1 /*
2 * Copyright (c) 2003, 2018, 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 "interpreter/interpreter.hpp"
32 #include "nativeInst_x86.hpp"
33 #include "oops/instanceOop.hpp"
34 #include "oops/method.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/methodHandles.hpp"
38 #include "runtime/frame.inline.hpp"
39 #include "runtime/handles.inline.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/stubCodeGenerator.hpp"
42 #include "runtime/stubRoutines.hpp"
43 #include "runtime/thread.inline.hpp"
44 #ifdef COMPILER2
45 #include "opto/runtime.hpp"
46 #endif
47 #if INCLUDE_ZGC
48 #include "gc/z/zThreadLocalData.hpp"
49 #endif
50
51 // Declaration and definition of StubGenerator (no .hpp file).
52 // For a more detailed description of the stub routine structure
53 // see the comment in stubRoutines.hpp
54
55 #define __ _masm->
56 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
57 #define a__ ((Assembler*)_masm)->
58
59 #ifdef PRODUCT
60 #define BLOCK_COMMENT(str) /* nothing */
61 #else
62 #define BLOCK_COMMENT(str) __ block_comment(str)
63 #endif
64
65 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
66 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
67
68 // Stub Code definitions
69
70 class StubGenerator: public StubCodeGenerator {
71 private:
72
73 #ifdef PRODUCT
74 #define inc_counter_np(counter) ((void)0)
75 #else
76 void inc_counter_np_(int& counter) {
77 // This can destroy rscratch1 if counter is far from the code cache
78 __ incrementl(ExternalAddress((address)&counter));
79 }
80 #define inc_counter_np(counter) \
81 BLOCK_COMMENT("inc_counter " #counter); \
82 inc_counter_np_(counter);
83 #endif
84
85 // Call stubs are used to call Java from C
86 //
87 // Linux Arguments:
88 // c_rarg0: call wrapper address address
89 // c_rarg1: result address
90 // c_rarg2: result type BasicType
91 // c_rarg3: method Method*
92 // c_rarg4: (interpreter) entry point address
93 // c_rarg5: parameters intptr_t*
94 // 16(rbp): parameter size (in words) int
95 // 24(rbp): thread Thread*
96 //
97 // [ return_from_Java ] <--- rsp
98 // [ argument word n ]
99 // ...
100 // -12 [ argument word 1 ]
101 // -11 [ saved r15 ] <--- rsp_after_call
102 // -10 [ saved r14 ]
103 // -9 [ saved r13 ]
104 // -8 [ saved r12 ]
105 // -7 [ saved rbx ]
106 // -6 [ call wrapper ]
107 // -5 [ result ]
108 // -4 [ result type ]
109 // -3 [ method ]
110 // -2 [ entry point ]
111 // -1 [ parameters ]
112 // 0 [ saved rbp ] <--- rbp
113 // 1 [ return address ]
114 // 2 [ parameter size ]
115 // 3 [ thread ]
116 //
117 // Windows Arguments:
118 // c_rarg0: call wrapper address address
119 // c_rarg1: result address
120 // c_rarg2: result type BasicType
121 // c_rarg3: method Method*
122 // 48(rbp): (interpreter) entry point address
123 // 56(rbp): parameters intptr_t*
124 // 64(rbp): parameter size (in words) int
125 // 72(rbp): thread Thread*
126 //
127 // [ return_from_Java ] <--- rsp
128 // [ argument word n ]
129 // ...
130 // -60 [ argument word 1 ]
131 // -59 [ saved xmm31 ] <--- rsp after_call
132 // [ saved xmm16-xmm30 ] (EVEX enabled, else the space is blank)
133 // -27 [ saved xmm15 ]
134 // [ saved xmm7-xmm14 ]
135 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
136 // -7 [ saved r15 ]
137 // -6 [ saved r14 ]
138 // -5 [ saved r13 ]
139 // -4 [ saved r12 ]
140 // -3 [ saved rdi ]
141 // -2 [ saved rsi ]
142 // -1 [ saved rbx ]
143 // 0 [ saved rbp ] <--- rbp
144 // 1 [ return address ]
145 // 2 [ call wrapper ]
146 // 3 [ result ]
147 // 4 [ result type ]
148 // 5 [ method ]
149 // 6 [ entry point ]
150 // 7 [ parameters ]
151 // 8 [ parameter size ]
152 // 9 [ thread ]
153 //
154 // Windows reserves the callers stack space for arguments 1-4.
155 // We spill c_rarg0-c_rarg3 to this space.
156
157 // Call stub stack layout word offsets from rbp
158 enum call_stub_layout {
159 #ifdef _WIN64
160 xmm_save_first = 6, // save from xmm6
161 xmm_save_last = 31, // to xmm31
162 xmm_save_base = -9,
163 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
164 r15_off = -7,
165 r14_off = -6,
166 r13_off = -5,
167 r12_off = -4,
168 rdi_off = -3,
169 rsi_off = -2,
170 rbx_off = -1,
171 rbp_off = 0,
172 retaddr_off = 1,
173 call_wrapper_off = 2,
174 result_off = 3,
175 result_type_off = 4,
176 method_off = 5,
177 entry_point_off = 6,
178 parameters_off = 7,
179 parameter_size_off = 8,
180 thread_off = 9
181 #else
182 rsp_after_call_off = -12,
183 mxcsr_off = rsp_after_call_off,
184 r15_off = -11,
185 r14_off = -10,
186 r13_off = -9,
187 r12_off = -8,
188 rbx_off = -7,
189 call_wrapper_off = -6,
190 result_off = -5,
191 result_type_off = -4,
192 method_off = -3,
193 entry_point_off = -2,
194 parameters_off = -1,
195 rbp_off = 0,
196 retaddr_off = 1,
197 parameter_size_off = 2,
198 thread_off = 3
199 #endif
200 };
201
202 #ifdef _WIN64
203 Address xmm_save(int reg) {
204 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
205 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
206 }
207 #endif
208
209 address generate_call_stub(address& return_address) {
210 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
211 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
212 "adjust this code");
213 StubCodeMark mark(this, "StubRoutines", "call_stub");
214 address start = __ pc();
215
216 // same as in generate_catch_exception()!
217 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
218
219 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
220 const Address result (rbp, result_off * wordSize);
221 const Address result_type (rbp, result_type_off * wordSize);
222 const Address method (rbp, method_off * wordSize);
223 const Address entry_point (rbp, entry_point_off * wordSize);
224 const Address parameters (rbp, parameters_off * wordSize);
225 const Address parameter_size(rbp, parameter_size_off * wordSize);
226
227 // same as in generate_catch_exception()!
228 const Address thread (rbp, thread_off * wordSize);
229
230 const Address r15_save(rbp, r15_off * wordSize);
231 const Address r14_save(rbp, r14_off * wordSize);
232 const Address r13_save(rbp, r13_off * wordSize);
233 const Address r12_save(rbp, r12_off * wordSize);
234 const Address rbx_save(rbp, rbx_off * wordSize);
235
236 // stub code
237 __ enter();
238 __ subptr(rsp, -rsp_after_call_off * wordSize);
239
240 // save register parameters
241 #ifndef _WIN64
242 __ movptr(parameters, c_rarg5); // parameters
243 __ movptr(entry_point, c_rarg4); // entry_point
244 #endif
245
246 __ movptr(method, c_rarg3); // method
247 __ movl(result_type, c_rarg2); // result type
248 __ movptr(result, c_rarg1); // result
249 __ movptr(call_wrapper, c_rarg0); // call wrapper
250
251 // save regs belonging to calling function
252 __ movptr(rbx_save, rbx);
253 __ movptr(r12_save, r12);
254 __ movptr(r13_save, r13);
255 __ movptr(r14_save, r14);
256 __ movptr(r15_save, r15);
257 if (UseAVX > 2) {
258 __ movl(rbx, 0xffff);
259 __ kmovwl(k1, rbx);
260 }
261 #ifdef _WIN64
262 int last_reg = 15;
263 if (UseAVX > 2) {
264 last_reg = 31;
265 }
266 if (VM_Version::supports_evex()) {
267 for (int i = xmm_save_first; i <= last_reg; i++) {
268 __ vextractf32x4(xmm_save(i), as_XMMRegister(i), 0);
269 }
270 } else {
271 for (int i = xmm_save_first; i <= last_reg; i++) {
272 __ movdqu(xmm_save(i), as_XMMRegister(i));
273 }
274 }
275
276 const Address rdi_save(rbp, rdi_off * wordSize);
277 const Address rsi_save(rbp, rsi_off * wordSize);
278
279 __ movptr(rsi_save, rsi);
280 __ movptr(rdi_save, rdi);
281 #else
282 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
283 {
284 Label skip_ldmx;
285 __ stmxcsr(mxcsr_save);
286 __ movl(rax, mxcsr_save);
287 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
288 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
289 __ cmp32(rax, mxcsr_std);
290 __ jcc(Assembler::equal, skip_ldmx);
291 __ ldmxcsr(mxcsr_std);
292 __ bind(skip_ldmx);
293 }
294 #endif
295
296 // Load up thread register
297 __ movptr(r15_thread, thread);
298 __ reinit_heapbase();
299
300 #ifdef ASSERT
301 // make sure we have no pending exceptions
302 {
303 Label L;
304 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
305 __ jcc(Assembler::equal, L);
306 __ stop("StubRoutines::call_stub: entered with pending exception");
307 __ bind(L);
308 }
309 #endif
310
311 // pass parameters if any
312 BLOCK_COMMENT("pass parameters if any");
313 Label parameters_done;
314 __ movl(c_rarg3, parameter_size);
315 __ testl(c_rarg3, c_rarg3);
316 __ jcc(Assembler::zero, parameters_done);
317
318 Label loop;
319 __ movptr(c_rarg2, parameters); // parameter pointer
320 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
321 __ BIND(loop);
322 __ movptr(rax, Address(c_rarg2, 0));// get parameter
323 __ addptr(c_rarg2, wordSize); // advance to next parameter
324 __ decrementl(c_rarg1); // decrement counter
325 __ push(rax); // pass parameter
326 __ jcc(Assembler::notZero, loop);
327
328 // call Java function
329 __ BIND(parameters_done);
330 __ movptr(rbx, method); // get Method*
331 __ movptr(c_rarg1, entry_point); // get entry_point
332 __ mov(r13, rsp); // set sender sp
333 BLOCK_COMMENT("call Java function");
334 __ call(c_rarg1);
335
336 BLOCK_COMMENT("call_stub_return_address:");
337 return_address = __ pc();
338
339 // store result depending on type (everything that is not
340 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
341 __ movptr(c_rarg0, result);
342 Label is_long, is_float, is_double, exit;
343 __ movl(c_rarg1, result_type);
344 __ cmpl(c_rarg1, T_OBJECT);
345 __ jcc(Assembler::equal, is_long);
346 __ cmpl(c_rarg1, T_LONG);
347 __ jcc(Assembler::equal, is_long);
348 __ cmpl(c_rarg1, T_FLOAT);
349 __ jcc(Assembler::equal, is_float);
350 __ cmpl(c_rarg1, T_DOUBLE);
351 __ jcc(Assembler::equal, is_double);
352
353 // handle T_INT case
354 __ movl(Address(c_rarg0, 0), rax);
355
356 __ BIND(exit);
357
358 // pop parameters
359 __ lea(rsp, rsp_after_call);
360
361 #ifdef ASSERT
362 // verify that threads correspond
363 {
364 Label L1, L2, L3;
365 __ cmpptr(r15_thread, thread);
366 __ jcc(Assembler::equal, L1);
367 __ stop("StubRoutines::call_stub: r15_thread is corrupted");
368 __ bind(L1);
369 __ get_thread(rbx);
370 __ cmpptr(r15_thread, thread);
371 __ jcc(Assembler::equal, L2);
372 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
373 __ bind(L2);
374 __ cmpptr(r15_thread, rbx);
375 __ jcc(Assembler::equal, L3);
376 __ stop("StubRoutines::call_stub: threads must correspond");
377 __ bind(L3);
378 }
379 #endif
380
381 // restore regs belonging to calling function
382 #ifdef _WIN64
383 // emit the restores for xmm regs
384 if (VM_Version::supports_evex()) {
385 for (int i = xmm_save_first; i <= last_reg; i++) {
386 __ vinsertf32x4(as_XMMRegister(i), as_XMMRegister(i), xmm_save(i), 0);
387 }
388 } else {
389 for (int i = xmm_save_first; i <= last_reg; i++) {
390 __ movdqu(as_XMMRegister(i), xmm_save(i));
391 }
392 }
393 #endif
394 __ movptr(r15, r15_save);
395 __ movptr(r14, r14_save);
396 __ movptr(r13, r13_save);
397 __ movptr(r12, r12_save);
398 __ movptr(rbx, rbx_save);
399
400 #ifdef _WIN64
401 __ movptr(rdi, rdi_save);
402 __ movptr(rsi, rsi_save);
403 #else
404 __ ldmxcsr(mxcsr_save);
405 #endif
406
407 // restore rsp
408 __ addptr(rsp, -rsp_after_call_off * wordSize);
409
410 // return
411 __ vzeroupper();
412 __ pop(rbp);
413 __ ret(0);
414
415 // handle return types different from T_INT
416 __ BIND(is_long);
417 __ movq(Address(c_rarg0, 0), rax);
418 __ jmp(exit);
419
420 __ BIND(is_float);
421 __ movflt(Address(c_rarg0, 0), xmm0);
422 __ jmp(exit);
423
424 __ BIND(is_double);
425 __ movdbl(Address(c_rarg0, 0), xmm0);
426 __ jmp(exit);
427
428 return start;
429 }
430
431 // Return point for a Java call if there's an exception thrown in
432 // Java code. The exception is caught and transformed into a
433 // pending exception stored in JavaThread that can be tested from
434 // within the VM.
435 //
436 // Note: Usually the parameters are removed by the callee. In case
437 // of an exception crossing an activation frame boundary, that is
438 // not the case if the callee is compiled code => need to setup the
439 // rsp.
440 //
441 // rax: exception oop
442
443 address generate_catch_exception() {
444 StubCodeMark mark(this, "StubRoutines", "catch_exception");
445 address start = __ pc();
446
447 // same as in generate_call_stub():
448 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
449 const Address thread (rbp, thread_off * wordSize);
450
451 #ifdef ASSERT
452 // verify that threads correspond
453 {
454 Label L1, L2, L3;
455 __ cmpptr(r15_thread, thread);
456 __ jcc(Assembler::equal, L1);
457 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
458 __ bind(L1);
459 __ get_thread(rbx);
460 __ cmpptr(r15_thread, thread);
461 __ jcc(Assembler::equal, L2);
462 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
463 __ bind(L2);
464 __ cmpptr(r15_thread, rbx);
465 __ jcc(Assembler::equal, L3);
466 __ stop("StubRoutines::catch_exception: threads must correspond");
467 __ bind(L3);
468 }
469 #endif
470
471 // set pending exception
472 __ verify_oop(rax);
473
474 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
475 __ lea(rscratch1, ExternalAddress((address)__FILE__));
476 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
477 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
478
479 // complete return to VM
480 assert(StubRoutines::_call_stub_return_address != NULL,
481 "_call_stub_return_address must have been generated before");
482 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
483
484 return start;
485 }
486
487 // Continuation point for runtime calls returning with a pending
488 // exception. The pending exception check happened in the runtime
489 // or native call stub. The pending exception in Thread is
490 // converted into a Java-level exception.
491 //
492 // Contract with Java-level exception handlers:
493 // rax: exception
494 // rdx: throwing pc
495 //
496 // NOTE: At entry of this stub, exception-pc must be on stack !!
497
498 address generate_forward_exception() {
499 StubCodeMark mark(this, "StubRoutines", "forward exception");
500 address start = __ pc();
501
502 // Upon entry, the sp points to the return address returning into
503 // Java (interpreted or compiled) code; i.e., the return address
504 // becomes the throwing pc.
505 //
506 // Arguments pushed before the runtime call are still on the stack
507 // but the exception handler will reset the stack pointer ->
508 // ignore them. A potential result in registers can be ignored as
509 // well.
510
511 #ifdef ASSERT
512 // make sure this code is only executed if there is a pending exception
513 {
514 Label L;
515 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
516 __ jcc(Assembler::notEqual, L);
517 __ stop("StubRoutines::forward exception: no pending exception (1)");
518 __ bind(L);
519 }
520 #endif
521
522 // compute exception handler into rbx
523 __ movptr(c_rarg0, Address(rsp, 0));
524 BLOCK_COMMENT("call exception_handler_for_return_address");
525 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
526 SharedRuntime::exception_handler_for_return_address),
527 r15_thread, c_rarg0);
528 __ mov(rbx, rax);
529
530 // setup rax & rdx, remove return address & clear pending exception
531 __ pop(rdx);
532 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
533 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
534
535 #ifdef ASSERT
536 // make sure exception is set
537 {
538 Label L;
539 __ testptr(rax, rax);
540 __ jcc(Assembler::notEqual, L);
541 __ stop("StubRoutines::forward exception: no pending exception (2)");
542 __ bind(L);
543 }
544 #endif
545
546 // continue at exception handler (return address removed)
547 // rax: exception
548 // rbx: exception handler
549 // rdx: throwing pc
550 __ verify_oop(rax);
551 __ jmp(rbx);
552
553 return start;
554 }
555
556 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
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 // Support for intptr_t atomic::xchg_long(jlong exchange_value, volatile jlong* dest)
576 //
577 // Arguments :
578 // c_rarg0: exchange_value
579 // c_rarg1: dest
580 //
581 // Result:
582 // *dest <- ex, return (orig *dest)
583 address generate_atomic_xchg_long() {
584 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_long");
585 address start = __ pc();
586
587 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
588 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
589 __ ret(0);
590
591 return start;
592 }
593
594 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
595 // jint compare_value)
596 //
597 // Arguments :
598 // c_rarg0: exchange_value
599 // c_rarg1: dest
600 // c_rarg2: compare_value
601 //
602 // Result:
603 // if ( compare_value == *dest ) {
604 // *dest = exchange_value
605 // return compare_value;
606 // else
607 // return *dest;
608 address generate_atomic_cmpxchg() {
609 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
610 address start = __ pc();
611
612 __ movl(rax, c_rarg2);
613 if ( os::is_MP() ) __ lock();
614 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
615 __ ret(0);
616
617 return start;
618 }
619
620 // Support for int8_t atomic::atomic_cmpxchg(int8_t exchange_value, volatile int8_t* dest,
621 // int8_t compare_value)
622 //
623 // Arguments :
624 // c_rarg0: exchange_value
625 // c_rarg1: dest
626 // c_rarg2: compare_value
627 //
628 // Result:
629 // if ( compare_value == *dest ) {
630 // *dest = exchange_value
631 // return compare_value;
632 // else
633 // return *dest;
634 address generate_atomic_cmpxchg_byte() {
635 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_byte");
636 address start = __ pc();
637
638 __ movsbq(rax, c_rarg2);
639 if ( os::is_MP() ) __ lock();
640 __ cmpxchgb(c_rarg0, Address(c_rarg1, 0));
641 __ ret(0);
642
643 return start;
644 }
645
646 // Support for int64_t atomic::atomic_cmpxchg(int64_t exchange_value,
647 // volatile int64_t* dest,
648 // int64_t compare_value)
649 // Arguments :
650 // c_rarg0: exchange_value
651 // c_rarg1: dest
652 // c_rarg2: compare_value
653 //
654 // Result:
655 // if ( compare_value == *dest ) {
656 // *dest = exchange_value
657 // return compare_value;
658 // else
659 // return *dest;
660 address generate_atomic_cmpxchg_long() {
661 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
662 address start = __ pc();
663
664 __ movq(rax, c_rarg2);
665 if ( os::is_MP() ) __ lock();
666 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
667 __ ret(0);
668
669 return start;
670 }
671
672 // Support for jint atomic::add(jint add_value, volatile jint* dest)
673 //
674 // Arguments :
675 // c_rarg0: add_value
676 // c_rarg1: dest
677 //
678 // Result:
679 // *dest += add_value
680 // return *dest;
681 address generate_atomic_add() {
682 StubCodeMark mark(this, "StubRoutines", "atomic_add");
683 address start = __ pc();
684
685 __ movl(rax, c_rarg0);
686 if ( os::is_MP() ) __ lock();
687 __ xaddl(Address(c_rarg1, 0), c_rarg0);
688 __ addl(rax, c_rarg0);
689 __ ret(0);
690
691 return start;
692 }
693
694 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
695 //
696 // Arguments :
697 // c_rarg0: add_value
698 // c_rarg1: dest
699 //
700 // Result:
701 // *dest += add_value
702 // return *dest;
703 address generate_atomic_add_long() {
704 StubCodeMark mark(this, "StubRoutines", "atomic_add_long");
705 address start = __ pc();
706
707 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
708 if ( os::is_MP() ) __ lock();
709 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
710 __ addptr(rax, c_rarg0);
711 __ ret(0);
712
713 return start;
714 }
715
716 // Support for intptr_t OrderAccess::fence()
717 //
718 // Arguments :
719 //
720 // Result:
721 address generate_orderaccess_fence() {
722 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
723 address start = __ pc();
724 __ membar(Assembler::StoreLoad);
725 __ ret(0);
726
727 return start;
728 }
729
730 // Support for intptr_t get_previous_fp()
731 //
732 // This routine is used to find the previous frame pointer for the
733 // caller (current_frame_guess). This is used as part of debugging
734 // ps() is seemingly lost trying to find frames.
735 // This code assumes that caller current_frame_guess) has a frame.
736 address generate_get_previous_fp() {
737 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
738 const Address old_fp(rbp, 0);
739 const Address older_fp(rax, 0);
740 address start = __ pc();
741
742 __ enter();
743 __ movptr(rax, old_fp); // callers fp
744 __ movptr(rax, older_fp); // the frame for ps()
745 __ pop(rbp);
746 __ ret(0);
747
748 return start;
749 }
750
751 // Support for intptr_t get_previous_sp()
752 //
753 // This routine is used to find the previous stack pointer for the
754 // caller.
755 address generate_get_previous_sp() {
756 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
757 address start = __ pc();
758
759 __ movptr(rax, rsp);
760 __ addptr(rax, 8); // return address is at the top of the stack.
761 __ ret(0);
762
763 return start;
764 }
765
766 //----------------------------------------------------------------------------------------------------
767 // Support for void verify_mxcsr()
768 //
769 // This routine is used with -Xcheck:jni to verify that native
770 // JNI code does not return to Java code without restoring the
771 // MXCSR register to our expected state.
772
773 address generate_verify_mxcsr() {
774 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
775 address start = __ pc();
776
777 const Address mxcsr_save(rsp, 0);
778
779 if (CheckJNICalls) {
780 Label ok_ret;
781 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
782 __ push(rax);
783 __ subptr(rsp, wordSize); // allocate a temp location
784 __ stmxcsr(mxcsr_save);
785 __ movl(rax, mxcsr_save);
786 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
787 __ cmp32(rax, mxcsr_std);
788 __ jcc(Assembler::equal, ok_ret);
789
790 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
791
792 __ ldmxcsr(mxcsr_std);
793
794 __ bind(ok_ret);
795 __ addptr(rsp, wordSize);
796 __ pop(rax);
797 }
798
799 __ ret(0);
800
801 return start;
802 }
803
804 address generate_f2i_fixup() {
805 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
806 Address inout(rsp, 5 * wordSize); // return address + 4 saves
807
808 address start = __ pc();
809
810 Label L;
811
812 __ push(rax);
813 __ push(c_rarg3);
814 __ push(c_rarg2);
815 __ push(c_rarg1);
816
817 __ movl(rax, 0x7f800000);
818 __ xorl(c_rarg3, c_rarg3);
819 __ movl(c_rarg2, inout);
820 __ movl(c_rarg1, c_rarg2);
821 __ andl(c_rarg1, 0x7fffffff);
822 __ cmpl(rax, c_rarg1); // NaN? -> 0
823 __ jcc(Assembler::negative, L);
824 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
825 __ movl(c_rarg3, 0x80000000);
826 __ movl(rax, 0x7fffffff);
827 __ cmovl(Assembler::positive, c_rarg3, rax);
828
829 __ bind(L);
830 __ movptr(inout, c_rarg3);
831
832 __ pop(c_rarg1);
833 __ pop(c_rarg2);
834 __ pop(c_rarg3);
835 __ pop(rax);
836
837 __ ret(0);
838
839 return start;
840 }
841
842 address generate_f2l_fixup() {
843 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
844 Address inout(rsp, 5 * wordSize); // return address + 4 saves
845 address start = __ pc();
846
847 Label L;
848
849 __ push(rax);
850 __ push(c_rarg3);
851 __ push(c_rarg2);
852 __ push(c_rarg1);
853
854 __ movl(rax, 0x7f800000);
855 __ xorl(c_rarg3, c_rarg3);
856 __ movl(c_rarg2, inout);
857 __ movl(c_rarg1, c_rarg2);
858 __ andl(c_rarg1, 0x7fffffff);
859 __ cmpl(rax, c_rarg1); // NaN? -> 0
860 __ jcc(Assembler::negative, L);
861 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
862 __ mov64(c_rarg3, 0x8000000000000000);
863 __ mov64(rax, 0x7fffffffffffffff);
864 __ cmov(Assembler::positive, c_rarg3, rax);
865
866 __ bind(L);
867 __ movptr(inout, c_rarg3);
868
869 __ pop(c_rarg1);
870 __ pop(c_rarg2);
871 __ pop(c_rarg3);
872 __ pop(rax);
873
874 __ ret(0);
875
876 return start;
877 }
878
879 address generate_d2i_fixup() {
880 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
881 Address inout(rsp, 6 * wordSize); // return address + 5 saves
882
883 address start = __ pc();
884
885 Label L;
886
887 __ push(rax);
888 __ push(c_rarg3);
889 __ push(c_rarg2);
890 __ push(c_rarg1);
891 __ push(c_rarg0);
892
893 __ movl(rax, 0x7ff00000);
894 __ movq(c_rarg2, inout);
895 __ movl(c_rarg3, c_rarg2);
896 __ mov(c_rarg1, c_rarg2);
897 __ mov(c_rarg0, c_rarg2);
898 __ negl(c_rarg3);
899 __ shrptr(c_rarg1, 0x20);
900 __ orl(c_rarg3, c_rarg2);
901 __ andl(c_rarg1, 0x7fffffff);
902 __ xorl(c_rarg2, c_rarg2);
903 __ shrl(c_rarg3, 0x1f);
904 __ orl(c_rarg1, c_rarg3);
905 __ cmpl(rax, c_rarg1);
906 __ jcc(Assembler::negative, L); // NaN -> 0
907 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
908 __ movl(c_rarg2, 0x80000000);
909 __ movl(rax, 0x7fffffff);
910 __ cmov(Assembler::positive, c_rarg2, rax);
911
912 __ bind(L);
913 __ movptr(inout, c_rarg2);
914
915 __ pop(c_rarg0);
916 __ pop(c_rarg1);
917 __ pop(c_rarg2);
918 __ pop(c_rarg3);
919 __ pop(rax);
920
921 __ ret(0);
922
923 return start;
924 }
925
926 address generate_d2l_fixup() {
927 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
928 Address inout(rsp, 6 * wordSize); // return address + 5 saves
929
930 address start = __ pc();
931
932 Label L;
933
934 __ push(rax);
935 __ push(c_rarg3);
936 __ push(c_rarg2);
937 __ push(c_rarg1);
938 __ push(c_rarg0);
939
940 __ movl(rax, 0x7ff00000);
941 __ movq(c_rarg2, inout);
942 __ movl(c_rarg3, c_rarg2);
943 __ mov(c_rarg1, c_rarg2);
944 __ mov(c_rarg0, c_rarg2);
945 __ negl(c_rarg3);
946 __ shrptr(c_rarg1, 0x20);
947 __ orl(c_rarg3, c_rarg2);
948 __ andl(c_rarg1, 0x7fffffff);
949 __ xorl(c_rarg2, c_rarg2);
950 __ shrl(c_rarg3, 0x1f);
951 __ orl(c_rarg1, c_rarg3);
952 __ cmpl(rax, c_rarg1);
953 __ jcc(Assembler::negative, L); // NaN -> 0
954 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
955 __ mov64(c_rarg2, 0x8000000000000000);
956 __ mov64(rax, 0x7fffffffffffffff);
957 __ cmovq(Assembler::positive, c_rarg2, rax);
958
959 __ bind(L);
960 __ movq(inout, c_rarg2);
961
962 __ pop(c_rarg0);
963 __ pop(c_rarg1);
964 __ pop(c_rarg2);
965 __ pop(c_rarg3);
966 __ pop(rax);
967
968 __ ret(0);
969
970 return start;
971 }
972
973 address generate_fp_mask(const char *stub_name, int64_t mask) {
974 __ align(CodeEntryAlignment);
975 StubCodeMark mark(this, "StubRoutines", stub_name);
976 address start = __ pc();
977
978 __ emit_data64( mask, relocInfo::none );
979 __ emit_data64( mask, relocInfo::none );
980
981 return start;
982 }
983
984 // Non-destructive plausibility checks for oops
985 //
986 // Arguments:
987 // all args on stack!
988 //
989 // Stack after saving c_rarg3:
990 // [tos + 0]: saved c_rarg3
991 // [tos + 1]: saved c_rarg2
992 // [tos + 2]: saved r12 (several TemplateTable methods use it)
993 // [tos + 3]: saved flags
994 // [tos + 4]: return address
995 // * [tos + 5]: error message (char*)
996 // * [tos + 6]: object to verify (oop)
997 // * [tos + 7]: saved rax - saved by caller and bashed
998 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
999 // * = popped on exit
1000 address generate_verify_oop() {
1001 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1002 address start = __ pc();
1003
1004 Label exit, error;
1005
1006 __ pushf();
1007 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1008
1009 __ push(r12);
1010
1011 // save c_rarg2 and c_rarg3
1012 __ push(c_rarg2);
1013 __ push(c_rarg3);
1014
1015 enum {
1016 // After previous pushes.
1017 oop_to_verify = 6 * wordSize,
1018 saved_rax = 7 * wordSize,
1019 saved_r10 = 8 * wordSize,
1020
1021 // Before the call to MacroAssembler::debug(), see below.
1022 return_addr = 16 * wordSize,
1023 error_msg = 17 * wordSize
1024 };
1025
1026 // get object
1027 __ movptr(rax, Address(rsp, oop_to_verify));
1028
1029 // make sure object is 'reasonable'
1030 __ testptr(rax, rax);
1031 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1032
1033 #if INCLUDE_ZGC
1034 if (UseZGC) {
1035 // Check if metadata bits indicate a bad oop
1036 __ testptr(rax, Address(r15_thread, ZThreadLocalData::address_bad_mask_offset()));
1037 __ jcc(Assembler::notZero, error);
1038 }
1039 #endif
1040
1041 // Check if the oop is in the right area of memory
1042 __ movptr(c_rarg2, rax);
1043 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1044 __ andptr(c_rarg2, c_rarg3);
1045 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1046 __ cmpptr(c_rarg2, c_rarg3);
1047 __ jcc(Assembler::notZero, error);
1048
1049 // set r12 to heapbase for load_klass()
1050 __ reinit_heapbase();
1051
1052 // make sure klass is 'reasonable', which is not zero.
1053 __ load_klass(rax, rax); // get klass
1054 __ testptr(rax, rax);
1055 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1056
1057 // return if everything seems ok
1058 __ bind(exit);
1059 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1060 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1061 __ pop(c_rarg3); // restore c_rarg3
1062 __ pop(c_rarg2); // restore c_rarg2
1063 __ pop(r12); // restore r12
1064 __ popf(); // restore flags
1065 __ ret(4 * wordSize); // pop caller saved stuff
1066
1067 // handle errors
1068 __ bind(error);
1069 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1070 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1071 __ pop(c_rarg3); // get saved c_rarg3 back
1072 __ pop(c_rarg2); // get saved c_rarg2 back
1073 __ pop(r12); // get saved r12 back
1074 __ popf(); // get saved flags off stack --
1075 // will be ignored
1076
1077 __ pusha(); // push registers
1078 // (rip is already
1079 // already pushed)
1080 // debug(char* msg, int64_t pc, int64_t regs[])
1081 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1082 // pushed all the registers, so now the stack looks like:
1083 // [tos + 0] 16 saved registers
1084 // [tos + 16] return address
1085 // * [tos + 17] error message (char*)
1086 // * [tos + 18] object to verify (oop)
1087 // * [tos + 19] saved rax - saved by caller and bashed
1088 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1089 // * = popped on exit
1090
1091 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1092 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1093 __ movq(c_rarg2, rsp); // pass address of regs on stack
1094 __ mov(r12, rsp); // remember rsp
1095 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1096 __ andptr(rsp, -16); // align stack as required by ABI
1097 BLOCK_COMMENT("call MacroAssembler::debug");
1098 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1099 __ mov(rsp, r12); // restore rsp
1100 __ popa(); // pop registers (includes r12)
1101 __ ret(4 * wordSize); // pop caller saved stuff
1102
1103 return start;
1104 }
1105
1106 //
1107 // Verify that a register contains clean 32-bits positive value
1108 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1109 //
1110 // Input:
1111 // Rint - 32-bits value
1112 // Rtmp - scratch
1113 //
1114 void assert_clean_int(Register Rint, Register Rtmp) {
1115 #ifdef ASSERT
1116 Label L;
1117 assert_different_registers(Rtmp, Rint);
1118 __ movslq(Rtmp, Rint);
1119 __ cmpq(Rtmp, Rint);
1120 __ jcc(Assembler::equal, L);
1121 __ stop("high 32-bits of int value are not 0");
1122 __ bind(L);
1123 #endif
1124 }
1125
1126 // Generate overlap test for array copy stubs
1127 //
1128 // Input:
1129 // c_rarg0 - from
1130 // c_rarg1 - to
1131 // c_rarg2 - element count
1132 //
1133 // Output:
1134 // rax - &from[element count - 1]
1135 //
1136 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1137 assert(no_overlap_target != NULL, "must be generated");
1138 array_overlap_test(no_overlap_target, NULL, sf);
1139 }
1140 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1141 array_overlap_test(NULL, &L_no_overlap, sf);
1142 }
1143 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1144 const Register from = c_rarg0;
1145 const Register to = c_rarg1;
1146 const Register count = c_rarg2;
1147 const Register end_from = rax;
1148
1149 __ cmpptr(to, from);
1150 __ lea(end_from, Address(from, count, sf, 0));
1151 if (NOLp == NULL) {
1152 ExternalAddress no_overlap(no_overlap_target);
1153 __ jump_cc(Assembler::belowEqual, no_overlap);
1154 __ cmpptr(to, end_from);
1155 __ jump_cc(Assembler::aboveEqual, no_overlap);
1156 } else {
1157 __ jcc(Assembler::belowEqual, (*NOLp));
1158 __ cmpptr(to, end_from);
1159 __ jcc(Assembler::aboveEqual, (*NOLp));
1160 }
1161 }
1162
1163 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1164 //
1165 // Outputs:
1166 // rdi - rcx
1167 // rsi - rdx
1168 // rdx - r8
1169 // rcx - r9
1170 //
1171 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1172 // are non-volatile. r9 and r10 should not be used by the caller.
1173 //
1174 void setup_arg_regs(int nargs = 3) {
1175 const Register saved_rdi = r9;
1176 const Register saved_rsi = r10;
1177 assert(nargs == 3 || nargs == 4, "else fix");
1178 #ifdef _WIN64
1179 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1180 "unexpected argument registers");
1181 if (nargs >= 4)
1182 __ mov(rax, r9); // r9 is also saved_rdi
1183 __ movptr(saved_rdi, rdi);
1184 __ movptr(saved_rsi, rsi);
1185 __ mov(rdi, rcx); // c_rarg0
1186 __ mov(rsi, rdx); // c_rarg1
1187 __ mov(rdx, r8); // c_rarg2
1188 if (nargs >= 4)
1189 __ mov(rcx, rax); // c_rarg3 (via rax)
1190 #else
1191 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1192 "unexpected argument registers");
1193 #endif
1194 }
1195
1196 void restore_arg_regs() {
1197 const Register saved_rdi = r9;
1198 const Register saved_rsi = r10;
1199 #ifdef _WIN64
1200 __ movptr(rdi, saved_rdi);
1201 __ movptr(rsi, saved_rsi);
1202 #endif
1203 }
1204
1205
1206 // Copy big chunks forward
1207 //
1208 // Inputs:
1209 // end_from - source arrays end address
1210 // end_to - destination array end address
1211 // qword_count - 64-bits element count, negative
1212 // to - scratch
1213 // L_copy_bytes - entry label
1214 // L_copy_8_bytes - exit label
1215 //
1216 void copy_bytes_forward(Register end_from, Register end_to,
1217 Register qword_count, Register to,
1218 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1219 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1220 Label L_loop;
1221 __ align(OptoLoopAlignment);
1222 if (UseUnalignedLoadStores) {
1223 Label L_end;
1224 if (UseAVX > 2) {
1225 __ movl(to, 0xffff);
1226 __ kmovwl(k1, to);
1227 }
1228 // Copy 64-bytes per iteration
1229 __ BIND(L_loop);
1230 if (UseAVX > 2) {
1231 __ evmovdqul(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1232 __ evmovdqul(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1233 } else if (UseAVX == 2) {
1234 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1235 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1236 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1237 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1238 } else {
1239 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1240 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1241 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1242 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1243 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1244 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1245 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1246 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1247 }
1248 __ BIND(L_copy_bytes);
1249 __ addptr(qword_count, 8);
1250 __ jcc(Assembler::lessEqual, L_loop);
1251 __ subptr(qword_count, 4); // sub(8) and add(4)
1252 __ jccb(Assembler::greater, L_end);
1253 // Copy trailing 32 bytes
1254 if (UseAVX >= 2) {
1255 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1256 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1257 } else {
1258 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1259 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1260 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1261 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1262 }
1263 __ addptr(qword_count, 4);
1264 __ BIND(L_end);
1265 if (UseAVX >= 2) {
1266 // clean upper bits of YMM registers
1267 __ vpxor(xmm0, xmm0);
1268 __ vpxor(xmm1, xmm1);
1269 }
1270 } else {
1271 // Copy 32-bytes per iteration
1272 __ BIND(L_loop);
1273 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1274 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1275 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1276 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1277 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1278 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1279 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1280 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1281
1282 __ BIND(L_copy_bytes);
1283 __ addptr(qword_count, 4);
1284 __ jcc(Assembler::lessEqual, L_loop);
1285 }
1286 __ subptr(qword_count, 4);
1287 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1288 }
1289
1290 // Copy big chunks backward
1291 //
1292 // Inputs:
1293 // from - source arrays address
1294 // dest - destination array address
1295 // qword_count - 64-bits element count
1296 // to - scratch
1297 // L_copy_bytes - entry label
1298 // L_copy_8_bytes - exit label
1299 //
1300 void copy_bytes_backward(Register from, Register dest,
1301 Register qword_count, Register to,
1302 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1303 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1304 Label L_loop;
1305 __ align(OptoLoopAlignment);
1306 if (UseUnalignedLoadStores) {
1307 Label L_end;
1308 if (UseAVX > 2) {
1309 __ movl(to, 0xffff);
1310 __ kmovwl(k1, to);
1311 }
1312 // Copy 64-bytes per iteration
1313 __ BIND(L_loop);
1314 if (UseAVX > 2) {
1315 __ evmovdqul(xmm0, Address(from, qword_count, Address::times_8, 0), Assembler::AVX_512bit);
1316 __ evmovdqul(Address(dest, qword_count, Address::times_8, 0), xmm0, Assembler::AVX_512bit);
1317 } else if (UseAVX == 2) {
1318 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1319 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1320 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1321 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1322 } else {
1323 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1324 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1325 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1326 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1327 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1328 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1329 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1330 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1331 }
1332 __ BIND(L_copy_bytes);
1333 __ subptr(qword_count, 8);
1334 __ jcc(Assembler::greaterEqual, L_loop);
1335
1336 __ addptr(qword_count, 4); // add(8) and sub(4)
1337 __ jccb(Assembler::less, L_end);
1338 // Copy trailing 32 bytes
1339 if (UseAVX >= 2) {
1340 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1341 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1342 } else {
1343 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1344 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1345 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1346 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1347 }
1348 __ subptr(qword_count, 4);
1349 __ BIND(L_end);
1350 if (UseAVX >= 2) {
1351 // clean upper bits of YMM registers
1352 __ vpxor(xmm0, xmm0);
1353 __ vpxor(xmm1, xmm1);
1354 }
1355 } else {
1356 // Copy 32-bytes per iteration
1357 __ BIND(L_loop);
1358 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1359 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1360 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1361 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1362 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1363 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1364 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1365 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1366
1367 __ BIND(L_copy_bytes);
1368 __ subptr(qword_count, 4);
1369 __ jcc(Assembler::greaterEqual, L_loop);
1370 }
1371 __ addptr(qword_count, 4);
1372 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1373 }
1374
1375
1376 // Arguments:
1377 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1378 // ignored
1379 // name - stub name string
1380 //
1381 // Inputs:
1382 // c_rarg0 - source array address
1383 // c_rarg1 - destination array address
1384 // c_rarg2 - element count, treated as ssize_t, can be zero
1385 //
1386 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1387 // we let the hardware handle it. The one to eight bytes within words,
1388 // dwords or qwords that span cache line boundaries will still be loaded
1389 // and stored atomically.
1390 //
1391 // Side Effects:
1392 // disjoint_byte_copy_entry is set to the no-overlap entry point
1393 // used by generate_conjoint_byte_copy().
1394 //
1395 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1396 __ align(CodeEntryAlignment);
1397 StubCodeMark mark(this, "StubRoutines", name);
1398 address start = __ pc();
1399
1400 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1401 Label L_copy_byte, L_exit;
1402 const Register from = rdi; // source array address
1403 const Register to = rsi; // destination array address
1404 const Register count = rdx; // elements count
1405 const Register byte_count = rcx;
1406 const Register qword_count = count;
1407 const Register end_from = from; // source array end address
1408 const Register end_to = to; // destination array end address
1409 // End pointers are inclusive, and if count is not zero they point
1410 // to the last unit copied: end_to[0] := end_from[0]
1411
1412 __ enter(); // required for proper stackwalking of RuntimeStub frame
1413 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1414
1415 if (entry != NULL) {
1416 *entry = __ pc();
1417 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1418 BLOCK_COMMENT("Entry:");
1419 }
1420
1421 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1422 // r9 and r10 may be used to save non-volatile registers
1423
1424 // 'from', 'to' and 'count' are now valid
1425 __ movptr(byte_count, count);
1426 __ shrptr(count, 3); // count => qword_count
1427
1428 // Copy from low to high addresses. Use 'to' as scratch.
1429 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1430 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1431 __ negptr(qword_count); // make the count negative
1432 __ jmp(L_copy_bytes);
1433
1434 // Copy trailing qwords
1435 __ BIND(L_copy_8_bytes);
1436 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1437 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1438 __ increment(qword_count);
1439 __ jcc(Assembler::notZero, L_copy_8_bytes);
1440
1441 // Check for and copy trailing dword
1442 __ BIND(L_copy_4_bytes);
1443 __ testl(byte_count, 4);
1444 __ jccb(Assembler::zero, L_copy_2_bytes);
1445 __ movl(rax, Address(end_from, 8));
1446 __ movl(Address(end_to, 8), rax);
1447
1448 __ addptr(end_from, 4);
1449 __ addptr(end_to, 4);
1450
1451 // Check for and copy trailing word
1452 __ BIND(L_copy_2_bytes);
1453 __ testl(byte_count, 2);
1454 __ jccb(Assembler::zero, L_copy_byte);
1455 __ movw(rax, Address(end_from, 8));
1456 __ movw(Address(end_to, 8), rax);
1457
1458 __ addptr(end_from, 2);
1459 __ addptr(end_to, 2);
1460
1461 // Check for and copy trailing byte
1462 __ BIND(L_copy_byte);
1463 __ testl(byte_count, 1);
1464 __ jccb(Assembler::zero, L_exit);
1465 __ movb(rax, Address(end_from, 8));
1466 __ movb(Address(end_to, 8), rax);
1467
1468 __ BIND(L_exit);
1469 restore_arg_regs();
1470 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1471 __ xorptr(rax, rax); // return 0
1472 __ vzeroupper();
1473 __ leave(); // required for proper stackwalking of RuntimeStub frame
1474 __ ret(0);
1475
1476 // Copy in multi-bytes chunks
1477 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1478 __ jmp(L_copy_4_bytes);
1479
1480 return start;
1481 }
1482
1483 // Arguments:
1484 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1485 // ignored
1486 // name - stub name string
1487 //
1488 // Inputs:
1489 // c_rarg0 - source array address
1490 // c_rarg1 - destination array address
1491 // c_rarg2 - element count, treated as ssize_t, can be zero
1492 //
1493 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1494 // we let the hardware handle it. The one to eight bytes within words,
1495 // dwords or qwords that span cache line boundaries will still be loaded
1496 // and stored atomically.
1497 //
1498 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1499 address* entry, const char *name) {
1500 __ align(CodeEntryAlignment);
1501 StubCodeMark mark(this, "StubRoutines", name);
1502 address start = __ pc();
1503
1504 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1505 const Register from = rdi; // source array address
1506 const Register to = rsi; // destination array address
1507 const Register count = rdx; // elements count
1508 const Register byte_count = rcx;
1509 const Register qword_count = count;
1510
1511 __ enter(); // required for proper stackwalking of RuntimeStub frame
1512 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1513
1514 if (entry != NULL) {
1515 *entry = __ pc();
1516 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1517 BLOCK_COMMENT("Entry:");
1518 }
1519
1520 array_overlap_test(nooverlap_target, Address::times_1);
1521 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1522 // r9 and r10 may be used to save non-volatile registers
1523
1524 // 'from', 'to' and 'count' are now valid
1525 __ movptr(byte_count, count);
1526 __ shrptr(count, 3); // count => qword_count
1527
1528 // Copy from high to low addresses.
1529
1530 // Check for and copy trailing byte
1531 __ testl(byte_count, 1);
1532 __ jcc(Assembler::zero, L_copy_2_bytes);
1533 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1534 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1535 __ decrement(byte_count); // Adjust for possible trailing word
1536
1537 // Check for and copy trailing word
1538 __ BIND(L_copy_2_bytes);
1539 __ testl(byte_count, 2);
1540 __ jcc(Assembler::zero, L_copy_4_bytes);
1541 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1542 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1543
1544 // Check for and copy trailing dword
1545 __ BIND(L_copy_4_bytes);
1546 __ testl(byte_count, 4);
1547 __ jcc(Assembler::zero, L_copy_bytes);
1548 __ movl(rax, Address(from, qword_count, Address::times_8));
1549 __ movl(Address(to, qword_count, Address::times_8), rax);
1550 __ jmp(L_copy_bytes);
1551
1552 // Copy trailing qwords
1553 __ BIND(L_copy_8_bytes);
1554 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1555 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1556 __ decrement(qword_count);
1557 __ jcc(Assembler::notZero, L_copy_8_bytes);
1558
1559 restore_arg_regs();
1560 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1561 __ xorptr(rax, rax); // return 0
1562 __ vzeroupper();
1563 __ leave(); // required for proper stackwalking of RuntimeStub frame
1564 __ ret(0);
1565
1566 // Copy in multi-bytes chunks
1567 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1568
1569 restore_arg_regs();
1570 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1571 __ xorptr(rax, rax); // return 0
1572 __ vzeroupper();
1573 __ leave(); // required for proper stackwalking of RuntimeStub frame
1574 __ ret(0);
1575
1576 return start;
1577 }
1578
1579 // Arguments:
1580 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1581 // ignored
1582 // name - stub name string
1583 //
1584 // Inputs:
1585 // c_rarg0 - source array address
1586 // c_rarg1 - destination array address
1587 // c_rarg2 - element count, treated as ssize_t, can be zero
1588 //
1589 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1590 // let the hardware handle it. The two or four words within dwords
1591 // or qwords that span cache line boundaries will still be loaded
1592 // and stored atomically.
1593 //
1594 // Side Effects:
1595 // disjoint_short_copy_entry is set to the no-overlap entry point
1596 // used by generate_conjoint_short_copy().
1597 //
1598 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1599 __ align(CodeEntryAlignment);
1600 StubCodeMark mark(this, "StubRoutines", name);
1601 address start = __ pc();
1602
1603 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1604 const Register from = rdi; // source array address
1605 const Register to = rsi; // destination array address
1606 const Register count = rdx; // elements count
1607 const Register word_count = rcx;
1608 const Register qword_count = count;
1609 const Register end_from = from; // source array end address
1610 const Register end_to = to; // destination array end address
1611 // End pointers are inclusive, and if count is not zero they point
1612 // to the last unit copied: end_to[0] := end_from[0]
1613
1614 __ enter(); // required for proper stackwalking of RuntimeStub frame
1615 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1616
1617 if (entry != NULL) {
1618 *entry = __ pc();
1619 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1620 BLOCK_COMMENT("Entry:");
1621 }
1622
1623 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1624 // r9 and r10 may be used to save non-volatile registers
1625
1626 // 'from', 'to' and 'count' are now valid
1627 __ movptr(word_count, count);
1628 __ shrptr(count, 2); // count => qword_count
1629
1630 // Copy from low to high addresses. Use 'to' as scratch.
1631 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1632 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1633 __ negptr(qword_count);
1634 __ jmp(L_copy_bytes);
1635
1636 // Copy trailing qwords
1637 __ BIND(L_copy_8_bytes);
1638 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1639 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1640 __ increment(qword_count);
1641 __ jcc(Assembler::notZero, L_copy_8_bytes);
1642
1643 // Original 'dest' is trashed, so we can't use it as a
1644 // base register for a possible trailing word copy
1645
1646 // Check for and copy trailing dword
1647 __ BIND(L_copy_4_bytes);
1648 __ testl(word_count, 2);
1649 __ jccb(Assembler::zero, L_copy_2_bytes);
1650 __ movl(rax, Address(end_from, 8));
1651 __ movl(Address(end_to, 8), rax);
1652
1653 __ addptr(end_from, 4);
1654 __ addptr(end_to, 4);
1655
1656 // Check for and copy trailing word
1657 __ BIND(L_copy_2_bytes);
1658 __ testl(word_count, 1);
1659 __ jccb(Assembler::zero, L_exit);
1660 __ movw(rax, Address(end_from, 8));
1661 __ movw(Address(end_to, 8), rax);
1662
1663 __ BIND(L_exit);
1664 restore_arg_regs();
1665 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1666 __ xorptr(rax, rax); // return 0
1667 __ vzeroupper();
1668 __ leave(); // required for proper stackwalking of RuntimeStub frame
1669 __ ret(0);
1670
1671 // Copy in multi-bytes chunks
1672 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1673 __ jmp(L_copy_4_bytes);
1674
1675 return start;
1676 }
1677
1678 address generate_fill(BasicType t, bool aligned, const char *name) {
1679 __ align(CodeEntryAlignment);
1680 StubCodeMark mark(this, "StubRoutines", name);
1681 address start = __ pc();
1682
1683 BLOCK_COMMENT("Entry:");
1684
1685 const Register to = c_rarg0; // source array address
1686 const Register value = c_rarg1; // value
1687 const Register count = c_rarg2; // elements count
1688
1689 __ enter(); // required for proper stackwalking of RuntimeStub frame
1690
1691 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1692
1693 __ vzeroupper();
1694 __ leave(); // required for proper stackwalking of RuntimeStub frame
1695 __ ret(0);
1696 return start;
1697 }
1698
1699 // Arguments:
1700 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1701 // ignored
1702 // name - stub name string
1703 //
1704 // Inputs:
1705 // c_rarg0 - source array address
1706 // c_rarg1 - destination array address
1707 // c_rarg2 - element count, treated as ssize_t, can be zero
1708 //
1709 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1710 // let the hardware handle it. The two or four words within dwords
1711 // or qwords that span cache line boundaries will still be loaded
1712 // and stored atomically.
1713 //
1714 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1715 address *entry, const char *name) {
1716 __ align(CodeEntryAlignment);
1717 StubCodeMark mark(this, "StubRoutines", name);
1718 address start = __ pc();
1719
1720 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1721 const Register from = rdi; // source array address
1722 const Register to = rsi; // destination array address
1723 const Register count = rdx; // elements count
1724 const Register word_count = rcx;
1725 const Register qword_count = count;
1726
1727 __ enter(); // required for proper stackwalking of RuntimeStub frame
1728 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1729
1730 if (entry != NULL) {
1731 *entry = __ pc();
1732 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1733 BLOCK_COMMENT("Entry:");
1734 }
1735
1736 array_overlap_test(nooverlap_target, Address::times_2);
1737 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1738 // r9 and r10 may be used to save non-volatile registers
1739
1740 // 'from', 'to' and 'count' are now valid
1741 __ movptr(word_count, count);
1742 __ shrptr(count, 2); // count => qword_count
1743
1744 // Copy from high to low addresses. Use 'to' as scratch.
1745
1746 // Check for and copy trailing word
1747 __ testl(word_count, 1);
1748 __ jccb(Assembler::zero, L_copy_4_bytes);
1749 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1750 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1751
1752 // Check for and copy trailing dword
1753 __ BIND(L_copy_4_bytes);
1754 __ testl(word_count, 2);
1755 __ jcc(Assembler::zero, L_copy_bytes);
1756 __ movl(rax, Address(from, qword_count, Address::times_8));
1757 __ movl(Address(to, qword_count, Address::times_8), rax);
1758 __ jmp(L_copy_bytes);
1759
1760 // Copy trailing qwords
1761 __ BIND(L_copy_8_bytes);
1762 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1763 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1764 __ decrement(qword_count);
1765 __ jcc(Assembler::notZero, L_copy_8_bytes);
1766
1767 restore_arg_regs();
1768 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1769 __ xorptr(rax, rax); // return 0
1770 __ vzeroupper();
1771 __ leave(); // required for proper stackwalking of RuntimeStub frame
1772 __ ret(0);
1773
1774 // Copy in multi-bytes chunks
1775 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1776
1777 restore_arg_regs();
1778 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1779 __ xorptr(rax, rax); // return 0
1780 __ vzeroupper();
1781 __ leave(); // required for proper stackwalking of RuntimeStub frame
1782 __ ret(0);
1783
1784 return start;
1785 }
1786
1787 // Arguments:
1788 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1789 // ignored
1790 // is_oop - true => oop array, so generate store check code
1791 // name - stub name string
1792 //
1793 // Inputs:
1794 // c_rarg0 - source array address
1795 // c_rarg1 - destination array address
1796 // c_rarg2 - element count, treated as ssize_t, can be zero
1797 //
1798 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1799 // the hardware handle it. The two dwords within qwords that span
1800 // cache line boundaries will still be loaded and stored atomicly.
1801 //
1802 // Side Effects:
1803 // disjoint_int_copy_entry is set to the no-overlap entry point
1804 // used by generate_conjoint_int_oop_copy().
1805 //
1806 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1807 const char *name, bool dest_uninitialized = false) {
1808 __ align(CodeEntryAlignment);
1809 StubCodeMark mark(this, "StubRoutines", name);
1810 address start = __ pc();
1811
1812 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1813 const Register from = rdi; // source array address
1814 const Register to = rsi; // destination array address
1815 const Register count = rdx; // elements count
1816 const Register dword_count = rcx;
1817 const Register qword_count = count;
1818 const Register end_from = from; // source array end address
1819 const Register end_to = to; // destination array end address
1820 // End pointers are inclusive, and if count is not zero they point
1821 // to the last unit copied: end_to[0] := end_from[0]
1822
1823 __ enter(); // required for proper stackwalking of RuntimeStub frame
1824 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1825
1826 if (entry != NULL) {
1827 *entry = __ pc();
1828 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1829 BLOCK_COMMENT("Entry:");
1830 }
1831
1832 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1833 // r9 and r10 may be used to save non-volatile registers
1834
1835 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
1836 if (dest_uninitialized) {
1837 decorators |= IS_DEST_UNINITIALIZED;
1838 }
1839 if (aligned) {
1840 decorators |= ARRAYCOPY_ALIGNED;
1841 }
1842
1843 BasicType type = is_oop ? T_OBJECT : T_INT;
1844 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1845 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1846
1847 // 'from', 'to' and 'count' are now valid
1848 __ movptr(dword_count, count);
1849 __ shrptr(count, 1); // count => qword_count
1850
1851 // Copy from low to high addresses. Use 'to' as scratch.
1852 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1853 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1854 __ negptr(qword_count);
1855 __ jmp(L_copy_bytes);
1856
1857 // Copy trailing qwords
1858 __ BIND(L_copy_8_bytes);
1859 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1860 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1861 __ increment(qword_count);
1862 __ jcc(Assembler::notZero, L_copy_8_bytes);
1863
1864 // Check for and copy trailing dword
1865 __ BIND(L_copy_4_bytes);
1866 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1867 __ jccb(Assembler::zero, L_exit);
1868 __ movl(rax, Address(end_from, 8));
1869 __ movl(Address(end_to, 8), rax);
1870
1871 __ BIND(L_exit);
1872 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
1873 restore_arg_regs();
1874 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1875 __ vzeroupper();
1876 __ xorptr(rax, rax); // return 0
1877 __ leave(); // required for proper stackwalking of RuntimeStub frame
1878 __ ret(0);
1879
1880 // Copy in multi-bytes chunks
1881 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1882 __ jmp(L_copy_4_bytes);
1883
1884 return start;
1885 }
1886
1887 // Arguments:
1888 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1889 // ignored
1890 // is_oop - true => oop array, so generate store check code
1891 // name - stub name string
1892 //
1893 // Inputs:
1894 // c_rarg0 - source array address
1895 // c_rarg1 - destination array address
1896 // c_rarg2 - element count, treated as ssize_t, can be zero
1897 //
1898 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1899 // the hardware handle it. The two dwords within qwords that span
1900 // cache line boundaries will still be loaded and stored atomicly.
1901 //
1902 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1903 address *entry, const char *name,
1904 bool dest_uninitialized = false) {
1905 __ align(CodeEntryAlignment);
1906 StubCodeMark mark(this, "StubRoutines", name);
1907 address start = __ pc();
1908
1909 Label L_copy_bytes, L_copy_8_bytes, L_exit;
1910 const Register from = rdi; // source array address
1911 const Register to = rsi; // destination array address
1912 const Register count = rdx; // elements count
1913 const Register dword_count = rcx;
1914 const Register qword_count = count;
1915
1916 __ enter(); // required for proper stackwalking of RuntimeStub frame
1917 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1918
1919 if (entry != NULL) {
1920 *entry = __ pc();
1921 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1922 BLOCK_COMMENT("Entry:");
1923 }
1924
1925 array_overlap_test(nooverlap_target, Address::times_4);
1926 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1927 // r9 and r10 may be used to save non-volatile registers
1928
1929 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
1930 if (dest_uninitialized) {
1931 decorators |= IS_DEST_UNINITIALIZED;
1932 }
1933 if (aligned) {
1934 decorators |= ARRAYCOPY_ALIGNED;
1935 }
1936
1937 BasicType type = is_oop ? T_OBJECT : T_INT;
1938 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1939 // no registers are destroyed by this call
1940 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1941
1942 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1943 // 'from', 'to' and 'count' are now valid
1944 __ movptr(dword_count, count);
1945 __ shrptr(count, 1); // count => qword_count
1946
1947 // Copy from high to low addresses. Use 'to' as scratch.
1948
1949 // Check for and copy trailing dword
1950 __ testl(dword_count, 1);
1951 __ jcc(Assembler::zero, L_copy_bytes);
1952 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1953 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1954 __ jmp(L_copy_bytes);
1955
1956 // Copy trailing qwords
1957 __ BIND(L_copy_8_bytes);
1958 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1959 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1960 __ decrement(qword_count);
1961 __ jcc(Assembler::notZero, L_copy_8_bytes);
1962
1963 if (is_oop) {
1964 __ jmp(L_exit);
1965 }
1966 restore_arg_regs();
1967 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1968 __ xorptr(rax, rax); // return 0
1969 __ vzeroupper();
1970 __ leave(); // required for proper stackwalking of RuntimeStub frame
1971 __ ret(0);
1972
1973 // Copy in multi-bytes chunks
1974 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1975
1976 __ BIND(L_exit);
1977 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
1978 restore_arg_regs();
1979 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1980 __ xorptr(rax, rax); // return 0
1981 __ vzeroupper();
1982 __ leave(); // required for proper stackwalking of RuntimeStub frame
1983 __ ret(0);
1984
1985 return start;
1986 }
1987
1988 // Arguments:
1989 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1990 // ignored
1991 // is_oop - true => oop array, so generate store check code
1992 // name - stub name string
1993 //
1994 // Inputs:
1995 // c_rarg0 - source array address
1996 // c_rarg1 - destination array address
1997 // c_rarg2 - element count, treated as ssize_t, can be zero
1998 //
1999 // Side Effects:
2000 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2001 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2002 //
2003 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2004 const char *name, bool dest_uninitialized = false) {
2005 __ align(CodeEntryAlignment);
2006 StubCodeMark mark(this, "StubRoutines", name);
2007 address start = __ pc();
2008
2009 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2010 const Register from = rdi; // source array address
2011 const Register to = rsi; // destination array address
2012 const Register qword_count = rdx; // elements count
2013 const Register end_from = from; // source array end address
2014 const Register end_to = rcx; // destination array end address
2015 const Register saved_count = r11;
2016 // End pointers are inclusive, and if count is not zero they point
2017 // to the last unit copied: end_to[0] := end_from[0]
2018
2019 __ enter(); // required for proper stackwalking of RuntimeStub frame
2020 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2021 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2022
2023 if (entry != NULL) {
2024 *entry = __ pc();
2025 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2026 BLOCK_COMMENT("Entry:");
2027 }
2028
2029 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2030 // r9 and r10 may be used to save non-volatile registers
2031 // 'from', 'to' and 'qword_count' are now valid
2032
2033 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2034 if (dest_uninitialized) {
2035 decorators |= IS_DEST_UNINITIALIZED;
2036 }
2037 if (aligned) {
2038 decorators |= ARRAYCOPY_ALIGNED;
2039 }
2040
2041 BasicType type = is_oop ? T_OBJECT : T_LONG;
2042 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2043 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2044
2045 // Copy from low to high addresses. Use 'to' as scratch.
2046 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2047 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2048 __ negptr(qword_count);
2049 __ jmp(L_copy_bytes);
2050
2051 // Copy trailing qwords
2052 __ BIND(L_copy_8_bytes);
2053 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2054 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2055 __ increment(qword_count);
2056 __ jcc(Assembler::notZero, L_copy_8_bytes);
2057
2058 if (is_oop) {
2059 __ jmp(L_exit);
2060 } else {
2061 restore_arg_regs();
2062 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2063 __ xorptr(rax, rax); // return 0
2064 __ vzeroupper();
2065 __ leave(); // required for proper stackwalking of RuntimeStub frame
2066 __ ret(0);
2067 }
2068
2069 // Copy in multi-bytes chunks
2070 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2071
2072 __ BIND(L_exit);
2073 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2074 restore_arg_regs();
2075 if (is_oop) {
2076 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2077 } else {
2078 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2079 }
2080 __ vzeroupper();
2081 __ xorptr(rax, rax); // return 0
2082 __ leave(); // required for proper stackwalking of RuntimeStub frame
2083 __ ret(0);
2084
2085 return start;
2086 }
2087
2088 // Arguments:
2089 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2090 // ignored
2091 // is_oop - true => oop array, so generate store check code
2092 // name - stub name string
2093 //
2094 // Inputs:
2095 // c_rarg0 - source array address
2096 // c_rarg1 - destination array address
2097 // c_rarg2 - element count, treated as ssize_t, can be zero
2098 //
2099 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2100 address nooverlap_target, address *entry,
2101 const char *name, bool dest_uninitialized = false) {
2102 __ align(CodeEntryAlignment);
2103 StubCodeMark mark(this, "StubRoutines", name);
2104 address start = __ pc();
2105
2106 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2107 const Register from = rdi; // source array address
2108 const Register to = rsi; // destination array address
2109 const Register qword_count = rdx; // elements count
2110 const Register saved_count = rcx;
2111
2112 __ enter(); // required for proper stackwalking of RuntimeStub frame
2113 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2114
2115 if (entry != NULL) {
2116 *entry = __ pc();
2117 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2118 BLOCK_COMMENT("Entry:");
2119 }
2120
2121 array_overlap_test(nooverlap_target, Address::times_8);
2122 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2123 // r9 and r10 may be used to save non-volatile registers
2124 // 'from', 'to' and 'qword_count' are now valid
2125
2126 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2127 if (dest_uninitialized) {
2128 decorators |= IS_DEST_UNINITIALIZED;
2129 }
2130 if (aligned) {
2131 decorators |= ARRAYCOPY_ALIGNED;
2132 }
2133
2134 BasicType type = is_oop ? T_OBJECT : T_LONG;
2135 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2136 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2137
2138 __ jmp(L_copy_bytes);
2139
2140 // Copy trailing qwords
2141 __ BIND(L_copy_8_bytes);
2142 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2143 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2144 __ decrement(qword_count);
2145 __ jcc(Assembler::notZero, L_copy_8_bytes);
2146
2147 if (is_oop) {
2148 __ jmp(L_exit);
2149 } else {
2150 restore_arg_regs();
2151 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2152 __ xorptr(rax, rax); // return 0
2153 __ vzeroupper();
2154 __ leave(); // required for proper stackwalking of RuntimeStub frame
2155 __ ret(0);
2156 }
2157
2158 // Copy in multi-bytes chunks
2159 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2160
2161 __ BIND(L_exit);
2162 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2163 restore_arg_regs();
2164 if (is_oop) {
2165 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2166 } else {
2167 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2168 }
2169 __ vzeroupper();
2170 __ xorptr(rax, rax); // return 0
2171 __ leave(); // required for proper stackwalking of RuntimeStub frame
2172 __ ret(0);
2173
2174 return start;
2175 }
2176
2177
2178 // Helper for generating a dynamic type check.
2179 // Smashes no registers.
2180 void generate_type_check(Register sub_klass,
2181 Register super_check_offset,
2182 Register super_klass,
2183 Label& L_success) {
2184 assert_different_registers(sub_klass, super_check_offset, super_klass);
2185
2186 BLOCK_COMMENT("type_check:");
2187
2188 Label L_miss;
2189
2190 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2191 super_check_offset);
2192 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2193
2194 // Fall through on failure!
2195 __ BIND(L_miss);
2196 }
2197
2198 //
2199 // Generate checkcasting array copy stub
2200 //
2201 // Input:
2202 // c_rarg0 - source array address
2203 // c_rarg1 - destination array address
2204 // c_rarg2 - element count, treated as ssize_t, can be zero
2205 // c_rarg3 - size_t ckoff (super_check_offset)
2206 // not Win64
2207 // c_rarg4 - oop ckval (super_klass)
2208 // Win64
2209 // rsp+40 - oop ckval (super_klass)
2210 //
2211 // Output:
2212 // rax == 0 - success
2213 // rax == -1^K - failure, where K is partial transfer count
2214 //
2215 address generate_checkcast_copy(const char *name, address *entry,
2216 bool dest_uninitialized = false) {
2217
2218 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2219
2220 // Input registers (after setup_arg_regs)
2221 const Register from = rdi; // source array address
2222 const Register to = rsi; // destination array address
2223 const Register length = rdx; // elements count
2224 const Register ckoff = rcx; // super_check_offset
2225 const Register ckval = r8; // super_klass
2226
2227 // Registers used as temps (r13, r14 are save-on-entry)
2228 const Register end_from = from; // source array end address
2229 const Register end_to = r13; // destination array end address
2230 const Register count = rdx; // -(count_remaining)
2231 const Register r14_length = r14; // saved copy of length
2232 // End pointers are inclusive, and if length is not zero they point
2233 // to the last unit copied: end_to[0] := end_from[0]
2234
2235 const Register rax_oop = rax; // actual oop copied
2236 const Register r11_klass = r11; // oop._klass
2237
2238 //---------------------------------------------------------------
2239 // Assembler stub will be used for this call to arraycopy
2240 // if the two arrays are subtypes of Object[] but the
2241 // destination array type is not equal to or a supertype
2242 // of the source type. Each element must be separately
2243 // checked.
2244
2245 __ align(CodeEntryAlignment);
2246 StubCodeMark mark(this, "StubRoutines", name);
2247 address start = __ pc();
2248
2249 __ enter(); // required for proper stackwalking of RuntimeStub frame
2250
2251 #ifdef ASSERT
2252 // caller guarantees that the arrays really are different
2253 // otherwise, we would have to make conjoint checks
2254 { Label L;
2255 array_overlap_test(L, TIMES_OOP);
2256 __ stop("checkcast_copy within a single array");
2257 __ bind(L);
2258 }
2259 #endif //ASSERT
2260
2261 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2262 // ckoff => rcx, ckval => r8
2263 // r9 and r10 may be used to save non-volatile registers
2264 #ifdef _WIN64
2265 // last argument (#4) is on stack on Win64
2266 __ movptr(ckval, Address(rsp, 6 * wordSize));
2267 #endif
2268
2269 // Caller of this entry point must set up the argument registers.
2270 if (entry != NULL) {
2271 *entry = __ pc();
2272 BLOCK_COMMENT("Entry:");
2273 }
2274
2275 // allocate spill slots for r13, r14
2276 enum {
2277 saved_r13_offset,
2278 saved_r14_offset,
2279 saved_rbp_offset
2280 };
2281 __ subptr(rsp, saved_rbp_offset * wordSize);
2282 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2283 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2284
2285 // check that int operands are properly extended to size_t
2286 assert_clean_int(length, rax);
2287 assert_clean_int(ckoff, rax);
2288
2289 #ifdef ASSERT
2290 BLOCK_COMMENT("assert consistent ckoff/ckval");
2291 // The ckoff and ckval must be mutually consistent,
2292 // even though caller generates both.
2293 { Label L;
2294 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2295 __ cmpl(ckoff, Address(ckval, sco_offset));
2296 __ jcc(Assembler::equal, L);
2297 __ stop("super_check_offset inconsistent");
2298 __ bind(L);
2299 }
2300 #endif //ASSERT
2301
2302 // Loop-invariant addresses. They are exclusive end pointers.
2303 Address end_from_addr(from, length, TIMES_OOP, 0);
2304 Address end_to_addr(to, length, TIMES_OOP, 0);
2305 // Loop-variant addresses. They assume post-incremented count < 0.
2306 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2307 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2308
2309 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST;
2310 if (dest_uninitialized) {
2311 decorators |= IS_DEST_UNINITIALIZED;
2312 }
2313
2314 BasicType type = T_OBJECT;
2315 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2316 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2317
2318 // Copy from low to high addresses, indexed from the end of each array.
2319 __ lea(end_from, end_from_addr);
2320 __ lea(end_to, end_to_addr);
2321 __ movptr(r14_length, length); // save a copy of the length
2322 assert(length == count, ""); // else fix next line:
2323 __ negptr(count); // negate and test the length
2324 __ jcc(Assembler::notZero, L_load_element);
2325
2326 // Empty array: Nothing to do.
2327 __ xorptr(rax, rax); // return 0 on (trivial) success
2328 __ jmp(L_done);
2329
2330 // ======== begin loop ========
2331 // (Loop is rotated; its entry is L_load_element.)
2332 // Loop control:
2333 // for (count = -count; count != 0; count++)
2334 // Base pointers src, dst are biased by 8*(count-1),to last element.
2335 __ align(OptoLoopAlignment);
2336
2337 __ BIND(L_store_element);
2338 __ store_heap_oop(to_element_addr, rax_oop, noreg, noreg, AS_RAW); // store the oop
2339 __ increment(count); // increment the count toward zero
2340 __ jcc(Assembler::zero, L_do_card_marks);
2341
2342 // ======== loop entry is here ========
2343 __ BIND(L_load_element);
2344 __ load_heap_oop(rax_oop, from_element_addr, noreg, noreg, AS_RAW); // load the oop
2345 __ testptr(rax_oop, rax_oop);
2346 __ jcc(Assembler::zero, L_store_element);
2347
2348 __ load_klass(r11_klass, rax_oop);// query the object klass
2349 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2350 // ======== end loop ========
2351
2352 // It was a real error; we must depend on the caller to finish the job.
2353 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2354 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2355 // and report their number to the caller.
2356 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2357 Label L_post_barrier;
2358 __ addptr(r14_length, count); // K = (original - remaining) oops
2359 __ movptr(rax, r14_length); // save the value
2360 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2361 __ jccb(Assembler::notZero, L_post_barrier);
2362 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2363
2364 // Come here on success only.
2365 __ BIND(L_do_card_marks);
2366 __ xorptr(rax, rax); // return 0 on success
2367
2368 __ BIND(L_post_barrier);
2369 bs->arraycopy_epilogue(_masm, decorators, type, from, to, r14_length);
2370
2371 // Common exit point (success or failure).
2372 __ BIND(L_done);
2373 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2374 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2375 restore_arg_regs();
2376 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2377 __ leave(); // required for proper stackwalking of RuntimeStub frame
2378 __ ret(0);
2379
2380 return start;
2381 }
2382
2383 //
2384 // Generate 'unsafe' array copy stub
2385 // Though just as safe as the other stubs, it takes an unscaled
2386 // size_t argument instead of an element count.
2387 //
2388 // Input:
2389 // c_rarg0 - source array address
2390 // c_rarg1 - destination array address
2391 // c_rarg2 - byte count, treated as ssize_t, can be zero
2392 //
2393 // Examines the alignment of the operands and dispatches
2394 // to a long, int, short, or byte copy loop.
2395 //
2396 address generate_unsafe_copy(const char *name,
2397 address byte_copy_entry, address short_copy_entry,
2398 address int_copy_entry, address long_copy_entry) {
2399
2400 Label L_long_aligned, L_int_aligned, L_short_aligned;
2401
2402 // Input registers (before setup_arg_regs)
2403 const Register from = c_rarg0; // source array address
2404 const Register to = c_rarg1; // destination array address
2405 const Register size = c_rarg2; // byte count (size_t)
2406
2407 // Register used as a temp
2408 const Register bits = rax; // test copy of low bits
2409
2410 __ align(CodeEntryAlignment);
2411 StubCodeMark mark(this, "StubRoutines", name);
2412 address start = __ pc();
2413
2414 __ enter(); // required for proper stackwalking of RuntimeStub frame
2415
2416 // bump this on entry, not on exit:
2417 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2418
2419 __ mov(bits, from);
2420 __ orptr(bits, to);
2421 __ orptr(bits, size);
2422
2423 __ testb(bits, BytesPerLong-1);
2424 __ jccb(Assembler::zero, L_long_aligned);
2425
2426 __ testb(bits, BytesPerInt-1);
2427 __ jccb(Assembler::zero, L_int_aligned);
2428
2429 __ testb(bits, BytesPerShort-1);
2430 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2431
2432 __ BIND(L_short_aligned);
2433 __ shrptr(size, LogBytesPerShort); // size => short_count
2434 __ jump(RuntimeAddress(short_copy_entry));
2435
2436 __ BIND(L_int_aligned);
2437 __ shrptr(size, LogBytesPerInt); // size => int_count
2438 __ jump(RuntimeAddress(int_copy_entry));
2439
2440 __ BIND(L_long_aligned);
2441 __ shrptr(size, LogBytesPerLong); // size => qword_count
2442 __ jump(RuntimeAddress(long_copy_entry));
2443
2444 return start;
2445 }
2446
2447 // Perform range checks on the proposed arraycopy.
2448 // Kills temp, but nothing else.
2449 // Also, clean the sign bits of src_pos and dst_pos.
2450 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2451 Register src_pos, // source position (c_rarg1)
2452 Register dst, // destination array oo (c_rarg2)
2453 Register dst_pos, // destination position (c_rarg3)
2454 Register length,
2455 Register temp,
2456 Label& L_failed) {
2457 BLOCK_COMMENT("arraycopy_range_checks:");
2458
2459 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2460 __ movl(temp, length);
2461 __ addl(temp, src_pos); // src_pos + length
2462 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2463 __ jcc(Assembler::above, L_failed);
2464
2465 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2466 __ movl(temp, length);
2467 __ addl(temp, dst_pos); // dst_pos + length
2468 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2469 __ jcc(Assembler::above, L_failed);
2470
2471 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2472 // Move with sign extension can be used since they are positive.
2473 __ movslq(src_pos, src_pos);
2474 __ movslq(dst_pos, dst_pos);
2475
2476 BLOCK_COMMENT("arraycopy_range_checks done");
2477 }
2478
2479 //
2480 // Generate generic array copy stubs
2481 //
2482 // Input:
2483 // c_rarg0 - src oop
2484 // c_rarg1 - src_pos (32-bits)
2485 // c_rarg2 - dst oop
2486 // c_rarg3 - dst_pos (32-bits)
2487 // not Win64
2488 // c_rarg4 - element count (32-bits)
2489 // Win64
2490 // rsp+40 - element count (32-bits)
2491 //
2492 // Output:
2493 // rax == 0 - success
2494 // rax == -1^K - failure, where K is partial transfer count
2495 //
2496 address generate_generic_copy(const char *name,
2497 address byte_copy_entry, address short_copy_entry,
2498 address int_copy_entry, address oop_copy_entry,
2499 address long_copy_entry, address checkcast_copy_entry) {
2500
2501 Label L_failed, L_failed_0, L_objArray;
2502 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2503
2504 // Input registers
2505 const Register src = c_rarg0; // source array oop
2506 const Register src_pos = c_rarg1; // source position
2507 const Register dst = c_rarg2; // destination array oop
2508 const Register dst_pos = c_rarg3; // destination position
2509 #ifndef _WIN64
2510 const Register length = c_rarg4;
2511 #else
2512 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2513 #endif
2514
2515 { int modulus = CodeEntryAlignment;
2516 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2517 int advance = target - (__ offset() % modulus);
2518 if (advance < 0) advance += modulus;
2519 if (advance > 0) __ nop(advance);
2520 }
2521 StubCodeMark mark(this, "StubRoutines", name);
2522
2523 // Short-hop target to L_failed. Makes for denser prologue code.
2524 __ BIND(L_failed_0);
2525 __ jmp(L_failed);
2526 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2527
2528 __ align(CodeEntryAlignment);
2529 address start = __ pc();
2530
2531 __ enter(); // required for proper stackwalking of RuntimeStub frame
2532
2533 // bump this on entry, not on exit:
2534 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2535
2536 //-----------------------------------------------------------------------
2537 // Assembler stub will be used for this call to arraycopy
2538 // if the following conditions are met:
2539 //
2540 // (1) src and dst must not be null.
2541 // (2) src_pos must not be negative.
2542 // (3) dst_pos must not be negative.
2543 // (4) length must not be negative.
2544 // (5) src klass and dst klass should be the same and not NULL.
2545 // (6) src and dst should be arrays.
2546 // (7) src_pos + length must not exceed length of src.
2547 // (8) dst_pos + length must not exceed length of dst.
2548 //
2549
2550 // if (src == NULL) return -1;
2551 __ testptr(src, src); // src oop
2552 size_t j1off = __ offset();
2553 __ jccb(Assembler::zero, L_failed_0);
2554
2555 // if (src_pos < 0) return -1;
2556 __ testl(src_pos, src_pos); // src_pos (32-bits)
2557 __ jccb(Assembler::negative, L_failed_0);
2558
2559 // if (dst == NULL) return -1;
2560 __ testptr(dst, dst); // dst oop
2561 __ jccb(Assembler::zero, L_failed_0);
2562
2563 // if (dst_pos < 0) return -1;
2564 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2565 size_t j4off = __ offset();
2566 __ jccb(Assembler::negative, L_failed_0);
2567
2568 // The first four tests are very dense code,
2569 // but not quite dense enough to put four
2570 // jumps in a 16-byte instruction fetch buffer.
2571 // That's good, because some branch predicters
2572 // do not like jumps so close together.
2573 // Make sure of this.
2574 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2575
2576 // registers used as temp
2577 const Register r11_length = r11; // elements count to copy
2578 const Register r10_src_klass = r10; // array klass
2579
2580 // if (length < 0) return -1;
2581 __ movl(r11_length, length); // length (elements count, 32-bits value)
2582 __ testl(r11_length, r11_length);
2583 __ jccb(Assembler::negative, L_failed_0);
2584
2585 __ load_klass(r10_src_klass, src);
2586 #ifdef ASSERT
2587 // assert(src->klass() != NULL);
2588 {
2589 BLOCK_COMMENT("assert klasses not null {");
2590 Label L1, L2;
2591 __ testptr(r10_src_klass, r10_src_klass);
2592 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2593 __ bind(L1);
2594 __ stop("broken null klass");
2595 __ bind(L2);
2596 __ load_klass(rax, dst);
2597 __ cmpq(rax, 0);
2598 __ jcc(Assembler::equal, L1); // this would be broken also
2599 BLOCK_COMMENT("} assert klasses not null done");
2600 }
2601 #endif
2602
2603 // Load layout helper (32-bits)
2604 //
2605 // |array_tag| | header_size | element_type | |log2_element_size|
2606 // 32 30 24 16 8 2 0
2607 //
2608 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2609 //
2610
2611 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2612
2613 // Handle objArrays completely differently...
2614 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2615 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2616 __ jcc(Assembler::equal, L_objArray);
2617
2618 // if (src->klass() != dst->klass()) return -1;
2619 __ load_klass(rax, dst);
2620 __ cmpq(r10_src_klass, rax);
2621 __ jcc(Assembler::notEqual, L_failed);
2622
2623 const Register rax_lh = rax; // layout helper
2624 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2625
2626 // if (!src->is_Array()) return -1;
2627 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2628 __ jcc(Assembler::greaterEqual, L_failed);
2629
2630 // At this point, it is known to be a typeArray (array_tag 0x3).
2631 #ifdef ASSERT
2632 {
2633 BLOCK_COMMENT("assert primitive array {");
2634 Label L;
2635 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2636 __ jcc(Assembler::greaterEqual, L);
2637 __ stop("must be a primitive array");
2638 __ bind(L);
2639 BLOCK_COMMENT("} assert primitive array done");
2640 }
2641 #endif
2642
2643 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2644 r10, L_failed);
2645
2646 // TypeArrayKlass
2647 //
2648 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2649 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2650 //
2651
2652 const Register r10_offset = r10; // array offset
2653 const Register rax_elsize = rax_lh; // element size
2654
2655 __ movl(r10_offset, rax_lh);
2656 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2657 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2658 __ addptr(src, r10_offset); // src array offset
2659 __ addptr(dst, r10_offset); // dst array offset
2660 BLOCK_COMMENT("choose copy loop based on element size");
2661 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2662
2663 // next registers should be set before the jump to corresponding stub
2664 const Register from = c_rarg0; // source array address
2665 const Register to = c_rarg1; // destination array address
2666 const Register count = c_rarg2; // elements count
2667
2668 // 'from', 'to', 'count' registers should be set in such order
2669 // since they are the same as 'src', 'src_pos', 'dst'.
2670
2671 __ BIND(L_copy_bytes);
2672 __ cmpl(rax_elsize, 0);
2673 __ jccb(Assembler::notEqual, L_copy_shorts);
2674 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2675 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2676 __ movl2ptr(count, r11_length); // length
2677 __ jump(RuntimeAddress(byte_copy_entry));
2678
2679 __ BIND(L_copy_shorts);
2680 __ cmpl(rax_elsize, LogBytesPerShort);
2681 __ jccb(Assembler::notEqual, L_copy_ints);
2682 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2683 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2684 __ movl2ptr(count, r11_length); // length
2685 __ jump(RuntimeAddress(short_copy_entry));
2686
2687 __ BIND(L_copy_ints);
2688 __ cmpl(rax_elsize, LogBytesPerInt);
2689 __ jccb(Assembler::notEqual, L_copy_longs);
2690 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2691 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2692 __ movl2ptr(count, r11_length); // length
2693 __ jump(RuntimeAddress(int_copy_entry));
2694
2695 __ BIND(L_copy_longs);
2696 #ifdef ASSERT
2697 {
2698 BLOCK_COMMENT("assert long copy {");
2699 Label L;
2700 __ cmpl(rax_elsize, LogBytesPerLong);
2701 __ jcc(Assembler::equal, L);
2702 __ stop("must be long copy, but elsize is wrong");
2703 __ bind(L);
2704 BLOCK_COMMENT("} assert long copy done");
2705 }
2706 #endif
2707 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2708 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2709 __ movl2ptr(count, r11_length); // length
2710 __ jump(RuntimeAddress(long_copy_entry));
2711
2712 // ObjArrayKlass
2713 __ BIND(L_objArray);
2714 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2715
2716 Label L_plain_copy, L_checkcast_copy;
2717 // test array classes for subtyping
2718 __ load_klass(rax, dst);
2719 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2720 __ jcc(Assembler::notEqual, L_checkcast_copy);
2721
2722 // Identically typed arrays can be copied without element-wise checks.
2723 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2724 r10, L_failed);
2725
2726 __ lea(from, Address(src, src_pos, TIMES_OOP,
2727 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2728 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2729 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2730 __ movl2ptr(count, r11_length); // length
2731 __ BIND(L_plain_copy);
2732 __ jump(RuntimeAddress(oop_copy_entry));
2733
2734 __ BIND(L_checkcast_copy);
2735 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2736 {
2737 // Before looking at dst.length, make sure dst is also an objArray.
2738 __ cmpl(Address(rax, lh_offset), objArray_lh);
2739 __ jcc(Assembler::notEqual, L_failed);
2740
2741 // It is safe to examine both src.length and dst.length.
2742 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2743 rax, L_failed);
2744
2745 const Register r11_dst_klass = r11;
2746 __ load_klass(r11_dst_klass, dst); // reload
2747
2748 // Marshal the base address arguments now, freeing registers.
2749 __ lea(from, Address(src, src_pos, TIMES_OOP,
2750 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2751 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2752 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2753 __ movl(count, length); // length (reloaded)
2754 Register sco_temp = c_rarg3; // this register is free now
2755 assert_different_registers(from, to, count, sco_temp,
2756 r11_dst_klass, r10_src_klass);
2757 assert_clean_int(count, sco_temp);
2758
2759 // Generate the type check.
2760 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2761 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2762 assert_clean_int(sco_temp, rax);
2763 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2764
2765 // Fetch destination element klass from the ObjArrayKlass header.
2766 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2767 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2768 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2769 assert_clean_int(sco_temp, rax);
2770
2771 // the checkcast_copy loop needs two extra arguments:
2772 assert(c_rarg3 == sco_temp, "#3 already in place");
2773 // Set up arguments for checkcast_copy_entry.
2774 setup_arg_regs(4);
2775 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2776 __ jump(RuntimeAddress(checkcast_copy_entry));
2777 }
2778
2779 __ BIND(L_failed);
2780 __ xorptr(rax, rax);
2781 __ notptr(rax); // return -1
2782 __ leave(); // required for proper stackwalking of RuntimeStub frame
2783 __ ret(0);
2784
2785 return start;
2786 }
2787
2788 void generate_arraycopy_stubs() {
2789 address entry;
2790 address entry_jbyte_arraycopy;
2791 address entry_jshort_arraycopy;
2792 address entry_jint_arraycopy;
2793 address entry_oop_arraycopy;
2794 address entry_jlong_arraycopy;
2795 address entry_checkcast_arraycopy;
2796
2797 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2798 "jbyte_disjoint_arraycopy");
2799 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2800 "jbyte_arraycopy");
2801
2802 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2803 "jshort_disjoint_arraycopy");
2804 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2805 "jshort_arraycopy");
2806
2807 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2808 "jint_disjoint_arraycopy");
2809 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2810 &entry_jint_arraycopy, "jint_arraycopy");
2811
2812 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2813 "jlong_disjoint_arraycopy");
2814 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2815 &entry_jlong_arraycopy, "jlong_arraycopy");
2816
2817
2818 if (UseCompressedOops) {
2819 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2820 "oop_disjoint_arraycopy");
2821 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2822 &entry_oop_arraycopy, "oop_arraycopy");
2823 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2824 "oop_disjoint_arraycopy_uninit",
2825 /*dest_uninitialized*/true);
2826 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2827 NULL, "oop_arraycopy_uninit",
2828 /*dest_uninitialized*/true);
2829 } else {
2830 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2831 "oop_disjoint_arraycopy");
2832 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2833 &entry_oop_arraycopy, "oop_arraycopy");
2834 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2835 "oop_disjoint_arraycopy_uninit",
2836 /*dest_uninitialized*/true);
2837 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2838 NULL, "oop_arraycopy_uninit",
2839 /*dest_uninitialized*/true);
2840 }
2841
2842 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2843 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2844 /*dest_uninitialized*/true);
2845
2846 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2847 entry_jbyte_arraycopy,
2848 entry_jshort_arraycopy,
2849 entry_jint_arraycopy,
2850 entry_jlong_arraycopy);
2851 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2852 entry_jbyte_arraycopy,
2853 entry_jshort_arraycopy,
2854 entry_jint_arraycopy,
2855 entry_oop_arraycopy,
2856 entry_jlong_arraycopy,
2857 entry_checkcast_arraycopy);
2858
2859 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2860 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2861 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2862 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2863 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2864 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2865
2866 // We don't generate specialized code for HeapWord-aligned source
2867 // arrays, so just use the code we've already generated
2868 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2869 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2870
2871 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2872 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2873
2874 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2875 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2876
2877 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2878 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2879
2880 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2881 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2882
2883 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2884 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2885 }
2886
2887 // AES intrinsic stubs
2888 enum {AESBlockSize = 16};
2889
2890 address generate_key_shuffle_mask() {
2891 __ align(16);
2892 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2893 address start = __ pc();
2894 __ emit_data64( 0x0405060700010203, relocInfo::none );
2895 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
2896 return start;
2897 }
2898
2899 address generate_counter_shuffle_mask() {
2900 __ align(16);
2901 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2902 address start = __ pc();
2903 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
2904 __ emit_data64(0x0001020304050607, relocInfo::none);
2905 return start;
2906 }
2907
2908 // Utility routine for loading a 128-bit key word in little endian format
2909 // can optionally specify that the shuffle mask is already in an xmmregister
2910 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2911 __ movdqu(xmmdst, Address(key, offset));
2912 if (xmm_shuf_mask != NULL) {
2913 __ pshufb(xmmdst, xmm_shuf_mask);
2914 } else {
2915 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2916 }
2917 }
2918
2919 // Utility routine for increase 128bit counter (iv in CTR mode)
2920 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
2921 __ pextrq(reg, xmmdst, 0x0);
2922 __ addq(reg, inc_delta);
2923 __ pinsrq(xmmdst, reg, 0x0);
2924 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2925 __ pextrq(reg, xmmdst, 0x01); // Carry
2926 __ addq(reg, 0x01);
2927 __ pinsrq(xmmdst, reg, 0x01); //Carry end
2928 __ BIND(next_block); // next instruction
2929 }
2930
2931 // Arguments:
2932 //
2933 // Inputs:
2934 // c_rarg0 - source byte array address
2935 // c_rarg1 - destination byte array address
2936 // c_rarg2 - K (key) in little endian int array
2937 //
2938 address generate_aescrypt_encryptBlock() {
2939 assert(UseAES, "need AES instructions and misaligned SSE support");
2940 __ align(CodeEntryAlignment);
2941 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2942 Label L_doLast;
2943 address start = __ pc();
2944
2945 const Register from = c_rarg0; // source array address
2946 const Register to = c_rarg1; // destination array address
2947 const Register key = c_rarg2; // key array address
2948 const Register keylen = rax;
2949
2950 const XMMRegister xmm_result = xmm0;
2951 const XMMRegister xmm_key_shuf_mask = xmm1;
2952 // On win64 xmm6-xmm15 must be preserved so don't use them.
2953 const XMMRegister xmm_temp1 = xmm2;
2954 const XMMRegister xmm_temp2 = xmm3;
2955 const XMMRegister xmm_temp3 = xmm4;
2956 const XMMRegister xmm_temp4 = xmm5;
2957
2958 __ enter(); // required for proper stackwalking of RuntimeStub frame
2959
2960 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2961 // context for the registers used, where all instructions below are using 128-bit mode
2962 // On EVEX without VL and BW, these instructions will all be AVX.
2963 if (VM_Version::supports_avx512vlbw()) {
2964 __ movl(rax, 0xffff);
2965 __ kmovql(k1, rax);
2966 }
2967
2968 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2969 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2970
2971 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2972 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2973
2974 // For encryption, the java expanded key ordering is just what we need
2975 // we don't know if the key is aligned, hence not using load-execute form
2976
2977 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2978 __ pxor(xmm_result, xmm_temp1);
2979
2980 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2981 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2982 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2983 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2984
2985 __ aesenc(xmm_result, xmm_temp1);
2986 __ aesenc(xmm_result, xmm_temp2);
2987 __ aesenc(xmm_result, xmm_temp3);
2988 __ aesenc(xmm_result, xmm_temp4);
2989
2990 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2991 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2992 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2993 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2994
2995 __ aesenc(xmm_result, xmm_temp1);
2996 __ aesenc(xmm_result, xmm_temp2);
2997 __ aesenc(xmm_result, xmm_temp3);
2998 __ aesenc(xmm_result, xmm_temp4);
2999
3000 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3001 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3002
3003 __ cmpl(keylen, 44);
3004 __ jccb(Assembler::equal, L_doLast);
3005
3006 __ aesenc(xmm_result, xmm_temp1);
3007 __ aesenc(xmm_result, xmm_temp2);
3008
3009 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3010 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3011
3012 __ cmpl(keylen, 52);
3013 __ jccb(Assembler::equal, L_doLast);
3014
3015 __ aesenc(xmm_result, xmm_temp1);
3016 __ aesenc(xmm_result, xmm_temp2);
3017
3018 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3019 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3020
3021 __ BIND(L_doLast);
3022 __ aesenc(xmm_result, xmm_temp1);
3023 __ aesenclast(xmm_result, xmm_temp2);
3024 __ movdqu(Address(to, 0), xmm_result); // store the result
3025 __ xorptr(rax, rax); // return 0
3026 __ leave(); // required for proper stackwalking of RuntimeStub frame
3027 __ ret(0);
3028
3029 return start;
3030 }
3031
3032
3033 // Arguments:
3034 //
3035 // Inputs:
3036 // c_rarg0 - source byte array address
3037 // c_rarg1 - destination byte array address
3038 // c_rarg2 - K (key) in little endian int array
3039 //
3040 address generate_aescrypt_decryptBlock() {
3041 assert(UseAES, "need AES instructions and misaligned SSE support");
3042 __ align(CodeEntryAlignment);
3043 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3044 Label L_doLast;
3045 address start = __ pc();
3046
3047 const Register from = c_rarg0; // source array address
3048 const Register to = c_rarg1; // destination array address
3049 const Register key = c_rarg2; // key array address
3050 const Register keylen = rax;
3051
3052 const XMMRegister xmm_result = xmm0;
3053 const XMMRegister xmm_key_shuf_mask = xmm1;
3054 // On win64 xmm6-xmm15 must be preserved so don't use them.
3055 const XMMRegister xmm_temp1 = xmm2;
3056 const XMMRegister xmm_temp2 = xmm3;
3057 const XMMRegister xmm_temp3 = xmm4;
3058 const XMMRegister xmm_temp4 = xmm5;
3059
3060 __ enter(); // required for proper stackwalking of RuntimeStub frame
3061
3062 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
3063 // context for the registers used, where all instructions below are using 128-bit mode
3064 // On EVEX without VL and BW, these instructions will all be AVX.
3065 if (VM_Version::supports_avx512vlbw()) {
3066 __ movl(rax, 0xffff);
3067 __ kmovql(k1, rax);
3068 }
3069
3070 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3071 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3072
3073 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3074 __ movdqu(xmm_result, Address(from, 0));
3075
3076 // for decryption java expanded key ordering is rotated one position from what we want
3077 // so we start from 0x10 here and hit 0x00 last
3078 // we don't know if the key is aligned, hence not using load-execute form
3079 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3080 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3081 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3082 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3083
3084 __ pxor (xmm_result, xmm_temp1);
3085 __ aesdec(xmm_result, xmm_temp2);
3086 __ aesdec(xmm_result, xmm_temp3);
3087 __ aesdec(xmm_result, xmm_temp4);
3088
3089 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3090 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3091 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3092 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3093
3094 __ aesdec(xmm_result, xmm_temp1);
3095 __ aesdec(xmm_result, xmm_temp2);
3096 __ aesdec(xmm_result, xmm_temp3);
3097 __ aesdec(xmm_result, xmm_temp4);
3098
3099 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3100 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3101 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3102
3103 __ cmpl(keylen, 44);
3104 __ jccb(Assembler::equal, L_doLast);
3105
3106 __ aesdec(xmm_result, xmm_temp1);
3107 __ aesdec(xmm_result, xmm_temp2);
3108
3109 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3110 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3111
3112 __ cmpl(keylen, 52);
3113 __ jccb(Assembler::equal, L_doLast);
3114
3115 __ aesdec(xmm_result, xmm_temp1);
3116 __ aesdec(xmm_result, xmm_temp2);
3117
3118 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3119 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3120
3121 __ BIND(L_doLast);
3122 __ aesdec(xmm_result, xmm_temp1);
3123 __ aesdec(xmm_result, xmm_temp2);
3124
3125 // for decryption the aesdeclast operation is always on key+0x00
3126 __ aesdeclast(xmm_result, xmm_temp3);
3127 __ movdqu(Address(to, 0), xmm_result); // store the result
3128 __ xorptr(rax, rax); // return 0
3129 __ leave(); // required for proper stackwalking of RuntimeStub frame
3130 __ ret(0);
3131
3132 return start;
3133 }
3134
3135
3136 // Arguments:
3137 //
3138 // Inputs:
3139 // c_rarg0 - source byte array address
3140 // c_rarg1 - destination byte array address
3141 // c_rarg2 - K (key) in little endian int array
3142 // c_rarg3 - r vector byte array address
3143 // c_rarg4 - input length
3144 //
3145 // Output:
3146 // rax - input length
3147 //
3148 address generate_cipherBlockChaining_encryptAESCrypt() {
3149 assert(UseAES, "need AES instructions and misaligned SSE support");
3150 __ align(CodeEntryAlignment);
3151 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3152 address start = __ pc();
3153
3154 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3155 const Register from = c_rarg0; // source array address
3156 const Register to = c_rarg1; // destination array address
3157 const Register key = c_rarg2; // key array address
3158 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3159 // and left with the results of the last encryption block
3160 #ifndef _WIN64
3161 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3162 #else
3163 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3164 const Register len_reg = r11; // pick the volatile windows register
3165 #endif
3166 const Register pos = rax;
3167
3168 // xmm register assignments for the loops below
3169 const XMMRegister xmm_result = xmm0;
3170 const XMMRegister xmm_temp = xmm1;
3171 // keys 0-10 preloaded into xmm2-xmm12
3172 const int XMM_REG_NUM_KEY_FIRST = 2;
3173 const int XMM_REG_NUM_KEY_LAST = 15;
3174 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3175 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3176 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3177 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3178 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3179
3180 __ enter(); // required for proper stackwalking of RuntimeStub frame
3181
3182 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
3183 // context for the registers used, where all instructions below are using 128-bit mode
3184 // On EVEX without VL and BW, these instructions will all be AVX.
3185 if (VM_Version::supports_avx512vlbw()) {
3186 __ movl(rax, 0xffff);
3187 __ kmovql(k1, rax);
3188 }
3189
3190 #ifdef _WIN64
3191 // on win64, fill len_reg from stack position
3192 __ movl(len_reg, len_mem);
3193 #else
3194 __ push(len_reg); // Save
3195 #endif
3196
3197 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3198 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3199 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3200 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3201 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3202 offset += 0x10;
3203 }
3204 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3205
3206 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3207 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3208 __ cmpl(rax, 44);
3209 __ jcc(Assembler::notEqual, L_key_192_256);
3210
3211 // 128 bit code follows here
3212 __ movptr(pos, 0);
3213 __ align(OptoLoopAlignment);
3214
3215 __ BIND(L_loopTop_128);
3216 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3217 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3218 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3219 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3220 __ aesenc(xmm_result, as_XMMRegister(rnum));
3221 }
3222 __ aesenclast(xmm_result, xmm_key10);
3223 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3224 // no need to store r to memory until we exit
3225 __ addptr(pos, AESBlockSize);
3226 __ subptr(len_reg, AESBlockSize);
3227 __ jcc(Assembler::notEqual, L_loopTop_128);
3228
3229 __ BIND(L_exit);
3230 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3231
3232 #ifdef _WIN64
3233 __ movl(rax, len_mem);
3234 #else
3235 __ pop(rax); // return length
3236 #endif
3237 __ leave(); // required for proper stackwalking of RuntimeStub frame
3238 __ ret(0);
3239
3240 __ BIND(L_key_192_256);
3241 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3242 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3243 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3244 __ cmpl(rax, 52);
3245 __ jcc(Assembler::notEqual, L_key_256);
3246
3247 // 192-bit code follows here (could be changed to use more xmm registers)
3248 __ movptr(pos, 0);
3249 __ align(OptoLoopAlignment);
3250
3251 __ BIND(L_loopTop_192);
3252 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3253 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3254 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3255 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3256 __ aesenc(xmm_result, as_XMMRegister(rnum));
3257 }
3258 __ aesenclast(xmm_result, xmm_key12);
3259 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3260 // no need to store r to memory until we exit
3261 __ addptr(pos, AESBlockSize);
3262 __ subptr(len_reg, AESBlockSize);
3263 __ jcc(Assembler::notEqual, L_loopTop_192);
3264 __ jmp(L_exit);
3265
3266 __ BIND(L_key_256);
3267 // 256-bit code follows here (could be changed to use more xmm registers)
3268 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3269 __ movptr(pos, 0);
3270 __ align(OptoLoopAlignment);
3271
3272 __ BIND(L_loopTop_256);
3273 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3274 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3275 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3276 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3277 __ aesenc(xmm_result, as_XMMRegister(rnum));
3278 }
3279 load_key(xmm_temp, key, 0xe0);
3280 __ aesenclast(xmm_result, xmm_temp);
3281 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3282 // no need to store r to memory until we exit
3283 __ addptr(pos, AESBlockSize);
3284 __ subptr(len_reg, AESBlockSize);
3285 __ jcc(Assembler::notEqual, L_loopTop_256);
3286 __ jmp(L_exit);
3287
3288 return start;
3289 }
3290
3291 // Safefetch stubs.
3292 void generate_safefetch(const char* name, int size, address* entry,
3293 address* fault_pc, address* continuation_pc) {
3294 // safefetch signatures:
3295 // int SafeFetch32(int* adr, int errValue);
3296 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3297 //
3298 // arguments:
3299 // c_rarg0 = adr
3300 // c_rarg1 = errValue
3301 //
3302 // result:
3303 // PPC_RET = *adr or errValue
3304
3305 StubCodeMark mark(this, "StubRoutines", name);
3306
3307 // Entry point, pc or function descriptor.
3308 *entry = __ pc();
3309
3310 // Load *adr into c_rarg1, may fault.
3311 *fault_pc = __ pc();
3312 switch (size) {
3313 case 4:
3314 // int32_t
3315 __ movl(c_rarg1, Address(c_rarg0, 0));
3316 break;
3317 case 8:
3318 // int64_t
3319 __ movq(c_rarg1, Address(c_rarg0, 0));
3320 break;
3321 default:
3322 ShouldNotReachHere();
3323 }
3324
3325 // return errValue or *adr
3326 *continuation_pc = __ pc();
3327 __ movq(rax, c_rarg1);
3328 __ ret(0);
3329 }
3330
3331 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3332 // to hide instruction latency
3333 //
3334 // Arguments:
3335 //
3336 // Inputs:
3337 // c_rarg0 - source byte array address
3338 // c_rarg1 - destination byte array address
3339 // c_rarg2 - K (key) in little endian int array
3340 // c_rarg3 - r vector byte array address
3341 // c_rarg4 - input length
3342 //
3343 // Output:
3344 // rax - input length
3345 //
3346 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3347 assert(UseAES, "need AES instructions and misaligned SSE support");
3348 __ align(CodeEntryAlignment);
3349 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3350 address start = __ pc();
3351
3352 const Register from = c_rarg0; // source array address
3353 const Register to = c_rarg1; // destination array address
3354 const Register key = c_rarg2; // key array address
3355 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3356 // and left with the results of the last encryption block
3357 #ifndef _WIN64
3358 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3359 #else
3360 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3361 const Register len_reg = r11; // pick the volatile windows register
3362 #endif
3363 const Register pos = rax;
3364
3365 const int PARALLEL_FACTOR = 4;
3366 const int ROUNDS[3] = { 10, 12, 14 }; // aes rounds for key128, key192, key256
3367
3368 Label L_exit;
3369 Label L_singleBlock_loopTopHead[3]; // 128, 192, 256
3370 Label L_singleBlock_loopTopHead2[3]; // 128, 192, 256
3371 Label L_singleBlock_loopTop[3]; // 128, 192, 256
3372 Label L_multiBlock_loopTopHead[3]; // 128, 192, 256
3373 Label L_multiBlock_loopTop[3]; // 128, 192, 256
3374
3375 // keys 0-10 preloaded into xmm5-xmm15
3376 const int XMM_REG_NUM_KEY_FIRST = 5;
3377 const int XMM_REG_NUM_KEY_LAST = 15;
3378 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3379 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3380
3381 __ enter(); // required for proper stackwalking of RuntimeStub frame
3382
3383 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
3384 // context for the registers used, where all instructions below are using 128-bit mode
3385 // On EVEX without VL and BW, these instructions will all be AVX.
3386 if (VM_Version::supports_avx512vlbw()) {
3387 __ movl(rax, 0xffff);
3388 __ kmovql(k1, rax);
3389 }
3390
3391 #ifdef _WIN64
3392 // on win64, fill len_reg from stack position
3393 __ movl(len_reg, len_mem);
3394 #else
3395 __ push(len_reg); // Save
3396 #endif
3397 __ push(rbx);
3398 // the java expanded key ordering is rotated one position from what we want
3399 // so we start from 0x10 here and hit 0x00 last
3400 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3401 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3402 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3403 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3404 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3405 offset += 0x10;
3406 }
3407 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3408
3409 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3410
3411 // registers holding the four results in the parallelized loop
3412 const XMMRegister xmm_result0 = xmm0;
3413 const XMMRegister xmm_result1 = xmm2;
3414 const XMMRegister xmm_result2 = xmm3;
3415 const XMMRegister xmm_result3 = xmm4;
3416
3417 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3418
3419 __ xorptr(pos, pos);
3420
3421 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3422 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3423 __ cmpl(rbx, 52);
3424 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[1]);
3425 __ cmpl(rbx, 60);
3426 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[2]);
3427
3428 #define DoFour(opc, src_reg) \
3429 __ opc(xmm_result0, src_reg); \
3430 __ opc(xmm_result1, src_reg); \
3431 __ opc(xmm_result2, src_reg); \
3432 __ opc(xmm_result3, src_reg); \
3433
3434 for (int k = 0; k < 3; ++k) {
3435 __ BIND(L_multiBlock_loopTopHead[k]);
3436 if (k != 0) {
3437 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3438 __ jcc(Assembler::less, L_singleBlock_loopTopHead2[k]);
3439 }
3440 if (k == 1) {
3441 __ subptr(rsp, 6 * wordSize);
3442 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3443 load_key(xmm15, key, 0xb0); // 0xb0; 192-bit key goes up to 0xc0
3444 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3445 load_key(xmm1, key, 0xc0); // 0xc0;
3446 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3447 } else if (k == 2) {
3448 __ subptr(rsp, 10 * wordSize);
3449 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3450 load_key(xmm15, key, 0xd0); // 0xd0; 256-bit key goes upto 0xe0
3451 __ movdqu(Address(rsp, 6 * wordSize), xmm15);
3452 load_key(xmm1, key, 0xe0); // 0xe0;
3453 __ movdqu(Address(rsp, 8 * wordSize), xmm1);
3454 load_key(xmm15, key, 0xb0); // 0xb0;
3455 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3456 load_key(xmm1, key, 0xc0); // 0xc0;
3457 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3458 }
3459 __ align(OptoLoopAlignment);
3460 __ BIND(L_multiBlock_loopTop[k]);
3461 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3462 __ jcc(Assembler::less, L_singleBlock_loopTopHead[k]);
3463
3464 if (k != 0) {
3465 __ movdqu(xmm15, Address(rsp, 2 * wordSize));
3466 __ movdqu(xmm1, Address(rsp, 4 * wordSize));
3467 }
3468
3469 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
3470 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3471 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3472 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3473
3474 DoFour(pxor, xmm_key_first);
3475 if (k == 0) {
3476 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
3477 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3478 }
3479 DoFour(aesdeclast, xmm_key_last);
3480 } else if (k == 1) {
3481 for (int rnum = 1; rnum <= ROUNDS[k]-2; rnum++) {
3482 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3483 }
3484 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3485 DoFour(aesdec, xmm1); // key : 0xc0
3486 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3487 DoFour(aesdeclast, xmm_key_last);
3488 } else if (k == 2) {
3489 for (int rnum = 1; rnum <= ROUNDS[k] - 4; rnum++) {
3490 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3491 }
3492 DoFour(aesdec, xmm1); // key : 0xc0
3493 __ movdqu(xmm15, Address(rsp, 6 * wordSize));
3494 __ movdqu(xmm1, Address(rsp, 8 * wordSize));
3495 DoFour(aesdec, xmm15); // key : 0xd0
3496 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3497 DoFour(aesdec, xmm1); // key : 0xe0
3498 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3499 DoFour(aesdeclast, xmm_key_last);
3500 }
3501
3502 // for each result, xor with the r vector of previous cipher block
3503 __ pxor(xmm_result0, xmm_prev_block_cipher);
3504 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3505 __ pxor(xmm_result1, xmm_prev_block_cipher);
3506 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3507 __ pxor(xmm_result2, xmm_prev_block_cipher);
3508 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3509 __ pxor(xmm_result3, xmm_prev_block_cipher);
3510 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
3511 if (k != 0) {
3512 __ movdqu(Address(rvec, 0x00), xmm_prev_block_cipher);
3513 }
3514
3515 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3516 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3517 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3518 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3519
3520 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize);
3521 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize);
3522 __ jmp(L_multiBlock_loopTop[k]);
3523
3524 // registers used in the non-parallelized loops
3525 // xmm register assignments for the loops below
3526 const XMMRegister xmm_result = xmm0;
3527 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3528 const XMMRegister xmm_key11 = xmm3;
3529 const XMMRegister xmm_key12 = xmm4;
3530 const XMMRegister key_tmp = xmm4;
3531
3532 __ BIND(L_singleBlock_loopTopHead[k]);
3533 if (k == 1) {
3534 __ addptr(rsp, 6 * wordSize);
3535 } else if (k == 2) {
3536 __ addptr(rsp, 10 * wordSize);
3537 }
3538 __ cmpptr(len_reg, 0); // any blocks left??
3539 __ jcc(Assembler::equal, L_exit);
3540 __ BIND(L_singleBlock_loopTopHead2[k]);
3541 if (k == 1) {
3542 load_key(xmm_key11, key, 0xb0); // 0xb0; 192-bit key goes upto 0xc0
3543 load_key(xmm_key12, key, 0xc0); // 0xc0; 192-bit key goes upto 0xc0
3544 }
3545 if (k == 2) {
3546 load_key(xmm_key11, key, 0xb0); // 0xb0; 256-bit key goes upto 0xe0
3547 }
3548 __ align(OptoLoopAlignment);
3549 __ BIND(L_singleBlock_loopTop[k]);
3550 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3551 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3552 __ pxor(xmm_result, xmm_key_first); // do the aes dec rounds
3553 for (int rnum = 1; rnum <= 9 ; rnum++) {
3554 __ aesdec(xmm_result, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3555 }
3556 if (k == 1) {
3557 __ aesdec(xmm_result, xmm_key11);
3558 __ aesdec(xmm_result, xmm_key12);
3559 }
3560 if (k == 2) {
3561 __ aesdec(xmm_result, xmm_key11);
3562 load_key(key_tmp, key, 0xc0);
3563 __ aesdec(xmm_result, key_tmp);
3564 load_key(key_tmp, key, 0xd0);
3565 __ aesdec(xmm_result, key_tmp);
3566 load_key(key_tmp, key, 0xe0);
3567 __ aesdec(xmm_result, key_tmp);
3568 }
3569
3570 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3571 __ pxor(xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3572 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3573 // no need to store r to memory until we exit
3574 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3575 __ addptr(pos, AESBlockSize);
3576 __ subptr(len_reg, AESBlockSize);
3577 __ jcc(Assembler::notEqual, L_singleBlock_loopTop[k]);
3578 if (k != 2) {
3579 __ jmp(L_exit);
3580 }
3581 } //for 128/192/256
3582
3583 __ BIND(L_exit);
3584 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3585 __ pop(rbx);
3586 #ifdef _WIN64
3587 __ movl(rax, len_mem);
3588 #else
3589 __ pop(rax); // return length
3590 #endif
3591 __ leave(); // required for proper stackwalking of RuntimeStub frame
3592 __ ret(0);
3593 return start;
3594 }
3595
3596 address generate_upper_word_mask() {
3597 __ align(64);
3598 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3599 address start = __ pc();
3600 __ emit_data64(0x0000000000000000, relocInfo::none);
3601 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
3602 return start;
3603 }
3604
3605 address generate_shuffle_byte_flip_mask() {
3606 __ align(64);
3607 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3608 address start = __ pc();
3609 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3610 __ emit_data64(0x0001020304050607, relocInfo::none);
3611 return start;
3612 }
3613
3614 // ofs and limit are use for multi-block byte array.
3615 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3616 address generate_sha1_implCompress(bool multi_block, const char *name) {
3617 __ align(CodeEntryAlignment);
3618 StubCodeMark mark(this, "StubRoutines", name);
3619 address start = __ pc();
3620
3621 Register buf = c_rarg0;
3622 Register state = c_rarg1;
3623 Register ofs = c_rarg2;
3624 Register limit = c_rarg3;
3625
3626 const XMMRegister abcd = xmm0;
3627 const XMMRegister e0 = xmm1;
3628 const XMMRegister e1 = xmm2;
3629 const XMMRegister msg0 = xmm3;
3630
3631 const XMMRegister msg1 = xmm4;
3632 const XMMRegister msg2 = xmm5;
3633 const XMMRegister msg3 = xmm6;
3634 const XMMRegister shuf_mask = xmm7;
3635
3636 __ enter();
3637
3638 __ subptr(rsp, 4 * wordSize);
3639
3640 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3641 buf, state, ofs, limit, rsp, multi_block);
3642
3643 __ addptr(rsp, 4 * wordSize);
3644
3645 __ leave();
3646 __ ret(0);
3647 return start;
3648 }
3649
3650 address generate_pshuffle_byte_flip_mask() {
3651 __ align(64);
3652 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3653 address start = __ pc();
3654 __ emit_data64(0x0405060700010203, relocInfo::none);
3655 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3656
3657 if (VM_Version::supports_avx2()) {
3658 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
3659 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3660 // _SHUF_00BA
3661 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3662 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3663 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3664 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3665 // _SHUF_DC00
3666 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3667 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3668 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3669 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3670 }
3671
3672 return start;
3673 }
3674
3675 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
3676 address generate_pshuffle_byte_flip_mask_sha512() {
3677 __ align(32);
3678 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
3679 address start = __ pc();
3680 if (VM_Version::supports_avx2()) {
3681 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
3682 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3683 __ emit_data64(0x1011121314151617, relocInfo::none);
3684 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
3685 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
3686 __ emit_data64(0x0000000000000000, relocInfo::none);
3687 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3688 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3689 }
3690
3691 return start;
3692 }
3693
3694 // ofs and limit are use for multi-block byte array.
3695 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3696 address generate_sha256_implCompress(bool multi_block, const char *name) {
3697 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
3698 __ align(CodeEntryAlignment);
3699 StubCodeMark mark(this, "StubRoutines", name);
3700 address start = __ pc();
3701
3702 Register buf = c_rarg0;
3703 Register state = c_rarg1;
3704 Register ofs = c_rarg2;
3705 Register limit = c_rarg3;
3706
3707 const XMMRegister msg = xmm0;
3708 const XMMRegister state0 = xmm1;
3709 const XMMRegister state1 = xmm2;
3710 const XMMRegister msgtmp0 = xmm3;
3711
3712 const XMMRegister msgtmp1 = xmm4;
3713 const XMMRegister msgtmp2 = xmm5;
3714 const XMMRegister msgtmp3 = xmm6;
3715 const XMMRegister msgtmp4 = xmm7;
3716
3717 const XMMRegister shuf_mask = xmm8;
3718
3719 __ enter();
3720
3721 __ subptr(rsp, 4 * wordSize);
3722
3723 if (VM_Version::supports_sha()) {
3724 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3725 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3726 } else if (VM_Version::supports_avx2()) {
3727 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3728 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3729 }
3730 __ addptr(rsp, 4 * wordSize);
3731 __ vzeroupper();
3732 __ leave();
3733 __ ret(0);
3734 return start;
3735 }
3736
3737 address generate_sha512_implCompress(bool multi_block, const char *name) {
3738 assert(VM_Version::supports_avx2(), "");
3739 assert(VM_Version::supports_bmi2(), "");
3740 __ align(CodeEntryAlignment);
3741 StubCodeMark mark(this, "StubRoutines", name);
3742 address start = __ pc();
3743
3744 Register buf = c_rarg0;
3745 Register state = c_rarg1;
3746 Register ofs = c_rarg2;
3747 Register limit = c_rarg3;
3748
3749 const XMMRegister msg = xmm0;
3750 const XMMRegister state0 = xmm1;
3751 const XMMRegister state1 = xmm2;
3752 const XMMRegister msgtmp0 = xmm3;
3753 const XMMRegister msgtmp1 = xmm4;
3754 const XMMRegister msgtmp2 = xmm5;
3755 const XMMRegister msgtmp3 = xmm6;
3756 const XMMRegister msgtmp4 = xmm7;
3757
3758 const XMMRegister shuf_mask = xmm8;
3759
3760 __ enter();
3761
3762 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3763 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3764
3765 __ vzeroupper();
3766 __ leave();
3767 __ ret(0);
3768 return start;
3769 }
3770
3771 // This is a version of CTR/AES crypt which does 6 blocks in a loop at a time
3772 // to hide instruction latency
3773 //
3774 // Arguments:
3775 //
3776 // Inputs:
3777 // c_rarg0 - source byte array address
3778 // c_rarg1 - destination byte array address
3779 // c_rarg2 - K (key) in little endian int array
3780 // c_rarg3 - counter vector byte array address
3781 // Linux
3782 // c_rarg4 - input length
3783 // c_rarg5 - saved encryptedCounter start
3784 // rbp + 6 * wordSize - saved used length
3785 // Windows
3786 // rbp + 6 * wordSize - input length
3787 // rbp + 7 * wordSize - saved encryptedCounter start
3788 // rbp + 8 * wordSize - saved used length
3789 //
3790 // Output:
3791 // rax - input length
3792 //
3793 address generate_counterMode_AESCrypt_Parallel() {
3794 assert(UseAES, "need AES instructions and misaligned SSE support");
3795 __ align(CodeEntryAlignment);
3796 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
3797 address start = __ pc();
3798 const Register from = c_rarg0; // source array address
3799 const Register to = c_rarg1; // destination array address
3800 const Register key = c_rarg2; // key array address
3801 const Register counter = c_rarg3; // counter byte array initialized from counter array address
3802 // and updated with the incremented counter in the end
3803 #ifndef _WIN64
3804 const Register len_reg = c_rarg4;
3805 const Register saved_encCounter_start = c_rarg5;
3806 const Register used_addr = r10;
3807 const Address used_mem(rbp, 2 * wordSize);
3808 const Register used = r11;
3809 #else
3810 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3811 const Address saved_encCounter_mem(rbp, 7 * wordSize); // length is on stack on Win64
3812 const Address used_mem(rbp, 8 * wordSize); // length is on stack on Win64
3813 const Register len_reg = r10; // pick the first volatile windows register
3814 const Register saved_encCounter_start = r11;
3815 const Register used_addr = r13;
3816 const Register used = r14;
3817 #endif
3818 const Register pos = rax;
3819
3820 const int PARALLEL_FACTOR = 6;
3821 const XMMRegister xmm_counter_shuf_mask = xmm0;
3822 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3823 const XMMRegister xmm_curr_counter = xmm2;
3824
3825 const XMMRegister xmm_key_tmp0 = xmm3;
3826 const XMMRegister xmm_key_tmp1 = xmm4;
3827
3828 // registers holding the four results in the parallelized loop
3829 const XMMRegister xmm_result0 = xmm5;
3830 const XMMRegister xmm_result1 = xmm6;
3831 const XMMRegister xmm_result2 = xmm7;
3832 const XMMRegister xmm_result3 = xmm8;
3833 const XMMRegister xmm_result4 = xmm9;
3834 const XMMRegister xmm_result5 = xmm10;
3835
3836 const XMMRegister xmm_from0 = xmm11;
3837 const XMMRegister xmm_from1 = xmm12;
3838 const XMMRegister xmm_from2 = xmm13;
3839 const XMMRegister xmm_from3 = xmm14; //the last one is xmm14. we have to preserve it on WIN64.
3840 const XMMRegister xmm_from4 = xmm3; //reuse xmm3~4. Because xmm_key_tmp0~1 are useless when loading input text
3841 const XMMRegister xmm_from5 = xmm4;
3842
3843 //for key_128, key_192, key_256
3844 const int rounds[3] = {10, 12, 14};
3845 Label L_exit_preLoop, L_preLoop_start;
3846 Label L_multiBlock_loopTop[3];
3847 Label L_singleBlockLoopTop[3];
3848 Label L__incCounter[3][6]; //for 6 blocks
3849 Label L__incCounter_single[3]; //for single block, key128, key192, key256
3850 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];
3851 Label L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
3852
3853 Label L_exit;
3854
3855 __ enter(); // required for proper stackwalking of RuntimeStub frame
3856
3857 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
3858 // context for the registers used, where all instructions below are using 128-bit mode
3859 // On EVEX without VL and BW, these instructions will all be AVX.
3860 if (VM_Version::supports_avx512vlbw()) {
3861 __ movl(rax, 0xffff);
3862 __ kmovql(k1, rax);
3863 }
3864
3865 #ifdef _WIN64
3866 // allocate spill slots for r13, r14
3867 enum {
3868 saved_r13_offset,
3869 saved_r14_offset
3870 };
3871 __ subptr(rsp, 2 * wordSize);
3872 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
3873 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
3874
3875 // on win64, fill len_reg from stack position
3876 __ movl(len_reg, len_mem);
3877 __ movptr(saved_encCounter_start, saved_encCounter_mem);
3878 __ movptr(used_addr, used_mem);
3879 __ movl(used, Address(used_addr, 0));
3880 #else
3881 __ push(len_reg); // Save
3882 __ movptr(used_addr, used_mem);
3883 __ movl(used, Address(used_addr, 0));
3884 #endif
3885
3886 __ push(rbx); // Save RBX
3887 __ movdqu(xmm_curr_counter, Address(counter, 0x00)); // initialize counter with initial counter
3888 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()), pos); // pos as scratch
3889 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled
3890 __ movptr(pos, 0);
3891
3892 // Use the partially used encrpyted counter from last invocation
3893 __ BIND(L_preLoop_start);
3894 __ cmpptr(used, 16);
3895 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
3896 __ cmpptr(len_reg, 0);
3897 __ jcc(Assembler::lessEqual, L_exit_preLoop);
3898 __ movb(rbx, Address(saved_encCounter_start, used));
3899 __ xorb(rbx, Address(from, pos));
3900 __ movb(Address(to, pos), rbx);
3901 __ addptr(pos, 1);
3902 __ addptr(used, 1);
3903 __ subptr(len_reg, 1);
3904
3905 __ jmp(L_preLoop_start);
3906
3907 __ BIND(L_exit_preLoop);
3908 __ movl(Address(used_addr, 0), used);
3909
3910 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
3911 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()), rbx); // rbx as scratch
3912 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3913 __ cmpl(rbx, 52);
3914 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
3915 __ cmpl(rbx, 60);
3916 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
3917
3918 #define CTR_DoSix(opc, src_reg) \
3919 __ opc(xmm_result0, src_reg); \
3920 __ opc(xmm_result1, src_reg); \
3921 __ opc(xmm_result2, src_reg); \
3922 __ opc(xmm_result3, src_reg); \
3923 __ opc(xmm_result4, src_reg); \
3924 __ opc(xmm_result5, src_reg);
3925
3926 // k == 0 : generate code for key_128
3927 // k == 1 : generate code for key_192
3928 // k == 2 : generate code for key_256
3929 for (int k = 0; k < 3; ++k) {
3930 //multi blocks starts here
3931 __ align(OptoLoopAlignment);
3932 __ BIND(L_multiBlock_loopTop[k]);
3933 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
3934 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
3935 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
3936
3937 //load, then increase counters
3938 CTR_DoSix(movdqa, xmm_curr_counter);
3939 inc_counter(rbx, xmm_result1, 0x01, L__incCounter[k][0]);
3940 inc_counter(rbx, xmm_result2, 0x02, L__incCounter[k][1]);
3941 inc_counter(rbx, xmm_result3, 0x03, L__incCounter[k][2]);
3942 inc_counter(rbx, xmm_result4, 0x04, L__incCounter[k][3]);
3943 inc_counter(rbx, xmm_result5, 0x05, L__incCounter[k][4]);
3944 inc_counter(rbx, xmm_curr_counter, 0x06, L__incCounter[k][5]);
3945 CTR_DoSix(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
3946 CTR_DoSix(pxor, xmm_key_tmp0); //PXOR with Round 0 key
3947
3948 //load two ROUND_KEYs at a time
3949 for (int i = 1; i < rounds[k]; ) {
3950 load_key(xmm_key_tmp1, key, (0x10 * i), xmm_key_shuf_mask);
3951 load_key(xmm_key_tmp0, key, (0x10 * (i+1)), xmm_key_shuf_mask);
3952 CTR_DoSix(aesenc, xmm_key_tmp1);
3953 i++;
3954 if (i != rounds[k]) {
3955 CTR_DoSix(aesenc, xmm_key_tmp0);
3956 } else {
3957 CTR_DoSix(aesenclast, xmm_key_tmp0);
3958 }
3959 i++;
3960 }
3961
3962 // get next PARALLEL_FACTOR blocks into xmm_result registers
3963 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3964 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3965 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3966 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3967 __ movdqu(xmm_from4, Address(from, pos, Address::times_1, 4 * AESBlockSize));
3968 __ movdqu(xmm_from5, Address(from, pos, Address::times_1, 5 * AESBlockSize));
3969
3970 __ pxor(xmm_result0, xmm_from0);
3971 __ pxor(xmm_result1, xmm_from1);
3972 __ pxor(xmm_result2, xmm_from2);
3973 __ pxor(xmm_result3, xmm_from3);
3974 __ pxor(xmm_result4, xmm_from4);
3975 __ pxor(xmm_result5, xmm_from5);
3976
3977 // store 6 results into the next 64 bytes of output
3978 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
3979 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3980 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3981 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3982 __ movdqu(Address(to, pos, Address::times_1, 4 * AESBlockSize), xmm_result4);
3983 __ movdqu(Address(to, pos, Address::times_1, 5 * AESBlockSize), xmm_result5);
3984
3985 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
3986 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
3987 __ jmp(L_multiBlock_loopTop[k]);
3988
3989 // singleBlock starts here
3990 __ align(OptoLoopAlignment);
3991 __ BIND(L_singleBlockLoopTop[k]);
3992 __ cmpptr(len_reg, 0);
3993 __ jcc(Assembler::lessEqual, L_exit);
3994 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
3995 __ movdqa(xmm_result0, xmm_curr_counter);
3996 inc_counter(rbx, xmm_curr_counter, 0x01, L__incCounter_single[k]);
3997 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
3998 __ pxor(xmm_result0, xmm_key_tmp0);
3999 for (int i = 1; i < rounds[k]; i++) {
4000 load_key(xmm_key_tmp0, key, (0x10 * i), xmm_key_shuf_mask);
4001 __ aesenc(xmm_result0, xmm_key_tmp0);
4002 }
4003 load_key(xmm_key_tmp0, key, (rounds[k] * 0x10), xmm_key_shuf_mask);
4004 __ aesenclast(xmm_result0, xmm_key_tmp0);
4005 __ cmpptr(len_reg, AESBlockSize);
4006 __ jcc(Assembler::less, L_processTail_insr[k]);
4007 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4008 __ pxor(xmm_result0, xmm_from0);
4009 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4010 __ addptr(pos, AESBlockSize);
4011 __ subptr(len_reg, AESBlockSize);
4012 __ jmp(L_singleBlockLoopTop[k]);
4013 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
4014 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
4015 __ testptr(len_reg, 8);
4016 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
4017 __ subptr(pos,8);
4018 __ pinsrq(xmm_from0, Address(from, pos), 0);
4019 __ BIND(L_processTail_4_insr[k]);
4020 __ testptr(len_reg, 4);
4021 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
4022 __ subptr(pos,4);
4023 __ pslldq(xmm_from0, 4);
4024 __ pinsrd(xmm_from0, Address(from, pos), 0);
4025 __ BIND(L_processTail_2_insr[k]);
4026 __ testptr(len_reg, 2);
4027 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
4028 __ subptr(pos, 2);
4029 __ pslldq(xmm_from0, 2);
4030 __ pinsrw(xmm_from0, Address(from, pos), 0);
4031 __ BIND(L_processTail_1_insr[k]);
4032 __ testptr(len_reg, 1);
4033 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
4034 __ subptr(pos, 1);
4035 __ pslldq(xmm_from0, 1);
4036 __ pinsrb(xmm_from0, Address(from, pos), 0);
4037 __ BIND(L_processTail_exit_insr[k]);
4038
4039 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
4040 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
4041
4042 __ testptr(len_reg, 8);
4043 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
4044 __ pextrq(Address(to, pos), xmm_result0, 0);
4045 __ psrldq(xmm_result0, 8);
4046 __ addptr(pos, 8);
4047 __ BIND(L_processTail_4_extr[k]);
4048 __ testptr(len_reg, 4);
4049 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
4050 __ pextrd(Address(to, pos), xmm_result0, 0);
4051 __ psrldq(xmm_result0, 4);
4052 __ addptr(pos, 4);
4053 __ BIND(L_processTail_2_extr[k]);
4054 __ testptr(len_reg, 2);
4055 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
4056 __ pextrw(Address(to, pos), xmm_result0, 0);
4057 __ psrldq(xmm_result0, 2);
4058 __ addptr(pos, 2);
4059 __ BIND(L_processTail_1_extr[k]);
4060 __ testptr(len_reg, 1);
4061 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
4062 __ pextrb(Address(to, pos), xmm_result0, 0);
4063
4064 __ BIND(L_processTail_exit_extr[k]);
4065 __ movl(Address(used_addr, 0), len_reg);
4066 __ jmp(L_exit);
4067
4068 }
4069
4070 __ BIND(L_exit);
4071 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
4072 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
4073 __ pop(rbx); // pop the saved RBX.
4074 #ifdef _WIN64
4075 __ movl(rax, len_mem);
4076 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
4077 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
4078 __ addptr(rsp, 2 * wordSize);
4079 #else
4080 __ pop(rax); // return 'len'
4081 #endif
4082 __ leave(); // required for proper stackwalking of RuntimeStub frame
4083 __ ret(0);
4084 return start;
4085 }
4086
4087 void roundDec(XMMRegister xmm_reg) {
4088 __ vaesdec(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4089 __ vaesdec(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4090 __ vaesdec(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4091 __ vaesdec(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4092 __ vaesdec(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4093 __ vaesdec(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4094 __ vaesdec(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4095 __ vaesdec(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4096 }
4097
4098 void roundDeclast(XMMRegister xmm_reg) {
4099 __ vaesdeclast(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4100 __ vaesdeclast(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4101 __ vaesdeclast(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4102 __ vaesdeclast(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4103 __ vaesdeclast(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4104 __ vaesdeclast(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4105 __ vaesdeclast(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4106 __ vaesdeclast(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4107 }
4108
4109 void ev_load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask = NULL) {
4110 __ movdqu(xmmdst, Address(key, offset));
4111 if (xmm_shuf_mask != NULL) {
4112 __ pshufb(xmmdst, xmm_shuf_mask);
4113 } else {
4114 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4115 }
4116 __ evshufi64x2(xmmdst, xmmdst, xmmdst, 0x0, Assembler::AVX_512bit);
4117
4118 }
4119
4120 address generate_cipherBlockChaining_decryptVectorAESCrypt() {
4121 assert(VM_Version::supports_vaes(), "need AES instructions and misaligned SSE support");
4122 __ align(CodeEntryAlignment);
4123 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
4124 address start = __ pc();
4125
4126 const Register from = c_rarg0; // source array address
4127 const Register to = c_rarg1; // destination array address
4128 const Register key = c_rarg2; // key array address
4129 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
4130 // and left with the results of the last encryption block
4131 #ifndef _WIN64
4132 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
4133 #else
4134 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4135 const Register len_reg = r11; // pick the volatile windows register
4136 #endif
4137
4138 Label Loop, Loop1, L_128, L_256, L_192, KEY_192, KEY_256, Loop2, Lcbc_dec_rem_loop,
4139 Lcbc_dec_rem_last, Lcbc_dec_ret, Lcbc_dec_rem, Lcbc_exit;
4140
4141 __ enter();
4142
4143 #ifdef _WIN64
4144 // on win64, fill len_reg from stack position
4145 __ movl(len_reg, len_mem);
4146 #else
4147 __ push(len_reg); // Save
4148 #endif
4149 __ push(rbx);
4150 __ vzeroupper();
4151
4152 // Temporary variable declaration for swapping key bytes
4153 const XMMRegister xmm_key_shuf_mask = xmm1;
4154 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4155
4156 // Calculate number of rounds from key size: 44 for 10-rounds, 52 for 12-rounds, 60 for 14-rounds
4157 const Register rounds = rbx;
4158 __ movl(rounds, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4159
4160 const XMMRegister IV = xmm0;
4161 // Load IV and broadcast value to 512-bits
4162 __ evbroadcasti64x2(IV, Address(rvec, 0), Assembler::AVX_512bit);
4163
4164 // Temporary variables for storing round keys
4165 const XMMRegister RK0 = xmm30;
4166 const XMMRegister RK1 = xmm9;
4167 const XMMRegister RK2 = xmm18;
4168 const XMMRegister RK3 = xmm19;
4169 const XMMRegister RK4 = xmm20;
4170 const XMMRegister RK5 = xmm21;
4171 const XMMRegister RK6 = xmm22;
4172 const XMMRegister RK7 = xmm23;
4173 const XMMRegister RK8 = xmm24;
4174 const XMMRegister RK9 = xmm25;
4175 const XMMRegister RK10 = xmm26;
4176
4177 // Load and shuffle key
4178 // the java expanded key ordering is rotated one position from what we want
4179 // so we start from 1*16 here and hit 0*16 last
4180 ev_load_key(RK1, key, 1 * 16, xmm_key_shuf_mask);
4181 ev_load_key(RK2, key, 2 * 16, xmm_key_shuf_mask);
4182 ev_load_key(RK3, key, 3 * 16, xmm_key_shuf_mask);
4183 ev_load_key(RK4, key, 4 * 16, xmm_key_shuf_mask);
4184 ev_load_key(RK5, key, 5 * 16, xmm_key_shuf_mask);
4185 ev_load_key(RK6, key, 6 * 16, xmm_key_shuf_mask);
4186 ev_load_key(RK7, key, 7 * 16, xmm_key_shuf_mask);
4187 ev_load_key(RK8, key, 8 * 16, xmm_key_shuf_mask);
4188 ev_load_key(RK9, key, 9 * 16, xmm_key_shuf_mask);
4189 ev_load_key(RK10, key, 10 * 16, xmm_key_shuf_mask);
4190 ev_load_key(RK0, key, 0*16, xmm_key_shuf_mask);
4191
4192 // Variables for storing source cipher text
4193 const XMMRegister S0 = xmm10;
4194 const XMMRegister S1 = xmm11;
4195 const XMMRegister S2 = xmm12;
4196 const XMMRegister S3 = xmm13;
4197 const XMMRegister S4 = xmm14;
4198 const XMMRegister S5 = xmm15;
4199 const XMMRegister S6 = xmm16;
4200 const XMMRegister S7 = xmm17;
4201
4202 // Variables for storing decrypted text
4203 const XMMRegister B0 = xmm1;
4204 const XMMRegister B1 = xmm2;
4205 const XMMRegister B2 = xmm3;
4206 const XMMRegister B3 = xmm4;
4207 const XMMRegister B4 = xmm5;
4208 const XMMRegister B5 = xmm6;
4209 const XMMRegister B6 = xmm7;
4210 const XMMRegister B7 = xmm8;
4211
4212 __ cmpl(rounds, 44);
4213 __ jcc(Assembler::greater, KEY_192);
4214 __ jmp(Loop);
4215
4216 __ BIND(KEY_192);
4217 const XMMRegister RK11 = xmm27;
4218 const XMMRegister RK12 = xmm28;
4219 ev_load_key(RK11, key, 11*16, xmm_key_shuf_mask);
4220 ev_load_key(RK12, key, 12*16, xmm_key_shuf_mask);
4221
4222 __ cmpl(rounds, 52);
4223 __ jcc(Assembler::greater, KEY_256);
4224 __ jmp(Loop);
4225
4226 __ BIND(KEY_256);
4227 const XMMRegister RK13 = xmm29;
4228 const XMMRegister RK14 = xmm31;
4229 ev_load_key(RK13, key, 13*16, xmm_key_shuf_mask);
4230 ev_load_key(RK14, key, 14*16, xmm_key_shuf_mask);
4231
4232 __ BIND(Loop);
4233 __ cmpl(len_reg, 512);
4234 __ jcc(Assembler::below, Lcbc_dec_rem);
4235 __ BIND(Loop1);
4236 __ subl(len_reg, 512);
4237 __ evmovdquq(S0, Address(from, 0 * 64), Assembler::AVX_512bit);
4238 __ evmovdquq(S1, Address(from, 1 * 64), Assembler::AVX_512bit);
4239 __ evmovdquq(S2, Address(from, 2 * 64), Assembler::AVX_512bit);
4240 __ evmovdquq(S3, Address(from, 3 * 64), Assembler::AVX_512bit);
4241 __ evmovdquq(S4, Address(from, 4 * 64), Assembler::AVX_512bit);
4242 __ evmovdquq(S5, Address(from, 5 * 64), Assembler::AVX_512bit);
4243 __ evmovdquq(S6, Address(from, 6 * 64), Assembler::AVX_512bit);
4244 __ evmovdquq(S7, Address(from, 7 * 64), Assembler::AVX_512bit);
4245 __ leaq(from, Address(from, 8 * 64));
4246
4247 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4248 __ evpxorq(B1, S1, RK1, Assembler::AVX_512bit);
4249 __ evpxorq(B2, S2, RK1, Assembler::AVX_512bit);
4250 __ evpxorq(B3, S3, RK1, Assembler::AVX_512bit);
4251 __ evpxorq(B4, S4, RK1, Assembler::AVX_512bit);
4252 __ evpxorq(B5, S5, RK1, Assembler::AVX_512bit);
4253 __ evpxorq(B6, S6, RK1, Assembler::AVX_512bit);
4254 __ evpxorq(B7, S7, RK1, Assembler::AVX_512bit);
4255
4256 __ evalignq(IV, S0, IV, 0x06);
4257 __ evalignq(S0, S1, S0, 0x06);
4258 __ evalignq(S1, S2, S1, 0x06);
4259 __ evalignq(S2, S3, S2, 0x06);
4260 __ evalignq(S3, S4, S3, 0x06);
4261 __ evalignq(S4, S5, S4, 0x06);
4262 __ evalignq(S5, S6, S5, 0x06);
4263 __ evalignq(S6, S7, S6, 0x06);
4264
4265 roundDec(RK2);
4266 roundDec(RK3);
4267 roundDec(RK4);
4268 roundDec(RK5);
4269 roundDec(RK6);
4270 roundDec(RK7);
4271 roundDec(RK8);
4272 roundDec(RK9);
4273 roundDec(RK10);
4274
4275 __ cmpl(rounds, 44);
4276 __ jcc(Assembler::belowEqual, L_128);
4277 roundDec(RK11);
4278 roundDec(RK12);
4279
4280 __ cmpl(rounds, 52);
4281 __ jcc(Assembler::belowEqual, L_192);
4282 roundDec(RK13);
4283 roundDec(RK14);
4284
4285 __ BIND(L_256);
4286 roundDeclast(RK0);
4287 __ jmp(Loop2);
4288
4289 __ BIND(L_128);
4290 roundDeclast(RK0);
4291 __ jmp(Loop2);
4292
4293 __ BIND(L_192);
4294 roundDeclast(RK0);
4295
4296 __ BIND(Loop2);
4297 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4298 __ evpxorq(B1, B1, S0, Assembler::AVX_512bit);
4299 __ evpxorq(B2, B2, S1, Assembler::AVX_512bit);
4300 __ evpxorq(B3, B3, S2, Assembler::AVX_512bit);
4301 __ evpxorq(B4, B4, S3, Assembler::AVX_512bit);
4302 __ evpxorq(B5, B5, S4, Assembler::AVX_512bit);
4303 __ evpxorq(B6, B6, S5, Assembler::AVX_512bit);
4304 __ evpxorq(B7, B7, S6, Assembler::AVX_512bit);
4305 __ evmovdquq(IV, S7, Assembler::AVX_512bit);
4306
4307 __ evmovdquq(Address(to, 0 * 64), B0, Assembler::AVX_512bit);
4308 __ evmovdquq(Address(to, 1 * 64), B1, Assembler::AVX_512bit);
4309 __ evmovdquq(Address(to, 2 * 64), B2, Assembler::AVX_512bit);
4310 __ evmovdquq(Address(to, 3 * 64), B3, Assembler::AVX_512bit);
4311 __ evmovdquq(Address(to, 4 * 64), B4, Assembler::AVX_512bit);
4312 __ evmovdquq(Address(to, 5 * 64), B5, Assembler::AVX_512bit);
4313 __ evmovdquq(Address(to, 6 * 64), B6, Assembler::AVX_512bit);
4314 __ evmovdquq(Address(to, 7 * 64), B7, Assembler::AVX_512bit);
4315 __ leaq(to, Address(to, 8 * 64));
4316 __ jmp(Loop);
4317
4318 __ BIND(Lcbc_dec_rem);
4319 __ evshufi64x2(IV, IV, IV, 0x03, Assembler::AVX_512bit);
4320
4321 __ BIND(Lcbc_dec_rem_loop);
4322 __ subl(len_reg, 16);
4323 __ jcc(Assembler::carrySet, Lcbc_dec_ret);
4324
4325 __ movdqu(S0, Address(from, 0));
4326 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4327 __ vaesdec(B0, B0, RK2, Assembler::AVX_512bit);
4328 __ vaesdec(B0, B0, RK3, Assembler::AVX_512bit);
4329 __ vaesdec(B0, B0, RK4, Assembler::AVX_512bit);
4330 __ vaesdec(B0, B0, RK5, Assembler::AVX_512bit);
4331 __ vaesdec(B0, B0, RK6, Assembler::AVX_512bit);
4332 __ vaesdec(B0, B0, RK7, Assembler::AVX_512bit);
4333 __ vaesdec(B0, B0, RK8, Assembler::AVX_512bit);
4334 __ vaesdec(B0, B0, RK9, Assembler::AVX_512bit);
4335 __ vaesdec(B0, B0, RK10, Assembler::AVX_512bit);
4336 __ cmpl(rounds, 44);
4337 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4338
4339 __ vaesdec(B0, B0, RK11, Assembler::AVX_512bit);
4340 __ vaesdec(B0, B0, RK12, Assembler::AVX_512bit);
4341 __ cmpl(rounds, 52);
4342 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4343
4344 __ vaesdec(B0, B0, RK13, Assembler::AVX_512bit);
4345 __ vaesdec(B0, B0, RK14, Assembler::AVX_512bit);
4346
4347 __ BIND(Lcbc_dec_rem_last);
4348 __ vaesdeclast(B0, B0, RK0, Assembler::AVX_512bit);
4349
4350 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4351 __ evmovdquq(IV, S0, Assembler::AVX_512bit);
4352 __ movdqu(Address(to, 0), B0);
4353 __ leaq(from, Address(from, 16));
4354 __ leaq(to, Address(to, 16));
4355 __ jmp(Lcbc_dec_rem_loop);
4356
4357 __ BIND(Lcbc_dec_ret);
4358 __ movdqu(Address(rvec, 0), IV);
4359
4360 // Zero out the round keys
4361 __ evpxorq(RK0, RK0, RK0, Assembler::AVX_512bit);
4362 __ evpxorq(RK1, RK1, RK1, Assembler::AVX_512bit);
4363 __ evpxorq(RK2, RK2, RK2, Assembler::AVX_512bit);
4364 __ evpxorq(RK3, RK3, RK3, Assembler::AVX_512bit);
4365 __ evpxorq(RK4, RK4, RK4, Assembler::AVX_512bit);
4366 __ evpxorq(RK5, RK5, RK5, Assembler::AVX_512bit);
4367 __ evpxorq(RK6, RK6, RK6, Assembler::AVX_512bit);
4368 __ evpxorq(RK7, RK7, RK7, Assembler::AVX_512bit);
4369 __ evpxorq(RK8, RK8, RK8, Assembler::AVX_512bit);
4370 __ evpxorq(RK9, RK9, RK9, Assembler::AVX_512bit);
4371 __ evpxorq(RK10, RK10, RK10, Assembler::AVX_512bit);
4372 __ cmpl(rounds, 44);
4373 __ jcc(Assembler::belowEqual, Lcbc_exit);
4374 __ evpxorq(RK11, RK11, RK11, Assembler::AVX_512bit);
4375 __ evpxorq(RK12, RK12, RK12, Assembler::AVX_512bit);
4376 __ cmpl(rounds, 52);
4377 __ jcc(Assembler::belowEqual, Lcbc_exit);
4378 __ evpxorq(RK13, RK13, RK13, Assembler::AVX_512bit);
4379 __ evpxorq(RK14, RK14, RK14, Assembler::AVX_512bit);
4380
4381 __ BIND(Lcbc_exit);
4382 __ pop(rbx);
4383 #ifdef _WIN64
4384 __ movl(rax, len_mem);
4385 #else
4386 __ pop(rax); // return length
4387 #endif
4388 __ leave(); // required for proper stackwalking of RuntimeStub frame
4389 __ ret(0);
4390 return start;
4391 }
4392
4393 // byte swap x86 long
4394 address generate_ghash_long_swap_mask() {
4395 __ align(CodeEntryAlignment);
4396 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
4397 address start = __ pc();
4398 __ emit_data64(0x0f0e0d0c0b0a0908, relocInfo::none );
4399 __ emit_data64(0x0706050403020100, relocInfo::none );
4400 return start;
4401 }
4402
4403 // byte swap x86 byte array
4404 address generate_ghash_byte_swap_mask() {
4405 __ align(CodeEntryAlignment);
4406 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
4407 address start = __ pc();
4408 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none );
4409 __ emit_data64(0x0001020304050607, relocInfo::none );
4410 return start;
4411 }
4412
4413 /* Single and multi-block ghash operations */
4414 address generate_ghash_processBlocks() {
4415 __ align(CodeEntryAlignment);
4416 Label L_ghash_loop, L_exit;
4417 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4418 address start = __ pc();
4419
4420 const Register state = c_rarg0;
4421 const Register subkeyH = c_rarg1;
4422 const Register data = c_rarg2;
4423 const Register blocks = c_rarg3;
4424
4425 const XMMRegister xmm_temp0 = xmm0;
4426 const XMMRegister xmm_temp1 = xmm1;
4427 const XMMRegister xmm_temp2 = xmm2;
4428 const XMMRegister xmm_temp3 = xmm3;
4429 const XMMRegister xmm_temp4 = xmm4;
4430 const XMMRegister xmm_temp5 = xmm5;
4431 const XMMRegister xmm_temp6 = xmm6;
4432 const XMMRegister xmm_temp7 = xmm7;
4433 const XMMRegister xmm_temp8 = xmm8;
4434 const XMMRegister xmm_temp9 = xmm9;
4435 const XMMRegister xmm_temp10 = xmm10;
4436
4437 __ enter();
4438
4439 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
4440 // context for the registers used, where all instructions below are using 128-bit mode
4441 // On EVEX without VL and BW, these instructions will all be AVX.
4442 if (VM_Version::supports_avx512vlbw()) {
4443 __ movl(rax, 0xffff);
4444 __ kmovql(k1, rax);
4445 }
4446
4447 __ movdqu(xmm_temp10, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
4448
4449 __ movdqu(xmm_temp0, Address(state, 0));
4450 __ pshufb(xmm_temp0, xmm_temp10);
4451
4452
4453 __ BIND(L_ghash_loop);
4454 __ movdqu(xmm_temp2, Address(data, 0));
4455 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
4456
4457 __ movdqu(xmm_temp1, Address(subkeyH, 0));
4458 __ pshufb(xmm_temp1, xmm_temp10);
4459
4460 __ pxor(xmm_temp0, xmm_temp2);
4461
4462 //
4463 // Multiply with the hash key
4464 //
4465 __ movdqu(xmm_temp3, xmm_temp0);
4466 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
4467 __ movdqu(xmm_temp4, xmm_temp0);
4468 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
4469
4470 __ movdqu(xmm_temp5, xmm_temp0);
4471 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
4472 __ movdqu(xmm_temp6, xmm_temp0);
4473 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
4474
4475 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
4476
4477 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
4478 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
4479 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
4480 __ pxor(xmm_temp3, xmm_temp5);
4481 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
4482 // of the carry-less multiplication of
4483 // xmm0 by xmm1.
4484
4485 // We shift the result of the multiplication by one bit position
4486 // to the left to cope for the fact that the bits are reversed.
4487 __ movdqu(xmm_temp7, xmm_temp3);
4488 __ movdqu(xmm_temp8, xmm_temp6);
4489 __ pslld(xmm_temp3, 1);
4490 __ pslld(xmm_temp6, 1);
4491 __ psrld(xmm_temp7, 31);
4492 __ psrld(xmm_temp8, 31);
4493 __ movdqu(xmm_temp9, xmm_temp7);
4494 __ pslldq(xmm_temp8, 4);
4495 __ pslldq(xmm_temp7, 4);
4496 __ psrldq(xmm_temp9, 12);
4497 __ por(xmm_temp3, xmm_temp7);
4498 __ por(xmm_temp6, xmm_temp8);
4499 __ por(xmm_temp6, xmm_temp9);
4500
4501 //
4502 // First phase of the reduction
4503 //
4504 // Move xmm3 into xmm7, xmm8, xmm9 in order to perform the shifts
4505 // independently.
4506 __ movdqu(xmm_temp7, xmm_temp3);
4507 __ movdqu(xmm_temp8, xmm_temp3);
4508 __ movdqu(xmm_temp9, xmm_temp3);
4509 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
4510 __ pslld(xmm_temp8, 30); // packed right shift shifting << 30
4511 __ pslld(xmm_temp9, 25); // packed right shift shifting << 25
4512 __ pxor(xmm_temp7, xmm_temp8); // xor the shifted versions
4513 __ pxor(xmm_temp7, xmm_temp9);
4514 __ movdqu(xmm_temp8, xmm_temp7);
4515 __ pslldq(xmm_temp7, 12);
4516 __ psrldq(xmm_temp8, 4);
4517 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
4518
4519 //
4520 // Second phase of the reduction
4521 //
4522 // Make 3 copies of xmm3 in xmm2, xmm4, xmm5 for doing these
4523 // shift operations.
4524 __ movdqu(xmm_temp2, xmm_temp3);
4525 __ movdqu(xmm_temp4, xmm_temp3);
4526 __ movdqu(xmm_temp5, xmm_temp3);
4527 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
4528 __ psrld(xmm_temp4, 2); // packed left shifting >> 2
4529 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
4530 __ pxor(xmm_temp2, xmm_temp4); // xor the shifted versions
4531 __ pxor(xmm_temp2, xmm_temp5);
4532 __ pxor(xmm_temp2, xmm_temp8);
4533 __ pxor(xmm_temp3, xmm_temp2);
4534 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
4535
4536 __ decrement(blocks);
4537 __ jcc(Assembler::zero, L_exit);
4538 __ movdqu(xmm_temp0, xmm_temp6);
4539 __ addptr(data, 16);
4540 __ jmp(L_ghash_loop);
4541
4542 __ BIND(L_exit);
4543 __ pshufb(xmm_temp6, xmm_temp10); // Byte swap 16-byte result
4544 __ movdqu(Address(state, 0), xmm_temp6); // store the result
4545 __ leave();
4546 __ ret(0);
4547 return start;
4548 }
4549
4550 //base64 character set
4551 address base64_charset_addr() {
4552 __ align(CodeEntryAlignment);
4553 StubCodeMark mark(this, "StubRoutines", "base64_charset");
4554 address start = __ pc();
4555 __ emit_data64(0x0000004200000041, relocInfo::none);
4556 __ emit_data64(0x0000004400000043, relocInfo::none);
4557 __ emit_data64(0x0000004600000045, relocInfo::none);
4558 __ emit_data64(0x0000004800000047, relocInfo::none);
4559 __ emit_data64(0x0000004a00000049, relocInfo::none);
4560 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4561 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4562 __ emit_data64(0x000000500000004f, relocInfo::none);
4563 __ emit_data64(0x0000005200000051, relocInfo::none);
4564 __ emit_data64(0x0000005400000053, relocInfo::none);
4565 __ emit_data64(0x0000005600000055, relocInfo::none);
4566 __ emit_data64(0x0000005800000057, relocInfo::none);
4567 __ emit_data64(0x0000005a00000059, relocInfo::none);
4568 __ emit_data64(0x0000006200000061, relocInfo::none);
4569 __ emit_data64(0x0000006400000063, relocInfo::none);
4570 __ emit_data64(0x0000006600000065, relocInfo::none);
4571 __ emit_data64(0x0000006800000067, relocInfo::none);
4572 __ emit_data64(0x0000006a00000069, relocInfo::none);
4573 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4574 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4575 __ emit_data64(0x000000700000006f, relocInfo::none);
4576 __ emit_data64(0x0000007200000071, relocInfo::none);
4577 __ emit_data64(0x0000007400000073, relocInfo::none);
4578 __ emit_data64(0x0000007600000075, relocInfo::none);
4579 __ emit_data64(0x0000007800000077, relocInfo::none);
4580 __ emit_data64(0x0000007a00000079, relocInfo::none);
4581 __ emit_data64(0x0000003100000030, relocInfo::none);
4582 __ emit_data64(0x0000003300000032, relocInfo::none);
4583 __ emit_data64(0x0000003500000034, relocInfo::none);
4584 __ emit_data64(0x0000003700000036, relocInfo::none);
4585 __ emit_data64(0x0000003900000038, relocInfo::none);
4586 __ emit_data64(0x0000002f0000002b, relocInfo::none);
4587 return start;
4588 }
4589
4590 //base64 url character set
4591 address base64url_charset_addr() {
4592 __ align(CodeEntryAlignment);
4593 StubCodeMark mark(this, "StubRoutines", "base64url_charset");
4594 address start = __ pc();
4595 __ emit_data64(0x0000004200000041, relocInfo::none);
4596 __ emit_data64(0x0000004400000043, relocInfo::none);
4597 __ emit_data64(0x0000004600000045, relocInfo::none);
4598 __ emit_data64(0x0000004800000047, relocInfo::none);
4599 __ emit_data64(0x0000004a00000049, relocInfo::none);
4600 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4601 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4602 __ emit_data64(0x000000500000004f, relocInfo::none);
4603 __ emit_data64(0x0000005200000051, relocInfo::none);
4604 __ emit_data64(0x0000005400000053, relocInfo::none);
4605 __ emit_data64(0x0000005600000055, relocInfo::none);
4606 __ emit_data64(0x0000005800000057, relocInfo::none);
4607 __ emit_data64(0x0000005a00000059, relocInfo::none);
4608 __ emit_data64(0x0000006200000061, relocInfo::none);
4609 __ emit_data64(0x0000006400000063, relocInfo::none);
4610 __ emit_data64(0x0000006600000065, relocInfo::none);
4611 __ emit_data64(0x0000006800000067, relocInfo::none);
4612 __ emit_data64(0x0000006a00000069, relocInfo::none);
4613 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4614 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4615 __ emit_data64(0x000000700000006f, relocInfo::none);
4616 __ emit_data64(0x0000007200000071, relocInfo::none);
4617 __ emit_data64(0x0000007400000073, relocInfo::none);
4618 __ emit_data64(0x0000007600000075, relocInfo::none);
4619 __ emit_data64(0x0000007800000077, relocInfo::none);
4620 __ emit_data64(0x0000007a00000079, relocInfo::none);
4621 __ emit_data64(0x0000003100000030, relocInfo::none);
4622 __ emit_data64(0x0000003300000032, relocInfo::none);
4623 __ emit_data64(0x0000003500000034, relocInfo::none);
4624 __ emit_data64(0x0000003700000036, relocInfo::none);
4625 __ emit_data64(0x0000003900000038, relocInfo::none);
4626 __ emit_data64(0x0000005f0000002d, relocInfo::none);
4627
4628 return start;
4629 }
4630
4631 address base64_bswap_mask_addr() {
4632 __ align(CodeEntryAlignment);
4633 StubCodeMark mark(this, "StubRoutines", "bswap_mask_base64");
4634 address start = __ pc();
4635 __ emit_data64(0x0504038002010080, relocInfo::none);
4636 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4637 __ emit_data64(0x0908078006050480, relocInfo::none);
4638 __ emit_data64(0x0f0e0d800c0b0a80, relocInfo::none);
4639 __ emit_data64(0x0605048003020180, relocInfo::none);
4640 __ emit_data64(0x0c0b0a8009080780, relocInfo::none);
4641 __ emit_data64(0x0504038002010080, relocInfo::none);
4642 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4643
4644 return start;
4645 }
4646
4647 address base64_right_shift_mask_addr() {
4648 __ align(CodeEntryAlignment);
4649 StubCodeMark mark(this, "StubRoutines", "right_shift_mask");
4650 address start = __ pc();
4651 __ emit_data64(0x0006000400020000, relocInfo::none);
4652 __ emit_data64(0x0006000400020000, relocInfo::none);
4653 __ emit_data64(0x0006000400020000, relocInfo::none);
4654 __ emit_data64(0x0006000400020000, relocInfo::none);
4655 __ emit_data64(0x0006000400020000, relocInfo::none);
4656 __ emit_data64(0x0006000400020000, relocInfo::none);
4657 __ emit_data64(0x0006000400020000, relocInfo::none);
4658 __ emit_data64(0x0006000400020000, relocInfo::none);
4659
4660 return start;
4661 }
4662
4663 address base64_left_shift_mask_addr() {
4664 __ align(CodeEntryAlignment);
4665 StubCodeMark mark(this, "StubRoutines", "left_shift_mask");
4666 address start = __ pc();
4667 __ emit_data64(0x0000000200040000, relocInfo::none);
4668 __ emit_data64(0x0000000200040000, relocInfo::none);
4669 __ emit_data64(0x0000000200040000, relocInfo::none);
4670 __ emit_data64(0x0000000200040000, relocInfo::none);
4671 __ emit_data64(0x0000000200040000, relocInfo::none);
4672 __ emit_data64(0x0000000200040000, relocInfo::none);
4673 __ emit_data64(0x0000000200040000, relocInfo::none);
4674 __ emit_data64(0x0000000200040000, relocInfo::none);
4675
4676 return start;
4677 }
4678
4679 address base64_and_mask_addr() {
4680 __ align(CodeEntryAlignment);
4681 StubCodeMark mark(this, "StubRoutines", "and_mask");
4682 address start = __ pc();
4683 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4684 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4685 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4686 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4687 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4688 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4689 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4690 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4691 return start;
4692 }
4693
4694 address base64_gather_mask_addr() {
4695 __ align(CodeEntryAlignment);
4696 StubCodeMark mark(this, "StubRoutines", "gather_mask");
4697 address start = __ pc();
4698 __ emit_data64(0xffffffffffffffff, relocInfo::none);
4699 return start;
4700 }
4701
4702 // Code for generating Base64 encoding.
4703 // Intrinsic function prototype in Base64.java:
4704 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL) {
4705 address generate_base64_encodeBlock() {
4706 __ align(CodeEntryAlignment);
4707 StubCodeMark mark(this, "StubRoutines", "implEncode");
4708 address start = __ pc();
4709 __ enter();
4710
4711 // Save callee-saved registers before using them
4712 __ push(r12);
4713 __ push(r13);
4714 __ push(r14);
4715 __ push(r15);
4716 __ push(rbx);
4717
4718 // arguments
4719 const Register source = c_rarg0; // Source Array
4720 const Register start_offset = c_rarg1; // start offset
4721 const Register end_offset = c_rarg2; // end offset
4722 const Register dest = c_rarg3; // destination array
4723
4724 #ifndef _WIN64
4725 const Register dp = c_rarg4; // Position for writing to dest array
4726 const Register isURL = c_rarg5;// Base64 or URL character set
4727 #else
4728 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
4729 const Address isURL_mem(rbp, 7 * wordSize);
4730 const Register isURL = r10; // pick the volatile windows register
4731 const Register dp = r12;
4732 __ movl(dp, dp_mem);
4733 __ movl(isURL, isURL_mem);
4734 #endif
4735
4736 const Register length = r14;
4737 Label L_process80, L_process32, L_process3, L_exit, L_processdata;
4738
4739 // calculate length from offsets
4740 __ movl(length, end_offset);
4741 __ subl(length, start_offset);
4742 __ cmpl(length, 0);
4743 __ jcc(Assembler::lessEqual, L_exit);
4744
4745 // Save k1 value in rbx
4746 __ kmovql(rbx, k1);
4747 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_charset_addr()));
4748 // check if base64 charset(isURL=0) or base64 url charset(isURL=1) needs to be loaded
4749 __ cmpl(isURL, 0);
4750 __ jcc(Assembler::equal, L_processdata);
4751 __ lea(r11, ExternalAddress(StubRoutines::x86::base64url_charset_addr()));
4752
4753 // load masks required for encoding data
4754 __ BIND(L_processdata);
4755 __ movdqu(xmm16, ExternalAddress(StubRoutines::x86::base64_gather_mask_addr()));
4756 // Set 64 bits of K register.
4757 __ evpcmpeqb(k1, xmm16, xmm16, Assembler::AVX_512bit);
4758 __ evmovdquq(xmm12, ExternalAddress(StubRoutines::x86::base64_bswap_mask_addr()), Assembler::AVX_256bit, r13);
4759 __ evmovdquq(xmm13, ExternalAddress(StubRoutines::x86::base64_right_shift_mask_addr()), Assembler::AVX_512bit, r13);
4760 __ evmovdquq(xmm14, ExternalAddress(StubRoutines::x86::base64_left_shift_mask_addr()), Assembler::AVX_512bit, r13);
4761 __ evmovdquq(xmm15, ExternalAddress(StubRoutines::x86::base64_and_mask_addr()), Assembler::AVX_512bit, r13);
4762
4763 // Vector Base64 implementation, producing 96 bytes of encoded data
4764 __ BIND(L_process80);
4765 __ cmpl(length, 80);
4766 __ jcc(Assembler::below, L_process32);
4767 __ evmovdquq(xmm0, Address(source, start_offset, Address::times_1, 0), Assembler::AVX_256bit);
4768 __ evmovdquq(xmm1, Address(source, start_offset, Address::times_1, 24), Assembler::AVX_256bit);
4769 __ evmovdquq(xmm2, Address(source, start_offset, Address::times_1, 48), Assembler::AVX_256bit);
4770
4771 //permute the input data in such a manner that we have continuity of the source
4772 __ vpermq(xmm3, xmm0, 148, Assembler::AVX_256bit);
4773 __ vpermq(xmm4, xmm1, 148, Assembler::AVX_256bit);
4774 __ vpermq(xmm5, xmm2, 148, Assembler::AVX_256bit);
4775
4776 //shuffle input and group 3 bytes of data and to it add 0 as the 4th byte.
4777 //we can deal with 12 bytes at a time in a 128 bit register
4778 __ vpshufb(xmm3, xmm3, xmm12, Assembler::AVX_256bit);
4779 __ vpshufb(xmm4, xmm4, xmm12, Assembler::AVX_256bit);
4780 __ vpshufb(xmm5, xmm5, xmm12, Assembler::AVX_256bit);
4781
4782 //convert byte to word. Each 128 bit register will have 6 bytes for processing
4783 __ vpmovzxbw(xmm3, xmm3, Assembler::AVX_512bit);
4784 __ vpmovzxbw(xmm4, xmm4, Assembler::AVX_512bit);
4785 __ vpmovzxbw(xmm5, xmm5, Assembler::AVX_512bit);
4786
4787 // Extract bits in the following pattern 6, 4+2, 2+4, 6 to convert 3, 8 bit numbers to 4, 6 bit numbers
4788 __ evpsrlvw(xmm0, xmm3, xmm13, Assembler::AVX_512bit);
4789 __ evpsrlvw(xmm1, xmm4, xmm13, Assembler::AVX_512bit);
4790 __ evpsrlvw(xmm2, xmm5, xmm13, Assembler::AVX_512bit);
4791
4792 __ evpsllvw(xmm3, xmm3, xmm14, Assembler::AVX_512bit);
4793 __ evpsllvw(xmm4, xmm4, xmm14, Assembler::AVX_512bit);
4794 __ evpsllvw(xmm5, xmm5, xmm14, Assembler::AVX_512bit);
4795
4796 __ vpsrlq(xmm0, xmm0, 8, Assembler::AVX_512bit);
4797 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
4798 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
4799
4800 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4801 __ vpsllq(xmm4, xmm4, 8, Assembler::AVX_512bit);
4802 __ vpsllq(xmm5, xmm5, 8, Assembler::AVX_512bit);
4803
4804 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
4805 __ vpandq(xmm4, xmm4, xmm15, Assembler::AVX_512bit);
4806 __ vpandq(xmm5, xmm5, xmm15, Assembler::AVX_512bit);
4807
4808 // Get the final 4*6 bits base64 encoding
4809 __ vporq(xmm3, xmm3, xmm0, Assembler::AVX_512bit);
4810 __ vporq(xmm4, xmm4, xmm1, Assembler::AVX_512bit);
4811 __ vporq(xmm5, xmm5, xmm2, Assembler::AVX_512bit);
4812
4813 // Shift
4814 __ vpsrlq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4815 __ vpsrlq(xmm4, xmm4, 8, Assembler::AVX_512bit);
4816 __ vpsrlq(xmm5, xmm5, 8, Assembler::AVX_512bit);
4817
4818 // look up 6 bits in the base64 character set to fetch the encoding
4819 // we are converting word to dword as gather instructions need dword indices for looking up encoding
4820 __ vextracti64x4(xmm6, xmm3, 0);
4821 __ vpmovzxwd(xmm0, xmm6, Assembler::AVX_512bit);
4822 __ vextracti64x4(xmm6, xmm3, 1);
4823 __ vpmovzxwd(xmm1, xmm6, Assembler::AVX_512bit);
4824
4825 __ vextracti64x4(xmm6, xmm4, 0);
4826 __ vpmovzxwd(xmm2, xmm6, Assembler::AVX_512bit);
4827 __ vextracti64x4(xmm6, xmm4, 1);
4828 __ vpmovzxwd(xmm3, xmm6, Assembler::AVX_512bit);
4829
4830 __ vextracti64x4(xmm4, xmm5, 0);
4831 __ vpmovzxwd(xmm6, xmm4, Assembler::AVX_512bit);
4832
4833 __ vextracti64x4(xmm4, xmm5, 1);
4834 __ vpmovzxwd(xmm7, xmm4, Assembler::AVX_512bit);
4835
4836 __ kmovql(k2, k1);
4837 __ evpgatherdd(xmm4, k2, Address(r11, xmm0, Address::times_4, 0), Assembler::AVX_512bit);
4838 __ kmovql(k2, k1);
4839 __ evpgatherdd(xmm5, k2, Address(r11, xmm1, Address::times_4, 0), Assembler::AVX_512bit);
4840 __ kmovql(k2, k1);
4841 __ evpgatherdd(xmm8, k2, Address(r11, xmm2, Address::times_4, 0), Assembler::AVX_512bit);
4842 __ kmovql(k2, k1);
4843 __ evpgatherdd(xmm9, k2, Address(r11, xmm3, Address::times_4, 0), Assembler::AVX_512bit);
4844 __ kmovql(k2, k1);
4845 __ evpgatherdd(xmm10, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
4846 __ kmovql(k2, k1);
4847 __ evpgatherdd(xmm11, k2, Address(r11, xmm7, Address::times_4, 0), Assembler::AVX_512bit);
4848
4849 //Down convert dword to byte. Final output is 16*6 = 96 bytes long
4850 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm4, Assembler::AVX_512bit);
4851 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm5, Assembler::AVX_512bit);
4852 __ evpmovdb(Address(dest, dp, Address::times_1, 32), xmm8, Assembler::AVX_512bit);
4853 __ evpmovdb(Address(dest, dp, Address::times_1, 48), xmm9, Assembler::AVX_512bit);
4854 __ evpmovdb(Address(dest, dp, Address::times_1, 64), xmm10, Assembler::AVX_512bit);
4855 __ evpmovdb(Address(dest, dp, Address::times_1, 80), xmm11, Assembler::AVX_512bit);
4856
4857 __ addq(dest, 96);
4858 __ addq(source, 72);
4859 __ subq(length, 72);
4860 __ jmp(L_process80);
4861
4862 // Vector Base64 implementation generating 32 bytes of encoded data
4863 __ BIND(L_process32);
4864 __ cmpl(length, 32);
4865 __ jcc(Assembler::below, L_process3);
4866 __ evmovdquq(xmm0, Address(source, start_offset), Assembler::AVX_256bit);
4867 __ vpermq(xmm0, xmm0, 148, Assembler::AVX_256bit);
4868 __ vpshufb(xmm6, xmm0, xmm12, Assembler::AVX_256bit);
4869 __ vpmovzxbw(xmm6, xmm6, Assembler::AVX_512bit);
4870 __ evpsrlvw(xmm2, xmm6, xmm13, Assembler::AVX_512bit);
4871 __ evpsllvw(xmm3, xmm6, xmm14, Assembler::AVX_512bit);
4872
4873 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
4874 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4875 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
4876 __ vporq(xmm1, xmm2, xmm3, Assembler::AVX_512bit);
4877 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
4878 __ vextracti64x4(xmm9, xmm1, 0);
4879 __ vpmovzxwd(xmm6, xmm9, Assembler::AVX_512bit);
4880 __ vextracti64x4(xmm9, xmm1, 1);
4881 __ vpmovzxwd(xmm5, xmm9, Assembler::AVX_512bit);
4882 __ kmovql(k2, k1);
4883 __ evpgatherdd(xmm8, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
4884 __ kmovql(k2, k1);
4885 __ evpgatherdd(xmm10, k2, Address(r11, xmm5, Address::times_4, 0), Assembler::AVX_512bit);
4886 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm8, Assembler::AVX_512bit);
4887 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm10, Assembler::AVX_512bit);
4888 __ subq(length, 24);
4889 __ addq(dest, 32);
4890 __ addq(source, 24);
4891 __ jmp(L_process32);
4892
4893 // Scalar data processing takes 3 bytes at a time and produces 4 bytes of encoded data
4894 /* This code corresponds to the scalar version of the following snippet in Base64.java
4895 ** int bits = (src[sp0++] & 0xff) << 16 |(src[sp0++] & 0xff) << 8 |(src[sp0++] & 0xff);
4896 ** dst[dp0++] = (byte)base64[(bits >> > 18) & 0x3f];
4897 ** dst[dp0++] = (byte)base64[(bits >> > 12) & 0x3f];
4898 ** dst[dp0++] = (byte)base64[(bits >> > 6) & 0x3f];
4899 ** dst[dp0++] = (byte)base64[bits & 0x3f];*/
4900 __ BIND(L_process3);
4901 __ cmpl(length, 3);
4902 __ jcc(Assembler::below, L_exit);
4903 // Read 1 byte at a time
4904 __ movzbl(rax, Address(source, start_offset));
4905 __ shll(rax, 0x10);
4906 __ movl(r15, rax);
4907 __ movzbl(rax, Address(source, start_offset, Address::times_1, 1));
4908 __ shll(rax, 0x8);
4909 __ movzwl(rax, rax);
4910 __ orl(r15, rax);
4911 __ movzbl(rax, Address(source, start_offset, Address::times_1, 2));
4912 __ orl(rax, r15);
4913 // Save 3 bytes read in r15
4914 __ movl(r15, rax);
4915 __ shrl(rax, 0x12);
4916 __ andl(rax, 0x3f);
4917 // rax contains the index, r11 contains base64 lookup table
4918 __ movb(rax, Address(r11, rax, Address::times_4));
4919 // Write the encoded byte to destination
4920 __ movb(Address(dest, dp, Address::times_1, 0), rax);
4921 __ movl(rax, r15);
4922 __ shrl(rax, 0xc);
4923 __ andl(rax, 0x3f);
4924 __ movb(rax, Address(r11, rax, Address::times_4));
4925 __ movb(Address(dest, dp, Address::times_1, 1), rax);
4926 __ movl(rax, r15);
4927 __ shrl(rax, 0x6);
4928 __ andl(rax, 0x3f);
4929 __ movb(rax, Address(r11, rax, Address::times_4));
4930 __ movb(Address(dest, dp, Address::times_1, 2), rax);
4931 __ movl(rax, r15);
4932 __ andl(rax, 0x3f);
4933 __ movb(rax, Address(r11, rax, Address::times_4));
4934 __ movb(Address(dest, dp, Address::times_1, 3), rax);
4935 __ subl(length, 3);
4936 __ addq(dest, 4);
4937 __ addq(source, 3);
4938 __ jmp(L_process3);
4939 __ BIND(L_exit);
4940 // restore k1 register value
4941 __ kmovql(k1, rbx);
4942 __ pop(rbx);
4943 __ pop(r15);
4944 __ pop(r14);
4945 __ pop(r13);
4946 __ pop(r12);
4947 __ leave();
4948 __ ret(0);
4949 return start;
4950 }
4951
4952 /**
4953 * Arguments:
4954 *
4955 * Inputs:
4956 * c_rarg0 - int crc
4957 * c_rarg1 - byte* buf
4958 * c_rarg2 - int length
4959 *
4960 * Ouput:
4961 * rax - int crc result
4962 */
4963 address generate_updateBytesCRC32() {
4964 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
4965
4966 __ align(CodeEntryAlignment);
4967 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
4968
4969 address start = __ pc();
4970 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
4971 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
4972 // rscratch1: r10
4973 const Register crc = c_rarg0; // crc
4974 const Register buf = c_rarg1; // source java byte array address
4975 const Register len = c_rarg2; // length
4976 const Register table = c_rarg3; // crc_table address (reuse register)
4977 const Register tmp = r11;
4978 assert_different_registers(crc, buf, len, table, tmp, rax);
4979
4980 BLOCK_COMMENT("Entry:");
4981 __ enter(); // required for proper stackwalking of RuntimeStub frame
4982
4983 __ kernel_crc32(crc, buf, len, table, tmp);
4984
4985 __ movl(rax, crc);
4986 __ vzeroupper();
4987 __ leave(); // required for proper stackwalking of RuntimeStub frame
4988 __ ret(0);
4989
4990 return start;
4991 }
4992
4993 /**
4994 * Arguments:
4995 *
4996 * Inputs:
4997 * c_rarg0 - int crc
4998 * c_rarg1 - byte* buf
4999 * c_rarg2 - long length
5000 * c_rarg3 - table_start - optional (present only when doing a library_call,
5001 * not used by x86 algorithm)
5002 *
5003 * Ouput:
5004 * rax - int crc result
5005 */
5006 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
5007 assert(UseCRC32CIntrinsics, "need SSE4_2");
5008 __ align(CodeEntryAlignment);
5009 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
5010 address start = __ pc();
5011 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
5012 //Windows RCX RDX R8 R9 none none XMM0..XMM3
5013 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
5014 const Register crc = c_rarg0; // crc
5015 const Register buf = c_rarg1; // source java byte array address
5016 const Register len = c_rarg2; // length
5017 const Register a = rax;
5018 const Register j = r9;
5019 const Register k = r10;
5020 const Register l = r11;
5021 #ifdef _WIN64
5022 const Register y = rdi;
5023 const Register z = rsi;
5024 #else
5025 const Register y = rcx;
5026 const Register z = r8;
5027 #endif
5028 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
5029
5030 BLOCK_COMMENT("Entry:");
5031 __ enter(); // required for proper stackwalking of RuntimeStub frame
5032 #ifdef _WIN64
5033 __ push(y);
5034 __ push(z);
5035 #endif
5036 __ crc32c_ipl_alg2_alt2(crc, buf, len,
5037 a, j, k,
5038 l, y, z,
5039 c_farg0, c_farg1, c_farg2,
5040 is_pclmulqdq_supported);
5041 __ movl(rax, crc);
5042 #ifdef _WIN64
5043 __ pop(z);
5044 __ pop(y);
5045 #endif
5046 __ vzeroupper();
5047 __ leave(); // required for proper stackwalking of RuntimeStub frame
5048 __ ret(0);
5049
5050 return start;
5051 }
5052
5053 /**
5054 * Arguments:
5055 *
5056 * Input:
5057 * c_rarg0 - x address
5058 * c_rarg1 - x length
5059 * c_rarg2 - y address
5060 * c_rarg3 - y length
5061 * not Win64
5062 * c_rarg4 - z address
5063 * c_rarg5 - z length
5064 * Win64
5065 * rsp+40 - z address
5066 * rsp+48 - z length
5067 */
5068 address generate_multiplyToLen() {
5069 __ align(CodeEntryAlignment);
5070 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
5071
5072 address start = __ pc();
5073 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5074 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5075 const Register x = rdi;
5076 const Register xlen = rax;
5077 const Register y = rsi;
5078 const Register ylen = rcx;
5079 const Register z = r8;
5080 const Register zlen = r11;
5081
5082 // Next registers will be saved on stack in multiply_to_len().
5083 const Register tmp1 = r12;
5084 const Register tmp2 = r13;
5085 const Register tmp3 = r14;
5086 const Register tmp4 = r15;
5087 const Register tmp5 = rbx;
5088
5089 BLOCK_COMMENT("Entry:");
5090 __ enter(); // required for proper stackwalking of RuntimeStub frame
5091
5092 #ifndef _WIN64
5093 __ movptr(zlen, r9); // Save r9 in r11 - zlen
5094 #endif
5095 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
5096 // ylen => rcx, z => r8, zlen => r11
5097 // r9 and r10 may be used to save non-volatile registers
5098 #ifdef _WIN64
5099 // last 2 arguments (#4, #5) are on stack on Win64
5100 __ movptr(z, Address(rsp, 6 * wordSize));
5101 __ movptr(zlen, Address(rsp, 7 * wordSize));
5102 #endif
5103
5104 __ movptr(xlen, rsi);
5105 __ movptr(y, rdx);
5106 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
5107
5108 restore_arg_regs();
5109
5110 __ leave(); // required for proper stackwalking of RuntimeStub frame
5111 __ ret(0);
5112
5113 return start;
5114 }
5115
5116 /**
5117 * Arguments:
5118 *
5119 * Input:
5120 * c_rarg0 - obja address
5121 * c_rarg1 - objb address
5122 * c_rarg3 - length length
5123 * c_rarg4 - scale log2_array_indxscale
5124 *
5125 * Output:
5126 * rax - int >= mismatched index, < 0 bitwise complement of tail
5127 */
5128 address generate_vectorizedMismatch() {
5129 __ align(CodeEntryAlignment);
5130 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
5131 address start = __ pc();
5132
5133 BLOCK_COMMENT("Entry:");
5134 __ enter();
5135
5136 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5137 const Register scale = c_rarg0; //rcx, will exchange with r9
5138 const Register objb = c_rarg1; //rdx
5139 const Register length = c_rarg2; //r8
5140 const Register obja = c_rarg3; //r9
5141 __ xchgq(obja, scale); //now obja and scale contains the correct contents
5142
5143 const Register tmp1 = r10;
5144 const Register tmp2 = r11;
5145 #endif
5146 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5147 const Register obja = c_rarg0; //U:rdi
5148 const Register objb = c_rarg1; //U:rsi
5149 const Register length = c_rarg2; //U:rdx
5150 const Register scale = c_rarg3; //U:rcx
5151 const Register tmp1 = r8;
5152 const Register tmp2 = r9;
5153 #endif
5154 const Register result = rax; //return value
5155 const XMMRegister vec0 = xmm0;
5156 const XMMRegister vec1 = xmm1;
5157 const XMMRegister vec2 = xmm2;
5158
5159 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
5160
5161 __ vzeroupper();
5162 __ leave();
5163 __ ret(0);
5164
5165 return start;
5166 }
5167
5168 /**
5169 * Arguments:
5170 *
5171 // Input:
5172 // c_rarg0 - x address
5173 // c_rarg1 - x length
5174 // c_rarg2 - z address
5175 // c_rarg3 - z lenth
5176 *
5177 */
5178 address generate_squareToLen() {
5179
5180 __ align(CodeEntryAlignment);
5181 StubCodeMark mark(this, "StubRoutines", "squareToLen");
5182
5183 address start = __ pc();
5184 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5185 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
5186 const Register x = rdi;
5187 const Register len = rsi;
5188 const Register z = r8;
5189 const Register zlen = rcx;
5190
5191 const Register tmp1 = r12;
5192 const Register tmp2 = r13;
5193 const Register tmp3 = r14;
5194 const Register tmp4 = r15;
5195 const Register tmp5 = rbx;
5196
5197 BLOCK_COMMENT("Entry:");
5198 __ enter(); // required for proper stackwalking of RuntimeStub frame
5199
5200 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
5201 // zlen => rcx
5202 // r9 and r10 may be used to save non-volatile registers
5203 __ movptr(r8, rdx);
5204 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5205
5206 restore_arg_regs();
5207
5208 __ leave(); // required for proper stackwalking of RuntimeStub frame
5209 __ ret(0);
5210
5211 return start;
5212 }
5213
5214 /**
5215 * Arguments:
5216 *
5217 * Input:
5218 * c_rarg0 - out address
5219 * c_rarg1 - in address
5220 * c_rarg2 - offset
5221 * c_rarg3 - len
5222 * not Win64
5223 * c_rarg4 - k
5224 * Win64
5225 * rsp+40 - k
5226 */
5227 address generate_mulAdd() {
5228 __ align(CodeEntryAlignment);
5229 StubCodeMark mark(this, "StubRoutines", "mulAdd");
5230
5231 address start = __ pc();
5232 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5233 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5234 const Register out = rdi;
5235 const Register in = rsi;
5236 const Register offset = r11;
5237 const Register len = rcx;
5238 const Register k = r8;
5239
5240 // Next registers will be saved on stack in mul_add().
5241 const Register tmp1 = r12;
5242 const Register tmp2 = r13;
5243 const Register tmp3 = r14;
5244 const Register tmp4 = r15;
5245 const Register tmp5 = rbx;
5246
5247 BLOCK_COMMENT("Entry:");
5248 __ enter(); // required for proper stackwalking of RuntimeStub frame
5249
5250 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
5251 // len => rcx, k => r8
5252 // r9 and r10 may be used to save non-volatile registers
5253 #ifdef _WIN64
5254 // last argument is on stack on Win64
5255 __ movl(k, Address(rsp, 6 * wordSize));
5256 #endif
5257 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
5258 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5259
5260 restore_arg_regs();
5261
5262 __ leave(); // required for proper stackwalking of RuntimeStub frame
5263 __ ret(0);
5264
5265 return start;
5266 }
5267
5268 address generate_libmExp() {
5269 StubCodeMark mark(this, "StubRoutines", "libmExp");
5270
5271 address start = __ pc();
5272
5273 const XMMRegister x0 = xmm0;
5274 const XMMRegister x1 = xmm1;
5275 const XMMRegister x2 = xmm2;
5276 const XMMRegister x3 = xmm3;
5277
5278 const XMMRegister x4 = xmm4;
5279 const XMMRegister x5 = xmm5;
5280 const XMMRegister x6 = xmm6;
5281 const XMMRegister x7 = xmm7;
5282
5283 const Register tmp = r11;
5284
5285 BLOCK_COMMENT("Entry:");
5286 __ enter(); // required for proper stackwalking of RuntimeStub frame
5287
5288 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5289
5290 __ leave(); // required for proper stackwalking of RuntimeStub frame
5291 __ ret(0);
5292
5293 return start;
5294
5295 }
5296
5297 address generate_libmLog() {
5298 StubCodeMark mark(this, "StubRoutines", "libmLog");
5299
5300 address start = __ pc();
5301
5302 const XMMRegister x0 = xmm0;
5303 const XMMRegister x1 = xmm1;
5304 const XMMRegister x2 = xmm2;
5305 const XMMRegister x3 = xmm3;
5306
5307 const XMMRegister x4 = xmm4;
5308 const XMMRegister x5 = xmm5;
5309 const XMMRegister x6 = xmm6;
5310 const XMMRegister x7 = xmm7;
5311
5312 const Register tmp1 = r11;
5313 const Register tmp2 = r8;
5314
5315 BLOCK_COMMENT("Entry:");
5316 __ enter(); // required for proper stackwalking of RuntimeStub frame
5317
5318 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2);
5319
5320 __ leave(); // required for proper stackwalking of RuntimeStub frame
5321 __ ret(0);
5322
5323 return start;
5324
5325 }
5326
5327 address generate_libmLog10() {
5328 StubCodeMark mark(this, "StubRoutines", "libmLog10");
5329
5330 address start = __ pc();
5331
5332 const XMMRegister x0 = xmm0;
5333 const XMMRegister x1 = xmm1;
5334 const XMMRegister x2 = xmm2;
5335 const XMMRegister x3 = xmm3;
5336
5337 const XMMRegister x4 = xmm4;
5338 const XMMRegister x5 = xmm5;
5339 const XMMRegister x6 = xmm6;
5340 const XMMRegister x7 = xmm7;
5341
5342 const Register tmp = r11;
5343
5344 BLOCK_COMMENT("Entry:");
5345 __ enter(); // required for proper stackwalking of RuntimeStub frame
5346
5347 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5348
5349 __ leave(); // required for proper stackwalking of RuntimeStub frame
5350 __ ret(0);
5351
5352 return start;
5353
5354 }
5355
5356 address generate_libmPow() {
5357 StubCodeMark mark(this, "StubRoutines", "libmPow");
5358
5359 address start = __ pc();
5360
5361 const XMMRegister x0 = xmm0;
5362 const XMMRegister x1 = xmm1;
5363 const XMMRegister x2 = xmm2;
5364 const XMMRegister x3 = xmm3;
5365
5366 const XMMRegister x4 = xmm4;
5367 const XMMRegister x5 = xmm5;
5368 const XMMRegister x6 = xmm6;
5369 const XMMRegister x7 = xmm7;
5370
5371 const Register tmp1 = r8;
5372 const Register tmp2 = r9;
5373 const Register tmp3 = r10;
5374 const Register tmp4 = r11;
5375
5376 BLOCK_COMMENT("Entry:");
5377 __ enter(); // required for proper stackwalking of RuntimeStub frame
5378
5379 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5380
5381 __ leave(); // required for proper stackwalking of RuntimeStub frame
5382 __ ret(0);
5383
5384 return start;
5385
5386 }
5387
5388 address generate_libmSin() {
5389 StubCodeMark mark(this, "StubRoutines", "libmSin");
5390
5391 address start = __ pc();
5392
5393 const XMMRegister x0 = xmm0;
5394 const XMMRegister x1 = xmm1;
5395 const XMMRegister x2 = xmm2;
5396 const XMMRegister x3 = xmm3;
5397
5398 const XMMRegister x4 = xmm4;
5399 const XMMRegister x5 = xmm5;
5400 const XMMRegister x6 = xmm6;
5401 const XMMRegister x7 = xmm7;
5402
5403 const Register tmp1 = r8;
5404 const Register tmp2 = r9;
5405 const Register tmp3 = r10;
5406 const Register tmp4 = r11;
5407
5408 BLOCK_COMMENT("Entry:");
5409 __ enter(); // required for proper stackwalking of RuntimeStub frame
5410
5411 #ifdef _WIN64
5412 __ push(rsi);
5413 __ push(rdi);
5414 #endif
5415 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5416
5417 #ifdef _WIN64
5418 __ pop(rdi);
5419 __ pop(rsi);
5420 #endif
5421
5422 __ leave(); // required for proper stackwalking of RuntimeStub frame
5423 __ ret(0);
5424
5425 return start;
5426
5427 }
5428
5429 address generate_libmCos() {
5430 StubCodeMark mark(this, "StubRoutines", "libmCos");
5431
5432 address start = __ pc();
5433
5434 const XMMRegister x0 = xmm0;
5435 const XMMRegister x1 = xmm1;
5436 const XMMRegister x2 = xmm2;
5437 const XMMRegister x3 = xmm3;
5438
5439 const XMMRegister x4 = xmm4;
5440 const XMMRegister x5 = xmm5;
5441 const XMMRegister x6 = xmm6;
5442 const XMMRegister x7 = xmm7;
5443
5444 const Register tmp1 = r8;
5445 const Register tmp2 = r9;
5446 const Register tmp3 = r10;
5447 const Register tmp4 = r11;
5448
5449 BLOCK_COMMENT("Entry:");
5450 __ enter(); // required for proper stackwalking of RuntimeStub frame
5451
5452 #ifdef _WIN64
5453 __ push(rsi);
5454 __ push(rdi);
5455 #endif
5456 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5457
5458 #ifdef _WIN64
5459 __ pop(rdi);
5460 __ pop(rsi);
5461 #endif
5462
5463 __ leave(); // required for proper stackwalking of RuntimeStub frame
5464 __ ret(0);
5465
5466 return start;
5467
5468 }
5469
5470 address generate_libmTan() {
5471 StubCodeMark mark(this, "StubRoutines", "libmTan");
5472
5473 address start = __ pc();
5474
5475 const XMMRegister x0 = xmm0;
5476 const XMMRegister x1 = xmm1;
5477 const XMMRegister x2 = xmm2;
5478 const XMMRegister x3 = xmm3;
5479
5480 const XMMRegister x4 = xmm4;
5481 const XMMRegister x5 = xmm5;
5482 const XMMRegister x6 = xmm6;
5483 const XMMRegister x7 = xmm7;
5484
5485 const Register tmp1 = r8;
5486 const Register tmp2 = r9;
5487 const Register tmp3 = r10;
5488 const Register tmp4 = r11;
5489
5490 BLOCK_COMMENT("Entry:");
5491 __ enter(); // required for proper stackwalking of RuntimeStub frame
5492
5493 #ifdef _WIN64
5494 __ push(rsi);
5495 __ push(rdi);
5496 #endif
5497 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5498
5499 #ifdef _WIN64
5500 __ pop(rdi);
5501 __ pop(rsi);
5502 #endif
5503
5504 __ leave(); // required for proper stackwalking of RuntimeStub frame
5505 __ ret(0);
5506
5507 return start;
5508
5509 }
5510
5511 #undef __
5512 #define __ masm->
5513
5514 // Continuation point for throwing of implicit exceptions that are
5515 // not handled in the current activation. Fabricates an exception
5516 // oop and initiates normal exception dispatching in this
5517 // frame. Since we need to preserve callee-saved values (currently
5518 // only for C2, but done for C1 as well) we need a callee-saved oop
5519 // map and therefore have to make these stubs into RuntimeStubs
5520 // rather than BufferBlobs. If the compiler needs all registers to
5521 // be preserved between the fault point and the exception handler
5522 // then it must assume responsibility for that in
5523 // AbstractCompiler::continuation_for_implicit_null_exception or
5524 // continuation_for_implicit_division_by_zero_exception. All other
5525 // implicit exceptions (e.g., NullPointerException or
5526 // AbstractMethodError on entry) are either at call sites or
5527 // otherwise assume that stack unwinding will be initiated, so
5528 // caller saved registers were assumed volatile in the compiler.
5529 address generate_throw_exception(const char* name,
5530 address runtime_entry,
5531 Register arg1 = noreg,
5532 Register arg2 = noreg) {
5533 // Information about frame layout at time of blocking runtime call.
5534 // Note that we only have to preserve callee-saved registers since
5535 // the compilers are responsible for supplying a continuation point
5536 // if they expect all registers to be preserved.
5537 enum layout {
5538 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
5539 rbp_off2,
5540 return_off,
5541 return_off2,
5542 framesize // inclusive of return address
5543 };
5544
5545 int insts_size = 512;
5546 int locs_size = 64;
5547
5548 CodeBuffer code(name, insts_size, locs_size);
5549 OopMapSet* oop_maps = new OopMapSet();
5550 MacroAssembler* masm = new MacroAssembler(&code);
5551
5552 address start = __ pc();
5553
5554 // This is an inlined and slightly modified version of call_VM
5555 // which has the ability to fetch the return PC out of
5556 // thread-local storage and also sets up last_Java_sp slightly
5557 // differently than the real call_VM
5558
5559 __ enter(); // required for proper stackwalking of RuntimeStub frame
5560
5561 assert(is_even(framesize/2), "sp not 16-byte aligned");
5562
5563 // return address and rbp are already in place
5564 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
5565
5566 int frame_complete = __ pc() - start;
5567
5568 // Set up last_Java_sp and last_Java_fp
5569 address the_pc = __ pc();
5570 __ set_last_Java_frame(rsp, rbp, the_pc);
5571 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
5572
5573 // Call runtime
5574 if (arg1 != noreg) {
5575 assert(arg2 != c_rarg1, "clobbered");
5576 __ movptr(c_rarg1, arg1);
5577 }
5578 if (arg2 != noreg) {
5579 __ movptr(c_rarg2, arg2);
5580 }
5581 __ movptr(c_rarg0, r15_thread);
5582 BLOCK_COMMENT("call runtime_entry");
5583 __ call(RuntimeAddress(runtime_entry));
5584
5585 // Generate oop map
5586 OopMap* map = new OopMap(framesize, 0);
5587
5588 oop_maps->add_gc_map(the_pc - start, map);
5589
5590 __ reset_last_Java_frame(true);
5591
5592 __ leave(); // required for proper stackwalking of RuntimeStub frame
5593
5594 // check for pending exceptions
5595 #ifdef ASSERT
5596 Label L;
5597 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
5598 (int32_t) NULL_WORD);
5599 __ jcc(Assembler::notEqual, L);
5600 __ should_not_reach_here();
5601 __ bind(L);
5602 #endif // ASSERT
5603 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5604
5605
5606 // codeBlob framesize is in words (not VMRegImpl::slot_size)
5607 RuntimeStub* stub =
5608 RuntimeStub::new_runtime_stub(name,
5609 &code,
5610 frame_complete,
5611 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
5612 oop_maps, false);
5613 return stub->entry_point();
5614 }
5615
5616 void create_control_words() {
5617 // Round to nearest, 53-bit mode, exceptions masked
5618 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
5619 // Round to zero, 53-bit mode, exception mased
5620 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
5621 // Round to nearest, 24-bit mode, exceptions masked
5622 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
5623 // Round to nearest, 64-bit mode, exceptions masked
5624 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
5625 // Round to nearest, 64-bit mode, exceptions masked
5626 StubRoutines::_mxcsr_std = 0x1F80;
5627 // Note: the following two constants are 80-bit values
5628 // layout is critical for correct loading by FPU.
5629 // Bias for strict fp multiply/divide
5630 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
5631 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
5632 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
5633 // Un-Bias for strict fp multiply/divide
5634 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
5635 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
5636 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
5637 }
5638
5639 // Initialization
5640 void generate_initial() {
5641 // Generates all stubs and initializes the entry points
5642
5643 // This platform-specific settings are needed by generate_call_stub()
5644 create_control_words();
5645
5646 // entry points that exist in all platforms Note: This is code
5647 // that could be shared among different platforms - however the
5648 // benefit seems to be smaller than the disadvantage of having a
5649 // much more complicated generator structure. See also comment in
5650 // stubRoutines.hpp.
5651
5652 StubRoutines::_forward_exception_entry = generate_forward_exception();
5653
5654 StubRoutines::_call_stub_entry =
5655 generate_call_stub(StubRoutines::_call_stub_return_address);
5656
5657 // is referenced by megamorphic call
5658 StubRoutines::_catch_exception_entry = generate_catch_exception();
5659
5660 // atomic calls
5661 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
5662 StubRoutines::_atomic_xchg_long_entry = generate_atomic_xchg_long();
5663 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
5664 StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
5665 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
5666 StubRoutines::_atomic_add_entry = generate_atomic_add();
5667 StubRoutines::_atomic_add_long_entry = generate_atomic_add_long();
5668 StubRoutines::_fence_entry = generate_orderaccess_fence();
5669
5670 // platform dependent
5671 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
5672 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
5673
5674 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
5675
5676 // Build this early so it's available for the interpreter.
5677 StubRoutines::_throw_StackOverflowError_entry =
5678 generate_throw_exception("StackOverflowError throw_exception",
5679 CAST_FROM_FN_PTR(address,
5680 SharedRuntime::
5681 throw_StackOverflowError));
5682 StubRoutines::_throw_delayed_StackOverflowError_entry =
5683 generate_throw_exception("delayed StackOverflowError throw_exception",
5684 CAST_FROM_FN_PTR(address,
5685 SharedRuntime::
5686 throw_delayed_StackOverflowError));
5687 if (UseCRC32Intrinsics) {
5688 // set table address before stub generation which use it
5689 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
5690 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
5691 }
5692
5693 if (UseCRC32CIntrinsics) {
5694 bool supports_clmul = VM_Version::supports_clmul();
5695 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
5696 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
5697 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
5698 }
5699 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
5700 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
5701 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
5702 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
5703 StubRoutines::x86::_ONEHALF_adr = (address)StubRoutines::x86::_ONEHALF;
5704 StubRoutines::x86::_P_2_adr = (address)StubRoutines::x86::_P_2;
5705 StubRoutines::x86::_SC_4_adr = (address)StubRoutines::x86::_SC_4;
5706 StubRoutines::x86::_Ctable_adr = (address)StubRoutines::x86::_Ctable;
5707 StubRoutines::x86::_SC_2_adr = (address)StubRoutines::x86::_SC_2;
5708 StubRoutines::x86::_SC_3_adr = (address)StubRoutines::x86::_SC_3;
5709 StubRoutines::x86::_SC_1_adr = (address)StubRoutines::x86::_SC_1;
5710 StubRoutines::x86::_PI_INV_TABLE_adr = (address)StubRoutines::x86::_PI_INV_TABLE;
5711 StubRoutines::x86::_PI_4_adr = (address)StubRoutines::x86::_PI_4;
5712 StubRoutines::x86::_PI32INV_adr = (address)StubRoutines::x86::_PI32INV;
5713 StubRoutines::x86::_SIGN_MASK_adr = (address)StubRoutines::x86::_SIGN_MASK;
5714 StubRoutines::x86::_P_1_adr = (address)StubRoutines::x86::_P_1;
5715 StubRoutines::x86::_P_3_adr = (address)StubRoutines::x86::_P_3;
5716 StubRoutines::x86::_NEG_ZERO_adr = (address)StubRoutines::x86::_NEG_ZERO;
5717 }
5718 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
5719 StubRoutines::_dexp = generate_libmExp();
5720 }
5721 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
5722 StubRoutines::_dlog = generate_libmLog();
5723 }
5724 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
5725 StubRoutines::_dlog10 = generate_libmLog10();
5726 }
5727 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
5728 StubRoutines::_dpow = generate_libmPow();
5729 }
5730 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
5731 StubRoutines::_dsin = generate_libmSin();
5732 }
5733 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
5734 StubRoutines::_dcos = generate_libmCos();
5735 }
5736 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
5737 StubRoutines::_dtan = generate_libmTan();
5738 }
5739 }
5740 }
5741
5742 void generate_all() {
5743 // Generates all stubs and initializes the entry points
5744
5745 // These entry points require SharedInfo::stack0 to be set up in
5746 // non-core builds and need to be relocatable, so they each
5747 // fabricate a RuntimeStub internally.
5748 StubRoutines::_throw_AbstractMethodError_entry =
5749 generate_throw_exception("AbstractMethodError throw_exception",
5750 CAST_FROM_FN_PTR(address,
5751 SharedRuntime::
5752 throw_AbstractMethodError));
5753
5754 StubRoutines::_throw_IncompatibleClassChangeError_entry =
5755 generate_throw_exception("IncompatibleClassChangeError throw_exception",
5756 CAST_FROM_FN_PTR(address,
5757 SharedRuntime::
5758 throw_IncompatibleClassChangeError));
5759
5760 StubRoutines::_throw_NullPointerException_at_call_entry =
5761 generate_throw_exception("NullPointerException at call throw_exception",
5762 CAST_FROM_FN_PTR(address,
5763 SharedRuntime::
5764 throw_NullPointerException_at_call));
5765
5766 // entry points that are platform specific
5767 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
5768 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
5769 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
5770 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
5771
5772 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
5773 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
5774 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
5775 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
5776
5777 // support for verify_oop (must happen after universe_init)
5778 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
5779
5780 // arraycopy stubs used by compilers
5781 generate_arraycopy_stubs();
5782
5783 // don't bother generating these AES intrinsic stubs unless global flag is set
5784 if (UseAESIntrinsics) {
5785 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
5786 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
5787 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
5788 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
5789 if (VM_Version::supports_vaes() && VM_Version::supports_avx512vl() && VM_Version::supports_avx512dq() ) {
5790 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptVectorAESCrypt();
5791 } else {
5792 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
5793 }
5794 }
5795 if (UseAESCTRIntrinsics){
5796 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
5797 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
5798 }
5799
5800 if (UseSHA1Intrinsics) {
5801 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
5802 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
5803 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
5804 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
5805 }
5806 if (UseSHA256Intrinsics) {
5807 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
5808 char* dst = (char*)StubRoutines::x86::_k256_W;
5809 char* src = (char*)StubRoutines::x86::_k256;
5810 for (int ii = 0; ii < 16; ++ii) {
5811 memcpy(dst + 32 * ii, src + 16 * ii, 16);
5812 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
5813 }
5814 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
5815 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
5816 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
5817 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
5818 }
5819 if (UseSHA512Intrinsics) {
5820 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
5821 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
5822 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
5823 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
5824 }
5825
5826 // Generate GHASH intrinsics code
5827 if (UseGHASHIntrinsics) {
5828 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
5829 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
5830 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
5831 }
5832
5833 if (UseBASE64Intrinsics) {
5834 StubRoutines::x86::_and_mask = base64_and_mask_addr();
5835 StubRoutines::x86::_bswap_mask = base64_bswap_mask_addr();
5836 StubRoutines::x86::_base64_charset = base64_charset_addr();
5837 StubRoutines::x86::_url_charset = base64url_charset_addr();
5838 StubRoutines::x86::_gather_mask = base64_gather_mask_addr();
5839 StubRoutines::x86::_left_shift_mask = base64_left_shift_mask_addr();
5840 StubRoutines::x86::_right_shift_mask = base64_right_shift_mask_addr();
5841 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
5842 }
5843
5844 // Safefetch stubs.
5845 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
5846 &StubRoutines::_safefetch32_fault_pc,
5847 &StubRoutines::_safefetch32_continuation_pc);
5848 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
5849 &StubRoutines::_safefetchN_fault_pc,
5850 &StubRoutines::_safefetchN_continuation_pc);
5851 #ifdef COMPILER2
5852 if (UseMultiplyToLenIntrinsic) {
5853 StubRoutines::_multiplyToLen = generate_multiplyToLen();
5854 }
5855 if (UseSquareToLenIntrinsic) {
5856 StubRoutines::_squareToLen = generate_squareToLen();
5857 }
5858 if (UseMulAddIntrinsic) {
5859 StubRoutines::_mulAdd = generate_mulAdd();
5860 }
5861 #ifndef _WINDOWS
5862 if (UseMontgomeryMultiplyIntrinsic) {
5863 StubRoutines::_montgomeryMultiply
5864 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
5865 }
5866 if (UseMontgomerySquareIntrinsic) {
5867 StubRoutines::_montgomerySquare
5868 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
5869 }
5870 #endif // WINDOWS
5871 #endif // COMPILER2
5872
5873 if (UseVectorizedMismatchIntrinsic) {
5874 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
5875 }
5876 }
5877
5878 public:
5879 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
5880 if (all) {
5881 generate_all();
5882 } else {
5883 generate_initial();
5884 }
5885 }
5886 }; // end class declaration
5887
5888 void StubGenerator_generate(CodeBuffer* code, bool all) {
5889 StubGenerator g(code, all);
5890 }