
2677 address long_copy_entry, address checkcast_copy_entry) {
2678
2679 Label L_failed, L_failed_0, L_objArray;
2680 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2681
2682 // Input registers
2683 const Register src = c_rarg0; // source array oop
2684 const Register src_pos = c_rarg1; // source position
2685 const Register dst = c_rarg2; // destination array oop
2686 const Register dst_pos = c_rarg3; // destination position
2687 #ifndef _WIN64
2688 const Register length = c_rarg4;
2689 const Register rklass_tmp = r9; // load_klass
2690 #else
2691 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2692 const Register rklass_tmp = rdi; // load_klass
2693 #endif
2694
2695 { int modulus = CodeEntryAlignment;
2696 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2697 int advance = target - (__ offset() % modulus);
2698 if (advance < 0) advance += modulus;
2699 if (advance > 0) __ nop(advance);
2700 }
2701 StubCodeMark mark(this, "StubRoutines", name);
2702
2703 // Short-hop target to L_failed. Makes for denser prologue code.
2704 __ BIND(L_failed_0);
2705 __ jmp(L_failed);
2706 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2707
2708 __ align(CodeEntryAlignment);
2709 address start = __ pc();
2710
2711 __ enter(); // required for proper stackwalking of RuntimeStub frame
2712
2713 // bump this on entry, not on exit:
2714 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2715
2716 //-----------------------------------------------------------------------
2717 // Assembler stub will be used for this call to arraycopy
2718 // if the following conditions are met:
2719 //
2720 // (1) src and dst must not be null.
2721 // (2) src_pos must not be negative.
2722 // (3) dst_pos must not be negative.
2723 // (4) length must not be negative.
2724 // (5) src klass and dst klass should be the same and not NULL.
2725 // (6) src and dst should be arrays.
2726 // (7) src_pos + length must not exceed length of src.
2727 // (8) dst_pos + length must not exceed length of dst.
2728 //
2729
2730 // if (src == NULL) return -1;
2731 __ testptr(src, src); // src oop
2732 size_t j1off = __ offset();
2733 __ jccb(Assembler::zero, L_failed_0);
2734
2735 // if (src_pos < 0) return -1;
2736 __ testl(src_pos, src_pos); // src_pos (32-bits)
2737 __ jccb(Assembler::negative, L_failed_0);
2738
2739 // if (dst == NULL) return -1;
2740 __ testptr(dst, dst); // dst oop
2741 __ jccb(Assembler::zero, L_failed_0);
2742
2743 // if (dst_pos < 0) return -1;
2744 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2745 size_t j4off = __ offset();
2746 __ jccb(Assembler::negative, L_failed_0);
2747
2748 // The first four tests are very dense code,
2749 // but not quite dense enough to put four
2750 // jumps in a 16-byte instruction fetch buffer.
2751 // That's good, because some branch predicters
2752 // do not like jumps so close together.
2753 // Make sure of this.
2754 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2755
2756 // registers used as temp
2757 const Register r11_length = r11; // elements count to copy
2758 const Register r10_src_klass = r10; // array klass
2759
2760 // if (length < 0) return -1;
2761 __ movl(r11_length, length); // length (elements count, 32-bits value)
2762 __ testl(r11_length, r11_length);
2763 __ jccb(Assembler::negative, L_failed_0);
2764
2765 __ load_klass(r10_src_klass, src, rklass_tmp);
2766 #ifdef ASSERT
2767 // assert(src->klass() != NULL);
2768 {
2769 BLOCK_COMMENT("assert klasses not null {");
2770 Label L1, L2;
2771 __ testptr(r10_src_klass, r10_src_klass);
2772 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2773 __ bind(L1);
2774 __ stop("broken null klass");
2775 __ bind(L2);
2776 __ load_klass(rax, dst, rklass_tmp);
2777 __ cmpq(rax, 0);
2778 __ jcc(Assembler::equal, L1); // this would be broken also
2779 BLOCK_COMMENT("} assert klasses not null done");
2780 }
2781 #endif
2782
2783 // Load layout helper (32-bits)
2784 //
2785 // |array_tag| | header_size | element_type | |log2_element_size|
2786 // 32 30 24 16 8 2 0
2787 //
2788 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2789 //
2790
2791 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2792
2793 // Handle objArrays completely differently...
2794 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2795 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2796 __ jcc(Assembler::equal, L_objArray);
2797
2798 // if (src->klass() != dst->klass()) return -1;
2799 __ load_klass(rax, dst, rklass_tmp);
2800 __ cmpq(r10_src_klass, rax);
2801 __ jcc(Assembler::notEqual, L_failed);
2802
2803 const Register rax_lh = rax; // layout helper
2804 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2805
2806 // if (!src->is_Array()) return -1;
2807 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2808 __ jcc(Assembler::greaterEqual, L_failed);
2809
2810 // At this point, it is known to be a typeArray (array_tag 0x3).
2811 #ifdef ASSERT
2812 {
2813 BLOCK_COMMENT("assert primitive array {");
2814 Label L;
2815 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2816 __ jcc(Assembler::greaterEqual, L);
2817 __ stop("must be a primitive array");
2818 __ bind(L);
2819 BLOCK_COMMENT("} assert primitive array done");
2820 }
2821 #endif
2822
2823 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2824 r10, L_failed);
2825
2826 // TypeArrayKlass
2827 //
2828 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2829 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2830 //
2831
2832 const Register r10_offset = r10; // array offset
2833 const Register rax_elsize = rax_lh; // element size
2834
2835 __ movl(r10_offset, rax_lh);
2836 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2837 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2838 __ addptr(src, r10_offset); // src array offset
2839 __ addptr(dst, r10_offset); // dst array offset
2840 BLOCK_COMMENT("choose copy loop based on element size");
2841 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2842
2843 // next registers should be set before the jump to corresponding stub
2844 const Register from = c_rarg0; // source array address
2845 const Register to = c_rarg1; // destination array address
2846 const Register count = c_rarg2; // elements count
2847
2848 // 'from', 'to', 'count' registers should be set in such order
2849 // since they are the same as 'src', 'src_pos', 'dst'.
2850
2851 __ BIND(L_copy_bytes);
2852 __ cmpl(rax_elsize, 0);
2853 __ jccb(Assembler::notEqual, L_copy_shorts);
2854 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2855 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2856 __ movl2ptr(count, r11_length); // length
2857 __ jump(RuntimeAddress(byte_copy_entry));
2858
2859 __ BIND(L_copy_shorts);
2860 __ cmpl(rax_elsize, LogBytesPerShort);
2861 __ jccb(Assembler::notEqual, L_copy_ints);
2862 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2863 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2864 __ movl2ptr(count, r11_length); // length
2865 __ jump(RuntimeAddress(short_copy_entry));
2866
2867 __ BIND(L_copy_ints);
2868 __ cmpl(rax_elsize, LogBytesPerInt);
2869 __ jccb(Assembler::notEqual, L_copy_longs);
2870 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2871 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2872 __ movl2ptr(count, r11_length); // length
2873 __ jump(RuntimeAddress(int_copy_entry));
2874
2875 __ BIND(L_copy_longs);
2876 #ifdef ASSERT
2877 {
2878 BLOCK_COMMENT("assert long copy {");
2879 Label L;
2880 __ cmpl(rax_elsize, LogBytesPerLong);
2881 __ jcc(Assembler::equal, L);
2882 __ stop("must be long copy, but elsize is wrong");
2883 __ bind(L);
2884 BLOCK_COMMENT("} assert long copy done");
2885 }
2886 #endif
2887 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2888 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2889 __ movl2ptr(count, r11_length); // length
2890 __ jump(RuntimeAddress(long_copy_entry));
2891
2892 // ObjArrayKlass
2893 __ BIND(L_objArray);
2894 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2895
2896 Label L_plain_copy, L_checkcast_copy;
2897 // test array classes for subtyping
2898 __ load_klass(rax, dst, rklass_tmp);
2899 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2900 __ jcc(Assembler::notEqual, L_checkcast_copy);
2901
2902 // Identically typed arrays can be copied without element-wise checks.
2903 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2904 r10, L_failed);
2905
2906 __ lea(from, Address(src, src_pos, TIMES_OOP,
2907 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2908 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2909 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2910 __ movl2ptr(count, r11_length); // length
2911 __ BIND(L_plain_copy);
2912 __ jump(RuntimeAddress(oop_copy_entry));
2913
2914 __ BIND(L_checkcast_copy);
2915 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2916 {
2917 // Before looking at dst.length, make sure dst is also an objArray.
2918 __ cmpl(Address(rax, lh_offset), objArray_lh);
2919 __ jcc(Assembler::notEqual, L_failed);
2920
2921 // It is safe to examine both src.length and dst.length.
2922 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2923 rax, L_failed);
2924
2925 const Register r11_dst_klass = r11;
2926 __ load_klass(r11_dst_klass, dst, rklass_tmp); // reload
2927
2928 // Marshal the base address arguments now, freeing registers.
2929 __ lea(from, Address(src, src_pos, TIMES_OOP,
2930 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2931 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2932 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2933 __ movl(count, length); // length (reloaded)
2934 Register sco_temp = c_rarg3; // this register is free now
2935 assert_different_registers(from, to, count, sco_temp,
2936 r11_dst_klass, r10_src_klass);
2937 assert_clean_int(count, sco_temp);
2938
2939 // Generate the type check.
2940 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2941 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2942 assert_clean_int(sco_temp, rax);
2943 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2944
2945 // Fetch destination element klass from the ObjArrayKlass header.
2946 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2947 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2948 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2949 assert_clean_int(sco_temp, rax);
2950
2951 // the checkcast_copy loop needs two extra arguments:
2952 assert(c_rarg3 == sco_temp, "#3 already in place");
2953 // Set up arguments for checkcast_copy_entry.
2954 setup_arg_regs(4);
2955 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2956 __ jump(RuntimeAddress(checkcast_copy_entry));
2957 }
2958
2959 __ BIND(L_failed);
2960 __ xorptr(rax, rax);
2961 __ notptr(rax); // return -1
2962 __ leave(); // required for proper stackwalking of RuntimeStub frame
2963 __ ret(0);
2964
2965 return start;
2966 }
2967
2968 address generate_data_cache_writeback() {
2969 const Register src = c_rarg0; // source address
2970
2971 __ align(CodeEntryAlignment);
2972
2973 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback");
2974
2975 address start = __ pc();
2976 __ enter();
2977 __ cache_wb(Address(src, 0));
2978 __ leave();
2979 __ ret(0);
2980
2981 return start;
2982 }
2983
2984 address generate_data_cache_writeback_sync() {
2985 const Register is_pre = c_rarg0; // pre or post sync
2986
2987 __ align(CodeEntryAlignment);
2988
2989 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback_sync");
2990
2991 // pre wbsync is a no-op
2992 // post wbsync translates to an sfence
2993
2994 Label skip;
2995 address start = __ pc();
2996 __ enter();
2997 __ cmpl(is_pre, 0);
2998 __ jcc(Assembler::notEqual, skip);
2999 __ cache_wbsync(false);
3000 __ bind(skip);
3001 __ leave();
3002 __ ret(0);
3003
3004 return start;
3005 }
3006
3007 void generate_arraycopy_stubs() {
3008 address entry;
3009 address entry_jbyte_arraycopy;
3010 address entry_jshort_arraycopy;
3011 address entry_jint_arraycopy;
3012 address entry_oop_arraycopy;
3013 address entry_jlong_arraycopy;
3014 address entry_checkcast_arraycopy;
3015
3016 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
3017 "jbyte_disjoint_arraycopy");
3018 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
3019 "jbyte_arraycopy");
3020
3021 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
3022 "jshort_disjoint_arraycopy");
3023 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
3024 "jshort_arraycopy");
3025
3026 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
3027 "jint_disjoint_arraycopy");
3028 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
3029 &entry_jint_arraycopy, "jint_arraycopy");
3030
3031 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
3032 "jlong_disjoint_arraycopy");
3033 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
3034 &entry_jlong_arraycopy, "jlong_arraycopy");
3035
3036
3037 if (UseCompressedOops) {
3038 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
3039 "oop_disjoint_arraycopy");
3040 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
3041 &entry_oop_arraycopy, "oop_arraycopy");
3042 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
3043 "oop_disjoint_arraycopy_uninit",
3044 /*dest_uninitialized*/true);
3045 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
3046 NULL, "oop_arraycopy_uninit",
3047 /*dest_uninitialized*/true);
3048 } else {
3049 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
3050 "oop_disjoint_arraycopy");
3051 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
3052 &entry_oop_arraycopy, "oop_arraycopy");
3053 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
3054 "oop_disjoint_arraycopy_uninit",
3055 /*dest_uninitialized*/true);
3056 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
3057 NULL, "oop_arraycopy_uninit",
3058 /*dest_uninitialized*/true);
3059 }
3060
3061 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
3062 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
3063 /*dest_uninitialized*/true);
3064
3065 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
3066 entry_jbyte_arraycopy,
3067 entry_jshort_arraycopy,
3068 entry_jint_arraycopy,
3069 entry_jlong_arraycopy);
3070 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
3071 entry_jbyte_arraycopy,
3072 entry_jshort_arraycopy,
3073 entry_jint_arraycopy,
3074 entry_oop_arraycopy,
3075 entry_jlong_arraycopy,
3076 entry_checkcast_arraycopy);
3077
3078 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
3079 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
3080 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
3081 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
3082 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
3083 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
3084
3085 // We don't generate specialized code for HeapWord-aligned source
3086 // arrays, so just use the code we've already generated
3087 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
3088 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
3089
3090 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
3091 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
3092
3093 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
3094 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
3095
3096 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
3097 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
3098
3099 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
3100 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
3101
3102 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
3103 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
3104 }
3105
3106 // AES intrinsic stubs
3107 enum {AESBlockSize = 16};
3108
3109 address generate_key_shuffle_mask() {
3110 __ align(16);
3111 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
3112 address start = __ pc();
3113 __ emit_data64( 0x0405060700010203, relocInfo::none );
3114 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
3115 return start;
3116 }
3117
3118 address generate_counter_shuffle_mask() {
3119 __ align(16);
3120 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
3121 address start = __ pc();
3122 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3123 __ emit_data64(0x0001020304050607, relocInfo::none);
3124 return start;
3125 }
3126
3127 // Utility routine for loading a 128-bit key word in little endian format
3128 // can optionally specify that the shuffle mask is already in an xmmregister
3129 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
3130 __ movdqu(xmmdst, Address(key, offset));
3131 if (xmm_shuf_mask != NULL) {
3132 __ pshufb(xmmdst, xmm_shuf_mask);
3133 } else {
3134 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3135 }
3136 }
3137
3138 // Utility routine for increase 128bit counter (iv in CTR mode)
3139 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
3140 __ pextrq(reg, xmmdst, 0x0);
3141 __ addq(reg, inc_delta);
3142 __ pinsrq(xmmdst, reg, 0x0);
3143 __ jcc(Assembler::carryClear, next_block); // jump if no carry
3144 __ pextrq(reg, xmmdst, 0x01); // Carry
3145 __ addq(reg, 0x01);
3146 __ pinsrq(xmmdst, reg, 0x01); //Carry end
3147 __ BIND(next_block); // next instruction
3148 }
3149
3150 // Arguments:
3151 //
3152 // Inputs:
3153 // c_rarg0 - source byte array address
3154 // c_rarg1 - destination byte array address
3155 // c_rarg2 - K (key) in little endian int array
3156 //
3157 address generate_aescrypt_encryptBlock() {
3158 assert(UseAES, "need AES instructions and misaligned SSE support");
3159 __ align(CodeEntryAlignment);
3160 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
3161 Label L_doLast;
3162 address start = __ pc();
3163
3164 const Register from = c_rarg0; // source array address
3165 const Register to = c_rarg1; // destination array address
3166 const Register key = c_rarg2; // key array address
3167 const Register keylen = rax;
3168
3169 const XMMRegister xmm_result = xmm0;
3170 const XMMRegister xmm_key_shuf_mask = xmm1;
3171 // On win64 xmm6-xmm15 must be preserved so don't use them.
3172 const XMMRegister xmm_temp1 = xmm2;
3173 const XMMRegister xmm_temp2 = xmm3;
3174 const XMMRegister xmm_temp3 = xmm4;
3175 const XMMRegister xmm_temp4 = xmm5;
3176
3177 __ enter(); // required for proper stackwalking of RuntimeStub frame
3178
3179 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3180 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3181
3182 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3183 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
3184
3185 // For encryption, the java expanded key ordering is just what we need
3186 // we don't know if the key is aligned, hence not using load-execute form
3187
3188 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3189 __ pxor(xmm_result, xmm_temp1);
3190
3191 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3192 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3193 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3194 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3195
3196 __ aesenc(xmm_result, xmm_temp1);
3197 __ aesenc(xmm_result, xmm_temp2);
3198 __ aesenc(xmm_result, xmm_temp3);
3199 __ aesenc(xmm_result, xmm_temp4);
3200
3201 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3202 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3203 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3204 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3205
3206 __ aesenc(xmm_result, xmm_temp1);
3207 __ aesenc(xmm_result, xmm_temp2);
3208 __ aesenc(xmm_result, xmm_temp3);
3209 __ aesenc(xmm_result, xmm_temp4);
3210
3211 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3212 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3213
3214 __ cmpl(keylen, 44);
3215 __ jccb(Assembler::equal, L_doLast);
3216
3217 __ aesenc(xmm_result, xmm_temp1);
3218 __ aesenc(xmm_result, xmm_temp2);
3219
3220 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3221 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3222
3223 __ cmpl(keylen, 52);
3224 __ jccb(Assembler::equal, L_doLast);
3225
3226 __ aesenc(xmm_result, xmm_temp1);
3227 __ aesenc(xmm_result, xmm_temp2);
3228
3229 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3230 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3231
3232 __ BIND(L_doLast);
3233 __ aesenc(xmm_result, xmm_temp1);
3234 __ aesenclast(xmm_result, xmm_temp2);
3235 __ movdqu(Address(to, 0), xmm_result); // store the result
3236 __ xorptr(rax, rax); // return 0
3237 __ leave(); // required for proper stackwalking of RuntimeStub frame
3238 __ ret(0);
3239
3240 return start;
3241 }
3242
3243
3244 // Arguments:
3245 //
3246 // Inputs:
3247 // c_rarg0 - source byte array address
3248 // c_rarg1 - destination byte array address
3249 // c_rarg2 - K (key) in little endian int array
3250 //
3251 address generate_aescrypt_decryptBlock() {
3252 assert(UseAES, "need AES instructions and misaligned SSE support");
3253 __ align(CodeEntryAlignment);
3254 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3255 Label L_doLast;
3256 address start = __ pc();
3257
3258 const Register from = c_rarg0; // source array address
3259 const Register to = c_rarg1; // destination array address
3260 const Register key = c_rarg2; // key array address
3261 const Register keylen = rax;
3262
3263 const XMMRegister xmm_result = xmm0;
3264 const XMMRegister xmm_key_shuf_mask = xmm1;
3265 // On win64 xmm6-xmm15 must be preserved so don't use them.
3266 const XMMRegister xmm_temp1 = xmm2;
3267 const XMMRegister xmm_temp2 = xmm3;
3268 const XMMRegister xmm_temp3 = xmm4;
3269 const XMMRegister xmm_temp4 = xmm5;
3270
3271 __ enter(); // required for proper stackwalking of RuntimeStub frame
3272
3273 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3274 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3275
3276 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3277 __ movdqu(xmm_result, Address(from, 0));
3278
3279 // for decryption java expanded key ordering is rotated one position from what we want
3280 // so we start from 0x10 here and hit 0x00 last
3281 // we don't know if the key is aligned, hence not using load-execute form
3282 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3283 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3284 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3285 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3286
3287 __ pxor (xmm_result, xmm_temp1);
3288 __ aesdec(xmm_result, xmm_temp2);
3289 __ aesdec(xmm_result, xmm_temp3);
3290 __ aesdec(xmm_result, xmm_temp4);
3291
3292 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3293 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3294 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3295 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3296
3297 __ aesdec(xmm_result, xmm_temp1);
3298 __ aesdec(xmm_result, xmm_temp2);
3299 __ aesdec(xmm_result, xmm_temp3);
3300 __ aesdec(xmm_result, xmm_temp4);
3301
3302 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3303 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3304 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3305
3306 __ cmpl(keylen, 44);
3307 __ jccb(Assembler::equal, L_doLast);
3308
3309 __ aesdec(xmm_result, xmm_temp1);
3310 __ aesdec(xmm_result, xmm_temp2);
3311
3312 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3313 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3314
3315 __ cmpl(keylen, 52);
3316 __ jccb(Assembler::equal, L_doLast);
3317
3318 __ aesdec(xmm_result, xmm_temp1);
3319 __ aesdec(xmm_result, xmm_temp2);
3320
3321 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3322 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3323
3324 __ BIND(L_doLast);
3325 __ aesdec(xmm_result, xmm_temp1);
3326 __ aesdec(xmm_result, xmm_temp2);
3327
3328 // for decryption the aesdeclast operation is always on key+0x00
3329 __ aesdeclast(xmm_result, xmm_temp3);
3330 __ movdqu(Address(to, 0), xmm_result); // store the result
3331 __ xorptr(rax, rax); // return 0
3332 __ leave(); // required for proper stackwalking of RuntimeStub frame
3333 __ ret(0);
3334
3335 return start;
3336 }
3337
3338
3339 // Arguments:
3340 //
3341 // Inputs:
3342 // c_rarg0 - source byte array address
3343 // c_rarg1 - destination byte array address
3344 // c_rarg2 - K (key) in little endian int array
3345 // c_rarg3 - r vector byte array address
3346 // c_rarg4 - input length
3347 //
3348 // Output:
3349 // rax - input length
3350 //
3351 address generate_cipherBlockChaining_encryptAESCrypt() {
3352 assert(UseAES, "need AES instructions and misaligned SSE support");
3353 __ align(CodeEntryAlignment);
3354 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3355 address start = __ pc();
3356
3357 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3358 const Register from = c_rarg0; // source array address
3359 const Register to = c_rarg1; // destination array address
3360 const Register key = c_rarg2; // key array address
3361 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3362 // and left with the results of the last encryption block
3363 #ifndef _WIN64
3364 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3365 #else
3366 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3367 const Register len_reg = r11; // pick the volatile windows register
3368 #endif
3369 const Register pos = rax;
3370
3371 // xmm register assignments for the loops below
3372 const XMMRegister xmm_result = xmm0;
3373 const XMMRegister xmm_temp = xmm1;
3374 // keys 0-10 preloaded into xmm2-xmm12
3375 const int XMM_REG_NUM_KEY_FIRST = 2;
3376 const int XMM_REG_NUM_KEY_LAST = 15;
3377 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3378 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3379 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3380 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3381 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3382
3383 __ enter(); // required for proper stackwalking of RuntimeStub frame
3384
3385 #ifdef _WIN64
3386 // on win64, fill len_reg from stack position
3387 __ movl(len_reg, len_mem);
3388 #else
3389 __ push(len_reg); // Save
3390 #endif
3391
3392 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3393 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3394 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3395 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3396 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3397 offset += 0x10;
3398 }
3399 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3400
3401 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3402 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3403 __ cmpl(rax, 44);
3404 __ jcc(Assembler::notEqual, L_key_192_256);
3405
3406 // 128 bit code follows here
3407 __ movptr(pos, 0);
3408 __ align(OptoLoopAlignment);
3409
3410 __ BIND(L_loopTop_128);
3411 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3412 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3413 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3414 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3415 __ aesenc(xmm_result, as_XMMRegister(rnum));
3416 }
3417 __ aesenclast(xmm_result, xmm_key10);
3418 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3419 // no need to store r to memory until we exit
3420 __ addptr(pos, AESBlockSize);
3421 __ subptr(len_reg, AESBlockSize);
3422 __ jcc(Assembler::notEqual, L_loopTop_128);
3423
3424 __ BIND(L_exit);
3425 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3426
3427 #ifdef _WIN64
3428 __ movl(rax, len_mem);
3429 #else
3430 __ pop(rax); // return length
3431 #endif
3432 __ leave(); // required for proper stackwalking of RuntimeStub frame
3433 __ ret(0);
3434
3435 __ BIND(L_key_192_256);
3436 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3437 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3438 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3439 __ cmpl(rax, 52);
3440 __ jcc(Assembler::notEqual, L_key_256);
3441
3442 // 192-bit code follows here (could be changed to use more xmm registers)
3443 __ movptr(pos, 0);
3444 __ align(OptoLoopAlignment);
3445
3446 __ BIND(L_loopTop_192);
3447 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3448 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3449 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3450 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3451 __ aesenc(xmm_result, as_XMMRegister(rnum));
3452 }
3453 __ aesenclast(xmm_result, xmm_key12);
3454 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3455 // no need to store r to memory until we exit
3456 __ addptr(pos, AESBlockSize);
3457 __ subptr(len_reg, AESBlockSize);
3458 __ jcc(Assembler::notEqual, L_loopTop_192);
3459 __ jmp(L_exit);
3460
3461 __ BIND(L_key_256);
3462 // 256-bit code follows here (could be changed to use more xmm registers)
3463 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3464 __ movptr(pos, 0);
3465 __ align(OptoLoopAlignment);
3466
3467 __ BIND(L_loopTop_256);
3468 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3469 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3470 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3471 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3472 __ aesenc(xmm_result, as_XMMRegister(rnum));
3473 }
3474 load_key(xmm_temp, key, 0xe0);
3475 __ aesenclast(xmm_result, xmm_temp);
3476 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3477 // no need to store r to memory until we exit
3478 __ addptr(pos, AESBlockSize);
3479 __ subptr(len_reg, AESBlockSize);
3480 __ jcc(Assembler::notEqual, L_loopTop_256);
3481 __ jmp(L_exit);
3482
3483 return start;
3484 }
3485
3486 // Safefetch stubs.
3487 void generate_safefetch(const char* name, int size, address* entry,
3488 address* fault_pc, address* continuation_pc) {
3489 // safefetch signatures:
3490 // int SafeFetch32(int* adr, int errValue);
3491 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3492 //
3493 // arguments:
3494 // c_rarg0 = adr
3495 // c_rarg1 = errValue
3496 //
3497 // result:
3498 // PPC_RET = *adr or errValue
3499
3500 StubCodeMark mark(this, "StubRoutines", name);
3501
3502 // Entry point, pc or function descriptor.
3503 *entry = __ pc();
3504
3505 // Load *adr into c_rarg1, may fault.
3506 *fault_pc = __ pc();
3507 switch (size) {
3508 case 4:
3509 // int32_t
3510 __ movl(c_rarg1, Address(c_rarg0, 0));
3511 break;
3512 case 8:
3513 // int64_t
3514 __ movq(c_rarg1, Address(c_rarg0, 0));
3515 break;
3516 default:
3517 ShouldNotReachHere();
3518 }
3519
3520 // return errValue or *adr
3521 *continuation_pc = __ pc();
3522 __ movq(rax, c_rarg1);
3523 __ ret(0);
3524 }
3525
3526 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3527 // to hide instruction latency
3528 //
3529 // Arguments:
3530 //
3531 // Inputs:
3532 // c_rarg0 - source byte array address
3533 // c_rarg1 - destination byte array address
3534 // c_rarg2 - K (key) in little endian int array
3535 // c_rarg3 - r vector byte array address
3536 // c_rarg4 - input length
3537 //
3538 // Output:
3539 // rax - input length
3540 //
3541 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3542 assert(UseAES, "need AES instructions and misaligned SSE support");
3543 __ align(CodeEntryAlignment);
3544 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3545 address start = __ pc();
3546
3547 const Register from = c_rarg0; // source array address
3548 const Register to = c_rarg1; // destination array address
3549 const Register key = c_rarg2; // key array address
3550 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3551 // and left with the results of the last encryption block
3552 #ifndef _WIN64
3553 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3554 #else
3555 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3556 const Register len_reg = r11; // pick the volatile windows register
3557 #endif
3558 const Register pos = rax;
3559
3560 const int PARALLEL_FACTOR = 4;
3561 const int ROUNDS[3] = { 10, 12, 14 }; // aes rounds for key128, key192, key256
3562
3563 Label L_exit;
3564 Label L_singleBlock_loopTopHead[3]; // 128, 192, 256
3565 Label L_singleBlock_loopTopHead2[3]; // 128, 192, 256
3566 Label L_singleBlock_loopTop[3]; // 128, 192, 256
3567 Label L_multiBlock_loopTopHead[3]; // 128, 192, 256
3568 Label L_multiBlock_loopTop[3]; // 128, 192, 256
3569
3570 // keys 0-10 preloaded into xmm5-xmm15
3571 const int XMM_REG_NUM_KEY_FIRST = 5;
3572 const int XMM_REG_NUM_KEY_LAST = 15;
3573 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3574 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3575
3576 __ enter(); // required for proper stackwalking of RuntimeStub frame
3577
3578 #ifdef _WIN64
3579 // on win64, fill len_reg from stack position
3580 __ movl(len_reg, len_mem);
3581 #else
3582 __ push(len_reg); // Save
3583 #endif
3584 __ push(rbx);
3585 // the java expanded key ordering is rotated one position from what we want
3586 // so we start from 0x10 here and hit 0x00 last
3587 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3588 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3589 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3590 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3591 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3592 offset += 0x10;
3593 }
3594 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3595
3596 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3597
3598 // registers holding the four results in the parallelized loop
3599 const XMMRegister xmm_result0 = xmm0;
3600 const XMMRegister xmm_result1 = xmm2;
3601 const XMMRegister xmm_result2 = xmm3;
3602 const XMMRegister xmm_result3 = xmm4;
3603
3604 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3605
3606 __ xorptr(pos, pos);
3607
3608 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3609 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3610 __ cmpl(rbx, 52);
3611 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[1]);
3612 __ cmpl(rbx, 60);
3613 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[2]);
3614
3615 #define DoFour(opc, src_reg) \
3616 __ opc(xmm_result0, src_reg); \
3617 __ opc(xmm_result1, src_reg); \
3618 __ opc(xmm_result2, src_reg); \
3619 __ opc(xmm_result3, src_reg); \
3620
3621 for (int k = 0; k < 3; ++k) {
3622 __ BIND(L_multiBlock_loopTopHead[k]);
3623 if (k != 0) {
3624 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3625 __ jcc(Assembler::less, L_singleBlock_loopTopHead2[k]);
3626 }
3627 if (k == 1) {
3628 __ subptr(rsp, 6 * wordSize);
3629 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3630 load_key(xmm15, key, 0xb0); // 0xb0; 192-bit key goes up to 0xc0
3631 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3632 load_key(xmm1, key, 0xc0); // 0xc0;
3633 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3634 } else if (k == 2) {
3635 __ subptr(rsp, 10 * wordSize);
3636 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3637 load_key(xmm15, key, 0xd0); // 0xd0; 256-bit key goes upto 0xe0
3638 __ movdqu(Address(rsp, 6 * wordSize), xmm15);
3639 load_key(xmm1, key, 0xe0); // 0xe0;
3640 __ movdqu(Address(rsp, 8 * wordSize), xmm1);
3641 load_key(xmm15, key, 0xb0); // 0xb0;
3642 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3643 load_key(xmm1, key, 0xc0); // 0xc0;
3644 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3645 }
3646 __ align(OptoLoopAlignment);
3647 __ BIND(L_multiBlock_loopTop[k]);
3648 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3649 __ jcc(Assembler::less, L_singleBlock_loopTopHead[k]);
3650
3651 if (k != 0) {
3652 __ movdqu(xmm15, Address(rsp, 2 * wordSize));
3653 __ movdqu(xmm1, Address(rsp, 4 * wordSize));
3654 }
3655
3656 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
3657 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3658 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3659 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3660
3661 DoFour(pxor, xmm_key_first);
3662 if (k == 0) {
3663 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
3664 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3665 }
3666 DoFour(aesdeclast, xmm_key_last);
3667 } else if (k == 1) {
3668 for (int rnum = 1; rnum <= ROUNDS[k]-2; rnum++) {
3669 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3670 }
3671 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3672 DoFour(aesdec, xmm1); // key : 0xc0
3673 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3674 DoFour(aesdeclast, xmm_key_last);
3675 } else if (k == 2) {
3676 for (int rnum = 1; rnum <= ROUNDS[k] - 4; rnum++) {
3677 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3678 }
3679 DoFour(aesdec, xmm1); // key : 0xc0
3680 __ movdqu(xmm15, Address(rsp, 6 * wordSize));
3681 __ movdqu(xmm1, Address(rsp, 8 * wordSize));
3682 DoFour(aesdec, xmm15); // key : 0xd0
3683 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3684 DoFour(aesdec, xmm1); // key : 0xe0
3685 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3686 DoFour(aesdeclast, xmm_key_last);
3687 }
3688
3689 // for each result, xor with the r vector of previous cipher block
3690 __ pxor(xmm_result0, xmm_prev_block_cipher);
3691 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3692 __ pxor(xmm_result1, xmm_prev_block_cipher);
3693 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3694 __ pxor(xmm_result2, xmm_prev_block_cipher);
3695 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3696 __ pxor(xmm_result3, xmm_prev_block_cipher);
3697 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
3698 if (k != 0) {
3699 __ movdqu(Address(rvec, 0x00), xmm_prev_block_cipher);
3700 }
3701
3702 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3703 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3704 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3705 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3706
3707 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize);
3708 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize);
3709 __ jmp(L_multiBlock_loopTop[k]);
3710
3711 // registers used in the non-parallelized loops
3712 // xmm register assignments for the loops below
3713 const XMMRegister xmm_result = xmm0;
3714 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3715 const XMMRegister xmm_key11 = xmm3;
3716 const XMMRegister xmm_key12 = xmm4;
3717 const XMMRegister key_tmp = xmm4;
3718
3719 __ BIND(L_singleBlock_loopTopHead[k]);
3720 if (k == 1) {
3721 __ addptr(rsp, 6 * wordSize);
3722 } else if (k == 2) {
3723 __ addptr(rsp, 10 * wordSize);
3724 }
3725 __ cmpptr(len_reg, 0); // any blocks left??
3726 __ jcc(Assembler::equal, L_exit);
3727 __ BIND(L_singleBlock_loopTopHead2[k]);
3728 if (k == 1) {
3729 load_key(xmm_key11, key, 0xb0); // 0xb0; 192-bit key goes upto 0xc0
3730 load_key(xmm_key12, key, 0xc0); // 0xc0; 192-bit key goes upto 0xc0
3731 }
3732 if (k == 2) {
3733 load_key(xmm_key11, key, 0xb0); // 0xb0; 256-bit key goes upto 0xe0
3734 }
3735 __ align(OptoLoopAlignment);
3736 __ BIND(L_singleBlock_loopTop[k]);
3737 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3738 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3739 __ pxor(xmm_result, xmm_key_first); // do the aes dec rounds
3740 for (int rnum = 1; rnum <= 9 ; rnum++) {
3741 __ aesdec(xmm_result, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3742 }
3743 if (k == 1) {
3744 __ aesdec(xmm_result, xmm_key11);
3745 __ aesdec(xmm_result, xmm_key12);
3746 }
3747 if (k == 2) {
3748 __ aesdec(xmm_result, xmm_key11);
3749 load_key(key_tmp, key, 0xc0);
3750 __ aesdec(xmm_result, key_tmp);
3751 load_key(key_tmp, key, 0xd0);
3752 __ aesdec(xmm_result, key_tmp);
3753 load_key(key_tmp, key, 0xe0);
3754 __ aesdec(xmm_result, key_tmp);
3755 }
3756
3757 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3758 __ pxor(xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3759 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3760 // no need to store r to memory until we exit
3761 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3762 __ addptr(pos, AESBlockSize);
3763 __ subptr(len_reg, AESBlockSize);
3764 __ jcc(Assembler::notEqual, L_singleBlock_loopTop[k]);
3765 if (k != 2) {
3766 __ jmp(L_exit);
3767 }
3768 } //for 128/192/256
3769
3770 __ BIND(L_exit);
3771 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3772 __ pop(rbx);
3773 #ifdef _WIN64
3774 __ movl(rax, len_mem);
3775 #else
3776 __ pop(rax); // return length
3777 #endif
3778 __ leave(); // required for proper stackwalking of RuntimeStub frame
3779 __ ret(0);
3780 return start;
3781 }
3782
3783 address generate_electronicCodeBook_encryptAESCrypt() {
3784 __ align(CodeEntryAlignment);
3785 StubCodeMark mark(this, "StubRoutines", "electronicCodeBook_encryptAESCrypt");
3786 address start = __ pc();
3787 const Register from = c_rarg0; // source array address
3788 const Register to = c_rarg1; // destination array address
3789 const Register key = c_rarg2; // key array address
3790 const Register len = c_rarg3; // src len (must be multiple of blocksize 16)
3791 __ enter(); // required for proper stackwalking of RuntimeStub frame
3792 __ aesecb_encrypt(from, to, key, len);
3793 __ leave(); // required for proper stackwalking of RuntimeStub frame
3794 __ ret(0);
3795 return start;
3796 }
3797
3798 address generate_electronicCodeBook_decryptAESCrypt() {
3799 __ align(CodeEntryAlignment);
3800 StubCodeMark mark(this, "StubRoutines", "electronicCodeBook_decryptAESCrypt");
3801 address start = __ pc();
3802 const Register from = c_rarg0; // source array address
3803 const Register to = c_rarg1; // destination array address
3804 const Register key = c_rarg2; // key array address
3805 const Register len = c_rarg3; // src len (must be multiple of blocksize 16)
3806 __ enter(); // required for proper stackwalking of RuntimeStub frame
3807 __ aesecb_decrypt(from, to, key, len);
3808 __ leave(); // required for proper stackwalking of RuntimeStub frame
3809 __ ret(0);
3810 return start;
3811 }
3812
3813 address generate_upper_word_mask() {
3814 __ align(64);
3815 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3816 address start = __ pc();
3817 __ emit_data64(0x0000000000000000, relocInfo::none);
3818 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
3819 return start;
3820 }
3821
3822 address generate_shuffle_byte_flip_mask() {
3823 __ align(64);
3824 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3825 address start = __ pc();
3826 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3827 __ emit_data64(0x0001020304050607, relocInfo::none);
3828 return start;
3829 }
3830
3831 // ofs and limit are use for multi-block byte array.
3832 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3833 address generate_sha1_implCompress(bool multi_block, const char *name) {
3834 __ align(CodeEntryAlignment);
3835 StubCodeMark mark(this, "StubRoutines", name);
3836 address start = __ pc();
3837
3838 Register buf = c_rarg0;
3839 Register state = c_rarg1;
3840 Register ofs = c_rarg2;
3841 Register limit = c_rarg3;
3842
3843 const XMMRegister abcd = xmm0;
3844 const XMMRegister e0 = xmm1;
3845 const XMMRegister e1 = xmm2;
3846 const XMMRegister msg0 = xmm3;
3847
3848 const XMMRegister msg1 = xmm4;
3849 const XMMRegister msg2 = xmm5;
3850 const XMMRegister msg3 = xmm6;
3851 const XMMRegister shuf_mask = xmm7;
3852
3853 __ enter();
3854
3855 __ subptr(rsp, 4 * wordSize);
3856
3857 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3858 buf, state, ofs, limit, rsp, multi_block);
3859
3860 __ addptr(rsp, 4 * wordSize);
3861
3862 __ leave();
3863 __ ret(0);
3864 return start;
3865 }
3866
3867 address generate_pshuffle_byte_flip_mask() {
3868 __ align(64);
3869 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3870 address start = __ pc();
3871 __ emit_data64(0x0405060700010203, relocInfo::none);
3872 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3873
3874 if (VM_Version::supports_avx2()) {
3875 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
3876 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3877 // _SHUF_00BA
3878 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3879 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3880 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3881 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3882 // _SHUF_DC00
3883 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3884 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3885 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3886 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3887 }
3888
3889 return start;
3890 }
3891
3892 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
3893 address generate_pshuffle_byte_flip_mask_sha512() {
3894 __ align(32);
3895 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
3896 address start = __ pc();
3897 if (VM_Version::supports_avx2()) {
3898 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
3899 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3900 __ emit_data64(0x1011121314151617, relocInfo::none);
3901 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
3902 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
3903 __ emit_data64(0x0000000000000000, relocInfo::none);
3904 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3905 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3906 }
3907
3908 return start;
3909 }
3910
3911 // ofs and limit are use for multi-block byte array.
3912 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3913 address generate_sha256_implCompress(bool multi_block, const char *name) {
3914 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
3915 __ align(CodeEntryAlignment);
3916 StubCodeMark mark(this, "StubRoutines", name);
3917 address start = __ pc();
3918
3919 Register buf = c_rarg0;
3920 Register state = c_rarg1;
3921 Register ofs = c_rarg2;
3922 Register limit = c_rarg3;
3923
3924 const XMMRegister msg = xmm0;
3925 const XMMRegister state0 = xmm1;
3926 const XMMRegister state1 = xmm2;
3927 const XMMRegister msgtmp0 = xmm3;
3928
3929 const XMMRegister msgtmp1 = xmm4;
3930 const XMMRegister msgtmp2 = xmm5;
3931 const XMMRegister msgtmp3 = xmm6;
3932 const XMMRegister msgtmp4 = xmm7;
3933
3934 const XMMRegister shuf_mask = xmm8;
3935
3936 __ enter();
3937
3938 __ subptr(rsp, 4 * wordSize);
3939
3940 if (VM_Version::supports_sha()) {
3941 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3942 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3943 } else if (VM_Version::supports_avx2()) {
3944 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3945 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3946 }
3947 __ addptr(rsp, 4 * wordSize);
3948 __ vzeroupper();
3949 __ leave();
3950 __ ret(0);
3951 return start;
3952 }
3953
3954 address generate_sha512_implCompress(bool multi_block, const char *name) {
3955 assert(VM_Version::supports_avx2(), "");
3956 assert(VM_Version::supports_bmi2(), "");
3957 __ align(CodeEntryAlignment);
3958 StubCodeMark mark(this, "StubRoutines", name);
3959 address start = __ pc();
3960
3961 Register buf = c_rarg0;
3962 Register state = c_rarg1;
3963 Register ofs = c_rarg2;
3964 Register limit = c_rarg3;
3965
3966 const XMMRegister msg = xmm0;
3967 const XMMRegister state0 = xmm1;
3968 const XMMRegister state1 = xmm2;
3969 const XMMRegister msgtmp0 = xmm3;
3970 const XMMRegister msgtmp1 = xmm4;
3971 const XMMRegister msgtmp2 = xmm5;
3972 const XMMRegister msgtmp3 = xmm6;
3973 const XMMRegister msgtmp4 = xmm7;
3974
3975 const XMMRegister shuf_mask = xmm8;
3976
3977 __ enter();
3978
3979 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3980 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3981
3982 __ vzeroupper();
3983 __ leave();
3984 __ ret(0);
3985 return start;
3986 }
3987
3988 // This mask is used for incrementing counter value(linc0, linc4, etc.)
3989 address counter_mask_addr() {
3990 __ align(64);
3991 StubCodeMark mark(this, "StubRoutines", "counter_mask_addr");
3992 address start = __ pc();
3993 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);//lbswapmask
3994 __ emit_data64(0x0001020304050607, relocInfo::none);
3995 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3996 __ emit_data64(0x0001020304050607, relocInfo::none);
3997 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3998 __ emit_data64(0x0001020304050607, relocInfo::none);
3999 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
4000 __ emit_data64(0x0001020304050607, relocInfo::none);
4001 __ emit_data64(0x0000000000000000, relocInfo::none);//linc0 = counter_mask_addr+64
4002 __ emit_data64(0x0000000000000000, relocInfo::none);
4003 __ emit_data64(0x0000000000000001, relocInfo::none);//counter_mask_addr() + 80
4004 __ emit_data64(0x0000000000000000, relocInfo::none);
4005 __ emit_data64(0x0000000000000002, relocInfo::none);
4006 __ emit_data64(0x0000000000000000, relocInfo::none);
4007 __ emit_data64(0x0000000000000003, relocInfo::none);
4008 __ emit_data64(0x0000000000000000, relocInfo::none);
4009 __ emit_data64(0x0000000000000004, relocInfo::none);//linc4 = counter_mask_addr() + 128
4010 __ emit_data64(0x0000000000000000, relocInfo::none);
4011 __ emit_data64(0x0000000000000004, relocInfo::none);
4012 __ emit_data64(0x0000000000000000, relocInfo::none);
4013 __ emit_data64(0x0000000000000004, relocInfo::none);
4014 __ emit_data64(0x0000000000000000, relocInfo::none);
4015 __ emit_data64(0x0000000000000004, relocInfo::none);
4016 __ emit_data64(0x0000000000000000, relocInfo::none);
4017 __ emit_data64(0x0000000000000008, relocInfo::none);//linc8 = counter_mask_addr() + 192
4018 __ emit_data64(0x0000000000000000, relocInfo::none);
4019 __ emit_data64(0x0000000000000008, relocInfo::none);
4020 __ emit_data64(0x0000000000000000, relocInfo::none);
4021 __ emit_data64(0x0000000000000008, relocInfo::none);
4022 __ emit_data64(0x0000000000000000, relocInfo::none);
4023 __ emit_data64(0x0000000000000008, relocInfo::none);
4024 __ emit_data64(0x0000000000000000, relocInfo::none);
4025 __ emit_data64(0x0000000000000020, relocInfo::none);//linc32 = counter_mask_addr() + 256
4026 __ emit_data64(0x0000000000000000, relocInfo::none);
4027 __ emit_data64(0x0000000000000020, relocInfo::none);
4028 __ emit_data64(0x0000000000000000, relocInfo::none);
4029 __ emit_data64(0x0000000000000020, relocInfo::none);
4030 __ emit_data64(0x0000000000000000, relocInfo::none);
4031 __ emit_data64(0x0000000000000020, relocInfo::none);
4032 __ emit_data64(0x0000000000000000, relocInfo::none);
4033 __ emit_data64(0x0000000000000010, relocInfo::none);//linc16 = counter_mask_addr() + 320
4034 __ emit_data64(0x0000000000000000, relocInfo::none);
4035 __ emit_data64(0x0000000000000010, relocInfo::none);
4036 __ emit_data64(0x0000000000000000, relocInfo::none);
4037 __ emit_data64(0x0000000000000010, relocInfo::none);
4038 __ emit_data64(0x0000000000000000, relocInfo::none);
4039 __ emit_data64(0x0000000000000010, relocInfo::none);
4040 __ emit_data64(0x0000000000000000, relocInfo::none);
4041 return start;
4042 }
4043
4044 // Vector AES Counter implementation
4045 address generate_counterMode_VectorAESCrypt() {
4046 __ align(CodeEntryAlignment);
4047 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
4048 address start = __ pc();
4049 const Register from = c_rarg0; // source array address
4050 const Register to = c_rarg1; // destination array address
4051 const Register key = c_rarg2; // key array address r8
4052 const Register counter = c_rarg3; // counter byte array initialized from counter array address
4053 // and updated with the incremented counter in the end
4054 #ifndef _WIN64
4055 const Register len_reg = c_rarg4;
4056 const Register saved_encCounter_start = c_rarg5;
4057 const Register used_addr = r10;
4058 const Address used_mem(rbp, 2 * wordSize);
4059 const Register used = r11;
4060 #else
4061 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4062 const Address saved_encCounter_mem(rbp, 7 * wordSize); // saved encrypted counter is on stack on Win64
4063 const Address used_mem(rbp, 8 * wordSize); // used length is on stack on Win64
4064 const Register len_reg = r10; // pick the first volatile windows register
4065 const Register saved_encCounter_start = r11;
4066 const Register used_addr = r13;
4067 const Register used = r14;
4068 #endif
4069 __ enter();
4070 // Save state before entering routine
4071 __ push(r12);
4072 __ push(r13);
4073 __ push(r14);
4074 __ push(r15);
4075 #ifdef _WIN64
4076 // on win64, fill len_reg from stack position
4077 __ movl(len_reg, len_mem);
4078 __ movptr(saved_encCounter_start, saved_encCounter_mem);
4079 __ movptr(used_addr, used_mem);
4080 __ movl(used, Address(used_addr, 0));
4081 #else
4082 __ push(len_reg); // Save
4083 __ movptr(used_addr, used_mem);
4084 __ movl(used, Address(used_addr, 0));
4085 #endif
4086 __ push(rbx);
4087 __ aesctr_encrypt(from, to, key, counter, len_reg, used, used_addr, saved_encCounter_start);
4088 // Restore state before leaving routine
4089 __ pop(rbx);
4090 #ifdef _WIN64
4091 __ movl(rax, len_mem); // return length
4092 #else
4093 __ pop(rax); // return length
4094 #endif
4095 __ pop(r15);
4096 __ pop(r14);
4097 __ pop(r13);
4098 __ pop(r12);
4099
4100 __ leave(); // required for proper stackwalking of RuntimeStub frame
4101 __ ret(0);
4102 return start;
4103 }
4104
4105 // This is a version of CTR/AES crypt which does 6 blocks in a loop at a time
4106 // to hide instruction latency
4107 //
4108 // Arguments:
4109 //
4110 // Inputs:
4111 // c_rarg0 - source byte array address
4112 // c_rarg1 - destination byte array address
4113 // c_rarg2 - K (key) in little endian int array
4114 // c_rarg3 - counter vector byte array address
4115 // Linux
4116 // c_rarg4 - input length
4117 // c_rarg5 - saved encryptedCounter start
4118 // rbp + 6 * wordSize - saved used length
4119 // Windows
4120 // rbp + 6 * wordSize - input length
4121 // rbp + 7 * wordSize - saved encryptedCounter start
4122 // rbp + 8 * wordSize - saved used length
4123 //
4124 // Output:
4125 // rax - input length
4126 //
4127 address generate_counterMode_AESCrypt_Parallel() {
4128 assert(UseAES, "need AES instructions and misaligned SSE support");
4129 __ align(CodeEntryAlignment);
4130 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
4131 address start = __ pc();
4132 const Register from = c_rarg0; // source array address
4133 const Register to = c_rarg1; // destination array address
4134 const Register key = c_rarg2; // key array address
4135 const Register counter = c_rarg3; // counter byte array initialized from counter array address
4136 // and updated with the incremented counter in the end
4137 #ifndef _WIN64
4138 const Register len_reg = c_rarg4;
4139 const Register saved_encCounter_start = c_rarg5;
4140 const Register used_addr = r10;
4141 const Address used_mem(rbp, 2 * wordSize);
4142 const Register used = r11;
4143 #else
4144 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4145 const Address saved_encCounter_mem(rbp, 7 * wordSize); // length is on stack on Win64
4146 const Address used_mem(rbp, 8 * wordSize); // length is on stack on Win64
4147 const Register len_reg = r10; // pick the first volatile windows register
4148 const Register saved_encCounter_start = r11;
4149 const Register used_addr = r13;
4150 const Register used = r14;
4151 #endif
4152 const Register pos = rax;
4153
4154 const int PARALLEL_FACTOR = 6;
4155 const XMMRegister xmm_counter_shuf_mask = xmm0;
4156 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
4157 const XMMRegister xmm_curr_counter = xmm2;
4158
4159 const XMMRegister xmm_key_tmp0 = xmm3;
4160 const XMMRegister xmm_key_tmp1 = xmm4;
4161
4162 // registers holding the four results in the parallelized loop
4163 const XMMRegister xmm_result0 = xmm5;
4164 const XMMRegister xmm_result1 = xmm6;
4165 const XMMRegister xmm_result2 = xmm7;
4166 const XMMRegister xmm_result3 = xmm8;
4167 const XMMRegister xmm_result4 = xmm9;
4168 const XMMRegister xmm_result5 = xmm10;
4169
4170 const XMMRegister xmm_from0 = xmm11;
4171 const XMMRegister xmm_from1 = xmm12;
4172 const XMMRegister xmm_from2 = xmm13;
4173 const XMMRegister xmm_from3 = xmm14; //the last one is xmm14. we have to preserve it on WIN64.
4174 const XMMRegister xmm_from4 = xmm3; //reuse xmm3~4. Because xmm_key_tmp0~1 are useless when loading input text
4175 const XMMRegister xmm_from5 = xmm4;
4176
4177 //for key_128, key_192, key_256
4178 const int rounds[3] = {10, 12, 14};
4179 Label L_exit_preLoop, L_preLoop_start;
4180 Label L_multiBlock_loopTop[3];
4181 Label L_singleBlockLoopTop[3];
4182 Label L__incCounter[3][6]; //for 6 blocks
4183 Label L__incCounter_single[3]; //for single block, key128, key192, key256
4184 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];
4185 Label L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
4186
4187 Label L_exit;
4188
4189 __ enter(); // required for proper stackwalking of RuntimeStub frame
4190
4191 #ifdef _WIN64
4192 // allocate spill slots for r13, r14
4193 enum {
4194 saved_r13_offset,
4195 saved_r14_offset
4196 };
4197 __ subptr(rsp, 2 * wordSize);
4198 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
4199 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
4200
4201 // on win64, fill len_reg from stack position
4202 __ movl(len_reg, len_mem);
4203 __ movptr(saved_encCounter_start, saved_encCounter_mem);
4204 __ movptr(used_addr, used_mem);
4205 __ movl(used, Address(used_addr, 0));
4206 #else
4207 __ push(len_reg); // Save
4208 __ movptr(used_addr, used_mem);
4209 __ movl(used, Address(used_addr, 0));
4210 #endif
4211
4212 __ push(rbx); // Save RBX
4213 __ movdqu(xmm_curr_counter, Address(counter, 0x00)); // initialize counter with initial counter
4214 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()), pos); // pos as scratch
4215 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled
4216 __ movptr(pos, 0);
4217
4218 // Use the partially used encrpyted counter from last invocation
4219 __ BIND(L_preLoop_start);
4220 __ cmpptr(used, 16);
4221 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
4222 __ cmpptr(len_reg, 0);
4223 __ jcc(Assembler::lessEqual, L_exit_preLoop);
4224 __ movb(rbx, Address(saved_encCounter_start, used));
4225 __ xorb(rbx, Address(from, pos));
4226 __ movb(Address(to, pos), rbx);
4227 __ addptr(pos, 1);
4228 __ addptr(used, 1);
4229 __ subptr(len_reg, 1);
4230
4231 __ jmp(L_preLoop_start);
4232
4233 __ BIND(L_exit_preLoop);
4234 __ movl(Address(used_addr, 0), used);
4235
4236 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
4237 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()), rbx); // rbx as scratch
4238 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4239 __ cmpl(rbx, 52);
4240 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
4241 __ cmpl(rbx, 60);
4242 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
4243
4244 #define CTR_DoSix(opc, src_reg) \
4245 __ opc(xmm_result0, src_reg); \
4246 __ opc(xmm_result1, src_reg); \
4247 __ opc(xmm_result2, src_reg); \
4248 __ opc(xmm_result3, src_reg); \
4249 __ opc(xmm_result4, src_reg); \
4250 __ opc(xmm_result5, src_reg);
4251
4252 // k == 0 : generate code for key_128
4253 // k == 1 : generate code for key_192
4254 // k == 2 : generate code for key_256
4255 for (int k = 0; k < 3; ++k) {
4256 //multi blocks starts here
4257 __ align(OptoLoopAlignment);
4258 __ BIND(L_multiBlock_loopTop[k]);
4259 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
4260 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
4261 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4262
4263 //load, then increase counters
4264 CTR_DoSix(movdqa, xmm_curr_counter);
4265 inc_counter(rbx, xmm_result1, 0x01, L__incCounter[k][0]);
4266 inc_counter(rbx, xmm_result2, 0x02, L__incCounter[k][1]);
4267 inc_counter(rbx, xmm_result3, 0x03, L__incCounter[k][2]);
4268 inc_counter(rbx, xmm_result4, 0x04, L__incCounter[k][3]);
4269 inc_counter(rbx, xmm_result5, 0x05, L__incCounter[k][4]);
4270 inc_counter(rbx, xmm_curr_counter, 0x06, L__incCounter[k][5]);
4271 CTR_DoSix(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
4272 CTR_DoSix(pxor, xmm_key_tmp0); //PXOR with Round 0 key
4273
4274 //load two ROUND_KEYs at a time
4275 for (int i = 1; i < rounds[k]; ) {
4276 load_key(xmm_key_tmp1, key, (0x10 * i), xmm_key_shuf_mask);
4277 load_key(xmm_key_tmp0, key, (0x10 * (i+1)), xmm_key_shuf_mask);
4278 CTR_DoSix(aesenc, xmm_key_tmp1);
4279 i++;
4280 if (i != rounds[k]) {
4281 CTR_DoSix(aesenc, xmm_key_tmp0);
4282 } else {
4283 CTR_DoSix(aesenclast, xmm_key_tmp0);
4284 }
4285 i++;
4286 }
4287
4288 // get next PARALLEL_FACTOR blocks into xmm_result registers
4289 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4290 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
4291 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
4292 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
4293 __ movdqu(xmm_from4, Address(from, pos, Address::times_1, 4 * AESBlockSize));
4294 __ movdqu(xmm_from5, Address(from, pos, Address::times_1, 5 * AESBlockSize));
4295
4296 __ pxor(xmm_result0, xmm_from0);
4297 __ pxor(xmm_result1, xmm_from1);
4298 __ pxor(xmm_result2, xmm_from2);
4299 __ pxor(xmm_result3, xmm_from3);
4300 __ pxor(xmm_result4, xmm_from4);
4301 __ pxor(xmm_result5, xmm_from5);
4302
4303 // store 6 results into the next 64 bytes of output
4304 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4305 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
4306 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
4307 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
4308 __ movdqu(Address(to, pos, Address::times_1, 4 * AESBlockSize), xmm_result4);
4309 __ movdqu(Address(to, pos, Address::times_1, 5 * AESBlockSize), xmm_result5);
4310
4311 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
4312 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
4313 __ jmp(L_multiBlock_loopTop[k]);
4314
4315 // singleBlock starts here
4316 __ align(OptoLoopAlignment);
4317 __ BIND(L_singleBlockLoopTop[k]);
4318 __ cmpptr(len_reg, 0);
4319 __ jcc(Assembler::lessEqual, L_exit);
4320 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4321 __ movdqa(xmm_result0, xmm_curr_counter);
4322 inc_counter(rbx, xmm_curr_counter, 0x01, L__incCounter_single[k]);
4323 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
4324 __ pxor(xmm_result0, xmm_key_tmp0);
4325 for (int i = 1; i < rounds[k]; i++) {
4326 load_key(xmm_key_tmp0, key, (0x10 * i), xmm_key_shuf_mask);
4327 __ aesenc(xmm_result0, xmm_key_tmp0);
4328 }
4329 load_key(xmm_key_tmp0, key, (rounds[k] * 0x10), xmm_key_shuf_mask);
4330 __ aesenclast(xmm_result0, xmm_key_tmp0);
4331 __ cmpptr(len_reg, AESBlockSize);
4332 __ jcc(Assembler::less, L_processTail_insr[k]);
4333 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4334 __ pxor(xmm_result0, xmm_from0);
4335 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4336 __ addptr(pos, AESBlockSize);
4337 __ subptr(len_reg, AESBlockSize);
4338 __ jmp(L_singleBlockLoopTop[k]);
4339 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
4340 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
4341 __ testptr(len_reg, 8);
4342 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
4343 __ subptr(pos,8);
4344 __ pinsrq(xmm_from0, Address(from, pos), 0);
4345 __ BIND(L_processTail_4_insr[k]);
4346 __ testptr(len_reg, 4);
4347 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
4348 __ subptr(pos,4);
4349 __ pslldq(xmm_from0, 4);
4350 __ pinsrd(xmm_from0, Address(from, pos), 0);
4351 __ BIND(L_processTail_2_insr[k]);
4352 __ testptr(len_reg, 2);
4353 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
4354 __ subptr(pos, 2);
4355 __ pslldq(xmm_from0, 2);
4356 __ pinsrw(xmm_from0, Address(from, pos), 0);
4357 __ BIND(L_processTail_1_insr[k]);
4358 __ testptr(len_reg, 1);
4359 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
4360 __ subptr(pos, 1);
4361 __ pslldq(xmm_from0, 1);
4362 __ pinsrb(xmm_from0, Address(from, pos), 0);
4363 __ BIND(L_processTail_exit_insr[k]);
4364
4365 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
4366 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
4367
4368 __ testptr(len_reg, 8);
4369 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
4370 __ pextrq(Address(to, pos), xmm_result0, 0);
4371 __ psrldq(xmm_result0, 8);
4372 __ addptr(pos, 8);
4373 __ BIND(L_processTail_4_extr[k]);
4374 __ testptr(len_reg, 4);
4375 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
4376 __ pextrd(Address(to, pos), xmm_result0, 0);
4377 __ psrldq(xmm_result0, 4);
4378 __ addptr(pos, 4);
4379 __ BIND(L_processTail_2_extr[k]);
4380 __ testptr(len_reg, 2);
4381 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
4382 __ pextrw(Address(to, pos), xmm_result0, 0);
4383 __ psrldq(xmm_result0, 2);
4384 __ addptr(pos, 2);
4385 __ BIND(L_processTail_1_extr[k]);
4386 __ testptr(len_reg, 1);
4387 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
4388 __ pextrb(Address(to, pos), xmm_result0, 0);
4389
4390 __ BIND(L_processTail_exit_extr[k]);
4391 __ movl(Address(used_addr, 0), len_reg);
4392 __ jmp(L_exit);
4393
4394 }
4395
4396 __ BIND(L_exit);
4397 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
4398 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
4399 __ pop(rbx); // pop the saved RBX.
4400 #ifdef _WIN64
4401 __ movl(rax, len_mem);
4402 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
4403 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
4404 __ addptr(rsp, 2 * wordSize);
4405 #else
4406 __ pop(rax); // return 'len'
4407 #endif
4408 __ leave(); // required for proper stackwalking of RuntimeStub frame
4409 __ ret(0);
4410 return start;
4411 }
4412
4413 void roundDec(XMMRegister xmm_reg) {
4414 __ vaesdec(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4415 __ vaesdec(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4416 __ vaesdec(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4417 __ vaesdec(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4418 __ vaesdec(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4419 __ vaesdec(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4420 __ vaesdec(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4421 __ vaesdec(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4422 }
4423
4424 void roundDeclast(XMMRegister xmm_reg) {
4425 __ vaesdeclast(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4426 __ vaesdeclast(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4427 __ vaesdeclast(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4428 __ vaesdeclast(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4429 __ vaesdeclast(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4430 __ vaesdeclast(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4431 __ vaesdeclast(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4432 __ vaesdeclast(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4433 }
4434
4435 void ev_load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask = NULL) {
4436 __ movdqu(xmmdst, Address(key, offset));
4437 if (xmm_shuf_mask != NULL) {
4438 __ pshufb(xmmdst, xmm_shuf_mask);
4439 } else {
4440 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4441 }
4442 __ evshufi64x2(xmmdst, xmmdst, xmmdst, 0x0, Assembler::AVX_512bit);
4443
4444 }
4445
4446 address generate_cipherBlockChaining_decryptVectorAESCrypt() {
4447 assert(VM_Version::supports_avx512_vaes(), "need AES instructions and misaligned SSE support");
4448 __ align(CodeEntryAlignment);
4449 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
4450 address start = __ pc();
4451
4452 const Register from = c_rarg0; // source array address
4453 const Register to = c_rarg1; // destination array address
4454 const Register key = c_rarg2; // key array address
4455 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
4456 // and left with the results of the last encryption block
4457 #ifndef _WIN64
4458 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
4459 #else
4460 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4461 const Register len_reg = r11; // pick the volatile windows register
4462 #endif
4463
4464 Label Loop, Loop1, L_128, L_256, L_192, KEY_192, KEY_256, Loop2, Lcbc_dec_rem_loop,
4465 Lcbc_dec_rem_last, Lcbc_dec_ret, Lcbc_dec_rem, Lcbc_exit;
4466
4467 __ enter();
4468
4469 #ifdef _WIN64
4470 // on win64, fill len_reg from stack position
4471 __ movl(len_reg, len_mem);
4472 #else
4473 __ push(len_reg); // Save
4474 #endif
4475 __ push(rbx);
4476 __ vzeroupper();
4477
4478 // Temporary variable declaration for swapping key bytes
4479 const XMMRegister xmm_key_shuf_mask = xmm1;
4480 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4481
4482 // Calculate number of rounds from key size: 44 for 10-rounds, 52 for 12-rounds, 60 for 14-rounds
4483 const Register rounds = rbx;
4484 __ movl(rounds, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4485
4486 const XMMRegister IV = xmm0;
4487 // Load IV and broadcast value to 512-bits
4488 __ evbroadcasti64x2(IV, Address(rvec, 0), Assembler::AVX_512bit);
4489
4490 // Temporary variables for storing round keys
4491 const XMMRegister RK0 = xmm30;
4492 const XMMRegister RK1 = xmm9;
4493 const XMMRegister RK2 = xmm18;
4494 const XMMRegister RK3 = xmm19;
4495 const XMMRegister RK4 = xmm20;
4496 const XMMRegister RK5 = xmm21;
4497 const XMMRegister RK6 = xmm22;
4498 const XMMRegister RK7 = xmm23;
4499 const XMMRegister RK8 = xmm24;
4500 const XMMRegister RK9 = xmm25;
4501 const XMMRegister RK10 = xmm26;
4502
4503 // Load and shuffle key
4504 // the java expanded key ordering is rotated one position from what we want
4505 // so we start from 1*16 here and hit 0*16 last
4506 ev_load_key(RK1, key, 1 * 16, xmm_key_shuf_mask);
4507 ev_load_key(RK2, key, 2 * 16, xmm_key_shuf_mask);
4508 ev_load_key(RK3, key, 3 * 16, xmm_key_shuf_mask);
4509 ev_load_key(RK4, key, 4 * 16, xmm_key_shuf_mask);
4510 ev_load_key(RK5, key, 5 * 16, xmm_key_shuf_mask);
4511 ev_load_key(RK6, key, 6 * 16, xmm_key_shuf_mask);
4512 ev_load_key(RK7, key, 7 * 16, xmm_key_shuf_mask);
4513 ev_load_key(RK8, key, 8 * 16, xmm_key_shuf_mask);
4514 ev_load_key(RK9, key, 9 * 16, xmm_key_shuf_mask);
4515 ev_load_key(RK10, key, 10 * 16, xmm_key_shuf_mask);
4516 ev_load_key(RK0, key, 0*16, xmm_key_shuf_mask);
4517
4518 // Variables for storing source cipher text
4519 const XMMRegister S0 = xmm10;
4520 const XMMRegister S1 = xmm11;
4521 const XMMRegister S2 = xmm12;
4522 const XMMRegister S3 = xmm13;
4523 const XMMRegister S4 = xmm14;
4524 const XMMRegister S5 = xmm15;
4525 const XMMRegister S6 = xmm16;
4526 const XMMRegister S7 = xmm17;
4527
4528 // Variables for storing decrypted text
4529 const XMMRegister B0 = xmm1;
4530 const XMMRegister B1 = xmm2;
4531 const XMMRegister B2 = xmm3;
4532 const XMMRegister B3 = xmm4;
4533 const XMMRegister B4 = xmm5;
4534 const XMMRegister B5 = xmm6;
4535 const XMMRegister B6 = xmm7;
4536 const XMMRegister B7 = xmm8;
4537
4538 __ cmpl(rounds, 44);
4539 __ jcc(Assembler::greater, KEY_192);
4540 __ jmp(Loop);
4541
4542 __ BIND(KEY_192);
4543 const XMMRegister RK11 = xmm27;
4544 const XMMRegister RK12 = xmm28;
4545 ev_load_key(RK11, key, 11*16, xmm_key_shuf_mask);
4546 ev_load_key(RK12, key, 12*16, xmm_key_shuf_mask);
4547
4548 __ cmpl(rounds, 52);
4549 __ jcc(Assembler::greater, KEY_256);
4550 __ jmp(Loop);
4551
4552 __ BIND(KEY_256);
4553 const XMMRegister RK13 = xmm29;
4554 const XMMRegister RK14 = xmm31;
4555 ev_load_key(RK13, key, 13*16, xmm_key_shuf_mask);
4556 ev_load_key(RK14, key, 14*16, xmm_key_shuf_mask);
4557
4558 __ BIND(Loop);
4559 __ cmpl(len_reg, 512);
4560 __ jcc(Assembler::below, Lcbc_dec_rem);
4561 __ BIND(Loop1);
4562 __ subl(len_reg, 512);
4563 __ evmovdquq(S0, Address(from, 0 * 64), Assembler::AVX_512bit);
4564 __ evmovdquq(S1, Address(from, 1 * 64), Assembler::AVX_512bit);
4565 __ evmovdquq(S2, Address(from, 2 * 64), Assembler::AVX_512bit);
4566 __ evmovdquq(S3, Address(from, 3 * 64), Assembler::AVX_512bit);
4567 __ evmovdquq(S4, Address(from, 4 * 64), Assembler::AVX_512bit);
4568 __ evmovdquq(S5, Address(from, 5 * 64), Assembler::AVX_512bit);
4569 __ evmovdquq(S6, Address(from, 6 * 64), Assembler::AVX_512bit);
4570 __ evmovdquq(S7, Address(from, 7 * 64), Assembler::AVX_512bit);
4571 __ leaq(from, Address(from, 8 * 64));
4572
4573 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4574 __ evpxorq(B1, S1, RK1, Assembler::AVX_512bit);
4575 __ evpxorq(B2, S2, RK1, Assembler::AVX_512bit);
4576 __ evpxorq(B3, S3, RK1, Assembler::AVX_512bit);
4577 __ evpxorq(B4, S4, RK1, Assembler::AVX_512bit);
4578 __ evpxorq(B5, S5, RK1, Assembler::AVX_512bit);
4579 __ evpxorq(B6, S6, RK1, Assembler::AVX_512bit);
4580 __ evpxorq(B7, S7, RK1, Assembler::AVX_512bit);
4581
4582 __ evalignq(IV, S0, IV, 0x06);
4583 __ evalignq(S0, S1, S0, 0x06);
4584 __ evalignq(S1, S2, S1, 0x06);
4585 __ evalignq(S2, S3, S2, 0x06);
4586 __ evalignq(S3, S4, S3, 0x06);
4587 __ evalignq(S4, S5, S4, 0x06);
4588 __ evalignq(S5, S6, S5, 0x06);
4589 __ evalignq(S6, S7, S6, 0x06);
4590
4591 roundDec(RK2);
4592 roundDec(RK3);
4593 roundDec(RK4);
4594 roundDec(RK5);
4595 roundDec(RK6);
4596 roundDec(RK7);
4597 roundDec(RK8);
4598 roundDec(RK9);
4599 roundDec(RK10);
4600
4601 __ cmpl(rounds, 44);
4602 __ jcc(Assembler::belowEqual, L_128);
4603 roundDec(RK11);
4604 roundDec(RK12);
4605
4606 __ cmpl(rounds, 52);
4607 __ jcc(Assembler::belowEqual, L_192);
4608 roundDec(RK13);
4609 roundDec(RK14);
4610
4611 __ BIND(L_256);
4612 roundDeclast(RK0);
4613 __ jmp(Loop2);
4614
4615 __ BIND(L_128);
4616 roundDeclast(RK0);
4617 __ jmp(Loop2);
4618
4619 __ BIND(L_192);
4620 roundDeclast(RK0);
4621
4622 __ BIND(Loop2);
4623 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4624 __ evpxorq(B1, B1, S0, Assembler::AVX_512bit);
4625 __ evpxorq(B2, B2, S1, Assembler::AVX_512bit);
4626 __ evpxorq(B3, B3, S2, Assembler::AVX_512bit);
4627 __ evpxorq(B4, B4, S3, Assembler::AVX_512bit);
4628 __ evpxorq(B5, B5, S4, Assembler::AVX_512bit);
4629 __ evpxorq(B6, B6, S5, Assembler::AVX_512bit);
4630 __ evpxorq(B7, B7, S6, Assembler::AVX_512bit);
4631 __ evmovdquq(IV, S7, Assembler::AVX_512bit);
4632
4633 __ evmovdquq(Address(to, 0 * 64), B0, Assembler::AVX_512bit);
4634 __ evmovdquq(Address(to, 1 * 64), B1, Assembler::AVX_512bit);
4635 __ evmovdquq(Address(to, 2 * 64), B2, Assembler::AVX_512bit);
4636 __ evmovdquq(Address(to, 3 * 64), B3, Assembler::AVX_512bit);
4637 __ evmovdquq(Address(to, 4 * 64), B4, Assembler::AVX_512bit);
4638 __ evmovdquq(Address(to, 5 * 64), B5, Assembler::AVX_512bit);
4639 __ evmovdquq(Address(to, 6 * 64), B6, Assembler::AVX_512bit);
4640 __ evmovdquq(Address(to, 7 * 64), B7, Assembler::AVX_512bit);
4641 __ leaq(to, Address(to, 8 * 64));
4642 __ jmp(Loop);
4643
4644 __ BIND(Lcbc_dec_rem);
4645 __ evshufi64x2(IV, IV, IV, 0x03, Assembler::AVX_512bit);
4646
4647 __ BIND(Lcbc_dec_rem_loop);
4648 __ subl(len_reg, 16);
4649 __ jcc(Assembler::carrySet, Lcbc_dec_ret);
4650
4651 __ movdqu(S0, Address(from, 0));
4652 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4653 __ vaesdec(B0, B0, RK2, Assembler::AVX_512bit);
4654 __ vaesdec(B0, B0, RK3, Assembler::AVX_512bit);
4655 __ vaesdec(B0, B0, RK4, Assembler::AVX_512bit);
4656 __ vaesdec(B0, B0, RK5, Assembler::AVX_512bit);
4657 __ vaesdec(B0, B0, RK6, Assembler::AVX_512bit);
4658 __ vaesdec(B0, B0, RK7, Assembler::AVX_512bit);
4659 __ vaesdec(B0, B0, RK8, Assembler::AVX_512bit);
4660 __ vaesdec(B0, B0, RK9, Assembler::AVX_512bit);
4661 __ vaesdec(B0, B0, RK10, Assembler::AVX_512bit);
4662 __ cmpl(rounds, 44);
4663 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4664
4665 __ vaesdec(B0, B0, RK11, Assembler::AVX_512bit);
4666 __ vaesdec(B0, B0, RK12, Assembler::AVX_512bit);
4667 __ cmpl(rounds, 52);
4668 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4669
4670 __ vaesdec(B0, B0, RK13, Assembler::AVX_512bit);
4671 __ vaesdec(B0, B0, RK14, Assembler::AVX_512bit);
4672
4673 __ BIND(Lcbc_dec_rem_last);
4674 __ vaesdeclast(B0, B0, RK0, Assembler::AVX_512bit);
4675
4676 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4677 __ evmovdquq(IV, S0, Assembler::AVX_512bit);
4678 __ movdqu(Address(to, 0), B0);
4679 __ leaq(from, Address(from, 16));
4680 __ leaq(to, Address(to, 16));
4681 __ jmp(Lcbc_dec_rem_loop);
4682
4683 __ BIND(Lcbc_dec_ret);
4684 __ movdqu(Address(rvec, 0), IV);
4685
4686 // Zero out the round keys
4687 __ evpxorq(RK0, RK0, RK0, Assembler::AVX_512bit);
4688 __ evpxorq(RK1, RK1, RK1, Assembler::AVX_512bit);
4689 __ evpxorq(RK2, RK2, RK2, Assembler::AVX_512bit);
4690 __ evpxorq(RK3, RK3, RK3, Assembler::AVX_512bit);
4691 __ evpxorq(RK4, RK4, RK4, Assembler::AVX_512bit);
4692 __ evpxorq(RK5, RK5, RK5, Assembler::AVX_512bit);
4693 __ evpxorq(RK6, RK6, RK6, Assembler::AVX_512bit);
4694 __ evpxorq(RK7, RK7, RK7, Assembler::AVX_512bit);
4695 __ evpxorq(RK8, RK8, RK8, Assembler::AVX_512bit);
4696 __ evpxorq(RK9, RK9, RK9, Assembler::AVX_512bit);
4697 __ evpxorq(RK10, RK10, RK10, Assembler::AVX_512bit);
4698 __ cmpl(rounds, 44);
4699 __ jcc(Assembler::belowEqual, Lcbc_exit);
4700 __ evpxorq(RK11, RK11, RK11, Assembler::AVX_512bit);
4701 __ evpxorq(RK12, RK12, RK12, Assembler::AVX_512bit);
4702 __ cmpl(rounds, 52);
4703 __ jcc(Assembler::belowEqual, Lcbc_exit);
4704 __ evpxorq(RK13, RK13, RK13, Assembler::AVX_512bit);
4705 __ evpxorq(RK14, RK14, RK14, Assembler::AVX_512bit);
4706
4707 __ BIND(Lcbc_exit);
4708 __ pop(rbx);
4709 #ifdef _WIN64
4710 __ movl(rax, len_mem);
4711 #else
4712 __ pop(rax); // return length
4713 #endif
4714 __ leave(); // required for proper stackwalking of RuntimeStub frame
4715 __ ret(0);
4716 return start;
4717 }
4718
4719 // Polynomial x^128+x^127+x^126+x^121+1
4720 address ghash_polynomial_addr() {
4721 __ align(CodeEntryAlignment);
4722 StubCodeMark mark(this, "StubRoutines", "_ghash_poly_addr");
4723 address start = __ pc();
4724 __ emit_data64(0x0000000000000001, relocInfo::none);
4725 __ emit_data64(0xc200000000000000, relocInfo::none);
4726 return start;
4727 }
4728
4729 address ghash_shufflemask_addr() {
4730 __ align(CodeEntryAlignment);
4731 StubCodeMark mark(this, "StubRoutines", "_ghash_shuffmask_addr");
4732 address start = __ pc();
4733 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4734 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4735 return start;
4736 }
4737
4738 // Ghash single and multi block operations using AVX instructions
4739 address generate_avx_ghash_processBlocks() {
4740 __ align(CodeEntryAlignment);
4741
4742 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4743 address start = __ pc();
4744
4745 // arguments
4746 const Register state = c_rarg0;
4747 const Register htbl = c_rarg1;
4748 const Register data = c_rarg2;
4749 const Register blocks = c_rarg3;
4750 __ enter();
4751 // Save state before entering routine
4752 __ avx_ghash(state, htbl, data, blocks);
4753 __ leave(); // required for proper stackwalking of RuntimeStub frame
4754 __ ret(0);
4755 return start;
4756 }
4757
4758 // byte swap x86 long
4759 address generate_ghash_long_swap_mask() {
4760 __ align(CodeEntryAlignment);
4761 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
4762 address start = __ pc();
4763 __ emit_data64(0x0f0e0d0c0b0a0908, relocInfo::none );
4764 __ emit_data64(0x0706050403020100, relocInfo::none );
4765 return start;
4766 }
4767
4768 // byte swap x86 byte array
4769 address generate_ghash_byte_swap_mask() {
4770 __ align(CodeEntryAlignment);
4771 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
4772 address start = __ pc();
4773 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none );
4774 __ emit_data64(0x0001020304050607, relocInfo::none );
4775 return start;
4776 }
4777
4778 /* Single and multi-block ghash operations */
4779 address generate_ghash_processBlocks() {
4780 __ align(CodeEntryAlignment);
4781 Label L_ghash_loop, L_exit;
4782 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4783 address start = __ pc();
4784
4785 const Register state = c_rarg0;
4786 const Register subkeyH = c_rarg1;
4787 const Register data = c_rarg2;
4788 const Register blocks = c_rarg3;
4789
4790 const XMMRegister xmm_temp0 = xmm0;
4791 const XMMRegister xmm_temp1 = xmm1;
4792 const XMMRegister xmm_temp2 = xmm2;
4793 const XMMRegister xmm_temp3 = xmm3;
4794 const XMMRegister xmm_temp4 = xmm4;
4795 const XMMRegister xmm_temp5 = xmm5;
4796 const XMMRegister xmm_temp6 = xmm6;
4797 const XMMRegister xmm_temp7 = xmm7;
4798 const XMMRegister xmm_temp8 = xmm8;
4799 const XMMRegister xmm_temp9 = xmm9;
4800 const XMMRegister xmm_temp10 = xmm10;
4801
4802 __ enter();
4803
4804 __ movdqu(xmm_temp10, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
4805
4806 __ movdqu(xmm_temp0, Address(state, 0));
4807 __ pshufb(xmm_temp0, xmm_temp10);
4808
4809
4810 __ BIND(L_ghash_loop);
4811 __ movdqu(xmm_temp2, Address(data, 0));
4812 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
4813
4814 __ movdqu(xmm_temp1, Address(subkeyH, 0));
4815 __ pshufb(xmm_temp1, xmm_temp10);
4816
4817 __ pxor(xmm_temp0, xmm_temp2);
4818
4819 //
4820 // Multiply with the hash key
4821 //
4822 __ movdqu(xmm_temp3, xmm_temp0);
4823 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
4824 __ movdqu(xmm_temp4, xmm_temp0);
4825 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
4826
4827 __ movdqu(xmm_temp5, xmm_temp0);
4828 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
4829 __ movdqu(xmm_temp6, xmm_temp0);
4830 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
4831
4832 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
4833
4834 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
4835 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
4836 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
4837 __ pxor(xmm_temp3, xmm_temp5);
4838 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
4839 // of the carry-less multiplication of
4840 // xmm0 by xmm1.
4841
4842 // We shift the result of the multiplication by one bit position
4843 // to the left to cope for the fact that the bits are reversed.
4844 __ movdqu(xmm_temp7, xmm_temp3);
4845 __ movdqu(xmm_temp8, xmm_temp6);
4846 __ pslld(xmm_temp3, 1);
4847 __ pslld(xmm_temp6, 1);
4848 __ psrld(xmm_temp7, 31);
4849 __ psrld(xmm_temp8, 31);
4850 __ movdqu(xmm_temp9, xmm_temp7);
4851 __ pslldq(xmm_temp8, 4);
4852 __ pslldq(xmm_temp7, 4);
4853 __ psrldq(xmm_temp9, 12);
4854 __ por(xmm_temp3, xmm_temp7);
4855 __ por(xmm_temp6, xmm_temp8);
4856 __ por(xmm_temp6, xmm_temp9);
4857
4858 //
4859 // First phase of the reduction
4860 //
4861 // Move xmm3 into xmm7, xmm8, xmm9 in order to perform the shifts
4862 // independently.
4863 __ movdqu(xmm_temp7, xmm_temp3);
4864 __ movdqu(xmm_temp8, xmm_temp3);
4865 __ movdqu(xmm_temp9, xmm_temp3);
4866 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
4867 __ pslld(xmm_temp8, 30); // packed right shift shifting << 30
4868 __ pslld(xmm_temp9, 25); // packed right shift shifting << 25
4869 __ pxor(xmm_temp7, xmm_temp8); // xor the shifted versions
4870 __ pxor(xmm_temp7, xmm_temp9);
4871 __ movdqu(xmm_temp8, xmm_temp7);
4872 __ pslldq(xmm_temp7, 12);
4873 __ psrldq(xmm_temp8, 4);
4874 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
4875
4876 //
4877 // Second phase of the reduction
4878 //
4879 // Make 3 copies of xmm3 in xmm2, xmm4, xmm5 for doing these
4880 // shift operations.
4881 __ movdqu(xmm_temp2, xmm_temp3);
4882 __ movdqu(xmm_temp4, xmm_temp3);
4883 __ movdqu(xmm_temp5, xmm_temp3);
4884 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
4885 __ psrld(xmm_temp4, 2); // packed left shifting >> 2
4886 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
4887 __ pxor(xmm_temp2, xmm_temp4); // xor the shifted versions
4888 __ pxor(xmm_temp2, xmm_temp5);
4889 __ pxor(xmm_temp2, xmm_temp8);
4890 __ pxor(xmm_temp3, xmm_temp2);
4891 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
4892
4893 __ decrement(blocks);
4894 __ jcc(Assembler::zero, L_exit);
4895 __ movdqu(xmm_temp0, xmm_temp6);
4896 __ addptr(data, 16);
4897 __ jmp(L_ghash_loop);
4898
4899 __ BIND(L_exit);
4900 __ pshufb(xmm_temp6, xmm_temp10); // Byte swap 16-byte result
4901 __ movdqu(Address(state, 0), xmm_temp6); // store the result
4902 __ leave();
4903 __ ret(0);
4904 return start;
4905 }
4906
4907 //base64 character set
4908 address base64_charset_addr() {
4909 __ align(CodeEntryAlignment);
4910 StubCodeMark mark(this, "StubRoutines", "base64_charset");
4911 address start = __ pc();
4912 __ emit_data64(0x0000004200000041, relocInfo::none);
4913 __ emit_data64(0x0000004400000043, relocInfo::none);
4914 __ emit_data64(0x0000004600000045, relocInfo::none);
4915 __ emit_data64(0x0000004800000047, relocInfo::none);
4916 __ emit_data64(0x0000004a00000049, relocInfo::none);
4917 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4918 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4919 __ emit_data64(0x000000500000004f, relocInfo::none);
4920 __ emit_data64(0x0000005200000051, relocInfo::none);
4921 __ emit_data64(0x0000005400000053, relocInfo::none);
4922 __ emit_data64(0x0000005600000055, relocInfo::none);
4923 __ emit_data64(0x0000005800000057, relocInfo::none);
4924 __ emit_data64(0x0000005a00000059, relocInfo::none);
4925 __ emit_data64(0x0000006200000061, relocInfo::none);
4926 __ emit_data64(0x0000006400000063, relocInfo::none);
4927 __ emit_data64(0x0000006600000065, relocInfo::none);
4928 __ emit_data64(0x0000006800000067, relocInfo::none);
4929 __ emit_data64(0x0000006a00000069, relocInfo::none);
4930 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4931 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4932 __ emit_data64(0x000000700000006f, relocInfo::none);
4933 __ emit_data64(0x0000007200000071, relocInfo::none);
4934 __ emit_data64(0x0000007400000073, relocInfo::none);
4935 __ emit_data64(0x0000007600000075, relocInfo::none);
4936 __ emit_data64(0x0000007800000077, relocInfo::none);
4937 __ emit_data64(0x0000007a00000079, relocInfo::none);
4938 __ emit_data64(0x0000003100000030, relocInfo::none);
4939 __ emit_data64(0x0000003300000032, relocInfo::none);
4940 __ emit_data64(0x0000003500000034, relocInfo::none);
4941 __ emit_data64(0x0000003700000036, relocInfo::none);
4942 __ emit_data64(0x0000003900000038, relocInfo::none);
4943 __ emit_data64(0x0000002f0000002b, relocInfo::none);
4944 return start;
4945 }
4946
4947 //base64 url character set
4948 address base64url_charset_addr() {
4949 __ align(CodeEntryAlignment);
4950 StubCodeMark mark(this, "StubRoutines", "base64url_charset");
4951 address start = __ pc();
4952 __ emit_data64(0x0000004200000041, relocInfo::none);
4953 __ emit_data64(0x0000004400000043, relocInfo::none);
4954 __ emit_data64(0x0000004600000045, relocInfo::none);
4955 __ emit_data64(0x0000004800000047, relocInfo::none);
4956 __ emit_data64(0x0000004a00000049, relocInfo::none);
4957 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4958 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4959 __ emit_data64(0x000000500000004f, relocInfo::none);
4960 __ emit_data64(0x0000005200000051, relocInfo::none);
4961 __ emit_data64(0x0000005400000053, relocInfo::none);
4962 __ emit_data64(0x0000005600000055, relocInfo::none);
4963 __ emit_data64(0x0000005800000057, relocInfo::none);
4964 __ emit_data64(0x0000005a00000059, relocInfo::none);
4965 __ emit_data64(0x0000006200000061, relocInfo::none);
4966 __ emit_data64(0x0000006400000063, relocInfo::none);
4967 __ emit_data64(0x0000006600000065, relocInfo::none);
4968 __ emit_data64(0x0000006800000067, relocInfo::none);
4969 __ emit_data64(0x0000006a00000069, relocInfo::none);
4970 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4971 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4972 __ emit_data64(0x000000700000006f, relocInfo::none);
4973 __ emit_data64(0x0000007200000071, relocInfo::none);
4974 __ emit_data64(0x0000007400000073, relocInfo::none);
4975 __ emit_data64(0x0000007600000075, relocInfo::none);
4976 __ emit_data64(0x0000007800000077, relocInfo::none);
4977 __ emit_data64(0x0000007a00000079, relocInfo::none);
4978 __ emit_data64(0x0000003100000030, relocInfo::none);
4979 __ emit_data64(0x0000003300000032, relocInfo::none);
4980 __ emit_data64(0x0000003500000034, relocInfo::none);
4981 __ emit_data64(0x0000003700000036, relocInfo::none);
4982 __ emit_data64(0x0000003900000038, relocInfo::none);
4983 __ emit_data64(0x0000005f0000002d, relocInfo::none);
4984
4985 return start;
4986 }
4987
4988 address base64_bswap_mask_addr() {
4989 __ align(CodeEntryAlignment);
4990 StubCodeMark mark(this, "StubRoutines", "bswap_mask_base64");
4991 address start = __ pc();
4992 __ emit_data64(0x0504038002010080, relocInfo::none);
4993 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4994 __ emit_data64(0x0908078006050480, relocInfo::none);
4995 __ emit_data64(0x0f0e0d800c0b0a80, relocInfo::none);
4996 __ emit_data64(0x0605048003020180, relocInfo::none);
4997 __ emit_data64(0x0c0b0a8009080780, relocInfo::none);
4998 __ emit_data64(0x0504038002010080, relocInfo::none);
4999 __ emit_data64(0x0b0a098008070680, relocInfo::none);
5000
5001 return start;
5002 }
5003
5004 address base64_right_shift_mask_addr() {
5005 __ align(CodeEntryAlignment);
5006 StubCodeMark mark(this, "StubRoutines", "right_shift_mask");
5007 address start = __ pc();
5008 __ emit_data64(0x0006000400020000, relocInfo::none);
5009 __ emit_data64(0x0006000400020000, relocInfo::none);
5010 __ emit_data64(0x0006000400020000, relocInfo::none);
5011 __ emit_data64(0x0006000400020000, relocInfo::none);
5012 __ emit_data64(0x0006000400020000, relocInfo::none);
5013 __ emit_data64(0x0006000400020000, relocInfo::none);
5014 __ emit_data64(0x0006000400020000, relocInfo::none);
5015 __ emit_data64(0x0006000400020000, relocInfo::none);
5016
5017 return start;
5018 }
5019
5020 address base64_left_shift_mask_addr() {
5021 __ align(CodeEntryAlignment);
5022 StubCodeMark mark(this, "StubRoutines", "left_shift_mask");
5023 address start = __ pc();
5024 __ emit_data64(0x0000000200040000, relocInfo::none);
5025 __ emit_data64(0x0000000200040000, relocInfo::none);
5026 __ emit_data64(0x0000000200040000, relocInfo::none);
5027 __ emit_data64(0x0000000200040000, relocInfo::none);
5028 __ emit_data64(0x0000000200040000, relocInfo::none);
5029 __ emit_data64(0x0000000200040000, relocInfo::none);
5030 __ emit_data64(0x0000000200040000, relocInfo::none);
5031 __ emit_data64(0x0000000200040000, relocInfo::none);
5032
5033 return start;
5034 }
5035
5036 address base64_and_mask_addr() {
5037 __ align(CodeEntryAlignment);
5038 StubCodeMark mark(this, "StubRoutines", "and_mask");
5039 address start = __ pc();
5040 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5041 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5042 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5043 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5044 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5045 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5046 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5047 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5048 return start;
5049 }
5050
5051 address base64_gather_mask_addr() {
5052 __ align(CodeEntryAlignment);
5053 StubCodeMark mark(this, "StubRoutines", "gather_mask");
5054 address start = __ pc();
5055 __ emit_data64(0xffffffffffffffff, relocInfo::none);
5056 return start;
5057 }
5058
5059 // Code for generating Base64 encoding.
5060 // Intrinsic function prototype in Base64.java:
5061 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL) {
5062 address generate_base64_encodeBlock() {
5063 __ align(CodeEntryAlignment);
5064 StubCodeMark mark(this, "StubRoutines", "implEncode");
5065 address start = __ pc();
5066 __ enter();
5067
5068 // Save callee-saved registers before using them
5069 __ push(r12);
5070 __ push(r13);
5071 __ push(r14);
5072 __ push(r15);
5073
5074 // arguments
5075 const Register source = c_rarg0; // Source Array
5076 const Register start_offset = c_rarg1; // start offset
5077 const Register end_offset = c_rarg2; // end offset
5078 const Register dest = c_rarg3; // destination array
5079
5080 #ifndef _WIN64
5081 const Register dp = c_rarg4; // Position for writing to dest array
5082 const Register isURL = c_rarg5;// Base64 or URL character set
5083 #else
5084 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
5085 const Address isURL_mem(rbp, 7 * wordSize);
5086 const Register isURL = r10; // pick the volatile windows register
5087 const Register dp = r12;
5088 __ movl(dp, dp_mem);
5089 __ movl(isURL, isURL_mem);
5090 #endif
5091
5092 const Register length = r14;
5093 Label L_process80, L_process32, L_process3, L_exit, L_processdata;
5094
5095 // calculate length from offsets
5096 __ movl(length, end_offset);
5097 __ subl(length, start_offset);
5098 __ cmpl(length, 0);
5099 __ jcc(Assembler::lessEqual, L_exit);
5100
5101 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_charset_addr()));
5102 // check if base64 charset(isURL=0) or base64 url charset(isURL=1) needs to be loaded
5103 __ cmpl(isURL, 0);
5104 __ jcc(Assembler::equal, L_processdata);
5105 __ lea(r11, ExternalAddress(StubRoutines::x86::base64url_charset_addr()));
5106
5107 // load masks required for encoding data
5108 __ BIND(L_processdata);
5109 __ movdqu(xmm16, ExternalAddress(StubRoutines::x86::base64_gather_mask_addr()));
5110 // Set 64 bits of K register.
5111 __ evpcmpeqb(k3, xmm16, xmm16, Assembler::AVX_512bit);
5112 __ evmovdquq(xmm12, ExternalAddress(StubRoutines::x86::base64_bswap_mask_addr()), Assembler::AVX_256bit, r13);
5113 __ evmovdquq(xmm13, ExternalAddress(StubRoutines::x86::base64_right_shift_mask_addr()), Assembler::AVX_512bit, r13);
5114 __ evmovdquq(xmm14, ExternalAddress(StubRoutines::x86::base64_left_shift_mask_addr()), Assembler::AVX_512bit, r13);
5115 __ evmovdquq(xmm15, ExternalAddress(StubRoutines::x86::base64_and_mask_addr()), Assembler::AVX_512bit, r13);
5116
5117 // Vector Base64 implementation, producing 96 bytes of encoded data
5118 __ BIND(L_process80);
5119 __ cmpl(length, 80);
5120 __ jcc(Assembler::below, L_process32);
5121 __ evmovdquq(xmm0, Address(source, start_offset, Address::times_1, 0), Assembler::AVX_256bit);
5122 __ evmovdquq(xmm1, Address(source, start_offset, Address::times_1, 24), Assembler::AVX_256bit);
5123 __ evmovdquq(xmm2, Address(source, start_offset, Address::times_1, 48), Assembler::AVX_256bit);
5124
5125 //permute the input data in such a manner that we have continuity of the source
5126 __ vpermq(xmm3, xmm0, 148, Assembler::AVX_256bit);
5127 __ vpermq(xmm4, xmm1, 148, Assembler::AVX_256bit);
5128 __ vpermq(xmm5, xmm2, 148, Assembler::AVX_256bit);
5129
5130 //shuffle input and group 3 bytes of data and to it add 0 as the 4th byte.
5131 //we can deal with 12 bytes at a time in a 128 bit register
5132 __ vpshufb(xmm3, xmm3, xmm12, Assembler::AVX_256bit);
5133 __ vpshufb(xmm4, xmm4, xmm12, Assembler::AVX_256bit);
5134 __ vpshufb(xmm5, xmm5, xmm12, Assembler::AVX_256bit);
5135
5136 //convert byte to word. Each 128 bit register will have 6 bytes for processing
5137 __ vpmovzxbw(xmm3, xmm3, Assembler::AVX_512bit);
5138 __ vpmovzxbw(xmm4, xmm4, Assembler::AVX_512bit);
5139 __ vpmovzxbw(xmm5, xmm5, Assembler::AVX_512bit);
5140
5141 // Extract bits in the following pattern 6, 4+2, 2+4, 6 to convert 3, 8 bit numbers to 4, 6 bit numbers
5142 __ evpsrlvw(xmm0, xmm3, xmm13, Assembler::AVX_512bit);
5143 __ evpsrlvw(xmm1, xmm4, xmm13, Assembler::AVX_512bit);
5144 __ evpsrlvw(xmm2, xmm5, xmm13, Assembler::AVX_512bit);
5145
5146 __ evpsllvw(xmm3, xmm3, xmm14, Assembler::AVX_512bit);
5147 __ evpsllvw(xmm4, xmm4, xmm14, Assembler::AVX_512bit);
5148 __ evpsllvw(xmm5, xmm5, xmm14, Assembler::AVX_512bit);
5149
5150 __ vpsrlq(xmm0, xmm0, 8, Assembler::AVX_512bit);
5151 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
5152 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
5153
5154 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5155 __ vpsllq(xmm4, xmm4, 8, Assembler::AVX_512bit);
5156 __ vpsllq(xmm5, xmm5, 8, Assembler::AVX_512bit);
5157
5158 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
5159 __ vpandq(xmm4, xmm4, xmm15, Assembler::AVX_512bit);
5160 __ vpandq(xmm5, xmm5, xmm15, Assembler::AVX_512bit);
5161
5162 // Get the final 4*6 bits base64 encoding
5163 __ vporq(xmm3, xmm3, xmm0, Assembler::AVX_512bit);
5164 __ vporq(xmm4, xmm4, xmm1, Assembler::AVX_512bit);
5165 __ vporq(xmm5, xmm5, xmm2, Assembler::AVX_512bit);
5166
5167 // Shift
5168 __ vpsrlq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5169 __ vpsrlq(xmm4, xmm4, 8, Assembler::AVX_512bit);
5170 __ vpsrlq(xmm5, xmm5, 8, Assembler::AVX_512bit);
5171
5172 // look up 6 bits in the base64 character set to fetch the encoding
5173 // we are converting word to dword as gather instructions need dword indices for looking up encoding
5174 __ vextracti64x4(xmm6, xmm3, 0);
5175 __ vpmovzxwd(xmm0, xmm6, Assembler::AVX_512bit);
5176 __ vextracti64x4(xmm6, xmm3, 1);
5177 __ vpmovzxwd(xmm1, xmm6, Assembler::AVX_512bit);
5178
5179 __ vextracti64x4(xmm6, xmm4, 0);
5180 __ vpmovzxwd(xmm2, xmm6, Assembler::AVX_512bit);
5181 __ vextracti64x4(xmm6, xmm4, 1);
5182 __ vpmovzxwd(xmm3, xmm6, Assembler::AVX_512bit);
5183
5184 __ vextracti64x4(xmm4, xmm5, 0);
5185 __ vpmovzxwd(xmm6, xmm4, Assembler::AVX_512bit);
5186
5187 __ vextracti64x4(xmm4, xmm5, 1);
5188 __ vpmovzxwd(xmm7, xmm4, Assembler::AVX_512bit);
5189
5190 __ kmovql(k2, k3);
5191 __ evpgatherdd(xmm4, k2, Address(r11, xmm0, Address::times_4, 0), Assembler::AVX_512bit);
5192 __ kmovql(k2, k3);
5193 __ evpgatherdd(xmm5, k2, Address(r11, xmm1, Address::times_4, 0), Assembler::AVX_512bit);
5194 __ kmovql(k2, k3);
5195 __ evpgatherdd(xmm8, k2, Address(r11, xmm2, Address::times_4, 0), Assembler::AVX_512bit);
5196 __ kmovql(k2, k3);
5197 __ evpgatherdd(xmm9, k2, Address(r11, xmm3, Address::times_4, 0), Assembler::AVX_512bit);
5198 __ kmovql(k2, k3);
5199 __ evpgatherdd(xmm10, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
5200 __ kmovql(k2, k3);
5201 __ evpgatherdd(xmm11, k2, Address(r11, xmm7, Address::times_4, 0), Assembler::AVX_512bit);
5202
5203 //Down convert dword to byte. Final output is 16*6 = 96 bytes long
5204 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm4, Assembler::AVX_512bit);
5205 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm5, Assembler::AVX_512bit);
5206 __ evpmovdb(Address(dest, dp, Address::times_1, 32), xmm8, Assembler::AVX_512bit);
5207 __ evpmovdb(Address(dest, dp, Address::times_1, 48), xmm9, Assembler::AVX_512bit);
5208 __ evpmovdb(Address(dest, dp, Address::times_1, 64), xmm10, Assembler::AVX_512bit);
5209 __ evpmovdb(Address(dest, dp, Address::times_1, 80), xmm11, Assembler::AVX_512bit);
5210
5211 __ addq(dest, 96);
5212 __ addq(source, 72);
5213 __ subq(length, 72);
5214 __ jmp(L_process80);
5215
5216 // Vector Base64 implementation generating 32 bytes of encoded data
5217 __ BIND(L_process32);
5218 __ cmpl(length, 32);
5219 __ jcc(Assembler::below, L_process3);
5220 __ evmovdquq(xmm0, Address(source, start_offset), Assembler::AVX_256bit);
5221 __ vpermq(xmm0, xmm0, 148, Assembler::AVX_256bit);
5222 __ vpshufb(xmm6, xmm0, xmm12, Assembler::AVX_256bit);
5223 __ vpmovzxbw(xmm6, xmm6, Assembler::AVX_512bit);
5224 __ evpsrlvw(xmm2, xmm6, xmm13, Assembler::AVX_512bit);
5225 __ evpsllvw(xmm3, xmm6, xmm14, Assembler::AVX_512bit);
5226
5227 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
5228 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5229 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
5230 __ vporq(xmm1, xmm2, xmm3, Assembler::AVX_512bit);
5231 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
5232 __ vextracti64x4(xmm9, xmm1, 0);
5233 __ vpmovzxwd(xmm6, xmm9, Assembler::AVX_512bit);
5234 __ vextracti64x4(xmm9, xmm1, 1);
5235 __ vpmovzxwd(xmm5, xmm9, Assembler::AVX_512bit);
5236 __ kmovql(k2, k3);
5237 __ evpgatherdd(xmm8, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
5238 __ kmovql(k2, k3);
5239 __ evpgatherdd(xmm10, k2, Address(r11, xmm5, Address::times_4, 0), Assembler::AVX_512bit);
5240 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm8, Assembler::AVX_512bit);
5241 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm10, Assembler::AVX_512bit);
5242 __ subq(length, 24);
5243 __ addq(dest, 32);
5244 __ addq(source, 24);
5245 __ jmp(L_process32);
5246
5247 // Scalar data processing takes 3 bytes at a time and produces 4 bytes of encoded data
5248 /* This code corresponds to the scalar version of the following snippet in Base64.java
5249 ** int bits = (src[sp0++] & 0xff) << 16 |(src[sp0++] & 0xff) << 8 |(src[sp0++] & 0xff);
5250 ** dst[dp0++] = (byte)base64[(bits >> > 18) & 0x3f];
5251 ** dst[dp0++] = (byte)base64[(bits >> > 12) & 0x3f];
5252 ** dst[dp0++] = (byte)base64[(bits >> > 6) & 0x3f];
5253 ** dst[dp0++] = (byte)base64[bits & 0x3f];*/
5254 __ BIND(L_process3);
5255 __ cmpl(length, 3);
5256 __ jcc(Assembler::below, L_exit);
5257 // Read 1 byte at a time
5258 __ movzbl(rax, Address(source, start_offset));
5259 __ shll(rax, 0x10);
5260 __ movl(r15, rax);
5261 __ movzbl(rax, Address(source, start_offset, Address::times_1, 1));
5262 __ shll(rax, 0x8);
5263 __ movzwl(rax, rax);
5264 __ orl(r15, rax);
5265 __ movzbl(rax, Address(source, start_offset, Address::times_1, 2));
5266 __ orl(rax, r15);
5267 // Save 3 bytes read in r15
5268 __ movl(r15, rax);
5269 __ shrl(rax, 0x12);
5270 __ andl(rax, 0x3f);
5271 // rax contains the index, r11 contains base64 lookup table
5272 __ movb(rax, Address(r11, rax, Address::times_4));
5273 // Write the encoded byte to destination
5274 __ movb(Address(dest, dp, Address::times_1, 0), rax);
5275 __ movl(rax, r15);
5276 __ shrl(rax, 0xc);
5277 __ andl(rax, 0x3f);
5278 __ movb(rax, Address(r11, rax, Address::times_4));
5279 __ movb(Address(dest, dp, Address::times_1, 1), rax);
5280 __ movl(rax, r15);
5281 __ shrl(rax, 0x6);
5282 __ andl(rax, 0x3f);
5283 __ movb(rax, Address(r11, rax, Address::times_4));
5284 __ movb(Address(dest, dp, Address::times_1, 2), rax);
5285 __ movl(rax, r15);
5286 __ andl(rax, 0x3f);
5287 __ movb(rax, Address(r11, rax, Address::times_4));
5288 __ movb(Address(dest, dp, Address::times_1, 3), rax);
5289 __ subl(length, 3);
5290 __ addq(dest, 4);
5291 __ addq(source, 3);
5292 __ jmp(L_process3);
5293 __ BIND(L_exit);
5294 __ pop(r15);
5295 __ pop(r14);
5296 __ pop(r13);
5297 __ pop(r12);
5298 __ leave();
5299 __ ret(0);
5300 return start;
5301 }
5302
5303 /**
5304 * Arguments:
5305 *
5306 * Inputs:
5307 * c_rarg0 - int crc
5308 * c_rarg1 - byte* buf
5309 * c_rarg2 - int length
5310 *
5311 * Ouput:
5312 * rax - int crc result
5313 */
5314 address generate_updateBytesCRC32() {
5315 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
5316
5317 __ align(CodeEntryAlignment);
5318 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
5319
5320 address start = __ pc();
5321 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5322 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5323 // rscratch1: r10
5324 const Register crc = c_rarg0; // crc
5325 const Register buf = c_rarg1; // source java byte array address
5326 const Register len = c_rarg2; // length
5327 const Register table = c_rarg3; // crc_table address (reuse register)
5328 const Register tmp1 = r11;
5329 const Register tmp2 = r10;
5330 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax);
5331
5332 BLOCK_COMMENT("Entry:");
5333 __ enter(); // required for proper stackwalking of RuntimeStub frame
5334
5335 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
5336 VM_Version::supports_avx512bw() &&
5337 VM_Version::supports_avx512vl()) {
5338 __ kernel_crc32_avx512(crc, buf, len, table, tmp1, tmp2);
5339 } else {
5340 __ kernel_crc32(crc, buf, len, table, tmp1);
5341 }
5342
5343 __ movl(rax, crc);
5344 __ vzeroupper();
5345 __ leave(); // required for proper stackwalking of RuntimeStub frame
5346 __ ret(0);
5347
5348 return start;
5349 }
5350
5351 /**
5352 * Arguments:
5353 *
5354 * Inputs:
5355 * c_rarg0 - int crc
5356 * c_rarg1 - byte* buf
5357 * c_rarg2 - long length
5358 * c_rarg3 - table_start - optional (present only when doing a library_call,
5359 * not used by x86 algorithm)
5360 *
5361 * Ouput:
5362 * rax - int crc result
5363 */
5364 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
5365 assert(UseCRC32CIntrinsics, "need SSE4_2");
5366 __ align(CodeEntryAlignment);
5367 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
5368 address start = __ pc();
5369 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
5370 //Windows RCX RDX R8 R9 none none XMM0..XMM3
5371 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
5372 const Register crc = c_rarg0; // crc
5373 const Register buf = c_rarg1; // source java byte array address
5374 const Register len = c_rarg2; // length
5375 const Register a = rax;
5376 const Register j = r9;
5377 const Register k = r10;
5378 const Register l = r11;
5379 #ifdef _WIN64
5380 const Register y = rdi;
5381 const Register z = rsi;
5382 #else
5383 const Register y = rcx;
5384 const Register z = r8;
5385 #endif
5386 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
5387
5388 BLOCK_COMMENT("Entry:");
5389 __ enter(); // required for proper stackwalking of RuntimeStub frame
5390 #ifdef _WIN64
5391 __ push(y);
5392 __ push(z);
5393 #endif
5394 __ crc32c_ipl_alg2_alt2(crc, buf, len,
5395 a, j, k,
5396 l, y, z,
5397 c_farg0, c_farg1, c_farg2,
5398 is_pclmulqdq_supported);
5399 __ movl(rax, crc);
5400 #ifdef _WIN64
5401 __ pop(z);
5402 __ pop(y);
5403 #endif
5404 __ vzeroupper();
5405 __ leave(); // required for proper stackwalking of RuntimeStub frame
5406 __ ret(0);
5407
5408 return start;
5409 }
5410
5411 /**
5412 * Arguments:
5413 *
5414 * Input:
5415 * c_rarg0 - x address
5416 * c_rarg1 - x length
5417 * c_rarg2 - y address
5418 * c_rarg3 - y length
5419 * not Win64
5420 * c_rarg4 - z address
5421 * c_rarg5 - z length
5422 * Win64
5423 * rsp+40 - z address
5424 * rsp+48 - z length
5425 */
5426 address generate_multiplyToLen() {
5427 __ align(CodeEntryAlignment);
5428 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
5429
5430 address start = __ pc();
5431 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5432 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5433 const Register x = rdi;
5434 const Register xlen = rax;
5435 const Register y = rsi;
5436 const Register ylen = rcx;
5437 const Register z = r8;
5438 const Register zlen = r11;
5439
5440 // Next registers will be saved on stack in multiply_to_len().
5441 const Register tmp1 = r12;
5442 const Register tmp2 = r13;
5443 const Register tmp3 = r14;
5444 const Register tmp4 = r15;
5445 const Register tmp5 = rbx;
5446
5447 BLOCK_COMMENT("Entry:");
5448 __ enter(); // required for proper stackwalking of RuntimeStub frame
5449
5450 #ifndef _WIN64
5451 __ movptr(zlen, r9); // Save r9 in r11 - zlen
5452 #endif
5453 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
5454 // ylen => rcx, z => r8, zlen => r11
5455 // r9 and r10 may be used to save non-volatile registers
5456 #ifdef _WIN64
5457 // last 2 arguments (#4, #5) are on stack on Win64
5458 __ movptr(z, Address(rsp, 6 * wordSize));
5459 __ movptr(zlen, Address(rsp, 7 * wordSize));
5460 #endif
5461
5462 __ movptr(xlen, rsi);
5463 __ movptr(y, rdx);
5464 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
5465
5466 restore_arg_regs();
5467
5468 __ leave(); // required for proper stackwalking of RuntimeStub frame
5469 __ ret(0);
5470
5471 return start;
5472 }
5473
5474 /**
5475 * Arguments:
5476 *
5477 * Input:
5478 * c_rarg0 - obja address
5479 * c_rarg1 - objb address
5480 * c_rarg3 - length length
5481 * c_rarg4 - scale log2_array_indxscale
5482 *
5483 * Output:
5484 * rax - int >= mismatched index, < 0 bitwise complement of tail
5485 */
5486 address generate_vectorizedMismatch() {
5487 __ align(CodeEntryAlignment);
5488 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
5489 address start = __ pc();
5490
5491 BLOCK_COMMENT("Entry:");
5492 __ enter();
5493
5494 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5495 const Register scale = c_rarg0; //rcx, will exchange with r9
5496 const Register objb = c_rarg1; //rdx
5497 const Register length = c_rarg2; //r8
5498 const Register obja = c_rarg3; //r9
5499 __ xchgq(obja, scale); //now obja and scale contains the correct contents
5500
5501 const Register tmp1 = r10;
5502 const Register tmp2 = r11;
5503 #endif
5504 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5505 const Register obja = c_rarg0; //U:rdi
5506 const Register objb = c_rarg1; //U:rsi
5507 const Register length = c_rarg2; //U:rdx
5508 const Register scale = c_rarg3; //U:rcx
5509 const Register tmp1 = r8;
5510 const Register tmp2 = r9;
5511 #endif
5512 const Register result = rax; //return value
5513 const XMMRegister vec0 = xmm0;
5514 const XMMRegister vec1 = xmm1;
5515 const XMMRegister vec2 = xmm2;
5516
5517 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
5518
5519 __ vzeroupper();
5520 __ leave();
5521 __ ret(0);
5522
5523 return start;
5524 }
5525
5526 /**
5527 * Arguments:
5528 *
5529 // Input:
5530 // c_rarg0 - x address
5531 // c_rarg1 - x length
5532 // c_rarg2 - z address
5533 // c_rarg3 - z lenth
5534 *
5535 */
5536 address generate_squareToLen() {
5537
5538 __ align(CodeEntryAlignment);
5539 StubCodeMark mark(this, "StubRoutines", "squareToLen");
5540
5541 address start = __ pc();
5542 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5543 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
5544 const Register x = rdi;
5545 const Register len = rsi;
5546 const Register z = r8;
5547 const Register zlen = rcx;
5548
5549 const Register tmp1 = r12;
5550 const Register tmp2 = r13;
5551 const Register tmp3 = r14;
5552 const Register tmp4 = r15;
5553 const Register tmp5 = rbx;
5554
5555 BLOCK_COMMENT("Entry:");
5556 __ enter(); // required for proper stackwalking of RuntimeStub frame
5557
5558 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
5559 // zlen => rcx
5560 // r9 and r10 may be used to save non-volatile registers
5561 __ movptr(r8, rdx);
5562 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5563
5564 restore_arg_regs();
5565
5566 __ leave(); // required for proper stackwalking of RuntimeStub frame
5567 __ ret(0);
5568
5569 return start;
5570 }
5571
5572 address generate_method_entry_barrier() {
5573 __ align(CodeEntryAlignment);
5574 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
5575
5576 Label deoptimize_label;
5577
5578 address start = __ pc();
5579
5580 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
5581
5582 BLOCK_COMMENT("Entry:");
5583 __ enter(); // save rbp
5584
5585 // save c_rarg0, because we want to use that value.
5586 // We could do without it but then we depend on the number of slots used by pusha
5587 __ push(c_rarg0);
5588
5589 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
5590
5591 __ pusha();
5592
5593 // The method may have floats as arguments, and we must spill them before calling
5594 // the VM runtime.
5595 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
5596 const int xmm_size = wordSize * 2;
5597 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
5598 __ subptr(rsp, xmm_spill_size);
5599 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
5600 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
5601 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
5602 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
5603 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
5604 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
5605 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
5606 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
5607
5608 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
5609
5610 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
5611 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
5612 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
5613 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
5614 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
5615 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
5616 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
5617 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
5618 __ addptr(rsp, xmm_spill_size);
5619
5620 __ cmpl(rax, 1); // 1 means deoptimize
5621 __ jcc(Assembler::equal, deoptimize_label);
5622
5623 __ popa();
5624 __ pop(c_rarg0);
5625
5626 __ leave();
5627
5628 __ addptr(rsp, 1 * wordSize); // cookie
5629 __ ret(0);
5630
5631
5632 __ BIND(deoptimize_label);
5633
5634 __ popa();
5635 __ pop(c_rarg0);
5636
5637 __ leave();
5638
5639 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
5640 // here while still having a correct stack is valuable
5641 __ testptr(rsp, Address(rsp, 0));
5642
5643 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
5644 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
5645
5646 return start;
5647 }
5648
5649 /**
5650 * Arguments:
5651 *
5652 * Input:
5653 * c_rarg0 - out address
5654 * c_rarg1 - in address
5655 * c_rarg2 - offset
5656 * c_rarg3 - len
5657 * not Win64
5658 * c_rarg4 - k
5659 * Win64
5660 * rsp+40 - k
5661 */
5662 address generate_mulAdd() {
5663 __ align(CodeEntryAlignment);
5664 StubCodeMark mark(this, "StubRoutines", "mulAdd");
5665
5666 address start = __ pc();
5667 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5668 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5669 const Register out = rdi;
5670 const Register in = rsi;
5671 const Register offset = r11;
5672 const Register len = rcx;
5673 const Register k = r8;
5674
5675 // Next registers will be saved on stack in mul_add().
5676 const Register tmp1 = r12;
5677 const Register tmp2 = r13;
5678 const Register tmp3 = r14;
5679 const Register tmp4 = r15;
5680 const Register tmp5 = rbx;
5681
5682 BLOCK_COMMENT("Entry:");
5683 __ enter(); // required for proper stackwalking of RuntimeStub frame
5684
5685 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
5686 // len => rcx, k => r8
5687 // r9 and r10 may be used to save non-volatile registers
5688 #ifdef _WIN64
5689 // last argument is on stack on Win64
5690 __ movl(k, Address(rsp, 6 * wordSize));
5691 #endif
5692 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
5693 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5694
5695 restore_arg_regs();
5696
5697 __ leave(); // required for proper stackwalking of RuntimeStub frame
5698 __ ret(0);
5699
5700 return start;
5701 }
5702
5703 address generate_bigIntegerRightShift() {
5704 __ align(CodeEntryAlignment);
5705 StubCodeMark mark(this, "StubRoutines", "bigIntegerRightShiftWorker");
5706
5707 address start = __ pc();
5708 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
5709 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
5710 const Register newArr = rdi;
5711 const Register oldArr = rsi;
5712 const Register newIdx = rdx;
5713 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
5714 const Register totalNumIter = r8;
5715
5716 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
5717 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
5718 const Register tmp1 = r11; // Caller save.
5719 const Register tmp2 = rax; // Caller save.
5720 const Register tmp3 = WINDOWS_ONLY(r12) NOT_WINDOWS(r9); // Windows: Callee save. Linux: Caller save.
5721 const Register tmp4 = WINDOWS_ONLY(r13) NOT_WINDOWS(r10); // Windows: Callee save. Linux: Caller save.
5722 const Register tmp5 = r14; // Callee save.
5723 const Register tmp6 = r15;
5724
5725 const XMMRegister x0 = xmm0;
5726 const XMMRegister x1 = xmm1;
5727 const XMMRegister x2 = xmm2;
5728
5729 BLOCK_COMMENT("Entry:");
5730 __ enter(); // required for proper stackwalking of RuntimeStub frame
5731
5732 #ifdef _WINDOWS
5733 setup_arg_regs(4);
5734 // For windows, since last argument is on stack, we need to move it to the appropriate register.
5735 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
5736 // Save callee save registers.
5737 __ push(tmp3);
5738 __ push(tmp4);
5739 #endif
5740 __ push(tmp5);
5741
5742 // Rename temps used throughout the code.
5743 const Register idx = tmp1;
5744 const Register nIdx = tmp2;
5745
5746 __ xorl(idx, idx);
5747
5748 // Start right shift from end of the array.
5749 // For example, if #iteration = 4 and newIdx = 1
5750 // then dest[4] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
5751 // if #iteration = 4 and newIdx = 0
5752 // then dest[3] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
5753 __ movl(idx, totalNumIter);
5754 __ movl(nIdx, idx);
5755 __ addl(nIdx, newIdx);
5756
5757 // If vectorization is enabled, check if the number of iterations is at least 64
5758 // If not, then go to ShifTwo processing 2 iterations
5759 if (VM_Version::supports_avx512_vbmi2()) {
5760 __ cmpptr(totalNumIter, (AVX3Threshold/64));
5761 __ jcc(Assembler::less, ShiftTwo);
5762
5763 if (AVX3Threshold < 16 * 64) {
5764 __ cmpl(totalNumIter, 16);
5765 __ jcc(Assembler::less, ShiftTwo);
5766 }
5767 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
5768 __ subl(idx, 16);
5769 __ subl(nIdx, 16);
5770 __ BIND(Shift512Loop);
5771 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 4), Assembler::AVX_512bit);
5772 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
5773 __ vpshrdvd(x2, x1, x0, Assembler::AVX_512bit);
5774 __ evmovdqul(Address(newArr, nIdx, Address::times_4), x2, Assembler::AVX_512bit);
5775 __ subl(nIdx, 16);
5776 __ subl(idx, 16);
5777 __ jcc(Assembler::greaterEqual, Shift512Loop);
5778 __ addl(idx, 16);
5779 __ addl(nIdx, 16);
5780 }
5781 __ BIND(ShiftTwo);
5782 __ cmpl(idx, 2);
5783 __ jcc(Assembler::less, ShiftOne);
5784 __ subl(idx, 2);
5785 __ subl(nIdx, 2);
5786 __ BIND(ShiftTwoLoop);
5787 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 8));
5788 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
5789 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
5790 __ shrdl(tmp5, tmp4);
5791 __ shrdl(tmp4, tmp3);
5792 __ movl(Address(newArr, nIdx, Address::times_4, 4), tmp5);
5793 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
5794 __ subl(nIdx, 2);
5795 __ subl(idx, 2);
5796 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
5797 __ addl(idx, 2);
5798 __ addl(nIdx, 2);
5799
5800 // Do the last iteration
5801 __ BIND(ShiftOne);
5802 __ cmpl(idx, 1);
5803 __ jcc(Assembler::less, Exit);
5804 __ subl(idx, 1);
5805 __ subl(nIdx, 1);
5806 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
5807 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
5808 __ shrdl(tmp4, tmp3);
5809 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
5810 __ BIND(Exit);
5811 // Restore callee save registers.
5812 __ pop(tmp5);
5813 #ifdef _WINDOWS
5814 __ pop(tmp4);
5815 __ pop(tmp3);
5816 restore_arg_regs();
5817 #endif
5818 __ leave(); // required for proper stackwalking of RuntimeStub frame
5819 __ ret(0);
5820 return start;
5821 }
5822
5823 /**
5824 * Arguments:
5825 *
5826 * Input:
5827 * c_rarg0 - newArr address
5828 * c_rarg1 - oldArr address
5829 * c_rarg2 - newIdx
5830 * c_rarg3 - shiftCount
5831 * not Win64
5832 * c_rarg4 - numIter
5833 * Win64
5834 * rsp40 - numIter
5835 */
5836 address generate_bigIntegerLeftShift() {
5837 __ align(CodeEntryAlignment);
5838 StubCodeMark mark(this, "StubRoutines", "bigIntegerLeftShiftWorker");
5839 address start = __ pc();
5840 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
5841 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
5842 const Register newArr = rdi;
5843 const Register oldArr = rsi;
5844 const Register newIdx = rdx;
5845 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
5846 const Register totalNumIter = r8;
5847 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
5848 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
5849 const Register tmp1 = r11; // Caller save.
5850 const Register tmp2 = rax; // Caller save.
5851 const Register tmp3 = WINDOWS_ONLY(r12) NOT_WINDOWS(r9); // Windows: Callee save. Linux: Caller save.
5852 const Register tmp4 = WINDOWS_ONLY(r13) NOT_WINDOWS(r10); // Windows: Callee save. Linux: Caller save.
5853 const Register tmp5 = r14; // Callee save.
5854
5855 const XMMRegister x0 = xmm0;
5856 const XMMRegister x1 = xmm1;
5857 const XMMRegister x2 = xmm2;
5858 BLOCK_COMMENT("Entry:");
5859 __ enter(); // required for proper stackwalking of RuntimeStub frame
5860
5861 #ifdef _WINDOWS
5862 setup_arg_regs(4);
5863 // For windows, since last argument is on stack, we need to move it to the appropriate register.
5864 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
5865 // Save callee save registers.
5866 __ push(tmp3);
5867 __ push(tmp4);
5868 #endif
5869 __ push(tmp5);
5870
5871 // Rename temps used throughout the code
5872 const Register idx = tmp1;
5873 const Register numIterTmp = tmp2;
5874
5875 // Start idx from zero.
5876 __ xorl(idx, idx);
5877 // Compute interior pointer for new array. We do this so that we can use same index for both old and new arrays.
5878 __ lea(newArr, Address(newArr, newIdx, Address::times_4));
5879 __ movl(numIterTmp, totalNumIter);
5880
5881 // If vectorization is enabled, check if the number of iterations is at least 64
5882 // If not, then go to ShiftTwo shifting two numbers at a time
5883 if (VM_Version::supports_avx512_vbmi2()) {
5884 __ cmpl(totalNumIter, (AVX3Threshold/64));
5885 __ jcc(Assembler::less, ShiftTwo);
5886
5887 if (AVX3Threshold < 16 * 64) {
5888 __ cmpl(totalNumIter, 16);
5889 __ jcc(Assembler::less, ShiftTwo);
5890 }
5891 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
5892 __ subl(numIterTmp, 16);
5893 __ BIND(Shift512Loop);
5894 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
5895 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 0x4), Assembler::AVX_512bit);
5896 __ vpshldvd(x1, x2, x0, Assembler::AVX_512bit);
5897 __ evmovdqul(Address(newArr, idx, Address::times_4), x1, Assembler::AVX_512bit);
5898 __ addl(idx, 16);
5899 __ subl(numIterTmp, 16);
5900 __ jcc(Assembler::greaterEqual, Shift512Loop);
5901 __ addl(numIterTmp, 16);
5902 }
5903 __ BIND(ShiftTwo);
5904 __ cmpl(totalNumIter, 1);
5905 __ jcc(Assembler::less, Exit);
5906 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
5907 __ subl(numIterTmp, 2);
5908 __ jcc(Assembler::less, ShiftOne);
5909
5910 __ BIND(ShiftTwoLoop);
5911 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
5912 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 0x8));
5913 __ shldl(tmp3, tmp4);
5914 __ shldl(tmp4, tmp5);
5915 __ movl(Address(newArr, idx, Address::times_4), tmp3);
5916 __ movl(Address(newArr, idx, Address::times_4, 0x4), tmp4);
5917 __ movl(tmp3, tmp5);
5918 __ addl(idx, 2);
5919 __ subl(numIterTmp, 2);
5920 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
5921
5922 // Do the last iteration
5923 __ BIND(ShiftOne);
5924 __ addl(numIterTmp, 2);
5925 __ cmpl(numIterTmp, 1);
5926 __ jcc(Assembler::less, Exit);
5927 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
5928 __ shldl(tmp3, tmp4);
5929 __ movl(Address(newArr, idx, Address::times_4), tmp3);
5930
5931 __ BIND(Exit);
5932 // Restore callee save registers.
5933 __ pop(tmp5);
5934 #ifdef _WINDOWS
5935 __ pop(tmp4);
5936 __ pop(tmp3);
5937 restore_arg_regs();
5938 #endif
5939 __ leave(); // required for proper stackwalking of RuntimeStub frame
5940 __ ret(0);
5941 return start;
5942 }
5943
5944 address generate_libmExp() {
5945 StubCodeMark mark(this, "StubRoutines", "libmExp");
5946
5947 address start = __ pc();
5948
5949 const XMMRegister x0 = xmm0;
5950 const XMMRegister x1 = xmm1;
5951 const XMMRegister x2 = xmm2;
5952 const XMMRegister x3 = xmm3;
5953
5954 const XMMRegister x4 = xmm4;
5955 const XMMRegister x5 = xmm5;
5956 const XMMRegister x6 = xmm6;
5957 const XMMRegister x7 = xmm7;
5958
5959 const Register tmp = r11;
5960
5961 BLOCK_COMMENT("Entry:");
5962 __ enter(); // required for proper stackwalking of RuntimeStub frame
5963
5964 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5965
5966 __ leave(); // required for proper stackwalking of RuntimeStub frame
5967 __ ret(0);
5968
5969 return start;
5970
5971 }
5972
5973 address generate_libmLog() {
5974 StubCodeMark mark(this, "StubRoutines", "libmLog");
5975
5976 address start = __ pc();
5977
5978 const XMMRegister x0 = xmm0;
5979 const XMMRegister x1 = xmm1;
5980 const XMMRegister x2 = xmm2;
5981 const XMMRegister x3 = xmm3;
5982
5983 const XMMRegister x4 = xmm4;
5984 const XMMRegister x5 = xmm5;
5985 const XMMRegister x6 = xmm6;
5986 const XMMRegister x7 = xmm7;
5987
5988 const Register tmp1 = r11;
5989 const Register tmp2 = r8;
5990
5991 BLOCK_COMMENT("Entry:");
5992 __ enter(); // required for proper stackwalking of RuntimeStub frame
5993
5994 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2);
5995
5996 __ leave(); // required for proper stackwalking of RuntimeStub frame
5997 __ ret(0);
5998
5999 return start;
6000
6001 }
6002
6003 address generate_libmLog10() {
6004 StubCodeMark mark(this, "StubRoutines", "libmLog10");
6005
6006 address start = __ pc();
6007
6008 const XMMRegister x0 = xmm0;
6009 const XMMRegister x1 = xmm1;
6010 const XMMRegister x2 = xmm2;
6011 const XMMRegister x3 = xmm3;
6012
6013 const XMMRegister x4 = xmm4;
6014 const XMMRegister x5 = xmm5;
6015 const XMMRegister x6 = xmm6;
6016 const XMMRegister x7 = xmm7;
6017
6018 const Register tmp = r11;
6019
6020 BLOCK_COMMENT("Entry:");
6021 __ enter(); // required for proper stackwalking of RuntimeStub frame
6022
6023 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
6024
6025 __ leave(); // required for proper stackwalking of RuntimeStub frame
6026 __ ret(0);
6027
6028 return start;
6029
6030 }
6031
6032 address generate_libmPow() {
6033 StubCodeMark mark(this, "StubRoutines", "libmPow");
6034
6035 address start = __ pc();
6036
6037 const XMMRegister x0 = xmm0;
6038 const XMMRegister x1 = xmm1;
6039 const XMMRegister x2 = xmm2;
6040 const XMMRegister x3 = xmm3;
6041
6042 const XMMRegister x4 = xmm4;
6043 const XMMRegister x5 = xmm5;
6044 const XMMRegister x6 = xmm6;
6045 const XMMRegister x7 = xmm7;
6046
6047 const Register tmp1 = r8;
6048 const Register tmp2 = r9;
6049 const Register tmp3 = r10;
6050 const Register tmp4 = r11;
6051
6052 BLOCK_COMMENT("Entry:");
6053 __ enter(); // required for proper stackwalking of RuntimeStub frame
6054
6055 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
6056
6057 __ leave(); // required for proper stackwalking of RuntimeStub frame
6058 __ ret(0);
6059
6060 return start;
6061
6062 }
6063
6064 address generate_libmSin() {
6065 StubCodeMark mark(this, "StubRoutines", "libmSin");
6066
6067 address start = __ pc();
6068
6069 const XMMRegister x0 = xmm0;
6070 const XMMRegister x1 = xmm1;
6071 const XMMRegister x2 = xmm2;
6072 const XMMRegister x3 = xmm3;
6073
6074 const XMMRegister x4 = xmm4;
6075 const XMMRegister x5 = xmm5;
6076 const XMMRegister x6 = xmm6;
6077 const XMMRegister x7 = xmm7;
6078
6079 const Register tmp1 = r8;
6080 const Register tmp2 = r9;
6081 const Register tmp3 = r10;
6082 const Register tmp4 = r11;
6083
6084 BLOCK_COMMENT("Entry:");
6085 __ enter(); // required for proper stackwalking of RuntimeStub frame
6086
6087 #ifdef _WIN64
6088 __ push(rsi);
6089 __ push(rdi);
6090 #endif
6091 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
6092
6093 #ifdef _WIN64
6094 __ pop(rdi);
6095 __ pop(rsi);
6096 #endif
6097
6098 __ leave(); // required for proper stackwalking of RuntimeStub frame
6099 __ ret(0);
6100
6101 return start;
6102
6103 }
6104
6105 address generate_libmCos() {
6106 StubCodeMark mark(this, "StubRoutines", "libmCos");
6107
6108 address start = __ pc();
6109
6110 const XMMRegister x0 = xmm0;
6111 const XMMRegister x1 = xmm1;
6112 const XMMRegister x2 = xmm2;
6113 const XMMRegister x3 = xmm3;
6114
6115 const XMMRegister x4 = xmm4;
6116 const XMMRegister x5 = xmm5;
6117 const XMMRegister x6 = xmm6;
6118 const XMMRegister x7 = xmm7;
6119
6120 const Register tmp1 = r8;
6121 const Register tmp2 = r9;
6122 const Register tmp3 = r10;
6123 const Register tmp4 = r11;
6124
6125 BLOCK_COMMENT("Entry:");
6126 __ enter(); // required for proper stackwalking of RuntimeStub frame
6127
6128 #ifdef _WIN64
6129 __ push(rsi);
6130 __ push(rdi);
6131 #endif
6132 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
6133
6134 #ifdef _WIN64
6135 __ pop(rdi);
6136 __ pop(rsi);
6137 #endif
6138
6139 __ leave(); // required for proper stackwalking of RuntimeStub frame
6140 __ ret(0);
6141
6142 return start;
6143
6144 }
6145
6146 address generate_libmTan() {
6147 StubCodeMark mark(this, "StubRoutines", "libmTan");
6148
6149 address start = __ pc();
6150
6151 const XMMRegister x0 = xmm0;
6152 const XMMRegister x1 = xmm1;
6153 const XMMRegister x2 = xmm2;
6154 const XMMRegister x3 = xmm3;
6155
6156 const XMMRegister x4 = xmm4;
6157 const XMMRegister x5 = xmm5;
6158 const XMMRegister x6 = xmm6;
6159 const XMMRegister x7 = xmm7;
6160
6161 const Register tmp1 = r8;
6162 const Register tmp2 = r9;
6163 const Register tmp3 = r10;
6164 const Register tmp4 = r11;
6165
6166 BLOCK_COMMENT("Entry:");
6167 __ enter(); // required for proper stackwalking of RuntimeStub frame
6168
6169 #ifdef _WIN64
6170 __ push(rsi);
6171 __ push(rdi);
6172 #endif
6173 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
6174
6175 #ifdef _WIN64
6176 __ pop(rdi);
6177 __ pop(rsi);
6178 #endif
6179
6180 __ leave(); // required for proper stackwalking of RuntimeStub frame
6181 __ ret(0);
6182
6183 return start;
6184
6185 }
6186
6187 #undef __
6188 #define __ masm->
6189
6190 // Continuation point for throwing of implicit exceptions that are
6191 // not handled in the current activation. Fabricates an exception
6192 // oop and initiates normal exception dispatching in this
6193 // frame. Since we need to preserve callee-saved values (currently
6194 // only for C2, but done for C1 as well) we need a callee-saved oop
6195 // map and therefore have to make these stubs into RuntimeStubs
6196 // rather than BufferBlobs. If the compiler needs all registers to
6197 // be preserved between the fault point and the exception handler
6198 // then it must assume responsibility for that in
6199 // AbstractCompiler::continuation_for_implicit_null_exception or
6200 // continuation_for_implicit_division_by_zero_exception. All other
6201 // implicit exceptions (e.g., NullPointerException or
6202 // AbstractMethodError on entry) are either at call sites or
6203 // otherwise assume that stack unwinding will be initiated, so
6204 // caller saved registers were assumed volatile in the compiler.
6205 address generate_throw_exception(const char* name,
6206 address runtime_entry,
6207 Register arg1 = noreg,
6208 Register arg2 = noreg) {
6209 // Information about frame layout at time of blocking runtime call.
6210 // Note that we only have to preserve callee-saved registers since
6211 // the compilers are responsible for supplying a continuation point
6212 // if they expect all registers to be preserved.
6213 enum layout {
6214 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
6215 rbp_off2,
6216 return_off,
6217 return_off2,
6218 framesize // inclusive of return address
6219 };
6220
6221 int insts_size = 512;
6222 int locs_size = 64;
6223
6224 CodeBuffer code(name, insts_size, locs_size);
6225 OopMapSet* oop_maps = new OopMapSet();
6226 MacroAssembler* masm = new MacroAssembler(&code);
6227
6228 address start = __ pc();
6229
6230 // This is an inlined and slightly modified version of call_VM
6231 // which has the ability to fetch the return PC out of
6232 // thread-local storage and also sets up last_Java_sp slightly
6233 // differently than the real call_VM
6234
6235 __ enter(); // required for proper stackwalking of RuntimeStub frame
6236
6237 assert(is_even(framesize/2), "sp not 16-byte aligned");
6238
6239 // return address and rbp are already in place
6240 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
6241
6242 int frame_complete = __ pc() - start;
6243
6244 // Set up last_Java_sp and last_Java_fp
6245 address the_pc = __ pc();
6246 __ set_last_Java_frame(rsp, rbp, the_pc);
6247 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
6248
6249 // Call runtime
6250 if (arg1 != noreg) {
6251 assert(arg2 != c_rarg1, "clobbered");
6252 __ movptr(c_rarg1, arg1);
6253 }
6254 if (arg2 != noreg) {
6255 __ movptr(c_rarg2, arg2);
6256 }
6257 __ movptr(c_rarg0, r15_thread);
6258 BLOCK_COMMENT("call runtime_entry");
6259 __ call(RuntimeAddress(runtime_entry));
6260
6261 // Generate oop map
6262 OopMap* map = new OopMap(framesize, 0);
6263
6264 oop_maps->add_gc_map(the_pc - start, map);
6265
6266 __ reset_last_Java_frame(true);
6267
6268 __ leave(); // required for proper stackwalking of RuntimeStub frame
6269
6270 // check for pending exceptions
6271 #ifdef ASSERT
6272 Label L;
6273 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
6274 (int32_t) NULL_WORD);
6275 __ jcc(Assembler::notEqual, L);
6276 __ should_not_reach_here();
6277 __ bind(L);
6278 #endif // ASSERT
6279 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
6280
6281
6282 // codeBlob framesize is in words (not VMRegImpl::slot_size)
6283 RuntimeStub* stub =
6284 RuntimeStub::new_runtime_stub(name,
6285 &code,
6286 frame_complete,
6287 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
6288 oop_maps, false);
6289 return stub->entry_point();
6290 }
6291
6292 void create_control_words() {
6293 // Round to nearest, 53-bit mode, exceptions masked
6294 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
6295 // Round to zero, 53-bit mode, exception mased
6296 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
6297 // Round to nearest, 24-bit mode, exceptions masked
6298 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
6299 // Round to nearest, 64-bit mode, exceptions masked
6300 StubRoutines::_mxcsr_std = 0x1F80;
6301 // Note: the following two constants are 80-bit values
6302 // layout is critical for correct loading by FPU.
6303 // Bias for strict fp multiply/divide
6304 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
6305 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
6306 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
6307 // Un-Bias for strict fp multiply/divide
6308 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
6309 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
6310 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
6311 }
6312
6313 // Initialization
6314 void generate_initial() {
6315 // Generates all stubs and initializes the entry points
6316
6317 // This platform-specific settings are needed by generate_call_stub()
6318 create_control_words();
6319
6320 // entry points that exist in all platforms Note: This is code
6321 // that could be shared among different platforms - however the
6322 // benefit seems to be smaller than the disadvantage of having a
6323 // much more complicated generator structure. See also comment in
6324 // stubRoutines.hpp.
6325
6326 StubRoutines::_forward_exception_entry = generate_forward_exception();
6327
6328 StubRoutines::_call_stub_entry =
6329 generate_call_stub(StubRoutines::_call_stub_return_address);
6330
6331 // is referenced by megamorphic call
6332 StubRoutines::_catch_exception_entry = generate_catch_exception();
6333
6334 // atomic calls
6335 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
6336 StubRoutines::_atomic_xchg_long_entry = generate_atomic_xchg_long();
6337 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
6338 StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
6339 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
6340 StubRoutines::_atomic_add_entry = generate_atomic_add();
6341 StubRoutines::_atomic_add_long_entry = generate_atomic_add_long();
6342 StubRoutines::_fence_entry = generate_orderaccess_fence();
6343
6344 // platform dependent
6345 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
6346 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
6347
6348 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
6349
6350 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
6351 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
6352 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
6353 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
6354
6355 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6356 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
6357 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6358 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
6359
6360 // Build this early so it's available for the interpreter.
6361 StubRoutines::_throw_StackOverflowError_entry =
6362 generate_throw_exception("StackOverflowError throw_exception",
6363 CAST_FROM_FN_PTR(address,
6364 SharedRuntime::
6365 throw_StackOverflowError));
6366 StubRoutines::_throw_delayed_StackOverflowError_entry =
6367 generate_throw_exception("delayed StackOverflowError throw_exception",
6368 CAST_FROM_FN_PTR(address,
6369 SharedRuntime::
6370 throw_delayed_StackOverflowError));
6371 if (UseCRC32Intrinsics) {
6372 // set table address before stub generation which use it
6373 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
6374 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
6375 }
6376
6377 if (UseCRC32CIntrinsics) {
6378 bool supports_clmul = VM_Version::supports_clmul();
6379 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
6380 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
6381 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
6382 }
6383 if (UseLibmIntrinsic && InlineIntrinsics) {
6384 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
6385 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
6386 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6387 StubRoutines::x86::_ONEHALF_adr = (address)StubRoutines::x86::_ONEHALF;
6388 StubRoutines::x86::_P_2_adr = (address)StubRoutines::x86::_P_2;
6389 StubRoutines::x86::_SC_4_adr = (address)StubRoutines::x86::_SC_4;
6390 StubRoutines::x86::_Ctable_adr = (address)StubRoutines::x86::_Ctable;
6391 StubRoutines::x86::_SC_2_adr = (address)StubRoutines::x86::_SC_2;
6392 StubRoutines::x86::_SC_3_adr = (address)StubRoutines::x86::_SC_3;
6393 StubRoutines::x86::_SC_1_adr = (address)StubRoutines::x86::_SC_1;
6394 StubRoutines::x86::_PI_INV_TABLE_adr = (address)StubRoutines::x86::_PI_INV_TABLE;
6395 StubRoutines::x86::_PI_4_adr = (address)StubRoutines::x86::_PI_4;
6396 StubRoutines::x86::_PI32INV_adr = (address)StubRoutines::x86::_PI32INV;
6397 StubRoutines::x86::_SIGN_MASK_adr = (address)StubRoutines::x86::_SIGN_MASK;
6398 StubRoutines::x86::_P_1_adr = (address)StubRoutines::x86::_P_1;
6399 StubRoutines::x86::_P_3_adr = (address)StubRoutines::x86::_P_3;
6400 StubRoutines::x86::_NEG_ZERO_adr = (address)StubRoutines::x86::_NEG_ZERO;
6401 }
6402 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
6403 StubRoutines::_dexp = generate_libmExp();
6404 }
6405 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
6406 StubRoutines::_dlog = generate_libmLog();
6407 }
6408 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
6409 StubRoutines::_dlog10 = generate_libmLog10();
6410 }
6411 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
6412 StubRoutines::_dpow = generate_libmPow();
6413 }
6414 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
6415 StubRoutines::_dsin = generate_libmSin();
6416 }
6417 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
6418 StubRoutines::_dcos = generate_libmCos();
6419 }
6420 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6421 StubRoutines::_dtan = generate_libmTan();
6422 }
6423 }
6424
6425 // Safefetch stubs.
6426 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
6427 &StubRoutines::_safefetch32_fault_pc,
6428 &StubRoutines::_safefetch32_continuation_pc);
6429 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
6430 &StubRoutines::_safefetchN_fault_pc,
6431 &StubRoutines::_safefetchN_continuation_pc);
6432 }
6433
6434 void generate_all() {
6435 // Generates all stubs and initializes the entry points
6436
6437 // These entry points require SharedInfo::stack0 to be set up in
6438 // non-core builds and need to be relocatable, so they each
6439 // fabricate a RuntimeStub internally.
6440 StubRoutines::_throw_AbstractMethodError_entry =
6441 generate_throw_exception("AbstractMethodError throw_exception",
6442 CAST_FROM_FN_PTR(address,
6443 SharedRuntime::
6444 throw_AbstractMethodError));
6445
6446 StubRoutines::_throw_IncompatibleClassChangeError_entry =
6447 generate_throw_exception("IncompatibleClassChangeError throw_exception",
6448 CAST_FROM_FN_PTR(address,
6449 SharedRuntime::
6450 throw_IncompatibleClassChangeError));
6451
6452 StubRoutines::_throw_NullPointerException_at_call_entry =
6453 generate_throw_exception("NullPointerException at call throw_exception",
6454 CAST_FROM_FN_PTR(address,
6455 SharedRuntime::
6456 throw_NullPointerException_at_call));
6457
6458 // entry points that are platform specific
6459 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6460 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
6461 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6462 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
6463 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
6464 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
6465 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
6466
6467 // support for verify_oop (must happen after universe_init)
6468 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
6469
6470 // data cache line writeback
6471 StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
6472 StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
6473
6474 // arraycopy stubs used by compilers
6475 generate_arraycopy_stubs();
6476
6477 // don't bother generating these AES intrinsic stubs unless global flag is set
6478 if (UseAESIntrinsics) {
6479 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
6480 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
6481 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
6482 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
6483 if (VM_Version::supports_avx512_vaes() && VM_Version::supports_avx512vl() && VM_Version::supports_avx512dq() ) {
6484 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptVectorAESCrypt();
6485 StubRoutines::_electronicCodeBook_encryptAESCrypt = generate_electronicCodeBook_encryptAESCrypt();
6486 StubRoutines::_electronicCodeBook_decryptAESCrypt = generate_electronicCodeBook_decryptAESCrypt();
6487 } else {
6488 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
6489 }
6490 }
6491 if (UseAESCTRIntrinsics) {
6492 if (VM_Version::supports_avx512_vaes() && VM_Version::supports_avx512bw() && VM_Version::supports_avx512vl()) {
6493 StubRoutines::x86::_counter_mask_addr = counter_mask_addr();
6494 StubRoutines::_counterMode_AESCrypt = generate_counterMode_VectorAESCrypt();
6495 } else {
6496 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
6497 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
6498 }
6499 }
6500
6501 if (UseSHA1Intrinsics) {
6502 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
6503 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
6504 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
6505 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
6506 }
6507 if (UseSHA256Intrinsics) {
6508 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
6509 char* dst = (char*)StubRoutines::x86::_k256_W;
6510 char* src = (char*)StubRoutines::x86::_k256;
6511 for (int ii = 0; ii < 16; ++ii) {
6512 memcpy(dst + 32 * ii, src + 16 * ii, 16);
6513 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
6514 }
6515 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
6516 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
6517 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
6518 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
6519 }
6520 if (UseSHA512Intrinsics) {
6521 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
6522 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
6523 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
6524 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
6525 }
6526
6527 // Generate GHASH intrinsics code
6528 if (UseGHASHIntrinsics) {
6529 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
6530 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
6531 if (VM_Version::supports_avx()) {
6532 StubRoutines::x86::_ghash_shuffmask_addr = ghash_shufflemask_addr();
6533 StubRoutines::x86::_ghash_poly_addr = ghash_polynomial_addr();
6534 StubRoutines::_ghash_processBlocks = generate_avx_ghash_processBlocks();
6535 } else {
6536 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
6537 }
6538 }
6539
6540 if (UseBASE64Intrinsics) {
6541 StubRoutines::x86::_and_mask = base64_and_mask_addr();
6542 StubRoutines::x86::_bswap_mask = base64_bswap_mask_addr();
6543 StubRoutines::x86::_base64_charset = base64_charset_addr();
6544 StubRoutines::x86::_url_charset = base64url_charset_addr();
6545 StubRoutines::x86::_gather_mask = base64_gather_mask_addr();
6546 StubRoutines::x86::_left_shift_mask = base64_left_shift_mask_addr();
6547 StubRoutines::x86::_right_shift_mask = base64_right_shift_mask_addr();
6548 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
6549 }
6550
6551 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
6552 if (bs_nm != NULL) {
6553 StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
6554 }
6555 #ifdef COMPILER2
6556 if (UseMultiplyToLenIntrinsic) {
6557 StubRoutines::_multiplyToLen = generate_multiplyToLen();
6558 }
6559 if (UseSquareToLenIntrinsic) {
6560 StubRoutines::_squareToLen = generate_squareToLen();
6561 }
6562 if (UseMulAddIntrinsic) {
6563 StubRoutines::_mulAdd = generate_mulAdd();
6564 }
6565 if (VM_Version::supports_avx512_vbmi2()) {
6566 StubRoutines::_bigIntegerRightShiftWorker = generate_bigIntegerRightShift();
6567 StubRoutines::_bigIntegerLeftShiftWorker = generate_bigIntegerLeftShift();
6568 }
6569 if (UseMontgomeryMultiplyIntrinsic) {
6570 StubRoutines::_montgomeryMultiply
6571 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
6572 }
6573 if (UseMontgomerySquareIntrinsic) {
6574 StubRoutines::_montgomerySquare
6575 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
6576 }
6577 #endif // COMPILER2
6578
6579 if (UseVectorizedMismatchIntrinsic) {
6580 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
6581 }
6582 }
6583
6584 public:
6585 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
6586 if (all) {
6587 generate_all();
6588 } else {
6589 generate_initial();
6590 }
6591 }
6592 }; // end class declaration
6593
6594 #define UCM_TABLE_MAX_ENTRIES 16
6595 void StubGenerator_generate(CodeBuffer* code, bool all) {
6596 if (UnsafeCopyMemory::_table == NULL) {
6597 UnsafeCopyMemory::create_table(UCM_TABLE_MAX_ENTRIES);
6598 }
6599 StubGenerator g(code, all);
6600 }